Analysis of Atmospheric Planetary Boundary Layer - Terrain Interactions. Wind Industry Implications

TESE DE DOUTORAMENTO ANALYSIS OF ATMOSPHERIC PLANETARY BOUNDARY LAYER - TERRAIN INTERACTIONS. WIND INDUSTRY IMPLICATIONS. Miguel Ángel Prósper Fernández ESCOLA DE DOUTORAMENTO INTERNACIONAL PROGRAMA DE DOUTORAMENTO Enerxías Renovables e Sustentabilidade Enerxética SANTIAGO DE COMPOSTELA ANO 2018 DECLARACIÓN DO AUTOR DA TESE Analysis of Atmospheric Planetary Boundary Layer - Terrain Interactions. Wind Industry Implications. D. Miguel Ángel Prósper Fernández Presento a miña tese, seguindo o procedemento axeitado ao Regulamento, e declaro que: 1) A tese abarca os resultados da elaboración do meu traballo. 2) De selo caso, na tese faise referencia ás colaboracións que tivo este traballo. 3) A tese é a versión definitiva presentada para a súa defensa e coincide coa versión enviada en formato electrónico. 4) Confirmo que a tese non incorre en ningún tipo de plaxio doutros autores nin de traballos presentados por min para a obtención doutros títulos. En Santiago de Compostela, a de de 20.. Asdo Miguel Ángel Prósper Fernández AUTORIZACIÓN DO DIRECTOR / TITOR DA TESE Analysis of Atmospheric Planetary Boundary Layer - Terrain Interactions. Wind Industry Implications. D. Gonzalo Miguez-Macho INFORMA: Que a presente tese, correspóndese co traballo realizado por D. Miguel Ángel Prósper Fernández baixo a miña dirección, e a utorizo a súa presentación , considerando que reúne os r equisitos esixidos no R egulamento de Estudos de Doutoramento da USC, e que como director desta non incorre nas causas de abstención establecidas na Lei 40/2015. En Santiago de Compostela, a de de 20.. Asdo Gonzalo Miguez-Macho A la estanquera de Rúa Nova Contents Acknowledgments 11 List of Figures 13 List of Tables 19 Summary 21 Resumo 23 Resumen 25 1 Introduction 27 1.1 Windenergyoutline................................ 27 1.2 Wind energy - Numerical modeling . . . . . . . . . . . . . . . . . . . . . . . 30 1.3 WRF model - Wind energy applications . . . . . . . . . . . . . . . . . . . . . 34 1.4 Motivation and thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 2 Wind resource forecasting 39 2.1 Introduction.................................... 39 2.2 Methodology ................................... 41 2.2.1 Wind farm characteristics . . . . . . . . . . . . . . . . . . . . . . . . 41 2.2.2 WRFconfiguration............................ 41 2.2.3 Experiments ............................... 43 2.2.4 Initial and boundary conditions . . . . . . . . . . . . . . . . . . . . . 43 2.2.5 Observations and error measures . . . . . . . . . . . . . . . . . . . . . 44 2.3 Resultsanddiscussion .............................. 45 2.3.1 Generalresults .............................. 45 2.3.2 Results by wind direction . . . . . . . . . . . . . . . . . . . . . . . . 52 2.4 Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 3 Wind nowcasting 59 3.1 Introduction.................................... 59 3.2 Methodologyanddata .............................. 60 3.2.1 Wind farm location and WRF configuration . . . . . . . . . . . . . . . 60 3.2.2 Hybrid Kalman Bayesian filter . . . . . . . . . . . . . . . . . . . . . . 62 3.2.3 Observational data and nowcasting experiments . . . . . . . . . . . . . 63 List of Figures 5.1 WRF nested domain configuration. (a) Coarser three domains with their number of grid points. (b) Higher resolution domains (d04 and d05), both with their respective topographies. MET1 and MET2 are the locations of the meteorological stations used as validation points. The two red lines represent the vertical cross-sections shown in Figures 5.5 and 5.7. . . . . . . . . . . . . . 92 5.2 (a) Temperature 850 hPa and sea level pressure from GFS 0.5 Analysis data at 2013-12-22 18:00 UTC. (b) Same as (a) at 2013-12-25 00:00 UTC. D01 simulation domain is represented with a white square in both cases. . . . . . . 94 5.3 2013-12-24 03:00 UTC (a) 850 hPa temperature, sea level pressure and wind arrows in the parent grid d01, (b) wind speed (values>10 m/s) and wind arrows at σ=3 (about 70m above ground) in d02, and (c) topography (contours) and potential temperature (shades) and wind arrows at σ=3 in d03. (d) Temperature sounding in the northeast Chivela Pass (NP point in (c)). . . . . . . . . . . . . 96 5.4 Observational wind speed time series from 2013-12-21 to 2013-12-31 in MET1 andMET2. .................................... 97 5.5 D04 vertical cross sections on 2013-12-23 15:30 UTC at MET1 of (a) potential temperature (contours), wind speed (shades), and wind arrows, (b) vertical wind component W and (c) V wind component isolines (positive in blue and negative in red) and turbulent kinetic energy (TKE, shades). (d-f) Same representations for cross sections at MET2. (g-l) Same as a-f on 2013-12-24 03:00 UTC. . . . 97 5.6 (a) Froude number in front of each station on the top of the mountain, TOP1 and TOP2. The grey zone represents night time. (b) Temperature vertical profile in MET1 and MET2 and (c) the same for their corresponding upstream points UP1 andUP2. ..................................... 99 5.7 As in Fig 5.5a and 5.5d, 2013-12-24 03:00 UTC, d05 vertical cross sections at (a) MET1 and (b) MET2 of potential temperature (contours, K), wind speed (shades, m/s), and wind arrows. (c) Vertical wind speed (m/s) in d05 at sigma level 17 (about 1400 m above ground) when the trapped lee wave pattern in the region is fully formed at 15:30 UTC 2013-12-24. . . . . . . . . . . . . . . . . 100 5.8 (a) Wind speed comparison among observations at MET1 (green), and model output from d04 (red) and, d05 (blue). (b) Same for station location MET2. . . 101 5.9 (a) Cloud water mixing ratio 3D representation in d05 at 2013-12-24 15:30 UTC. The two cross sections show the N-S wind profile at the longitudes of the meteorological stations MET1 and MET2. (b) Satellite image of the terrain in d05 and cloud water mixing ratio (white shades) and wind arrows at about 1.4 km above ground. (c) GOES-R satellite image on 2013-12-24 20:30 UTC, revealing very similar lenticular cloud formations in the same locations. . . . . 102 6.1 WRF nested domain configuration. (a) Coarser three domains with their number of grid points. (b) Higher resolution domains (d04 and d05), both with their respective topographies. MET1, MET2, and MET3 are the locations of the meteorological stations used as validation points (Subchapter 6.2.4). . . . . . . 107 6.2 (a) D05 original HGT variables, (b) D05 HGT with 12-pass-smoothing, (c) D05 HGT with zonal smoothing over unstable points at the centre of the domain. . . 110 16 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ 6.3 (a) 2017/10/09-2017/10/19 (UTC) Wind direction at M1 (80 meters height) for WRF-LES (blue) and observations (red). (b) Same as (a) at M2.(c) Same as (a) atM3........................................ 111 6.4 (a) 2017/10/09-2017/10/19 (UTC) Wind direction at M1 (80 meters height) for WRF-LES (blue) and observations (red). (b) Same as (a) at M2.(c) Same as (a) atM3........................................ 113 6.5 (a) 2017/10/09-2017/10/19 (UTC) Temperature at M1 (10 meters height) for WRF-LES (blue) and observations (red). (b) Same as (a) at M2. (c) Same as (a)atM3. ..................................... 114 6.6 (a) 2017/10/09-2017/10/19 UTC turbulent intensity at M1 (80 meters height) for WRF-LES (blue) and observations (red). (b) Same as (a) at M2.(c) Same as (a)atM3. ..................................... 116 6.7 (a) 2017/10/09-2017/10/19 (UTC) TITot (red), TILES (blue) and TISGS (yellow) at M1 (80 meters height) for WRF-LES (blue) and observations (red) . . . . . . 117 6.8 Instantaneous wind components at 2017-10-20 09:00 UTC for d05. (a) U plot along a longitudinal cut through M1. (b) Same as (a) for V component, (c) Same as (a) for W component. (d) U field at 80 m height. (e) V field at 80 m height. (f) W field at 80 m height. . . . . . . . . . . . . . . . . . . . . . . . . 118 17 List of Tables 2.1 Two main configurations, CASE 1 and 2 (with and without wake parameterization), with their variants. The identifier names represent the case characteristics, for example, WF-HI-1D: WF (Wind farm parameterization), HI (high resolution, 333 m) and 1D (24 hour lead time) or PO-LO-2D: PO (polynomial adjust), LO (low resolution, 1 km) and 2D (24 hour lead time). . . 44 2.2 Observed mean monthly wind speed normalized with observed mean annual windspeed. .................................... 48 2.3 The three wind farm areas with their wind speed annual MAE. . . . . . . . . . 50 2.4 Annual wind power NMAE and annual wind speed MAE, ME and RMSE calculated for all experiments. The lowest values among each configuration (WFandPO)arebold. .............................. 50 2.5 Annual wind power NMAE and annual wind speed MAE, ME and RMSE for all experiments for power ramp winds only. The lowest values among each configuration (WF and PO) are bold. . . . . . . . . . . . . . . . . . . . . . . . 52 2.6 Wind direction percentages in T07, T13 and T26. . . . . . . . . . . . . . . . . 53 3.1 The principal physical schemes used in the atmospheric model . . . . . . . . . 62 3.2 Short explanation of the experiments tested in this study. The bold names on the left are the identifiers used hereafter for each case. . . . . . . . . . . . . . . 64 3.3 (a) WS ME for all experiments and wind turbines, (b) same as (a) for RMSE. . 65 3.4 Observed mean monthly wind speed normalized with observed mean annual windspeed. .................................... 67 3.5 WD MAE for all the experiments and different persistence periods in power rampwinds..................................... 74 4.1 Parameterizations for the highest resolution domain, D04. . . . . . . . . . . . . 80 4.2 Annual wind speed ME, CC, RMSE and MAE for three different wind turbines (X27,X54,andX67)................................ 84 5.1 Main physic parameterizations used. . . . . . . . . . . . . . . . . . . . . . . . 93 5.2 Weather station positions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 5.3 Wind speed (WS), wind direction (WD), and temperature (T) mean errors (ME) and mean absolute errors (MAE) at MET1 and MET2 locations during the simulated period and for the two higher resolution domains (d04 and d05). . . . 101 6.1 Main parameterizations in higher resolution domains (D04 and D05). . . . . . 108 6.2 Meteorological station locations and heights. . . . . . . . . . . . . . . . . . . 110 List of Tables 6.3 Mean wind speed observed and forecasted for each validation point. Different mean wind speed errors measured in each point along the studied period: Correlation Coefficient (CC), Root Mean Square Error (RMSE), Mean Error (ME) and Mean Absolute Error (MAE). . . . . . . . . . . . . . . . . . . . . . 112 6.4 Mean wind direction Mean Absolute Error (MAE) in each point for the period studied. ...................................... 113 6.5 Mean temperature observed and forecasted for each validation point. Different mean wind speed errors measured at each point for the studied period: Correlation Coefficient (CC), Root Mean Square Error (RMSE), Mean Error (ME) and Mean Absolute Error (MAE). . . . . . . . . . . . . . . . . . . . . . 114 6.6 Mean turbulent intensity observed and forecasted for each validation point. Different mean TI errors measured in each point for the studied period: Correlation Coefficient (CC), Root Mean Square Error (RMSE), Mean Error (ME) and Mean Absolute Error (MAE). . . . . . . . . . . . . . . . . . . . . . 116 6.7 Mean values for TIobs,T ITot ,TILES and TISGS for the period studied in M1, M2,andM3. ................................... 117 20 Summary The development of wind energy has a direct effect on the reduction of carbon dioxide emissions from the energy sector. This industry is one of the largest anthropogenic contributors to the global problem of climate change. For this reason, and because it is the most expanded renewable energy source in the world, there is a clear need to optimize wind energy exploitation. Numerical modeling is a tool that forms part of the present and future of this sector; because it is able to reproduce the effect of wind farms on the atmosphere and to obtain its short-term production prediction. Besides, meteorological models allow us to improve wind resource analysis methods for any region of the planet. The present thesis aims to achieve a detailed quantification and understanding of the main interactions between atmospheric planet boundary layer and terrain, focusing on the behavior of wind flows at different scales. In addition, we intend to improve the tools, within numerical modeling, for the analysis of such mechanisms, contributing in this manner to the optimization of the use of wind as a resource. The primary tool used in this thesis is the WRF (Weather Research and Forecasting) model. It is a meso and microscale numerical prediction system designed for both atmospheric research and operational forecasting. In most parts of the research, high-resolution simulations are performed with the WRF model. This provides accurate information of near-surface wind fields and turbulent processes in a wide range of atmospheric stability conditions and areas of the planet, both in flat and complex terrain. Resumo O desenvolvemento da enerx´ ıa e´ olica ten un efecto directo na reducci´ on das emisi´ ons de di´ oxido de carbono procedentes do sector enerx´ etico. Dita industria ´ e o maior contribuinte antropox´ enico ´ o problema mundial do cambio clim´ atico. Por isto, e por ser a enerx´ ıa renovable m´ ais expandida a nivel global, existe una clara necesidade de optimizar a explotaci´ on e´ olica. A modelizaci´ on num´ erica ´ e unha ferramenta que forma parte do presente e do futuro deste sector, porque ´ e capaz de reproducir o efecto dos parques na atmosfera e de obtener a s´ ua predicci´ on de producci´ on a corto prazo. Ademais disto, os modelos meteorol´ oxicos perm´ ıtennos mellorar os m´ etodos de an´ alise do recurso e´ olico en calquera rexi´ on do planeta. A presente tese busca lograr unha detallada cuantificaci´ on e entendemento das principais interacci´ ons entre a capa l´ ımite da atmosfera e o terreo, centr´ andose no comportamento dos fluxos de vento a diferentes escalas. A maiores disto, pret´ endese mellorar os instrumentos, dentro da modelizaci´ on num´ erica, encargados de analizar ditos mecanismos. Desta forma, poderase contribuir ´ a optimizaci´ on do aproveitamento deste recurso e´ olico. A principal ferramenta empregada nesta tese ´ e o modelo WRF (Weather Reseach and Forecasting). Tr´ atase dun sistema de predicci´ on num´ erica a meso e microescala dese˜ nado tanto para a investigaci´ on atmosf´ erica como para predicci´ on operativa. Na maior´ ıa das partes da investigaci´ on realizaranse simulaci´ ons a alta resoluci´ on con dito modelo. Deste modo obterase informaci´ on precisa sobre campos de vento cercanos ´ a superficie e procesos turbulentos nunha ampla gama de condici´ ons de estabilidade atmosf´ erica e zonas do planeta, tanto en rexi´ ons chairas como en terreo complexo. Resumen El desarrollo de la energ´ ıa e´ olica tiene un efecto directo en la reducci´ on de las emisiones de di´ oxido de carbono procedentes del sector energ´ etico. Dicha industria es uno de los mayores contribuyentes antropog´ enicos al problema mundial del cambio clim´ atico. Por ello, y por ser la energ´ ıa renovable m´ as expandida a nivel global, existe una clara necesidad de optimizar la explotaci´ on e´ olica. La modelizaci´ on num´ erica es una herramienta que forma parte del presente y del futuro de este sector, porque es capaz de reproducir el efecto de los parques e´ olicos en la atm´ osfera y de obtener su predicci´ on de producci´ on a corto plazo. Adem´ as de esto, los modelos meteorol´ ogicos nos permiten mejorar los m´ etodos de an´ alisis de recurso e´ olico cualquier regi´ on del planeta. La presente tesis busca lograr una detallada cuantificaci´ on y entendimiento de las principales interacciones entre la capa l´ ımite de la atm´ osfera y la topograf´ ıa, centr´ andose en el comportamiento de los flujos de viento a diferentes escalas. A mayores de esto, se pretende mejorar los instrumentos, dentro de la modelizaci´ on num´ erica, encargados de analizar dichos mecanismos, contribuyendo de esta forma a la optimizaci´ on del aprovechamiento del recurso e´ olico. La principal herramienta empleada en esta tesis es el modelo WRF (Weather Reseach and Forecasting). Se trata de un sistema de predicci´ on num´ erica a meso y microescala dise˜ nado tanto para la investigaci´ on atmosf´ erica como para predicci´ on operativa. En la mayor´ ıa de las partes de la investigaci´ on se realizar´ an simulaciones a alta resoluci´ on con dicho modelo. De este modo se obtendr´ a informaci´ on precisa sobre campos de viento cercanos a la superficie y procesos turbulentos en una amplia gama de condiciones de estabilidad atmosf´ erica y zonas del planeta, tanto en regiones llanas como en terreno complejo. Chapter 1. Introduction are included in the Real Decreto 413/2014 of 6 June [23], and the autonomic normative of each community [24]. Focusing on the technical-meteorological aspects, when the location of the future installation is determined, and the legal permits are acquired, the wind company initiates a more accurate study of the region of interest. There is a previous observational data collection phase using meteorological stations with different heights designed explicitly for this purpose. However, it is more and more frequent for the installation companies to also use the capabilities of meteorological modeling. The exponential increase of computer power in HPC systems allows the development and faster use of these numerical tools. In this specific field, this means more runs with higher resolution and in a shorter time period. The results from these studies provide valuable information about local phenomenology in the analyzed area due to the good representation of the planetary boundary layer behavior from numerical modeling at high resolution. In the next figure, we can see a visual example of the capabilities of NWP within this field. It represents the annual mean wind speed field obtained from WRF model simulations in a densely exploited cluster region in Mexico (Chapter 5). Figure 1.5 depicts the annual mean wind speed at hub height obtained from high-resolution daily simulations over the area of interest. The image shows a noticeable mean resource difference between the two meteorological stations. These two stations are around 20 km from each other over flat terrain several kilometers away from the mountains. However, MET1 register a mean value close to 10 m/s and MET2 only reaches 4 m/s. It is easy to imagine the possible incorrect assumptions that could be made in a wind farm siting project if we just take into account the data from MET1 or only from MET2. Numerical meteorological modeling provides us with valuable information about all the area of interest, not only a few points; this is highly profitable if we take into account the large size that a wind farm can have. The conjunction of quality observational data and a high-resolution forecast with NWP, both for the same representative period, is the best-combined tool that wind resource analysts can have at their disposal. After the data collection analysis and the meteorological study of the area of interest, is when the wind farm can be definitively designed. Computational Fluid Dynamics (CFD) models are the current most extended tool for this purpose, due to their capacity for simulating accurate air flow fields, considering in detail the main microscale processes affecting the wind [25]. They can run over complex terrain using orography values from LiDAR data [26], and they also include specific turbine schemes to simulate wake effects on wind farm areas. In the design process, it is fundamental to take into account the wake that each turbine produces because it has a direct impact on the farm’s production and also affects the lifetime of the rest of the machines. When the wind farm is finally installed (Figure 1.6), the use of high-resolution meteorological modeling acquires more prominence. Local short-term forecasts can provide valuable information in an operational wind farm, as wind speed and direction predictions enable a better use of the resource. On the one hand, for electric grid operators, which distribute a vast amount of eolic energy that should be efficiently incorporated into the electrical grid for distribution, the prevention of ramp events effects [27] is fundamental. These extreme short-term situations produce important increases or drops in power production, which can have a major impact on the grid [28]. On the other hand, high-resolution meteorological modeling is able to provide an accurate wind energy production forecast for one and even two days ahead [29]. It can also protect the wind farm from extreme wind events which are potentially dangerous for the turbines and have a negative effect on productivity [30]. These reasons make 32 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Figure 1.5: Annual mean wind speed (70 meters height) from WRF 1 km resolution simulations in a wind area in Mexico (Chapter 5). The observational data used is obtained from the meteorological stations plotted in white. local short-term forecasting of great utility in the daily management of the installation. Numerical weather models can also be part of different applications in the very near future forecasting (nowcasting). Going a step forward, the combination of NWP and MOS (Model Output Statistics) is an effective technique for very short-term wind predictions [31]. MOS models are capable of minimizing NWP errors induced by sub-grid phenomena and local effects in a short time horizon. In this way, we can, for example, predict the optimal yaw orientation of the turbines a few hours ahead, obtaining higher performance and lower stress. 33 Chapter 1. Introduction Figure 1.6: Operative wind farm in a complex terrain area over Serra do Xistral, Galicia (Spain). WRF model - Wind energy applications In this thesis we deal with many of the issues described above, using the WRF model as a primary tool. This software is a mesoscale and microscale numerical prediction system designed for both atmospheric research and operational forecasting needs. This regional model uses global model inputs to define its boundary and initial conditions. It has been developed by the National Center for Atmospheric Research’s (NCAR), the National Oceanic and Atmospheric Administration’s (NOAA) and the National Centers for Environmental Prediction (NCEP) in collaboration with other state agencies and universities [32]. In general terms, WRF code V3 includes: WRF Software Framework (WSF), Advanced Research WRF (ARW) dynamic solver, the WRF Preprocessing System (WPS), WRF Data Assimilation (WRF-DA), numerous physics packages contributed by WRF partners and the research community and several graphics programs and conversion programs for other graphics tools [33]. The ARW solver is the key component of the modeling system, which has an equation set fully compressible, Eulerian and non-hydrostatic with a run-time hydrostatic option. This NWP model uses terrain-following, hydrostatic-pressure vertical coordinates with the top of the model being a constant pressure surface. The time integration scheme in the model uses the third-order Runge-Kutta scheme, and the spatial discretization employs 2nd to 6th order schemes. The model supports both idealized and real-data applications with various lateral boundary condition options. The model also supports one-way, two-way and moving nest options. It can run on single-processor or sharedand distributed-memory computers (http://www2.mmm.ucar.edu/wrf/users/model.html). The WRF model includes parameterizations for atmospheric processes that occur at scales unresolved by the model grid spacing. These schemes can be divided into five main groups: 1. Microphysics: Resolve the formation of meteors: steam, clouds, rain, ice, etc. 2. Cumulus convection: Resolve vertical moisture flows at scales smaller than those of the domain. Formation of cumulus and rains. 3. Planetary Boundary Layer: Solves the flows due to turbulence in the whole column of the atmospheric boundary layer. 34 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ 4. Surface: Calculate the humidity and heat flow at the surface level. It takes into account: the humidity of the soil, vegetation, accumulated snow, urban areas, etc. 5. Radiation: Thermal flows derived from the radiation of the Sun. Shortwave and long wave. To achieve high resolution with this model, it is necessary to use a nested configuration approach where successive child grids with increasing resolution are set within coarser parent domains. The corresponding parent provides boundary conditions for the nested grid along the simulation time. With this procedure, it is possible to perform high-resolution simulations starting from coarse global simulation data. In Figure 1.7 we show an example of a nesting domain configuration for the simulations performed in the study presented in Chapter 5. This capability is precious in wind energy applications because it allows us to reach the precision desired in any part of the world using global model data (many of them publicly available) and a nesting domain configuration Figure 1.7: Wind field representations over domain nesting configuration for an area studied in Mexico (Chapter 5). The following figure shows a flowchart of the general WRF simulation process used in the studies presented in this thesis. The first step in a WRF simulation is the WRF Preprocessing System (WPS) (Figure 1.8, blue part), which consists of three programs that provide the input data to the WRF real program to run the simulation. Geogrid defines the simulation domains and interpolates various terrestrial data sets to the model grids. Ungrib unpacks GRIB (GRIB1 and GRIB2) meteorological data (e.g. GFS, ERA5, ERA-Interim...) and packs it into an intermediate file format. Finally, metgrid horizontally interpolates the meteorological data onto your model domain, creating an output which is used as input to WRF. With the WPS complete, the next step is the real.exe, which creates initial and boundary condition files for real-data cases. After this, the data are ready to start the WRF running with wrf.exe. This is the main component; it performs the time integration controlled by the run-time selected namelist options. WRF outputs are formed by non-post-processed variables (e.g., T variation per level, decomposed wind components. .. ), so they need to be post-processed (Figure 1.8, orange part) to extract the information of interest, such as a wind field at hub height, or a wind speed series at a specific point. As commented above, the WRF model has many physical packages that can be used to resolve many different phenomena, such as those of atmospheric chemistry (WRF-Chem) or fire-related processes (WRF-Fire). In our case, we want to focus on the wind energy 35 Chapter 1. Introduction Figure 1.8: General WRF simulation process flowchart followed in the studies presented field. We are interested in the planetary boundary layerterrain interactions. That is why we need high-resolution horizontal and vertical simulations; we want to resolve the wind and temperature variations in the boundary layer accurately. Moreover, the WRF model also includes a wind farm parameterization [34] able to simulate the wake effects of a turbine on the close environment and to calculate its production according to a power curve, standard coefficients, and dimensions. This parameterization represents wind turbines as momentum sinks, transferring kinetic energy into turbulent kinetic energy and electricity [35]. The quality of the results obtained from high-resolution WRF simulations depends to a large extent on the representation that the model makes of large-scale eddies within the planetary boundary layer (PBL) and of how it treats subgrid-scale processes (SGS) [36]. By multiple nesting, as in Figure 1.7, it is possible to solve phenomenology on a turbulent scale and include the mesoscale effect in the domain of higher resolution. To be able to put this concept into practice, in many cases, it is necessary to reach very high levels of resolution (≤100m) by LES (Large Eddy Simulations). LES is an approach to directly solve the equations of motion within the atmospheric boundary layer by filtering the Navier-Stokes equations and directly solving the large turbulent eddies [37]. This type of simulation can be an accurate tool providing information about turbulent processes in a wide range of atmospheric stability situations, both in flat regions as well as complex terrain. 36 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Motivation and thesis outline In this thesis we aim to achieve a detailed quantification and understanding of the main interactions between the atmospheric planet boundary layer and terrain, focusing on the behavior of wind flows at different scales. In this way we intend to improve the tools, within numerical modeling, for the analysis of such mechanisms, contributing to the optimization of the use of the wind resource. For this purpose, we use the WRF (Weather Research and Forecasting) model as a primary tool. In most parts of the investigation, we configure it with high-resolution, reaching even the microscale through LES (Large Eddy Simulations). In this manner, we intend to provide accurate information of wind fields and different turbulent processes in a wide range of atmospheric stability conditions and areas of the planet, both in flat and complex terrain. In the first part of the manuscript, composed of the first two chapters, we present a broad study about wind resource forecasting for an onshore wind farm. The first chapter focuses on the case of wind production forecast and validation for a real onshore wind farm using high horizontal and vertical resolution WRF model simulations. The wind farm is located in Galicia, in the northwest of Spain, in a complex terrain region with high wind resource. Utilizing the Fitch scheme, specific for wind farms, a period of one year is simulated with a daily operational forecasting set-up. In the second chapter, we go a step forward in the short term forecasting issue combining the WRF prediction from the first study with a post-process statistical tool (Kalman Bayesian filter). With this technique, we reach the nowcasting temporal scale, with critical applications in the wind industry. The second part, presented in chapter three, is centered on the wake effect of the wind farms on their environment. We introduce the Annual Wake concept, which is the mean annual wind resource loss in the area due to the wind farm’s wake. We test this tool over the same wind farm from the first part and another one in China. The third part approaches the extreme wind study issue. Specifically, it is focused on investigating different mountain wave phenomena resulting from complex interaction between large-scale meteorological conditions and local orographic forcings in a wind farm cluster in Mexico. Finally, in the fourth and last part, in chapter five, we analyze the microscale turbulent phenomenology in a wind farm area in southern China using WRF-LES simulations. 37 Chapter 2 Wind resource forecasting Introduction There is a clear need to improve the exploitation of wind resource, given that the amount of eolic energy produced is rapidly increasing and it should be efficiently incorporated into the electrical grid for distribution. For this purpose, it is essential to have a clear knowledge of air flow structure, understanding its behavior in layers close to the surface, and to be able to deliver skillful local short-term forecasts in order to optimize wind farm construction and exploitation. Short-term wind power forecasting presents many challenges. In most cases, onshore farms are affected by microscale events that are especially complicated to predict because they have a turbulent phenomenology, with continuous wind direction and wind intensity changes in short periods of time. In recent years, due to the global push in wind power and the exponential increase in computing capacity, there has been an important development in short-term wind and wind power forecasting methods. The current state of the art of numerical modeling allows for improved wind farm characterization methods and for a more accurate and reliable study of their interaction with the atmosphere. Current methods of Computational Fluid Dynamics (CFD) are capable of simulating accurate air flow fields, considering in detail the main microscale processes affecting the wind, like buoyancy or turbulence diffusion effects [25]. They can run over complex terrain using orography height values from LiDAR data [26]. However, these models do not readily incorporate time-varying lateral boundary data from observations or analysis and, furthermore, thermal effects on turbulence and the wind flow are generally considered constant and hence, are not realistic, as those can vary substantially throughout the day. The current state of the art of meteorological numerical modeling allows for improved wind farm characterization methods and for a more accurate and reliable study of their interaction with the atmosphere. Mesoscale models such as WRF (Weather Research and Forecasting)[32] are able to perform simulations with a horizontal resolution of less than one hundred meters, Large Eddy Simulations (LES)[38] aiming at predicting and characterizing turbulent phenomena that affect wind farms and have direct impacts on energy production. In addition, wind farm output is also strongly affected by wake effects from the turbines themselves. There are several current studies with WRF LES simulating wind turbine wakes with high accuracy [39, 40, 41]. Notwithstanding, LES simulations still represent an excessive computational burden, and are not yet a viable tool for short term wind power operational forecasting. In studies combining statistical and numerical models, generally lower resolution simulations are used, such as Zhao et al. [42] with 3 km Chapter 2. Wind resource forecasting or Li et al. [43] with 6 km of horizontal resolution in the inner most domain. Che et al. [44] and Che and Xiao [45] developed a wind forecast system employing 0.5 km resolution WRF simulations with a Kalman filter. In regards to specific tools for numerical model predictions, WRF versions 3.3 or later include a wind farm parameterization [46] (WRF-WF hereafter) able to simulate the wake effects of a turbine on the close environment and to calculate its production according to a power curve, standard coefficients and dimensions. It is nevertheless difficult to find studies using WRF specifically for wind farms and directly comparing results with observations. Due to the lack of data, many investigations in this field examine the hypothetical impact of different farm configurations at several scales. For example AC Fitch [34, 35] and Vanderwende and Lundquist [47] study the theoretical impact of wind farms in regions of the U.S. Midwest on the close surroundings. For larger scales, the wake parameterization has been used in studies of the broader impact of wind farms on the global climate [48, 49, 50]. There are many studies focused on wake effect understanding using WRF-WF, as Santoni et al. [51], which couple a WRF simulation with WRF-WF and a WRF-LES in an installation in Texas. In this same U.S. state, Xia et al. [52] use WRF-WF in a one-month period simulation to analyze the effect of a large turbine cluster on the nearby near surface temperature. Kumar et al. [53] use the Fitch scheme to predict wind profiles and turbulent intensity in two wind farm locations in India and Scotland. There are several publications using WRF for the offshore farm of Horns Rev in Denmark. Jim´ enez PA et al. [54, 55] perform high-resolution simulations (333 m) to estimate the total farm power deficit due to the wake effects of the turbines. O Eriksson et al. have contrasted the wake parameterization with LES simulations using an actuator disk method in Rev Horns [56] and another farm in Sweden [57]. Another wake effect parameterization, explicit wake effect parametrization (EWP) designed by Volker P et al. has been compared to WRF-WF on this same farm [58]. More recently, in 2018, S. C. Pryor et al. have also assessed the differences between WRF-WF and EWP for several purposes [59, 60], with short-term simulations in the U.S. Central Plains area. Our main goal in this work is to use WRF at high resolution as a wind power forecasting tool for an actual on-shore farm, the Coruxeiras wind farm, for a whole year. We test the effectiveness of WRF in calculating the energy production of this farm in complex terrain, in real time. The availability of high quality in situ real data (at the nacelle of each turbine in the farm) allows us to perform an in detail validation of the simulations, in terms of power prediction and of the impact of the wakes from the turbines. In addition, we compare high-resolution with low-resolution simulations for different temporal scales to assess the potential of the method and the high resolution configuration. To the best of our knowledge, this is the first time that a thorough evaluation for such a long period of time and such detail has been performed. The article is organized as follows: In section II the methodology is explained in detail, in section III, the main results obtained are shown and finally in section IV the conclusions reached are discussed. 40 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Methodology Wind farm characteristics Coruxeiras wind farm is located in Serra do Xistral mountains in Galicia, in northwest Spain (Figure 2.1a). The climate in the region is temperate maritime. Galicia lies within the main North Atlantic storm track most of the year and winds are predominantly from the west-southwest. In summer, when the subtropical Azores high moves poleward, the region is only marginally affected by baroclinic storms within the westerlies, and northeasterly winds dominate. Because of the position of Serra do Xistral at the very northwest corner of the land mass of the Iberian Peninsula, the predominant southwest/northeast flows are accelerated and there is a very high potential for wind energy. Numerous wind farms exist in the area, which is one of the most productive in Europe [61, 62]. Norvento company is the developer and current operator of Coruxerias wind farm, installed in 2006. It is composed of 31 turbines ECOTECNIA74 [63] with 1.670 MW of nominal power (Figure 2.2). The hub height is 60 m and the rotor diameter is 74 m. They are located along a mountain crest of around 800 m elevation above sea level, one of the tallest in the area, and separated by a mean distance of 300 m. The orography in the farm’s surroundings is rather complex, with elevations reaching almost 1000 m descending rapidly to sea level in deep river valleys leading to the coast, which is in close proximity (Figure 2.1b and 2.1c). Figure 2.1: (a) WRF nested grid configuration, with the number of points for each domain indicated. (b) D04 is expanded showing topography. Coruxeiras wind farm is located in the central area of D04, on top of a hill (c). WRF configuration We use the Advanced Research WRF (ARW) model version 3.6 [32] (WRFV3.6) to perform the simulations. Based on a fully compressible and non-hydrostatic dynamic core [33], WRFV3.6 is a limited-area mesoscale model, with a terrain-following hydrostatic-pressure vertical coordinate, designed for operational forecasting, as well as research. The domain’s configuration meets the requirements recommended by Warner [64], including a parent (D01) and three nested grids (D02, D03 and D04) (Figure 2.1a) one-way interacting. D01 is centered at 43.29 N and 7.75 W (Figure 2.1c) with 121x121 grid points of 9 km of horizontal resolution. The horizontal resolutions of D02, D03 and D04 are 3 km 41 Chapter 2. Wind resource forecasting Mean Monthly WS / Mean annual WS Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 1.4 1.16 0.97 0.85 1.07 0.77 0.8 0.76 0.83 1.02 0.88 1.48 Table 2.2: Observed mean monthly wind speed normalized with observed mean annual wind speed. MAEs are relatively constant throughout the year, with values around 2m/s for all experiments, slightly larger in winter months, when mean wind speeds are higher, and lower in summer months. In general low resolution and second day forecast have worse skill than high resolution and first day forecasts, but performance degradation is not very significant in either case, except for the month of May. For each pair of experiments with (WF) and without (PO) wake parameterization, the latter has a slightly better skill for most months. MEs show a more marked seasonal cycle. They are mostly negative throughout the year, which indicate wind speed underestimations, and these are worse in late spring and early summer than in fall and winter months. NMAEs, with values between 13 and 25 %, and NMEs have qualitatively similar behaviors to MAEs and MEs, respectively. We examine next errors on a finer timescale. Figure 2.5 displays the mean hourly wind speed MAE, ME and NMAE for the 48h forecast periods of simulation covering the whole year. The observed mean wind speed variations in a day are also shown for reference (Figure 2.5d). Figure 2.5: (a) Annual wind speed MAE, hourly for the entire simulation period (48 hours). Each line represents 1D (0 to 24h) and 2D (24 to 48h) cases, WF-HI (blue), WF-LO (green); PO-HI (red) and PO-LO (cyan). (b) same as (a) but for wind speed ME.(c) same as (a) but for wind power NMAE. (d) Observed annual mean hourly wind speed normalized with observed mean annual wind speed. 48 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Errors in all cases show a noticeable daily cycle, with values of the MAE around noon clearly lower than during the night. For example, the MAE in WF-HI, already starts at 1.8 at 00UTC, then rise to 2.02 m/s at 06:00 UTC, to decrease to 1.60 m/s at 13:00 UTC , around midday in the area, which is in a GMT+1 time zone (GMT+2 in daylight saving time). Errors grow again in the afternoon and repeat a similar cycle for the later 24h in the forecast, with higher values due to the increasing lead time. There is a close correspondence between the diurnal cycle of MAE and that of NMAE. Minima and maxima values occur at the similar times of the day, with the lowest NMAE of 12.73 % in WF-HI, recorded at 13:00 UTC, the fist 24h of forecast. MAEs and NMAEs follow only loosely the daily cycle of wind speed (Figure 2.5d), and it is likely that there are other factors related to the representation of turbulence fluxes explaining the lower absolute errors at noon. MEs (Figure 2.5b) are negative at all times and for all experiments, indicating an underestimation of wind speeds. In the low resolution experiments, this negative bias shows a daily cycle closely following that of wind (Fig. 5d), with maximum and minimum errors corresponding to maximum and minimum wind velocity, at around sunrise and sunset, respectively. For high resolution experiments, the ME is largest at around 6 UTC, same as for MAEs and NMAEs, and decrease in general in the central hours of the day, but later than the time of minimum MAE and NMAE, which is around noon. Some of the persistent negatives MEs are likely due to an overrepresentation of turbulence in these simulations. With the high resolution used, at least part of the turbulent processes are dynamically resolved, in addition to being parametrized by the PBL scheme, leading to a ’double counting’ that increases mixing and slows down the flow. The general behaviour of errors throughout the day is likely the result of a better performance of the PBL parameterization during daytime hours, when there is more turbulence and better mixing, than during the night. The lower wind speed underestimation at the central hours of the day corresponds to a lower overestimation of the turbulence. In contrast, throughout the night the temperature vertical profiles can be more difficult to represent by the model due to their more stratified nature, and therefore, turbulent fluxes worse estimated. The daily cycle in wind speed forecast skill that we observe has been described previously in several studies [74]. Aside from the commented diurnal pattern, wind speed MAE and wind power NMAE grow gradually with forecast length. Power NMAE at 13 hours of day 1 is 12.73 % and at 13 hours of day 2 increases to 15.02 %. This degradation of skill, albeit small, is important to an operative forecast system, where each one per cent of error would suppose a problem to the electric system operator and an economic loss for the energy producer. With regard to the effect of spatial resolution, there is a clearly gap between HI and LO cases throughout the day. For example, in terms of wind speed MAE, WF-LO has always around 0.25 m/s higher values than WF-HI. The location of the turbines in the wind farm clustering in three areas (Figure 2.2c) allows naturally for a separate analysis for each section (Table 2.3) As shown in Figure 2.3a there are no significant differences in wind speed MAE among turbines, therefore it is possible to consider any of these areas as representative of the wind farm. To illustrate the day-to-day results of the WRF forecasting tool in terms of wind speed at hub height, we show next comparisons between WF-HI-1D, PO-HI-1D and observations for the central area of the farm for the entire months of December and June. These months have the highest and lowest MAE for WF-HI-1D (Figure 2.6), respectively. Mean monthly wind power NMAE and wind speed MAE for the area and the total wind farm are also presented aside. Notwithstanding some occasional large errors, in general, the predicted WF-HI-1D wind 49 Chapter 2. Wind resource forecasting Area Wind turbines WS MAE (m/s) North T01 to T10 1.81 Center T11 to T19 1.89 South T20 to T31 1.90 Table 2.3: The three wind farm areas with their wind speed annual MAE. coincides well with the observations, consistently and not just in the mean sense. PO-HI-1D forecast results are always very close to those of WF-HI-1D, in agreement with the similarity in all error measures between the two configurations shown in Figure 2.3. Lower MAE values in June and also in the rest of the summer (Figure 2.4) are likely due to lower wind resource during this season (Table 2.2). Wind speeds are significantly higher during winter, especially in December, which means that in winter months there are more situations where wind turbines are working at nominal power and therefore, even when wind speed MAEs are higher, wind power forecast errors remain quite similar during the whole year. An overall summary of the different error measures computed for the two configurations with and without wake effects, at high and low resolution and for first and second day forecasts is shown in Table 2.4. CASE POWER NMAE (%) MAE WS (m/s) ME WS (m/s) RMSE WS (m/s) WF-HI-1D 14.75 1.87 -0.58 2.46 WF-HI-2D 16.36 2.13 -0.66 2.81 WF-LO-1D 16.89 2.15 -0.93 2.82 WF-LO-2D 17.99 2.34 -0.98 3.10 PO-HI-1D 14.37 1.86 -0.18 2.43 PO-HI-2D 15.93 2.09 -0.26 2.79 PO-LO-1D 16.66 2.14 -0.72 2.79 PO-LO-2D 17.77 2.33 -0.78 3.08 Table 2.4: Annual wind power NMAE and annual wind speed MAE, ME and RMSE calculated for all experiments. The lowest values among each configuration (WF and PO) are bold. 50 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Figure 2.6: (a) December and (b) June mean wind speed (m/s) at hub height in the central area of the farm for the WF-HI-1D (blue) and PO-HI-1D (red) experiments and observations (green). Tables at right display monthly wind power NMAE and wind speed MAE for the area and for the total wind farm for WF-HI-1D. As discussed earlier in this section, low resolution (LO) and second day (2D) experiment errors are always higher than for high resolution (HI) and first day (1D) cases, respectively. With regard to the differences between simulations with (WF) or without wake effects (PO), errors are very similar, but in general, there is always a slightly better skill for the control (PO) experiments. Only MEs are significantly different for both configurations, with clearly more negative values for WF cases, which indicate a more pronounced underestimation of wind speeds. This suggests that the turbulent enhancement downwind of the turbines might be overestimated by the wake parameterization, leading to more mixing and moment loss in the simulated flow. 51 Chapter 2. Wind resource forecasting In a wind power forecast, the wind situations corresponding with the wind turbine power ramp, which is the ascending part of the wind curve before the nominal power (Figure 2.1a, between 5 and 13 m/s), have the biggest effect in forecast skill. Table 2.5 shows the same errors as in Table 2.4 but for observed power ramp winds only. CASE POWER NMAE (%) MAE WS (m/s) ME WS (m/s) RMSE WS (m/s) WF-HI-1D 19.79 1.94 -0.67 2.51 WF-HI-2D 21.38 2.19 -0.71 2.87 WF-LO-1D 22.64 2.28 -1.00 2.93 WF-LO-2D 23.43 2.46 -1.02 3.19 PO-HI-1D 18.76 1.87 -0.13 2.41 PO-HI-2D 20.53 2.13 -0.18 2.79 PO-LO-1D 21.96 2.24 -0.70 2.87 PO-LO-2D 22.76 2.42 -0.72 3.14 Table 2.5: Annual wind power NMAE and annual wind speed MAE, ME and RMSE for all experiments for power ramp winds only. The lowest values among each configuration (WF and PO) are bold. In general, results for power ramp winds are qualitatively similar to those for all wind speeds. Error values are however higher, indicating skill degradation. The increase in absolute value is around 5% for the NMAE and 0.1 m/s for the wind speed MAE, for all cases. The only improvement registered in comparison with the total data is in the wind speed ME for PO cases. The mean errors in Table 2.4 for HI-1D cases are lower than others from similar studies in complex terrain [44, 53, 75] and they are also slightly below other operative forecasting with WRF as [74]. Results by wind direction In the previous section we concluded that results in WF-HI-1D and PO-HI-1D experiments are quite similar in general for the considered wind farm. The main function of the WF-S is to simulate the wake produced by the wind turbines, and the effects of these wakes on the neighbouring wind turbines depend to a large extent on the incoming wind direction. Figure 2.7 shows the annual wind roses of the three wind farm areas obtained from observational data. 52 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Figure 2.7: Annual wind roses from observational data at hub height at turbines in the (a) north, (b) center and (c) south sections of the farm. DIR (%) N NE E SE S SW W NW North 6.95 18.66 15.82 2.92 12.45 27.43 12.11 3.66 Center 8.84 25.27 6.83 2.78 18.23 27.20 7.25 3.60 South 10.21 20.89 6.69 4.33 23.38 22.34 6.67 5.49 Table 2.6: Wind direction percentages in T07, T13 and T26. The wind direction registered is the same as the prevailing wind directions in Galicia (southwest-northeast). The perpendicular orientation of the farm (Figure 2.2b and Figure 2.2c) with respect to this axis of main wind directions explains why there are similar results between WF and PO cases. The wind farm orientation is optimal for its location and the wakes produced by the wind turbines do not have a significant effect on the wind farm itself. For this reason, in order to better study WF-S performance in these simulations, we analyze results relative to wind direction. The charts in Figure 2.8 show wind power NMAE as a function of direction for all experiment pairs with and without wake parameterization. Considering the impact of wakes reduces errors for north, northwest and southeast winds and most notably in the high resolution first day forecast (WF-HI-1D). Some small degradations of skill are obtained for south-westerly winds. Northerly winds, for which there is a clear positive impact of the wake parameterizaton, represent only between 7 and 10 % of cases, according to the roses in Figure 2.7, thus they contribute to a small improvement in overall skill scores. The slight negative contribution in the much more frequent southwest wind direction offsets these benefits. In spite of not being among the most frequent wind directions, the situations where WF-S is useful still encompass a large amount of data, since a 1 % of wind data is equivalent to around 90 hours. The dependence of the impact of the wake parameterization with wind direction is more evident when the three sectors of the farm are analyzed separately. Figure 2.9 displays charts similar to those in Figure 2.8, but for each area of the farm individually and only for high-resolution, first day cases (HI-1D). In the northern sector, wakes from the rest of the farm 53 Chapter 2. Wind resource forecasting Figure 2.8: Directional radar charts .comparing NMAE by direction, for the different pairs of experiments with (WF) and without (PO) wake parameterization: (a) HI-1D, (b) HI-2D, (c) LO-1D and (d) LO-2D cases. Figure 2.9: Directional radar charts as in Figure 8, but separately for each wind farm area ((a) north, (b) center, (c) south) and for HI-1D cases only. have an impact with southerly winds, which is when the WF-S scheme proves most useful, reducing errors significantly. For the central and southern sectors, on the contrary, it is with northerly winds that the turbines are mostly affected by wakes, and again in these situations the wake parameterization has a clear positive impact in skill scores. The southern area of the farm is at higher elevation than the rest, thus the central sector is not so affected by wakes with southerly winds. Figure 2.10a displays the wake produced by the wind farm with Northerly flow, which is the wind direction turning this wake onto to the farm itself. In contrast, Figure 2.10b depicts a southwesterly wind wake, which is, as commented above, the most common situation. Wakes are represented by the difference in wind velocity between the experiments with (WF-HI-1D) and without wake parameterization (PO-HI-1D) at hub height. In addition, time series of hourly mean wind speed for the selected two day periods are shown for observations, WF-HI-1D and PO-HI-1D simulations for turbines T03 and T29 (the third from the north and south ends of the farm, respectively). 54 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Figure 2.10: (a) Wind farm wake produced by a northerly wind, represented by wind speed losses. Two-day wind speed time series for two different wind turbines, (c) T03 (in the northern area) and (e) T29 (in the southern area). Arrows in each panel indicate wind direction and a vertical black line marks the time frame represented in (a) above. Curves display forecasted wind speed in WF-HI-1D (blue) and PO-HI-1D (red) experiments compared with observations (green). Panels b, d and f are similar to a, c, e, respectively, but for a case corresponding to a southwesterly wind period. Wakes in both cases extend for several kilometers, and they are more intense for northerly flows, when the perturbation from all turbines overlaps. They interact with the orography and produce non-homogenous areas of resource loss. Precisely, in the northerly wind case (Fig 10a), after moving across a valley and a terrain elevation, the wake reaches the simulation domain boundary, which implies a 15 km-long area of wind speed losses between 0.5 and 2.5 m/s. In contrast, the southwesterly wind wake shown in Figure 2.10b evidences that the total disturbance created in the most common situation for the area, produces a barely noticeable wind loss in the considered farm. This is due to the aforementioned optimal orientation of the turbines, and explains the similar results obtained between WF and PO experiments in the mean annual sense. 55 Chapter 2. Wind resource forecasting The time series in Figure 2.10 for turbines at both ends of the farm indicate that, for the case of a northerly wind wake, the simulation with wake parameterization (WF-HI-1D) is indeed clearly closer to observations at the south end of the farm, where the perturbation effect is mostly felt. Results for PO-HI-1D depart from the observed value much more than for other situations, and the difference between WF and PO experiments is around 2 m/s, which is quite significant. In all other cases, when the wake is not directly impacting the turbines, the differences between WF and PO simulations are minimal, and both experiments follow observations reasonably well. In summary, WF cases always achieve better results than PO in situations where the wakes clearly affect the wind farm itself, thus suggesting that the WF-S scheme provides a good estimation of wake effects in these simulations Summary and conclusions In the present study we evaluated the WRF model at high resolution as a wind energy forecasting tool for a real wind farm over complex terrain. In particular, we assessed the representation of wake effects with the WF-S scheme on this particular wind farm as well as on its nearby surroundings. For this purpose, 48h simulations encompassing one year, with two configurations with and without wake parameterization and at different spatial scales and lead times, have been performed and validated with real data. Our results show that wind speed annual mean absolute errors in 333 m resolution experiments considering wake effects are small, of only 1.87 m/s (1.94 m/s for power ramp winds only). These errors present a seasonal cycle, peaking in winter with values of 2.33 m/s in January and decreasing in summer with a minimum value of 1.55 m/s in July. The cycle follows that of wind speed in the region. Moreover, forecast skill scores show also a diurnal cycle, with bigger absolute errors in night-time hours (peak of 2.02 m/s at 6 UTC) than during the day (low of 1.6m/s at 13 UTC), perhaps related to both the daily cycle of wind speed and turbulence intensity. It is possible that at the high resolution used, the model overestimates turbulence, and hence moment loss, since some large turbulent eddies are starting to be resolved in addition to be parameterized. With regard to normalized mean absolute errors for power output, the behaviour of the results is qualitatively very similar to that of mean absolute errors for wind speed. Annual mean values for NMAE are of 14.75 % (19.75 % for power ramp winds only), and oscillate between 19 % inFebruary and and 12 % in August, and between 16.63% at 6:30 UTC (night) and 12.73% at 13 UTC (daytime). Second day forecasts present some skill degradation and similar trends as first day products. Wind speed mean absolute errors increase from 1.87 m/s the first day to 2.13 m/s on the second, and normalized mean absolute errors for power output raise to 16.36 % from 13.75 % the first 24h. These variations are relatively small, which suggest that the WRF modelling tool results with a 48 h lead time are still very valuable for a real application in the industry. Lowering resolution from 333 m to 1 km has very similar effects to increasing forecast lead time by 24h, Skill scores for the first day in low resolution experiments are almost identical to those of second day forecasts at high resolution, and they further degrade as simulation time progresses. The impact of resolution is important for this farm, given the complexity of the terrain of the area, which is much better resolved at 333m than at 1km grid space. Forecasts with and without considering wake effects have nearly the same mean errors. The spatial distribution of the turbines is optimal for the location, since they are oriented perpendicular to the prevailing wind directions in the southwest-northeast axis, hence minimizing wake impacts. Notwithstanding, forecast skill scores are clearly better when wakes are accounted for in the cases where the 56 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ farm is directly impacted by its own disturbance, which occurs mostly with northerly flows. However, these represent a small fraction of situations in the area, and the total reduction in errors is therefore not very relevant. 57 Chapter 3. Wind nowcasting like the U.S. National Weather Service stipulate that it pertains to lead times from zero to three hours. However, forecasts up to six hours are also considered nowcasts by different agencies [106]. To analyze the capabilities of our methodology, we evaluate the K-B filter with different nowcasting horizon periods, from 6 hours to 10 minutes (Table 3.2) for each one of the daily simulations. Results are compared with both observations and original non-post-processed WRF outputs. Experiment RAW WRF results without post-proccess K-B 6h K-B filter nowcasting used for 6 hours time horizon K-B 1h K-B filter nowcasting used for 1 hour time horizon K-B 30min K-B filter nowcasting used for 30 minutes time horizon K-B 10min K-B filter nowcasting used for 10 minutes time horizon Table 3.2: Short explanation of the experiments tested in this study. The bold names on the left are the identifiers used hereafter for each case. Results and discussion The next section is divided into three parts; the first one analyzes the results obtained with the nowcasting wind speed postprocess at different time horizons, from 6 hours to 10min in advance. The analysis is developed for different temporal periods, from annual mean results to the performance of the experiments at individual simulation timesteps. The second subsection shows the results of the K-B filter used to measured wind direction correction, same as with wind speed, with different nowcasting horizons and mean errors for several periods. Finally, subsection 3.3 shows an example of the improvement that K-B 1h applied in wind direction could effect in the daily management of the studied wind farm. Wind speed nowcasting This subsection employs several error measures at different temporal scales to provide a general view of the skill of all the experiments. The next equations show the statistical measures used in different temporal scales throughout the chapter. MAE =1 n n ∑ i=1|fi−obi|(3.5) ME =1 n n ∑ i=1 (fi−obi)(3.6) 64 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ RMSE =s∑n i=1(fi−obi)2 n(3.7) σ=s∑n i=1|xi−x|2 n(3.8) Equation (3.5) gives the Mean Absolute Error (MAE) between forecasts ( fi) and observations (obi), (3.6) the Mean Error (ME) estimating possible systematic biases, (3.7) the Root Mean Square Error (RMSE), where n is the sample size, and (3.8) the standard deviation (σ), where xiis the value, xthe mean value of a study period. Annual mean results are firstly shown in Figure 3.2, and Table 3.2 for each wind turbine studied. Figure 3.2 displays a bar chart with the annual WS MAE and standard deviation for each turbine and experiment performed. Table 3.3 a and b present the annual WS ME and RMSE, again, for each turbine and experiment. Figure 3.2: Barchart with annual WS MAE for all experiments and wind turbines analyzed. For each case, the standard deviation (σ) is represented by a black line on the top of each bar. Table 3.3: (a) WS ME for all experiments and wind turbines, (b) same as (a) for RMSE. Overall, a similar improvement concerning WS MAE for all the wind turbines and cases is recorded. Raw WRF forecast MAEs are between 1.91 and 1.74 m/s, except for WT26, which has a RAW MAE of 2.31 m/s. All of these results, regardless of wind farm area, present a small reduction of the WS MAE (of around 10%) in the longest nowcast period, K-B 6h (blue). The accuracy of the K-B filter is significantly enhanced for shorter lead times, with K-B 1h (pale blue) reaching a WS MAE of around 1 m/s in all cases. The shorter nowcasting periods, K-B 30m (green) and K-B 10m (pale green) show further reduction of error, with values around 0.86 and 0.72 m/s respectively. The standard deviations plotted on top of each bar (σ) are also reduced for all turbines, in parallel with the shortening of the nowcasting periods. RAW cases have a mean of 0.82 m/s, and K-B 10 min present a mean value for all of the machines of 0.33 m/s. The behavior of the WS ME is different as compared with MAE. All of the RAW WS MEs for the six wind turbines are negative, ranging from -0.33 m/s in WT13 to -1.40 in WT26. 65 Chapter 3. Wind nowcasting Unlike in WS MAEs, there is a clear difference between RAW and K-B 6h MEs. In K-B 6h cases, the ME practically disappears, with a value of 0.046 m/s. In the rest of the experiments, the ME is reduced even more; reaching a total ME of 0.005 m/s for the K-B 10min case. WS RMSEs show a similar tendency to that of WS MAEs, with close mean values of the RAW and K-B 6h experiments (2.54 and 2.36 m/s respectively) and a definite improvement between the three shorter nowcasting periods (K-B 1h-30min-10min) and K-B 6h. Figure 3.3 analyzes the relationship between the observed and simulated wind field in terms of module and direction. As also discussed previously, there are similar error values and patterns in all of the wind turbines, which allows us to group them by areas while maintaining the rigor of the validation. In Figure 3.3a-c-e the wind speed distribution is displayed for the observations and the RAW, K-B 1h and K-B 6h experiments, for the North, Center and South areas respectively. Radar charts in Figure 3.3b-d-e depict the WS MAE as a function of direction for all experiments and areas. Figure 3.3: (a) Wind speed distribution plot for observations, RAW, K-B 6h and K-B 1h time experiments in the north area of the wind farm. The standard deviation of each distribution is indicated in the legend. (b) Radar charts comparing WS errors by observed wind direction in the north area. (c) Same as (a) for the center area. (d) Same as (b) for the center area. (e) Same as (a) for the south area. (f) Same as (b) for the southern area. 66 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ The distribution plots in Figure 3.3a-c-e show a significant improvement from the K-B filter in the three areas. K-B 6h mitigates the main wind speed deviations in all the range of the distributions. In the north area (Figure 3.3a) K-B filter cases correct a general overestimation from 6 to 10 m/s and an underestimation from 11 to 22 m/s. This behavior is repeated in the Central and South areas. K-B experiments (even K-B 6h) obtain an important amelioration of the entire distribution shape, which is in agreement with the low ME errors shown in Table 3.2a. This demonstrates the reliability of the K-B filter in different wind situations, correcting underand over-estimations indistinctively, which is vital for wind energy applications. Charts in Figure 3.3b-d-f show the skill of the K-B tool depending on wind direction. In RAW cases, the three areas present variations, as the general orientation of each zone and their position relative to the rest of the farm change forecast performance. K-B 6h decreases mainly the higher directional errors as, for example, with southerly winds in the South area (Figure 3.3f). However, the K-B tool for 1h, 30 min, and 10 min tends to smooth out errors from all sources, improving results in all the prevailing wind directions. Apart from the low MAE values observed, the symmetry in the figures in the shorter K-B time ranges leads to the conclusion that the tool is robust, rectifying wind errors in any regime. It presents the same good accuracy correcting northeast flows, principally produced in summer, as it does with southwesterly situations, characteristic of winter months. The seasonal cycle of errors averaged for all the turbines is shown in Figure 3.4, comparing wind speed monthly MAE from RAW results with the same calculation from K-B experiments. Observed monthly mean wind speeds are also shown for reference in the table below Figure 3.4: Monthly wind speed MAE for the wind turbines with K-B model (6h (blue), 1h (dotted blue), 30min (green), and 10min (dotted green)) and RAW (red). Mean Monthly WS / Mean annual WS Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 1.4 1.16 0.97 0.85 1.07 0.77 0.8 0.76 0.83 1.02 0.88 1.48 Table 3.4: Observed mean monthly wind speed normalized with observed mean annual wind speed. 67 Chapter 3. Wind nowcasting RAW MAEs are relatively constant throughout the year, with values around 2m/s in all months. Slightly larger errors are observed in the winter period, when mean wind speeds are higher, and lower biases occur in summer months, when mean wind speeds are also lower. K-B 6h obtains the biggest improvements in the months when the RAW error is largest, such as December or February and minor improvements in the summertime. This tendency to better improve bigger RAW MAE is also reflected in previous comparisons such as for WT26 in Figure 3.2 or the south radar chart in Figure 3.3f. In K-B experiments for shorter lead times, the pattern shown in Figure 3.3 is repeated, with a significant error decrease relative to K-B 6h results and very similar outcomes in all K-B cases and months. The K-B 1h MAEs are all very close to 1 m/s, February registers the higher K-B 1h MAE, with 1.19 m/s and June the lowest, with 0.87 m/s. This equality among monthly errors is maintained in K-B 30m and 10m with mean values all year long around 0.88 and 0.73 m/s respectively. After investigating the intra-annual behavior and monthly error patterns of the experiments, we examine their accuracy on a finer timescale. Figure 3.5 displays the mean hourly wind speed MAE during the simulated year for all the wind turbines analyzed in each experiment. Figure 3.5: Wind speed MAE, hourly for the entire 24 h simulation period with RAW outputs (red) and K-B model cases: K-B 6h (blue), K-B 1h (dotted blue), K-B 30min (green) and K-B 10min (dotted green). RAW error results present a daily cycle, with values of the MAE around noon lower than during the night. These differences among the hourly WS MAEs are, however, not very significant, and the lower absolute errors at noon seem to be related with the better representation of turbulent fluxes during that part of the day [29]. In contrast, K-B 6h errors do not follow a daily cycle, attaining values below 1.75 m/s mostly during daytime, but obtaining a worse result than RAW at 01 and 13 UTC. K-B 1h, 30 min, and 10 min present a very flat error pattern all day long. There are practically no differences among hourly results in these experiments. The situation is very similar to that in the monthly error plot (Figure 3.4); shorter term K-B cases represent a significant improvement with respect to K-B 6h and RAW, and there is homogeneity in the entire error series in K-B 1h, 30 min, and 10 min. The skill limit for the shorter term cases corresponds to a non-systematic part of the error, which is unavoidable for the filter The general conclusion reached from the different time scale and wind regime comparisons presented above is that the K-B filter at short-term erases any source of error. It achieves lower errors in all conditions, regardless of the intensity and direction of the wind or the time of day 68 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ or season of the year. This uniformity in the corrections toward the elimination of the ME (Table 3.1) from short-term K-B cases, highlights the utility of this optimization module for wind energy purposes. To illustrate the day-to-day results of the K-B nowcasting tool regarding wind speed at hub height, we show next comparisons between RAW, K-B 1h, and observations for the Center area (WT13 and WT16 mean values) during the entire months of May and December. These months have the lowest and highest K-B 1h MAE in this area respectively. Figure 3.6: (a) Wind speed at hub height in the center area with observations (orange), RAW (dotted red), and K-B 1h (blue) in May. (b) Same as (a) in December. Wind speed MAE for each series is presented in the legend for both figures. Disregarding some occasional large errors, the original WRF output (RAW) generally yields a good performance during the months shown, achieving MAEs below 2 m/s in both cases. Even considering these low RAW errors, which are extensive to all the yearly series, the K-B 1h results represent an important improvement nonetheless. This is clearly apparent in these December and May plots (Figure 3.6), when the procedure successfully corrects forecast error by a factor of around 50 %. The nowcasting tool is able to produce better results at different time ranges regardless of the origin of the error, eliminating the most substantial deviations in any situation. Overall, the K-B filter exhibits a useful short-term operational forecasting performance, offering a stable improvement of the original WRF outputs during the whole year and for all the wind turbines. 69 Chapter 3. Wind nowcasting Wind direction nowcasting Following the discussion on the capabilities of the K-B filter regarding WS (wind speed) nowcasting, we review the statistical tool predicting WD (wind direction) following a similar structure to section 3.1. To perform the WD nowcasting post-process we obtain the zonal and meridional wind components (U and V) from WS and WD observations, and we use them to correct, with the K-B filter, RAW U and V variables directly extracted from the model output. With the corrected U and V, we recalculate the post-processed WD. Using the same process as in the WS analysis, we test the WD nowcasting for different short-term forecast time periods (Table 3.2). Figure 3.7 displays the annual WD MAE and standard deviation for each turbine and experiment. Figure 3.7: Barchart with annual WD MAE for all experiments and wind turbines analyzed. For each case, the standard deviation is represented by a black line () on the top of each bar. The bar chart depicts in red the results of WD MAE for each RAW wind turbine. All of them are above 20◦; WT7 with the lowest value, 20.69◦and WT24 with the highest, 28.51◦. The mean value of the MAEs for all the wind turbines in the RAW case is 25.10◦. Comparing K-B 6h with the unfiltered WRF results, we can see an important amelioration in most of the wind turbines; all of them are now below or practically at 20◦MAE, reducing the mean total K-B 6h MAE to 19.67◦. This value translates into a 22% improvement with respect to RAW results, a correction impact that doubles that obtained in wind speed nowcasting for this same comparison (WS RAW v.s. K-B 6h). This difference can be attributed to the origin of the observational data used. In WS nowcasting we correct one output data series, with another single data series (observed wind speed). However, in WD nowcasting we use WS and WD data from sonic anemometers to obtain the U and V wind components and calculate the post-processed WD. We are thus introducing more sources of error in this last step that the K-B filter seems nevertheless capable of offsetting. 70 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ In the shorter lead times of wind direction nowcasting (K-B 1h, 30min, and 10min) the behavior presents a similarity to that of wind speed nowcasting (Figure 3.2). All the turbines attain analogous error values in each experiment, with a total mean WD MAE of 12.16◦, 10.69◦ and 9.37◦for K-B 1h, K-B 30min and K-B 10min respectively. Aside from these low errors, there is also a corresponding reduction of the standard deviation. The K-B filter at short-term nowcasting increases prediction accuracy significantly and eliminates more substantial punctual errors from the original WRF forecasting, which is quite important for the use of these kinds of combined nowcasting systems in wind farm applications. The reason lies in the fact that during daily operations of these installations, wind turbines are continually being orientated depending on wind direction, and potential errors in this manoeuvre lead to machine overstress and production decrease. As in Section 3.1 for wind speed, the monthly errors for wind direction are shown in Figure 3.8a, and on a finer timescale, Figure 3.8b displays the mean hourly wind direction MAE, averaged for all the turbines in the farm. In both cases, RAW WRF outputs are compared with the results of the K-B experiments. Figure 3.8: (a) Monthly wind direction MAE for the wind farm with K-B model experiments and RAW. (b) Wind speed MAE, hourly for the entire 24 h simulation period. In both cases the comparison is between RAW outputs (red) and K-B model cases: K-B 6h (blue), K-B 1h (dotted blue), K-B 30min (green) and K-B 10min (dotted green). 71 Chapter 3. Wind nowcasting Monthly results in Figure 3.8a provide insights on the origin of the 22 % improvement in forecast skill of K-B 6h with respect to RAW, commented in Figure 3.7. From 02/15 to 03/15 and from 07/15 to 01/16 the effect of the K-B 6h is noticeable, with a decrease in errors of around 7.5◦during those periods. Nevertheless, the scenario is entirely different in spring, when K-B 6h practically does not refine the original RAW result. This lack of improvement can be partially explained by the fact that, during spring months, the original WRF outputs register the lowest U and V errors of all the year, and, similarly to previous results, the K-B filter has more difficulties correcting RAW forecast results with low errors. K-B 1h, 30min, and 10min follow the general tendency of the rest of the comparisons, presenting a substantial error decrease throughout the year. K-B 30min skill score is always below K-B 1h and the same for K-B 10 min with respect to K-B 30min, with all three time series showing a parallel behavior throughout the whole period. As opposed to the case of wind speed (Figure 3.4), winter months, particularly December and January, have the lowest errors of the series in all wind direction K-B results. For example, K-B 30min error is around 6.6◦during these winter months, but over 12◦in springtime. The lower WD MAE in winter months seems to be related with the higher mean wind speed registered during that part of the year (Table 3.3). This means that there are fewer periods of weak winds, which are associated with increased variability in wind direction, presenting, therefore, more difficulties for the K-B filter to handle. Changing the timescale of the analysis, Figure 3.8b shows the evolution of the WD MAE throughout the day. The RAW series (red line) does not present significant hourly changes; all the values are close to 25◦with a slight increase in the late hours of the day. K-B 6h differs from this homogeneity, displaying a slight daily cycle with more accurate predictions from 04 to 12 UTC, with a minimum MAE=16.64◦at 09 UTC increasing to a maximum of 21.76◦at 13 UTC. Same as in the daily analysis of wind speed (Figure 3.5), in the shortest nowcasting periods (1h, 30min, and 10min), the variability is practically eliminated. The K-B filter decreases errors and smooths out the series, with rather constant MAEs around 10◦throughout the day. The comparisons at different time scales among all the cases confirm that the K-B filter yields a valuable improvement in wind direction prediction. The shortest term nowcasting cases, starting from K-B 1h, lead to a significant correction of WRF outputs independently of the meteorological situation of the moment. To conclude this section, in Figure 3.9 we exemplify the capabilities of wind direction nowcasting with K-B filter. Specifically, a two-day WD MAE comparison is displayed for WT13 between RAW (red) and K-B 1h (blue). The arrows inside the circles show the instantaneous observed wind direction every four hours (black) and its respective RAW (red) and K-B 1h (blue) prediction 72 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Figure 3.9: Wind direction MAE with K-B 1h filter and RAW in W13 during a 48h period. Wind direction arrows with observations, K-B 1h, and RAW are displayed on the top every four hours. The two-day results presented, from 2015/05/15 to 2015/05/17, show different forecast skill patterns. In the first day, RAW and K-B 1h error series are close to each other in their first 12 hours. In the next hours, RAW maintains a more constant higher error (around 25◦) for 3-4 hours, which K-B 1h drastically corrects. During the second day, this difference between the time series increases in magnitude; RAW predictions go over MAE = 30◦for 12 hours, reaching up to 40◦in different moments. Throughout all this time period the K-B correction sharply eliminates these big MAEs to values below 10◦. Given these results, we can affirm that K-B 1h provides a good skill improvement in all the situations where the original RAW forecast presents large errors. As in the case of wind speed nowcasting, the correction of big WRF wind direction biases is crucial for wind farm operation applications. Application of the K-B filter for wind power forecasting In this last section, we test the capabilities of K-B 1h wind direction nowcasting tool on a real scenario of Coruxeiras wind farm in the studied year, with special attention to critical issues in power prediction, such as wind ramps. For this purpose, we focus on the results corresponding to the wind turbine power ramp, which is the ascending part of the wind power curve before nominal power (Figure 3.10). Wind speeds within this curve (in this case from 4 to 13 m/s) have the most significant effect in the forecast skill of any nowcasting tool in the wind energy field. In Figure 3.10, we present the relation between the wind power curve of the wind farm’s turbines (ECOTECNIA74 with 1.670 MW of nominal power [63]) and the WD MAE associated with each wind speed bin, both for RAW and K-B 1h cases. Table 3.5 displays the WD MAE in power ramp wind ranges for all the experiments and for each area of the wind farm. It also compares these results with persistence at 10min, 30min, and 1h. The persistence forecast is the assumption that the next timestep value in a prediction is going to be the same as the last measured value [107]. 73 Chapter 4. Annual Wake Physics Name Planetary Boundary Layer Mellor–Yamada Nakanishi Niino Level 3 Microphysics Thompson, Field, Rasmussen and Hall Cumulus Disabled Shortwave Radiation CAM Collins et al Longwave Radiation RRTMG (New version of RRTM) Land Surface Unified Noah Land Surface Model Surface Layer Nakanishi and Niino surface layer scheme Table 4.1: Parameterizations for the highest resolution domain, D04. Same as Coruxerias study (Chapter 2 and 3), the initial 6 h of each run are excluded from the forecast data series used in the analysis of the results, thus, only outputs after this spin-up time are considered. Annual Wake measurement As in Chapter 2, the Fitch scheme is used (WF-S hereafter) to estimate the effects of wind turbines. The WF-S represents wind turbines as momentum sinks, transferring one fraction of kinetic energy into turbulent kinetic energy (Equation 4.1) and another fraction into electricity (Equation 4.2)[35]. The MYNN 2.5 level scheme calculates the mixing resulting from the vertical wind shear induced by the momentum sink. MYNN also takes into account the effects of buoyancy and the impact of stability on the turbulent length scale. ∂TKEi jk ∂t=0.5Nti j(Cp−Ct)(|V|i jk)|V|3 i jk Ai jk (zk+1−zk)(4.1) ∂Pi jk ∂t=0.5Nti jCp(|V|i jk)|V|3 i jk Ai jk (zk+1−zk)(4.2) In equations 4.1 and 4.2: i and j are the grid cell coordinates, k is the vertical level, number of turbines per square meter, Cp is the power coefficient of the wind turbine, Ct is the thrust coefficient of the wind turbine, V is the horizontal velocity vector, A is the rotor swept area and z is the height of the model level. In the case studied, the wind farm is composed of two different kinds of turbines with different dimensions and power curves (Figure 4.2). The wind turbine scheme is capable of taking into account different types of turbines within the same simulation domain. We introduce in the running settings the characteristics of both machines (Figure 4.2). Each of them is going to have a different wake effect in the environment. 80 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ X wind turbines Y wind turbines Figure 4.2: (a) X wind turbine wind power curve. (b) Y wind turbine wind power curve. In both cases, the blue line represents power and the dotted green thrust coefficient. As shown in Chapter 2 (Figure 2.10) we can estimate the impact of wakes on the wind farm by taking the difference between simulations performed with and without wake parameterization. Similarly, in this study, we analyze the broader effect of the wind farm on its surroundings by calculating the mean wind speed percentage loss over all time steps throughout the year from the daily simulation difference between WF-S activated and WF-S deactivated experiments. The result is the mean “Annual Wake” displayed in the next section (Figure 4.6). Initial-boundary conditions and observations The study covers the period between 2016-05-01 and 2017-04-30. One year (2016-05-01 to 2017-04-30) of GFS analysis data from the National Center for Environmental Prediction (NCEP) is used as initial and boundary conditions, with a 3-h update interval. The horizontal resolution of this dataset for all variables is 0.25 x 0.25 deg, with 32 levels ranging from 1000 to 10 hPa. The observational data used in this work is provided by Xinjiang Goldwind Science & Technology Co. collected from anemometers located on top of each nacelle. One year ten-minute-interval wind data for three turbines are available. Each one is located in a different part of the installation, X27 in the north, X58 in the center an X68 in the southern part (Figure 4.1b). The next figure shows the annual wind rose for the three of them. Figure 4.3: Annual wind roses from observations for the three validation WT points X27, X58, and X67 at hub height. 81 Chapter 4. Annual Wake Results Jiangsu Annual Wake This section shows the Annual wake result for Jiangsu wind farm (4.3.1) and the validation of this same study using observational data introduced in the previous section (4.3.2). Apart from this, the last subsection presents the Annual Wake in Coruxeiras wind farm obtained from the results of the study presented in Chapter 2. The main objective of this study is to estimate the mean annual wake of the farm, as is explained in the previous section. This is obtained from the mean wind difference between WRF simulations with and without WF-S across all time steps. Each of these differences per timestep also gives us the wake effect at a specific instant. The next figures show three wakes obtained at different moments during the studied period. The mean instantaneous wind speed in the area (MWS) is also presented in each image. 2016-05-03 20:00 MWS = 11.43 m/s 2017-02-16 16:30 MWS = 14.12 m/s 2017-02-17 12:30 MWS = 5.52 m/s Figure 4.4: (a) Wake produced by a SW wind, (b) wake produced by a N wind, (c) wake produced by a NE wind. On the top of each image, the time instant represented and the mean WS at hub height in the area are indicated. In Figure 4.4a we can see the wake produced due to the interaction of a southwestern wind with around 11 m/s wind speed over the wind turbines. Wind speed losses of 1.5 m/s are registered in practically all the east part of the farm. This means that at that moment, 35 wind turbines are producing around 300 kW less than the other part of the park (Figure 4.2). Figure 4.4b displays a stronger wake impact on the farm itself created by a higher wind from the north (≈14 m/s). In this case, all the Y machines, and 20 of the X type are receiving wind at 10-11 m/s with a high turbulent regime. Changing the incident wind direction again, Figure 4.4c shows a resultant wake from a low wind (≈5 m/s) from the northeast. What becomes evidaent is how the wake, despite being the result of low wind, extends several kilometers (≈15 km), we can see how the wake even reaches the simulation domain boundary at the lower left corner. Continuing with the qualitative study of the wakes, the next group of figures displays three examples of daily mean wake effects measured during the period studied. 82 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ 2016-05-15 MWS = 7.5 m/s 2016-07-13 MWS = 5 m/s 2016-08-07 MWS = 9 m/s Figure 4.5: Daily mean wake effects for different days of the year (wind percentage loses). On the top of each image, the day represented, and the mean WS at hub height in the area are indicated. Depending on the prevailing winds of the day, daily mean wakes (Figure 4.5) are drawn over a different area and affect a different part of the wind farm, the picture for the 2016-05-15 case (Figure 4.5a) is the result of a mainly western and northwestern wind. In the figure 4.5b case the mean impact is softer due to the lower mean wind module registered for that day. The wake area on the right could be caused by interaction between the wind farm and the land breeze effect. The wake in Figure 4.5c shows a clear mean prevailing southeast wind case with a higher mean wind speed which provokes significant resource losses in all the north part of the wind farm. Just the same as in Figure 4.4c, the straight shape of the wakes at the last part of their edges are due to the contact with the boundary of the simulation domain. These pictures can be useful information in a wind farm siting project because they allow us to the see how the wake effect from each wind turbine interacts with the other neighbors’ wakes and affect other areas of the installation. The two main effects over downstream wind turbines, which are wind resource loses and stress to machines, are well resolved by this technique. A turbine can produce in its nominal power range being affected by a wake effect if the wind is high enough. However, this produces a more turbulent regime, making the machine suffer extra loads which are going to have a negative impact on it in the medium and long-term. In the same way as for the daily mean wakes, we can calculate the mean wind speed percentage loss over all the timesteps of the year, which is what we call “Annual Wake”. This annual measurement is displayed in the next figure for Jiangsu wind farm. The wind turbines of this farm have a significant separation between them, with around 800 m of mean separation distance all along the installation. Despite this configuration, the high variability of the wind direction (Figure 4.3) and flat orography of all the region (Figure 4.1) results in significant wind resource losses between wind turbines in several directions, and prolonged wake effect all around the wind farm. In figure 4.6b we can see the high percentage of wind resource loss around the wind turbines. The northern row of turbines (from X27 to X32) produce losses of over 7% 2 km away northbound. This high level of resource loss in the near environment not only has a direct effect on wind power production, but also on a machine’s lifetime. Figure 4.6a shows a more general view of the wind farm impact on the surroundings. Significant wake effects are distinguished 10 km away from the installation in different directions, especially pronounced in the west area of the farm. 83 Chapter 4. Annual Wake Figure 4.6: (a) Annual wind farm wake. (b) Zoom over the north of the wind farm. Jiangsu validation Observational data from three meteorological stations at the hub of three wind turbines is used to validate the wind output from the model used to estimate the Annual Wake. In the next table, we present some wind annual mean errors for each validation point. X27 (N) X54(C) X67(S) Mean ME (m/s) 0.932 0.997 0.964 0.964 CC 0.802 0.797 0.782 0.794 RMSE(m/s) 2.070 2.045 2.056 2.057 MAE(m/s) 1.592 1.565 1.580 1.579 Table 4.2: Annual wind speed ME, CC, RMSE and MAE for three different wind turbines (X27, X54, and X67). The first thing that stands out is the similarity between the three wind turbines in all the errors measured. This is also reflected in the total mean values (right column), which are practically the same as the ones for each point. This proximity between different turbine results could be in part due to the flat terrain of the region; this characteristic makes all the wind turbines 84 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ be affected to a similar degree by the wake effects from the rest of the farm. Furthermore, there is not a clear prevailing wind direction, as we can see in the wind roses (Figure 4.3) and the Annual Wake shape in Figure 4.6. The errors show a good performance of the simulations; the total MAE is 1.58 m/s which is below the same error calculated for theCoruxerias case (Chapter 2, Table 2.4, WT-HI-1D). The improvement concerning Coruxeiras seems to be related to the plain topography of this region, which makes forecasting easier without any interaction between complex terrain and wind to resolve. The total RMSE also presents a low value for the same reason. Apart from this, it is important to point out that in this case, we are using GFS analysis instead of GFS forecasting as in Coruxeiras, this has a direct effect on the quality of the simulation in all the domains. However, ME values are more significant than in the Galician case (WT-HI-1D ME= - 0.67 m/s), and with a positive value. The high-resolution simulations with WF-S for this area are overestimating the wind speed in all the region. The expected result for these kinds of simulations is an underestimation of the wind speed due to double counting of the turbulent processes as is explained in Chapter 2. This indicates that the error source can be related to the performance of the coarser resolution phenomena in the other domains. Similar overestimation in the wind for this region of China has been recorded in a Chinese 3km-grid WRF-RTFDDA forecasting system [117] where they attribute this bias to weak wind situations, especially in summer. To illustrate the day-to-day results of the WRF forecasting tool in terms of wind speed at hub height, we show the next comparison plots between model results and observations at hub height. Specifically, the following plots show the results throughout November for the three validation points tested. Figure 4.7: November wind speed at X27 hub height for WRF simulations (orange) and observations (blue). 85 Chapter 4. Annual Wake Figure 4.8: November wind speed at X58 hub height for WRF simulations (orange) and observations (blue). Figure 4.9: November wind speed at X67 hub height for WRF simulations (orange) and observations (blue). Apart from a few large punctual errors (e.g., 08-11-2016 or 22-11-2016), the wind forecast presents a good tendency in all the cases. As it is expected from mean errors in Table 2.2, the performance of the prediction is very similar in all the cases. The three of them have a good correlation across the whole period represented, which also concords well with the mean correlation values shown in Table 2.2, reaching a total mean value of CC = 0.79. 86 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Annual Wake Coruxeiras To finish with the presentation of results, in this subsection, we show the Annual Wake measurement for the Coruxeiras wind farm (Chapter 2), which lies in a densely exploited complex terrain region in Galicia. Just as in the Jiangsu case, a period of one year broken into 365 daily runnings is simulated at 333 m, in this case with a daily operational forecasting set-up, using GFS 0.25 Forecasting instead of the GFS 0.25 Analysis employed in Jiangsu. With the same technique as in 4.3.1., we estimate the Annual Wake for this wind farm. Figure 4.10: (a) Annual wind farm wake, represented in wind speed loss percentage. Other wind farm locations in the area are outlined in black squares. (b) Zoom of the NE of the installation. The wind farm marked on top suffers about a 2.25 % annual resource loss due to the presence of the Coruxeiras wind farm. The figure shows the significant disturbance that the wind farm produces on the area. Despite the attenuation of the effect caused by the complex terrain, considerable wind resource losses extend for several kilometers, with a pattern corresponding to the wind roses shown in Chapter 2 (Figure 2.7). The wake induces losses of around 0.5 % even at 17 km from the wind farm in the southwest-northeast axis corresponding to the prevailing winds in the area. The considered farm is also significantly affected by its own wake, especially with northerly winds. The central section is the most impacted, with a mean annual decrease in wind speed exceeding 6 %. The high wind resource in the area is intensively exploited by dozens of wind farms, outlined in black in Figure 4.10. From the estimated annual wake, it is apparent that the Coruxeiras wind farm affects others nearby. In turn, the combined wakes of neighboring farms likely decrease observed wind speeds in the location of study as well. Wake effects like those shown in Figure 4.10 would represent significant economic losses and should be taken into account in any initial wind farm plan. For this purpose, the modeling tool that we present in this study can be very useful, not just for exploitation planning, but also to assess the impacts of wind turbines on the environment. 87 Chapter 4. Annual Wake Conclusions The objective of this study is to calculate the mean annual wake or mean annual wind resource loss in the area due to wake disturbances originating from a wind farm in the Chinese province of Jiangsu. One year of high-resolution WRF with and without the Fitch wind turbine scheme are performed in order to estimate the Annual Wake. Observational data from three different wind turbines are used to validate the wind model results. Furthermore, the Annual wake calculation for the Coruxeiras wind farm (Chapter 2) is also presented. A measure of the instant wakes can be obtained from the difference between WRF with and without WF-S each timestep. These visual representations help to understand the effects over downstream wind turbines as wind resource losses and stress to machines, both well resolved by this model configuration. It can be interesting to correlate these visual wind results with the same ones for other variables to detect possible dangerous points for a turbine such as, for example, and specific area with high mean wind speed and a significant associated TKE. In the Jiangsu wind farm, despite the spread location of the turbines, the Annual Wake shows considerable wind resource losses among wind turbines in several directions and prolonged wake effects all around the wind farm. The percentage of wind resource loss is very high in the first kilometres away from the machines, reaching, for example, up to 7 % 2 km away northbound. In a more general view, The plain orography of the area allows significant wake effects to maintain themselves at large distances from the farm. Specifically, over 15 km away in different directions aas registered, especially pronounced in the west area of the farm. The validation presents a small MAE at the three observation points (MAE 1.6 m/s) practically without differences among them. This improvement with respect to the Coruxerias case is mainly due to the flat terrain where the farm is, and the global model analysis data used instead of the forecasting one used in the simulations for the Galician wind farm. The overestimation in the wind ME indicates that the error source can be related to the performance of the coarser resolution phenomena in the other domains. However, a more detailed analysis should be done to confirm this affirmation. In the Coruxeiras case, the Annual Wake length in complex terrain achieves longer distances than was expected. The environmental footprint extends for several kilometers in the southwest-northeast direction of the prevailing winds, with resource losses of 0.5% even at 17km from the turbines. This result suggests that the combined effect of the dozens of farms existing in the region could be considerable and that the disturbance of neighboring farms is likely causing a significant impact on power production. It would be interesting to carry out a thorough study of the effects of the wind farm cluster on the environment using WF-S and taking into account all the wind farms (Figure 4.10, black squares). Despite not reaching the scale of small eddies, the model resolution used in this study seems to be more than enough to resolve the majority of the most important processes related to wind farm and planetary boundary interaction. In view of the results, we conclude that this tool can be very useful in the wind energy industry. For instance, it can analyze beforehand the total impact of a future neighbor wind farm construction on an existent wind farm’s annual production. It can also provide valuable information to optimize farm designs, increasing the production and lifespan of the machines. 88 Chapter 5 Extreme events study Introduction The Isthmus of Tehuantepec, in the Mexican state of Oaxaca, is the narrowest stretch of land separating the Gulf of Mexico from the Pacific Ocean. The Sierra Madre mountains cross the isthmus from east to west, leaving, however, a pronounced gap in the middle (Chivela Pass), coinciding with the point of shortest distance between the two sea masses, of only 200 km. The elevation of the Chivela pass is 224 m, whereas mountain peaks on the side sierras reach 2000 m, creating ideal conditions to generate a powerful wind corridor [181]. In winter, cold high-pressure systems originating in North America move over the Gulf of Mexico in the wake of south-reaching cold fronts, and large pressure differences develop across the isthmus between the bay of Campeche and the Gulf of Tehuantepec, on the Pacific side. This pressure gradient results in a northerly flow situation, in which the wind is accelerated southward by cold air damming, traveling through Chivela Pass to finally blow violently outward into the Pacific Ocean. In the Gulf of Tehuantepec, the strong sea surface wind stress generates intense upwelling and vertical mixing in the upper ocean [179]. These powerful mountain gap winds are called Tehuantepecers or Tehuanos, and have been the focus of several previous studies [182, 183, 180] detailing the general setting, drivers and dynamics [185] of strong wind situations in the isthmus. The largest number of these events tend to occur in December, with a mean duration of 48h [182]. Little knowledge exists, however, on the fine-scale structure of the Tehuantepecer flow, which is prone to result in downslope windstorms (DSWS hereafter) potentially producing severe turbulent phenomena such as rotors and hydraulic jumps [155, 156] on the Pacific side of the isthmus, to the lee of local orography, particularly during the cold season. There is evidence from observations and earlier numerical studies [185] that the low elevation topography of Chivela pass can excite mountain waves, and also that these are not only restricted to the pass itself but extend into the much higher mountain crests to the west and especially to the east, as the cold air pool is often thick enough to surpass them. The ability to understand and forecast these events is very relevant, since the Isthmus has been an important development site for wind farms since the 2000’s [149]. Currently this region allocates 76.8 % of the wind power capacity installed in Mexico, with approximately 2360 MW [148], which is expected to double to 5076 MW by 2020 (https://www.amdee.org/mapas-eolicos). In addition, several accidents related to the strong winds are reported by the Oaxacas Civil Protection Commission [140, 141, 142, 143] every year during some Tehuano occurrences. Chapter 5. Extreme events study Figure 5.3: 2013-12-24 03:00 UTC (a) 850 hPa temperature, sea level pressure and wind arrows in the parent grid d01, (b) wind speed (values>10 m/s) and wind arrows at σ=3 (about 70m above ground) in d02, and (c) topography (contours) and potential temperature (shades) and wind arrows at σ=3 in d03. (d) Temperature sounding in the northeast Chivela Pass (NP point in (c)). Upstream-Downstream structure Figure 5.3 in the previous section shows the general structure of the Tehuano wind event, produced by the cold air intrusion from the north and orographic forcings at different scales. In this section, we focus on the fine scale structure of the flow acceleration across the Isthmus, and more precisely, on that occurring on the westernmost section of the Sierra Madre de Chiapas mountains bordering Chivela Pass, apart from the well-known strong gap wind jet. This is the area covered by the d04 domain of 1.3 km resolution (highlighted in green in Figure 5.3c), encompassing with 137 km from north to south the wind flow path before and after crossing the mountains. Figure 5.5 depicts from d04, latitudinal cross-sections (red lines in Figure 5.1b) facing east (south to the left, north to the right) of different variables at the longitude of the two validation points MET1 (Figure 5.5a-c 5.5g-i) and MET2 (Figure 5.5d-f 5.5j-l). The tight isentropes on the windward side of the mountains (to the right in Figure 5.5a 96 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Figure 5.4: Observational wind speed time series from 2013-12-21 to 2013-12-31 in MET1 and MET2. Figure 5.5: D04 vertical cross sections on 2013-12-23 15:30 UTC at MET1 of (a) potential temperature (contours), wind speed (shades), and wind arrows, (b) vertical wind component W and (c) V wind component isolines (positive in blue and negative in red) and turbulent kinetic energy (TKE, shades). (d-f) Same representations for cross sections at MET2. (g-l) Same as a-f on 2013-12-24 03:00 UTC. and 5.5d), indicate a quite similar stable stratification in the whole lower tropospheric column in both cases; stronger in the layers below about 2.5 km and weaker above. The temperature 97 Chapter 5. Extreme events study profile in Figure 5.3d evidences that the latter level corresponds with the depth of the cold air. Winds are northerly in these lower layers to the north of the mountains, with somewhat higher speeds above 20m/s upstream from MET1 than further east, north of MET2.In both cases, winds above the cool pool are much weaker, and back to easterly component at about 4000m. Markowski and Richardson [169] outline seven conditions conducive to DSWS, albeit not all of them absolutely necessary. These conditions are: a mountain with steeper lee slope (1) crossed by strong winds (>15m/s) (2) mostly normal to the barrier (3). A stable layer above the top and less stable above that (4) with cold air advection and large-scale subsidence to maintain the stability (5). Apart from this, reverse wind shear above (6) and no cool pool in the lee (7), is also desirable. These conditions are all perfectly met for both locations analyzed, as discussed previously, and indeed intense downslope windstorms occur in both cases. The stably stratified barrier cross flow displays wave activity from early on (Figure 5.5a), and wave breaking enhances turbulent mixing and yields a region of weak stability and reverse flow immediately downwind from the mountain crests (Figure 5.5a 5.5d). In both cases, isentropes on the windward side sink sharply under these layers of low stability on the lee side, much more pronouncedly for the tallest mountain (Figure 5.5d-e-f). Encompassing the well mixed region to the lee, a split streamline develops (Smith et al, 1985), and below its lower branch there is flow thinning and a significant increase in wind speed. The particular features existent on the lee side differ, however, depending of the height of the topographic obstacle. The strong accelerated flow bounded by intense turbulence extends for many kilometers downwind from the lowest mountain, while ending with a hydraulic jump and rotors at the foot when the barrier is higher. The formation of either of these lee wave events is related to the Froude number upstream [158]. Figure 5.6a plots the Froude number (Equation (1)), calculated using the average of the variables at the first 5 sigma levels, approximately between 16 and 120m above ground, at the crest, H=936 m for MET1 case and H= 1736 m for MET2. For the lowest mountain, Fr ≈2.5 during nighttime and even higher at some other times during the day, indicating supercritical conditions. The Brunt-V¨ ais¨ al¨ a frequency is between 0.020 and 0.025 1/s. The generated mountain waves have relatively short wavelength and a modest hydraulic jump that propagates downstream can be seen at the initial stages of the episode at 15 UTC December 23 (Figure 5.5a-f). 12 hours later, during nighttime, the aforementioned strong jet extending for tens of kilometers downwind is fully formed, with values above 35 m/s at about 750m above ground and strong turbulence at the surface and in the layers above the jet, where stability is much reduced. The temperature profile upwind (at 3 UTC December 24 and location UP1 in Figure 5.5) shows stable conditions for the lowest troposphere, specially marked at crest level, below the inversion at about 2000 m signaling the depth of the cool pool. At the observation location MET1, downwind from the mountain, the lowest layers up to about 750 m are well mixed, the result of the intense surface turbulence. Above the latter height and up to about 1.5 km elevation, a strongly stable layer exists corresponding with the aforementioned packing of the isentropes (and streamlines). This is where the highest wind speeds are found. Stability is much reduced further high, in the region encompassed by the dividing streamline. For the higher mountain north of MET2, Fr ≈1 consistently in the period, indicating a critical flow regime, prone to the formation of HJs [156]. The Brunt-V¨ ais¨ al¨ a frequency is between 0.012 and 0.015 1/s and the generated waves have higher amplitudes than in the MET1 case. Wave overturning and breaking is also much more pronounced (Figure 5.5a-c) and the forming well mixed region to the lee of the crest is deeper. Isentropes and streamlines 98 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ that sink underneath this region are packed in a very shallow layer on the lee slope of the mountain, generating an intense downslope windstorm with speeds above 35 m/s at the surface. These strong winds end abruptly at the foot of the hill, where the flow transitions to subcritical conditions and a marked stationary hydraulic jump forms, with vertical wind speeds of 6 m/s. A rotor extending from the jump to the location of observation station MET2 is also evident in Figure 5.5d-f. The temperature profile upstream is very similar to that of the MET1 case, but downwind from the mountain at location MET2, lacks the strongly stratified layer present in the case of the lowest mountain. Figure 5.6: (a) Froude number in front of each station on the top of the mountain, TOP1 and TOP2. The grey zone represents night time. (b) Temperature vertical profile in MET1 and MET2 and (c) the same for their corresponding upstream points UP1 and UP2. Results from the innermost nested grid d05 with the finest resolution (Figure 5.7) suggest that trapped lee waves develop in the MET1 case within the high stability layer where the strongest winds are found, just below the low stability region aloft that prevents their vertical propagation. This wavelike pattern is a common feature in DSWS periods [77, 157] and fully formed 12 h later (Fig 7c), extends for more than 100 km downstream aligned with the general orientation in the northwest-southeast direction of the Sierra in the region. Waves are absent further east in the MET2 cross section, where the topographic barrier is higher and a stationary HJ forms instead, as discussed above. 99 Chapter 5. Extreme events study Figure 5.7: As in Fig 5.5a and 5.5d, 2013-12-24 03:00 UTC, d05 vertical cross sections at (a) MET1 and (b) MET2 of potential temperature (contours, K), wind speed (shades, m/s), and wind arrows. (c) Vertical wind speed (m/s) in d05 at sigma level 17 (about 1400 m above ground) when the trapped lee wave pattern in the region is fully formed at 15:30 UTC 2013-12-24. Validation Finally, we contrast our simulation results with the very few data available for validation at meteorological stations MET1 and MET2. The two plots in Figure 5.8 compare the simulated wind speed (in d04 and d05) with observations from both stations. Wind speed results from d04 and d05 at location MET1 are quite similar, and compare well with observations (Figure 5.8a), slightly better so those from d05, with a mean absolute error (MAE) of 1.55 m/s (Table 5.2). The similarity in the low mean error between both simulations and their high correlation throughout the period are due to the nature of the event in that area, an intense and mostly steady jet that the d04 domain resolution (1.3 km) is already capable of resolving accurately. However, results in MET2, which registers the HJ situations, present more differences between d04 and d05 (Figure 5.8b) and there is a significant improvement in d05 with respect to its parent domain d04. Wind speeds in d04 are overestimated (Mean error ME = 2.76 m/s), and present a daily cycle that is absent or very subtle in the observations. The complexity and fast variability of HJs formation in this area are better resolved in the higher resolution grid, which perhaps reproduces more accurately the stagnant flow and rotor formations downstream from the HJ. With regard to wind direction, errors are small for MET1 and significantly higher for MET2, due to the same reasons. As for wind direction, results from the finer grid d05 are also better than in d04. Temperature errors are equal or below 1 K in both locations and domains, hence the surface thermal evolution is well captured. Lee waves can promote orographic cloud formation at different scales, depending on the amplitude of the wave and the elevation [176, 177]. Model results in d05 suggest that lenticular clouds form at the crests of the trapped lee waves depicted in Figure 5.7. Figure 5.9a shows 100 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Figure 5.8: (a) Wind speed comparison among observations at MET1 (green), and model output from d04 (red) and, d05 (blue). (b) Same for station location MET2. WS MAE (m/s) WS ME (m/s) WD MAE (◦) T MAE (K) T ME (K) MET1 D04 1.82 1.26 16.15 0.85 -0.80 MET1 D05 1.55 0.80 13.63 0.77 -0.71 MET2 D04 2.76 2.35 27.87 0.71 0.01 MET2 D05 1.31 0.18 24.07 1.02 0.61 Table 5.3: Wind speed (WS), wind direction (WD), and temperature (T) mean errors (ME) and mean absolute errors (MAE) at MET1 and MET2 locations during the simulated period and for the two higher resolution domains (d04 and d05). a 3D representation of the modeled cloud water mixing ratio at 2013-12-24 15:30 UTC in d05. Cross sections of wind speed at the surface observation locations MET1 and MET2 as in Figure 5.5 and Figure 5.7 are also included for reference. A 2D view of the same cloud mixing ratio variable and wind arrows at sigma level 17 (about 1.4 km above ground), revealing existing clouds, are depicted overlaying a satellite image of the area. An actual satellite image from Geostationary Operational Environmental Satellite - R Series (GOES-R) (http: //www.goes-r.gov/education/docs/fs_imagery.pdf) around the same time is shown for comparison, indicating that remarkably similar mountain wave cloud formations were indeed observed in the area. The actual existence of these lenticular clouds with the same location and pattern as in the simulation further validates the model results. 101 Chapter 5. Extreme events study Figure 5.9: (a) Cloud water mixing ratio 3D representation in d05 at 2013-12-24 15:30 UTC. The two cross sections show the N-S wind profile at the longitudes of the meteorological stations MET1 and MET2. (b) Satellite image of the terrain in d05 and cloud water mixing ratio (white shades) and wind arrows at about 1.4 km above ground. (c) GOES-R satellite image on 2013-12-24 20:30 UTC, revealing very similar lenticular cloud formations in the same locations. Summary and conclusions In the present work, we studied lee wave phenomena occurring during Tehuano events on the Pacific side of the Isthmus of Tehuantepec using WRF high-resolution simulations. Orographic forcings at different scales result in the well-known gap wind jet off Chivela pass, but also in downslope windstorms and hydraulic jumps in the neighboring mountains. We analyzed these phenomena in an episode in December 2013 having the typical genesis of Tehuantepecer wind events. An Arctic air mass in North America pushed as far south as the bay 102 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ of Campeche due to cold air damming east of the Rockies continuing to the east of the Sierra Madre Oriental range in Mexico. The displacement of the associated high-pressure system on the wake of the cold front created large pressure differences across the Isthmus of Tehuantepec, ultimately producing the strong mountain gap winds through the low elevation of Chivela pass. The model simulates these intense winds, blowing with speeds at the surface of more than 25 m/s that extend for many kilometers from the mountain gap, fanning out well into the Gulf of Tehuantepec, as it is commonly observed in Tehuano events [185]. The depth of the cold surge on the coastal plains of the Gulf of Mexico side of the Isthmus is about 2500m, therefore thick enough to surmount the lower elevations to the west, and especially to the east of Chivela pass. The flow over these mountains results in intense downslope wind storms to their lee, with the generation of intense turbulence, hydraulic jumps and rotors, depending on the particular height of the topography. We focus on two locations of different barrier elevation from where there are surface observations downwind: one of 963m closer to Chivela pass and another further east, with height increasing to 1736m. The thermodynamic characteristics of the air mass are rather uniform upwind both mountains, with strong stability within the cool pool and weaker above, and intense northerly winds that back and weaken to a more easterly component aloft. Mountain waves are generated in both cases, with smaller amplitudes for the lower mountain, where the Brunt-V¨ ais¨ al¨ a frequency at crest height is between 0.020 and 0.025 s-1, than for the higher mountain, where the Brunt-V¨ ais¨ al¨ a frequency is about half. Wave breaking produces mixing and generates a region of low stability to the lee of the mountains, which is deeper where the waves have higher amplitude. The Froude number is around 2.5 at crest height in the lower barrier and the flow presents a supercritical behavior. The region of low stability to the lee of the mountain lies above about 1500m and leads to a packing of the isentropes and streamlines underneath, resulting in strong stability and flow acceleration. An intense jet develops with wind speeds of 35 m/s at about 750m above ground extending for tens of kilometers downwind from the mountains. Wind speeds are reduced closer to the surface due to intense turbulence. Trapped lee waves form at about 1500m, just below the well mixed layer aloft that prevents their vertical propagation. The Froude number decreases to about 1 further east as elevation rises and the flow presents a critical regime. Isentropes on the windward side of the mountain sink much more pronouncedly under the wider mixed layer generated by wave breaking to the lee, and are tightly packed in a shallow layer above the surface. This generates an intense wind storm on the lee slope of the mountain, with surface wind speeds up to 35 m/s. The accelerated flow down the mountain ends abruptly at its foot, where the flow turns to subcritical state and a marked stationary hydraulic jump forms, with vertical velocities of 6 m/s. A rotor circulation develops further downstream from the jump. Only limited observations are available to validate our model results. Errors in surface wind speeds, directions and temperature are small at the only two stations available on the Pacific coastal plain downwind from the mountains. In addition, lenticular clouds similar in location and pattern to those produced by the model, are apparent in satellite imagery of the day of the event, provides valuable indication that the mountain lee wave phenomena simulated indeed corresponds to a real scenario. Our model results suggest that more extreme wind events develop in the area during Tehuano evends beyond the gap wind jet. This include downslope wind storms and hydraulic jumps, which are intense and highly turbulent flows that can have a substantial impact on the existent wind farm industry in the region. 103 Chapter 6 Microscale phenomena analysis Introduction The study of microscale wind flow phenomena has always been a point of interest within the wind industry. There are several well know wind events related with it that turbine designers and operations managers have to take into account, such as high operating gusts, extreme direction change, high coherent gust with direction change or extreme wind shear. All these turbulent wind events have been defined and categorized for a long time within the sector [118]. However, their capacity to, for example, cause torque reversals that can damage a turbine, or provoke decalibrations on hub sensors, have only recently been recognized and measured [30] A good part of these recent developments has been possible thanks to numerical modeling tools, which are continuously improving their techniques within turbulent phenomena characterization and forecasting. Specifically, this is one of the main strengths of CFD (Fluid Dynamic Simulations) models, as introduced in Chapter 2. These tools are useful and currently well established in the wind energy field to study future installation areas and design new wind farms [119, 120, 121]. However, as stated in previous chapters, they always work over stationary conditions, even with the best initial conditions, these simulations do not allow us to capture all the complexity of the phenomena at different scales that affect each region of study and which have a significant effect on turbulent fluxes that can flow in the area. Another modeling tool which is increasingly being used is WRF Large Eddy Simulations (LES). Based on a very high-resolution WRF simulation and a deactivation of the planetary boundary layer scheme, this WRF configuration can be utilized to simulate turbulent wind flows for wind energy applications [20]. WRF LES permits run simulations in two different ways, as an ideal case or within a nesting configuration fed by global model and mesoscale conditions in the boundaries (real-world). Many current WRF LES studies published are based on these ideal cases [122, 123]. This methodology assumes periodic boundary conditions and horizontal homogeneity [124] with the relatively uncomplicated surface. WRF LES ideal is extensively used for wind farm applications such as wakes studies, as commented in Chapter 4, or to analyze the performance of wind turbine schemes and boundary forcings [125, 20]. WRF LES real-world (LES-RW hereafter) is also each day the more commonly developed modeling technique. These simulations are fully coupled to the mesoscale parent domains giving realistic atmospheric forcing to the LES simulation. The high resolution of this methodology is capable of resolving an important part of the turbulent wind spectrum phenomena (Chapter 1, Figure 1.4). A representative example of this modeling tool is the Chapter 6. Microscale phenomena analysis Table 6.3: Mean wind speed observed and forecasted for each validation point. Different mean wind speed errors measured in each point along the studied period: Correlation Coefficient (CC), Root Mean Square Error (RMSE), Mean Error (ME) and Mean Absolute Error (MAE). The plot representations and its respective measured errors allow us to comprehend instactly the performance of the simulations. The northernmost station representation displays a significant overestimation in practically all the period studied (Figure 6.3a). This is also reflected in the WS ME, and consequently in the MAE and RSME (Table 6.3). Figure 6.3b and c show much better general results in the other two stations (M2 and M3), with an accurate wind field representation, reaching WS MAE of 1.43 and 1.46 m/s respectively. The ME is also distinctly lower than M1 in both of them (below 0.5 m/s). The significant M1 overestimation reflected in the WS ME and the relatively good CC suggests that at least part of the deviation obtained could be related to a misalignment between observation and model output heights. This could be due to an error in the measurement recording or a deviation in the height of the grid cell and/or its neighbors for this specific point in the model domain. All this leads to a total WS MAE = 1.89 m/s for the three stations, which can be considered a good result taking into account the complexity of the terrain. This MAE is practically the same as the one reached in Coruxeiras wind farm WRF forecasts (Chapter 2, Table 2.4). Leaving aside the mean errors, the three model plots (blue lines) reflect an increase in the variability of each recorded time period. The very high resolution and its respective small temporal scale (timestep) used to perform the simulation are able to reproduce smaller scale eddies (Chapter 1, Figure 1.4 green line) giving more reality to the simulations which is reflected in all the variables recorded. Continuing with the discussion about the capabilities of LES-RW, we review the model performance predicting wind direction. To do that, in the next figure we present the results of the three validation points following the same structure as with WS in Figure 6.3. The plot representations and the WD MAE results (Table 6.4) depict satisfactory results during the ten day period. The wind direction, which is maintained between 350◦and 100◦ across all the period, is well estimated by LES-RW, obtaining results close to the observation results across the whole series. This small deviation tendency is visible in the three plots and the MAE results with a total MAE of 17.49◦. Despite the prevailing N – NE wind during the whole the experiment (reason of the vertical lines in the center of the plots), there are no significant differences between stations. The southern station, M3, despite being more affected by the topography effects due to the direction of the incoming wind, does not register significant higher deviations (MAE=19.51◦) than the other two stations further north (15.36◦and 17.62◦). Two of the main factors of a proper RW-LES performance are coherent boundary turbulent forcing and a good turbulent phenomena reproduction within the innermost domain.These seem to be well resolved by the domains configurations and the parameterizations used. On the one hand, as it is reflected in the accurate WD M1 and M2 results, the d04 LES domain providesappropriate boundary forcings to the higher resolution domain. On the other hand, d05 maintains the turbulent behavior of the wind fields throughout the region, registering promising WD and WS results in M3, which is mainly affected by the local performance of its own domain. 112 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Figure 6.4: (a) 2017/10/09-2017/10/19 (UTC) Wind direction at M1 (80 meters height) for WRF-LES (blue) and observations (red). (b) Same as (a) at M2.(c) Same as (a) at M3. Table 6.4: Mean wind direction Mean Absolute Error (MAE) in each point for the period studied. As can be seen in the WS and WD results for the different stations, RW-LES with the several LES domains forcing technique is able to obtain realistic results across the majority of the area of the domain, even atpoint M3, which is relatively close to its border. This is an advantage with respect to boundary cell perturbation methods which need an extended relaxation zone in all the nearby areas to the horizontal boundary layers to obtain reliable turbulent representations [81]. This makes less area of the domain profitable, which is a problem for situations as in this study, where there are several locations of interest which need to be resolved within the same simulation domain. The only way for these methods to face the relaxation zone issue is increasing the domain area and, therefore, increasing computational costs. After analyzing one of the main topics of interest in this study, which is the wind representation by LES-RW, we examine the capacities of the tool for resolving the temperature fields near to surface. Specifically, continuing with Figure 6.4 and 6.5 structure, the next figure displays, a comparison between observed and forecasted T at 10-meter height in the three validation points and summary table with different statistical measures for the whole period. 113 Chapter 6. Microscale phenomena analysis Figure 6.5: (a) 2017/10/09-2017/10/19 (UTC) Temperature at M1 (10 meters height) for WRF-LES (blue) and observations (red). (b) Same as (a) at M2. (c) Same as (a) at M3. Table 6.5: Mean temperature observed and forecasted for each validation point. Different mean wind speed errors measured at each point for the studied period: Correlation Coefficient (CC), Root Mean Square Error (RMSE), Mean Error (ME) and Mean Absolute Error (MAE). Both plot series comparison for M1, M2, and M3 (Figure 6.5a, b, and c respectively), and mean errors in Table 6.5, show a proper general representation of T in the stations by LES simulations. The mean statistical values are relatively close between the three points. M1 recorded the best results, with the smallest MAE (1.15 ◦C) and RMSE (1.43 ◦C), and a practically inesistent ME (-0.03 ◦C). The southern station, M3, obtains the least accurate results, with higher MAE and RMSE, 1.59 ◦C and 1.95 ◦C respectively and a general T underestimation reflected in the series plot (Figure 6.5b), and its ME (-1.16 ◦C). In any case, these are not important variations between stations; all the T MAE are below 1.6 ◦C reaching a total mean value of 1.35 ◦C. The correlation coefficient is very high in the three locations, something which is also noticeable in the plot series representation. CC T results cannot be compared with the same statistical measure for WS or WD, the temporal scale of variability of T is different to WS. In any case, the total mean CC = 0.93 can be considered as an excellent result for a near-surface temperature. 114 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Temperature validation can be summarized by small MAEs, practically no ME in M1 and M3 and a very high CC in all cases. The accurate representation of T at the 10-meter height reflects that the simulations are well resolving that surface layer-terrain fluxes well and this has a direct effect on the correct reproduction of the microscale wind behavior that we are looking for in this study. Apart from an appropriate physical schemes configuration and a high horizontal resolution, the implementation of accurate land use data, as in this experiment (FROM CLC 30m), seems to have a major positive effect on the reproduction of T fields, especially at this low height. When we are running simulations at this high resolution, and we want variable series for one specific point, it is very important to check the aspect of the LU database input at the locations analyzed and their near surroundings. A mistake in the land use in our validation point can provoke a notable bias in the output during all the period simulated. After seeing the evaluation of the forecasting tool with WS, WD and T validations at point we check we can affirm that the LES-RW yields a good representation of an important part of the microscale phenomena in the area studied. The main statistical indicators used for each variable (Table 6.3, 6.4 and 6.5) score low errors at all the points. The more significant deviation from observations is registered in WS results for M1. Taking into account the excellent WS results in M2 and M3, and the low WD errors at all the points, the source of overestimation seems to be related with an inaccuracy in the orography data. The very good T ME registered in M1 partially discards the possibility of a mismatch between model output and observation height. However, there could still be incorrect height data at the north/northeast of the point which makes the wind blow over the point with a higher speed than observations have registered. Turbulence Intensity In this subsection, we address the turbulence intensity measurement from RW-LES. In particular, we obtain the forecasted TI following the calculation described in 6.2.2 in the validation points at 80 meters height and we analyze the results using the observational data from the meteorological stations at this same height. First, and continuing with the same format as in the previous subsection, Figure 6.6 displays a plot representation of TI forecasted and observed in the three locations and a summary table with different mean statistical values for each point and the total mean of them. In a first quantitative assessment, TI RW-LES results in Figure 6.6 reflect one common thing in the three plots, real behavior. The TI reproduced by the model presents a high variability throughout all the period which concords with the real data from observations. The TI obtained in WRF mesoscale simulations, using parameterized TKE values, is not able to reproduce these variations because the model configuration does not have enough spatial and temporal resolution to capture these motions. Aside from this, the simulations are also capable of reproducing the daily cycle that this variable uses to present higher values during the daytime (thermal influence) and lower ones at night. 115 Chapter 6. Microscale phenomena analysis Figure 6.6: (a) 2017/10/09-2017/10/19 UTC turbulent intensity at M1 (80 meters height) for WRF-LES (blue) and observations (red). (b) Same as (a) at M2.(c) Same as (a) at M3. Table 6.6: Mean turbulent intensity observed and forecasted for each validation point. Different mean TI errors measured in each point for the studied period: Correlation Coefficient (CC), Root Mean Square Error (RMSE), Mean Error (ME) and Mean Absolute Error (MAE). Focusing on the mean results (Table 6.6), M1 registers the best MAE (0.05) and M3 the worse (0.08), obtaining a total TI MAE of 0.06 taking into account the three locations. ME values register an underestimation in M1 and M3. This statistical measure is especially useful for analyzing this variable because a negative ME result can mean that the simulation is not resolving with enough energy the turbulent phenomena which meteorological stations are capturing. This is something common in this kind of LES-RW spectral energy analysis, previous studies at this resolution reveal a similar tendency [126, 125]. TI from Equation 6.1 does not take into account any eddy smaller than the resolution of d05 (98 m), leaving the last part of the wind energy spectrum (Figure 1.4, Chapter 1) to the parameterized TIsgs which is not capable of capturing the whole the range of microscale phenomena from the smaller eddies. However, taking these limitations into account, a total ME deviation of -0.02 can be considered as ans accurate result. Overall, ME, MAE and RMSE reflect reasonably good results for the TI forecasted by LES-RW. As described in 6.2.2, TI measured from LES-RW is the sum of the TI explicitly resolved by the model (TILES) and TI diagnosed by the subgrid-scale parameterization (T ISGS). The 116 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ first term is the result of the TKE produced by the larger eddies and the second is the TKE which represents the smaller motions. In order to analyze in more detail the behavior of these variables, the next figure displays the plot series for TITot,T ISGS and TILES in MET3. Figure 6.7: (a) 2017/10/09-2017/10/19 (UTC) TITot (red), TILES (blue) and TISGS (yellow) at M1 (80 meters height) for WRF-LES (blue) and observations (red) Table 6.7: Mean values for TIobs,TITot,TILES and TISGS for the period studied in M1, M2, and M3. The decomposition of TI in Figure 6.7 allows us to see the role of each part of TI throughout the 10 day period. T ILES is clearly more present during the daytime, representing most of the TITot registered during that part of the day. These periods are also the moments of the study with higher observed TI due to the influence of thermal interaction with the surface. TISGS, however, follows a different pattern. The subgrid-scale part of TI is present during the whole period. It represents the smaller eddies which affect the area in practically every moment of the day, without significant differences between day and night. The mean values in Table 6.7 highlight the importance of both sources of turbulence. The three stations register similar values, leaving a total mean amount of 66 % of TITot coming from the smaller eddies (<98m) estimated by TIsgs. It is evident that the primary objective of the TI forecasting is to achieve realistic results with small mean error, but, apart from this, being also able to characterize two different sources of turbulent kinetic energy with this method gives added values to this kind of tool. After the quantitative analysis of the capacities of RW-LES solving the turbulence which affects the validation points, we finish with this chapter, showing a qualitative view of the behavior of RW-LES flows along d05 domain. In particular, Figure 6.8 displays a vertical and a horizontal cut for each of the wind component (U, V and W) in a low wind instant the day after the studied period. 117 Chapter 6. Microscale phenomena analysis Figure 6.8: Instantaneous wind components at 2017-10-20 09:00 UTC for d05. (a) U plot along a longitudinal cut through M1. (b) Same as (a) for V component, (c) Same as (a) for W component. (d) U field at 80 m height. (e) V field at 80 m height. (f) W field at 80 m height. The general perspective of all the composition of pictures shows a clear turbulent regime in all the images. Vertical cuts depict turbulent structures in most of the longitudes of the cut drawing the end of the mixing layer around 500 meters height at that moment. U and V vertical plots present several convective rolls at right and left of the center elevation. These updraft and downdrafts motions which are related to thermal fluxes are also clearly visible in the W vertical cut. Focusing on the horizontal cut, we can distinguish turbulent structures throughout most of the domain in the three representations. The SW prevailing wind interacts with the orography of the region producing microscale wind fields of different sizes. We can see how, for example, the NE area which surrounds M1 is affected by a NE wind flow (blue area in Figure 6.8d and e) helped by the small mountain where M1 is located. The horizontal W representation in Figure 6.8f has a different aspect as U and V plots. It has finer structures which represent positive vertical winds resulting from the combination of thermal terrain-lower PBL interactions and wind field accelerations due to the complexity of the topography. 118 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ Conclusions The objective of this study was to develop a tool to analyze and characterize microscale phenomena in wind farm areas by using WRF-LES simulations. The model has been configured to simulate this type of simulations at very high resolution, a computational time optimization, and different numerical stability techniques were necessary to achieve that objective. The results have been tested using observational data from three different locations in the area. The following is a summary of the main conclusions to be drawn from experience. With LES-RW simulations we have sought to obtain accurate wind module, wind direction and turbulent intensity results in the wind farm area studied. The strong points of these simulations are related to their high horizontal and vertical resolution. Reaching this very high resolution, deactivating the PBL parameterization of the simulation and by selecting the appropriate settings, we are able to dynamically reproduce a large part of the turbulent phenomenology of the studied zones. The vertical resolution also acquires particular importance as it affects wind and temperature calculations in the lowest part of the PBL. A zonal smoothing tool was developed to avoid problems of instability due to the complex terrain. This method looks for the points of numerical instability in the domain and modifies the orography data around them. This gives stability to the simulation, modifying the domain as little as possible. The results have been very satisfactory; the zonal smoothing has been used successfully in several portions of the domain, making the simulations more stable and reducing computational time. Wind speed forecasting evaluation leaves a total MAE of 1.89 m/s. Excellent results in M2 and M3, and a prominent overestimation in M1 which seems to be related to some incorrect height data at the north/northeast of the point. RW-LES has accurately reproduced wind direction motion during the 10-day experiment, as the total MAE indicates (17.49◦). The small errors yielded by the model in the temperature validation at 10 m height (MAE = 1.35 K and ME= -0.45 K) reflect a good reproduction of the thermal fluxes near surface which have a major impact on the correct reproduction of the microscale wind behavior. TI measured from LES-RW was obtained from the sum of the TI explicitly resolved by the model and TKE diagnosed by the subgrid-scale parameterization (TISGS). TI results in the three validation points present a realistic performance of microscale phenomena by RW-LES, which achieves a total MAE of 0.06 and just a slight mean underestimation (ME=-0.02). Moreover, this method to estimate TI, allows us to characterize two different sources of turbulence adding extra value to the study. The qualitative analysis of the RW-LES displayed in horizontal and vertical wind field plots show realistic microscale flow all along the domain. We can conclude that the experience with LES-RW simulations has been satisfactory. It has also become clear that the use of WRF-LES as an operational tool requires a high technical level and significant computing resources. There are many lines of improvement on which we can continue working to make this tool more competitive regarding time and stability. In any case, it can be affirmed WRF-LES is a current and future tool within the wind industry. Throughout this experience, we have seen that WRF-LES is a field in continuous growth, with constant new developments and improvements in parameterizations in recent years. To this, must be added the exponential increase of the computing capacity, which allows ever quicker simulation and at a lower price. All this context promotes new ways of using this type of modeling tools, new products with promising uses in the sector. 119 Chapter 7 Main Conclusions The objective of this thesis has been the achievement of an improvement in the quantification and understanding of some of the main interactions between the atmospheric planetary boundary layer and the terrain, focusing on the behavior of wind flows at different scales. We intend to improve the tools, within numerical modeling, for the analysis of such mechanisms, contributing in this manner to the optimization of wind resource use. Throughout the thesis, we have addressed different topics within this field, starting from wind resource forecasting and nowcasting, continuing with the wind turbines wakes and extreme events studies and finishing with the microscale phenomena analysis in a wind farm area. The main conclusions reached from this thesis are summarized in the next points: •Wind resource forecasting is one of the main duties in daily wind farm operation. The WRF atmospheric model at high resolution has been tested for this purpose in a real wind farm over complex terrain. Results show that WRF yields good wind power operational predictions for this kind of wind farm, due to a good representation of the planetary boundary layer behavior of the region and the excellent performance of the Fitch scheme (wind turbine scheme) under these conditions. The study supports the concept of utilizing WRF high-resolution simulations as part of a wind speed and energy forecast tool within the wind energy industry. •The combination of the WRF model with a post-process technique which combines a non-linear Kalman filter and a Bayesian model has allowed us to perform very short-term wind predictions (nowcasting). The method obtains large improvements in wind speed and direction nowcasting with respect to the original forecast in different meteorological situations throughout an entire year. The accuracy and reliability of the results demonstrate the potential utility that this tool can have for a variety of applications in wind farm operations and energy markets. •The wind turbine wake effects present a reduction in the power output and the increased level of turbulent loads over the downstream turbines. The two Annual Wake analyses performed in this thesis show considerable wind resource losses among wind turbines in several directions and extensive wake effects all around the wind farm. This new technique can help to analyze beforehand the total impact of a future neighboring wind farm construction on an existent wind farm’s annual production and provide valuable information to optimize new farm designs. Chapter 8. Bibliography [66] J. A. Slater, B. Heady, G. Kroenung, W. Curtis, J. Haase, D. Hoegemann, C. Shockley, and K. Tracy, “Evaluation of the New ASTER Global Digital Elevation Model,” vol. 77, no. 4, pp. 335–349, 2009. [67] H. H. Shin, S.-Y. Hong, and J. Dudhia, “Impacts of the Lowest Model Level Height on the Performance of Planetary Boundary Layer Parameterizations,” Monthly Weather Review, vol. 140, no. 2, pp. 664–682, 2012. [68] E. J. Mlawer, S. J. Taubman, P. D. Brown, M. J. Iacono, and S. A. Clough, “Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the longwave,” Journal of Geophysical Research: Atmospheres, vol. 102, no. D14, pp. 16 663–16 682, 1997. [69] J. Dudhia, “Numerical Study of Convection Observed during the Winter Monsoon Experiment Using a Mesoscale Two-Dimensional Model,” pp. 3077–3107, 1989. [70] S. Hong and J. Lim, “The WRF single-moment 6-class microphysics scheme (WSM6),” pp. 129–151, 2006. [71] M. Tewari, F. Chen, W. Wang, J. Dudhia, M. A. Lemone, K. Mitchell, R. H. Cuenca, C. Springs, A. Force, and W. Agency, “Implementation and Verification of The Unified NOAH Land Surface Model In The WRF Model,” 20th Conference on Weather Analysis and Forecasting/16th Conference on Numerical Weather Prediction, 2004. [72] M. Nakanishi and H. Niino, “An improved Mellor-Yamada Level-3 model: Its numerical stability and application to a regional prediction of advection fog,” Boundary-Layer Meteorology, vol. 119, no. 2, pp. 397–407, 2006. [73] J. S. Kain, “The Kain–Fritsch Convective Parameterization: An Update,” Journal of Applied Meteorology, vol. 43, no. 1, pp. 170–181, 2004. [74] A. A. Wyszogrodzki, Y. Liu, N. Jacobs, P. Childs, Y. Zhang, G. Roux, and T. T. Warner, “Analysis of the surface temperature and wind forecast errors of the NCAR-AirDat operational CONUS 4-km WRF forecasting system,” Meteorology and Atmospheric Physics, vol. 122, no. 3-4, pp. 125–143, 2013. [75] D. Pozo, J. C. Mar´ ın, L. Illanes, M. Cur´ e, and D. Rabanus, “Validation of WRF forecasts for the Chajnantor region,” Monthly Notices of the Royal Astronomical Society, vol. 459, no. 1, pp. 419–426, 2016. [76] J. Zhao, Y. Guo, X. Xiao, J. Wang, D. Chi, and Z. Guo, “Multi-step wind speed and power forecasts based on a WRF simulation and an optimized association method,” Applied Energy, vol. 197, pp. 183–202, 2017. [77] B. Pokharel, B. Geerts, X. Chu, and P. Bergmaier, “Profiling radar observations and numerical simulations of a downslopewind storm and rotor on the lee of the Medicine Bow mountains in Wyoming,” Atmosphere, vol. 8, no. 2, 2017. [78] G. M. Wang, D. F. Liu, Y. P. Xu, T. Meng, and F. Zhu, “PET/CT imaging in diagnosing lymph node metastasis of esophageal carcinoma and its comparison with pathological 128 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ findings,” European Review for Medical and Pharmacological Sciences, vol. 20, no. 8, pp. 1495–1500, 2016. [79] M. O. Mughal, M. Lynch, F. Yu, and J. Sutton, “Forecasting and verification of winds in an East African complex terrain using coupled mesoscale - And micro-scale models,” Journal of Wind Engineering and Industrial Aerodynamics, vol. 176, no. December 2017, pp. 13–20, 2018. [80] C. L. Archer, H. P. Sim˜ ao, W. Kempton, W. B. Powell, and M. J. Dvorak, “The challenge of integrating offshore wind power in the U.S. electric grid. Part I: Wind forecast error,” Renewable Energy, vol. 103, pp. 346–360, 2017. [81] D. Mu˜ noz-Esparza, J. K. Lundquist, J. A. Sauer, B. Kosovi´ c, and R. R. Linn, “Coupled mesoscale-LES modeling of a diurnal cycle during the CWEX-13 field campaign: From weather to boundary-layer eddies,” Journal of Advances in Modeling Earth Systems, vol. 9, no. 3, pp. 1572–1594, 2017. [82] D. Mu˜ noz-Esparza, B. Kosovi´ c, J. Mirocha, and J. van Beeck, “Bridging the Transition from Mesoscale to Microscale Turbulence in Numerical Weather Prediction Models,” Boundary-Layer Meteorology, vol. 153, no. 3, pp. 409–440, 2014. [83] M. Huang, Y. Wang, W. Lou, and S. Cao, “Multi-scale simulation of time-varying wind fields for Hangzhou Jiubao Bridge during Typhoon Chan-hom,” Journal of Wind Engineering and Industrial Aerodynamics, vol. 179, no. April, pp. 419–437, 2018. [84] G. N. Kariniotakis and P. Pinson, “Evaluation of the MORE-CARE wind power prediction platform. Performance of the fuzzy logic based models,” no. June, 2003. [85] G. Kariniotakis, I. Mart´ ı, D. Casas, P. Pinson, T. S. Nielsen, H. Madsen, G. Giebel, J. Usaola, and I. Sanchez, “What performance can be expected by short-term wind power prediction models depending on site characteristics ?” EWC 2004 Conference, pp. 22–25, 2004. [86] E. Vanem, “Long-term time-dependent stochastic modelling of extreme waves,” Stochastic Environmental Research and Risk Assessment, vol. 25, no. 2, pp. 185–209, 2011. [87] G. Giebel, On the benefits of distributed generation of wind energy in Europe, PhD thesis from the Carl von Ossietzky Universitat Oldenburg, E. . Fortschritt-Berichte VDI. Reihe 6, Ed., Dusseldorf, Germany, 2001, no. August. [88] G. Resconi, “Geometry of risk analysis (morphogenetic system),” Stochastic Environmental Research and Risk Assessment, vol. 23, no. 4 SPEC. ISS., pp. 425–432, 2009. [89] P. S, G. G, and G. Kallos, “Solar and photovoltaic forecasting through post-processing of the global environmental multiscale numerical weather prediction model,” Progress in Photovoltaics: Research and Applications, 2011. 129 Chapter 8. Bibliography [90] G. Galanis, P. Louka, P. Katsafados, I. Pytharoulis, and G. Kallos, “Applications of Kalman filters based on non-linear functions to numerical weather predictions,” Annales Geophysicae, vol. 24, no. 10, pp. 2451–2460, 2006. [91] G. Galanis, G. Emmanouil, P. C. Chu, and G. Kallos, “A new methodology for the extension of the impact of data assimilation on ocean wave prediction,” Ocean Dynamics, vol. 59, no. 3, pp. 523–535, 2009. [92] P. Crochet, “Adaptive Kalman filtering of 2-metre temperature and 10-metre wind-speed forecasts in Iceland,” Meteorological Applications, vol. 11, no. 2, pp. 173–187, 2004. [93] G. Galanis and M. Anadranistakis, “A one-dimensional Kalman filter for the correction of near surface temperature forecasts,” Meteorological Applications, vol. 9, no. 4, pp. 437–441, 2002. [94] E. Kalnay, Atmospheric modeling, data assimilation, and predictability, 2003. [95] R. E. Kalman and R. S. Bucy, “New Results in Linear Filtering and Prediction Theory,” Journal of Basic Engineering, vol. 83, no. 1, p. 95, 1961. [96] R. E. Kalman, “A New Approach to Linear Filtering and Prediction Problems,” Journal of Basic Engineering, vol. 82, no. 1, p. 35, 1960. [97] C. Stathopoulos, A. Kaperoni, G. Galanis, and G. Kallos, “Wind power prediction based on numerical and statistical models,” Journal of Wind Engineering and Industrial Aerodynamics, vol. 112, pp. 25–38, 2013. [98] S. Hua, S. Wang, S. Jin, S. Feng, and B. Wang, “Wind speed optimisation method of numerical prediction for wind farm based on Kalman filter method,” The Journal of Engineering, vol. 2017, no. 13, pp. 1146–1149, 2017. [99] P. Louka, G. Galanis, N. Siebert, G. Kariniotakis, P. Katsafados, I. Pytharoulis, and G. Kallos, “Improvements in wind speed forecasts for wind power prediction purposes using Kalman filtering,” Journal of Wind Engineering and Industrial Aerodynamics, vol. 96, no. 12, pp. 2348–2362, 2008. [100] A. Zaher, S. D. McArthur, D. G. Infield, and Y. Patel, “Online wind turbine fault detection through automated SCADA data analysis,” Wind Energy, vol. 12, no. 6, pp. 574–593, 2009. [101] W. Skamarock, J. Klemp, J. Dudhi, D. Gill, D. Barker, M. Duda, X.-Y. Huang, W. Wang, and J. Powers, “A Description of the Advanced Research WRF Version 3,” NCAR, Tech. Rep. June, 2008. [102] A. C. Beljaars, “The parametrization of surface fluxes in largescale models under free convection,” Quarterly Journal of the Royal Meteorological Society, vol. 121, no. 522, pp. 255–270, 1995. [103] P. Patlakas, E. Drakaki, G. Galanis, C. Spyrou, and G. Kallos, “Wind gust estimation by combining a numerical weather prediction model and statistical post-processing,” in Energy Procedia, vol. 125. Elsevier B.V., 2017, pp. 190–198. 130 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ [104] T. G. Box G., Bayesian Inference in Statistical Analysis, J. Wiley and Sons, Eds., John Wiley and Sons. INC New York, USA, 1992. [105] J. Bernardo and A. Smith, Bayesian Theory, John Wiley and Sons. INC New York, USA, 2000. [106] I. Kuikka, “Wind Nowcasting: Optimizing runway in use,” Tech. Rep., 2009. [107] H. Utsumi, T. Misaka, and W. Moteki, “Prediction Model of Internal Relative Humidity During Self-desiccation in Hardened Cement Pastes,” Concrete Research and Technology, vol. 26, no. 3, pp. 11–19, 2015. [108] M. T. Van Dijk, J. W. Van Wingerden, T. Ashuri, Y. Li, and M. A. Rotea, “Yaw-Misalignment and its Impact on Wind Turbine Loads and Wind Farm Power Output,” in Journal of Physics: Conference Series, vol. 753, no. 6, 2016. [109] M. Churchfield and P. Fleming, “Wind Turbine Wake-Redirection Control at the Fishermen’s Atlantic City Windfarm,” Offshore Technology . . . , no. May, 2015. [110] P. Mckay, “Group Dynamics of Commercial Scale Wind Turbines,” 2011. [111] M. P. S. G. B. Sofia Elena Colescab, “Renewable and sustainable energy reviews,” Elsevier, vol. 56, pp. 156–170, 2016. [112] G. M. Wang, D. F. Liu, Y. P. Xu, T. Meng, and F. Zhu, “PET/CT imaging in diagnosing lymph node metastasis of esophageal carcinoma and its comparison with pathological findings,” European Review for Medical and Pharmacological Sciences, vol. 20, no. 8, pp. 1495–1500, 2016. [113] J. R. Marden, S. D. Ruben, and L. Y. Pao, “A model-free approach to wind farm control using game theoretic methods,” IEEE Transactions on Control Systems Technology, vol. 21, no. 4, pp. 1207–1214, 2013. [114] A. Alaimo, A. Esposito, A. Messineo, C. Orlando, and D. Tumino, “3D CFD analysis of a vertical axis wind turbine,” Energies, vol. 8, no. 4, pp. 3013–3033, 2015. [115] R. Krishnamurthy, S. Otarola-Bustos, and L. S. Leo, “WFIP2 project University of Notre Dame Halo Scanning Doppler LiDAR (lidar.z07) Quality Controlled Dataset ,” pp. 1–11, 2017. [116] J. Wilczak, C. Finley, J. Freedman, J. Cline, L. Bianco, J. Olson, I. Djalalova, L. Sheridan, M. Ahlstrom, J. Manobianco, J. Zack, J. R. Carley, S. Benjamin, R. Coulter, L. K. Berg, J. Mirocha, K. Clawson, E. Natenberg, and M. Marquis, “The wind forecast improvement project (WFIP): A public-private partnership addressing wind energy forecast needs,” Bulletin of the American Meteorological Society, vol. 96, no. 10, pp. 1699–1718, 2015. [117] L. Pan, Y. Liu, G. Roux, W. Cheng, Y. Liu, J. Hu, S. Jin, and S. Feng, “Seasonal variation of the surface wind forecast performance of the 3km-grid WRF-RTFDDA forecasting system over China .” Boulder, Colorado, 2018. 131 Chapter 8. Bibliography [118] C. Gong, “IEC International Standard 61400-1,” Order A Journal On The Theory Of Ordered Sets And Its Applications, pp. 1–9, 2008. [119] Y. Li, A. M. Castro, T. Sinokrot, W. Prescott, and P. M. Carrica, “Coupled multi-body dynamics and CFD for wind turbine simulation including explicit wind turbulence,” Renewable Energy, vol. 76, pp. 338–361, 2015. [120] S. C. Goh, S. R. Boopathy, C. Krishnaswami, and J. U. Schl¨ uter, “Tow testing of Savonius wind turbine above a bluff body complemented by CFD simulation,” Renewable Energy, vol. 87, pp. 332–345, 2016. [121] A. Rezaeiha, I. Kalkman, and B. Blocken, “CFD simulation of a vertical axis wind turbine operating at a moderate tip speed ratio: Guidelines for minimum domain size and azimuthal increment,” Renewable Energy, vol. 107, pp. 373–385, 2017. [122] R. J. Beare, M. K. Macvean, A. A. M. Holtslag, J. Cuxart, I. Esau, J. C. Golaz, M. A. Jimenez, M. Khairoutdinov, B. Kosovic, D. Lewellen, T. S. Lund, J. K. Lundquist, A. McCabe, A. F. Moene, Y. Noh, S. Raasch, and P. Sullivan, “An intercomparison of large-eddy simulations of the stable boundary layer,” Boundary-Layer Meteorology, vol. 118, no. 2, pp. 247–272, 2006. [123] F. K. Chow, R. L. Street, M. Xue, and J. H. Ferziger, “Explicit Filtering and Reconstruction Turbulence Modeling for Large-Eddy Simulation of Neutral Boundary Layer Flow,” Journal of the Atmospheric Sciences, vol. 62, no. 7, pp. 2058–2077, 2005. [124] N. Zhang, X. Wang, and Z. Peng, “Large-Eddy Simulation of Mesoscale Circulations Forced by Inhomogeneous Urban Heat Island,” Boundary-Layer Meteorology, vol. 151, no. 1, pp. 179–194, 2014. [125] K. R. Raj, H. Gopalan, and J. W. Naughton, “Effects of spatial and temporal resolution of the turbulent inflow on wind turbine performance estimation,” Wind Energy, vol. 20, no. 8, pp. 1495–1500, 2016. [126] P. Doubrawa, A. Montorn` es, R. J. Barthelmie, S. C. Pryor, and P. Casso, “Analysis of Different Gray Zone Treatments in WRF-LES Real Case Simulations,” Wind Energy Science, no. January, pp. 1–23, 2018. [127] J. C. WYNGAARD, “Toward Numerical Modeling in the “Terra Incognita” JOHN,” no. 2, pp. 3–4, 2004. [128] G. M. Wang, D. F. Liu, Y. P. Xu, T. Meng, and F. Zhu, “PET/CT imaging in diagnosing lymph node metastasis of esophageal carcinoma and its comparison with pathological findings,” European Review for Medical and Pharmacological Sciences, vol. 20, no. 8, pp. 1495–1500, 2016. [129] J. Mirocha, B. Kosovi´ c, and G. Kirkil, “Resolved Turbulence Characteristics in Large-Eddy Simulations Nested within Mesoscale Simulations Using the Weather Research and Forecasting Model,” Monthly Weather Review, vol. 142, no. 2, pp. 806–831, 2014. 132 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ [130] C. Talbot, E. Bou-Zeid, and J. Smith, “Nested Mesoscale Large-Eddy Simulations with WRF: Performance in Real Test Cases,” Journal of Hydrometeorology, vol. 13, no. 5, pp. 1421–1441, 2012. [131] Y. Liu, T. Warner, Y. Liu, C. Vincent, W. Wu, B. Mahoney, S. Swerdlin, K. Parks, and J. Boehnert, “Simultaneous nested modeling from the synoptic scale to the LES scale for wind energy applications,” Journal of Wind Engineering and Industrial Aerodynamics, vol. 99, no. 4, pp. 308–319, 2011. [132] D. Mu˜ noz-Esparza and B. Kosovi´ c, “Generation of Inflow Turbulence in Large-Eddy Simulations of Nonneutral Atmospheric Boundary Layers with the Cell Perturbation Method,” Monthly Weather Review, vol. 146, no. 6, pp. 1889–1909, 2018. [133] D. Mu˜ noz-Esparza, R. Sharman, J. Sauer, and B. Kosovi´ c, “Toward Low-Level Turbulence Forecasting at Eddy-Resolving Scales,” Geophysical Research Letters, vol. 45, no. 16, pp. 8655–8664, 2018. [134] C. Nie, H. Li, L. Yang, B. Ye, E. Dai, S. Wu, Y. Liu, and Y. Liao, “Spatial and temporal changes in extreme temperature and extreme precipitation in Guangxi,” Quaternary International, vol. 263, pp. 162–171, 2012. [135] P. Gong, J. Wang, L. Yu, Y. Zhao, Y. Zhao, L. Liang, Z. Niu, X. Huang, H. Fu, S. Liu, C. Li, X. Li, W. Fu, C. Liu, Y. Xu, X. Wang, Q. Cheng, L. Hu, W. Yao, H. Zhang, P. Zhu, Z. Zhao, H. Zhang, Y. Zheng, L. Ji, Y. Zhang, H. Chen, A. Yan, J. Guo, L. Yu, L. Wang, X. Liu, T. Shi, M. Zhu, Y. Chen, G. Yang, P. Tang, B. Xu, C. Giri, N. Clinton, Z. Zhu, J. Chen, and J. Chen, “Finer resolution observation and monitoring of global land cover: First mapping results with Landsat TM and ETM+ data,” International Journal of Remote Sensing, vol. 34, no. 7, pp. 2607–2654, 2013. [136] D. K. Lilly, “The Representation of Small-Scale Turbulence in Numerical Simulation Experiments.” Proceedings of the IBM Scientific Computing Symposium on Environmental Sciences, no. November, pp. 195–210, 1966. [137] S.-Y. Hong, Y. Noh, and J. Dudhia, “A New Vertical Diffusion Package with an Explicit Treatment of Entrainment Processes,” Monthly Weather Review, vol. 134, no. 9, pp. 2318–2341, 2006. [138] K. S. Hansen, R. J. Barthelmie, L. E. Jensen, and A. Sommer, “The impact of turbulence intensity and atmospheric stability on power deficits due to wind turbine wakes at Horns Rev wind farm,” Wind Energy, vol. 20, no. 8, pp. 1495–1500, 2012. [139] R. Baxter, N. Hastings, A. Law, and E. J. Glass, The Courant–Friedrichs–Lewy (CFL) Condition, 2008, vol. 39, no. 5. [140] Agust´ ın Santiago, “Atendi´ o Protecci´ on Civil m´ as de 20 reportes en el Istmo por fuertes vientos,” Salina Cruz, Oaxaca, 2018. [141] J. Hern´ andez, “Motorista pierde la vida en accidente sobre carretera 190,” Santiago Niltepec, Oaxaca, 2018. 133 Chapter 8. Bibliography [142] Rar, “Suspenden clases en escuelas del Istmo de Oaxaca por fuertes vientos,” Oaxaca, 2018. [143] ´ Oscar Rodr´ ıguez, “Viento provoca voladura de cami´ on en Oaxaca,” La Ventosa, 2018. [144] P. A. Jim´ enez, J. Dudhia, J. F. Gonz´ alez-Rouco, J. Navarro, J. P. Mont´ avez, and E. Garc´ ıa-Bustamante, “A Revised Scheme for the WRF Surface Layer Formulation,” Monthly Weather Review, vol. 140, no. 3, pp. 898–918, 2012. [145] M. Angel, G. Miguez, and F. Canoura, “W ind power forecasting for a real onshore wind farm on complex terrain using WRF high resolution simulations. .” [146] Z. Yang, Y. Li, and J. E. Seem, “Optimizing Energy Capture of Cascaded Wind Turbine Array With Nested-Loop Extremum Seeking Control,” Journal of Dynamic Systems, Measurement, and Control, vol. 137, no. 12, p. 121010, 2015. [147] A. Ecot and R. Boronat, “ecot` ecnia 74 , 80 1.67 & 80 2.0,” ALSTOM, Tech. Rep., 2008. [148] R. Baxter, N. Hastings, A. Law, and E. J. Glass, “Prospectiva de Energ´ ıas Renovables 2017-2031,” Tech. Rep. 5, 2017. [149] P. J. Coldwell, A. Ricardo, F. Quiroga, and F. Z. Reyes, “Reporte de Avance de Energias Limpias 2017,” Tech. Rep., 2017. [150] B. V. G. Associates, “The Power of Onshore Wind,” Tech. Rep. June, 2018. [151] R. B. Smith, “On Severe Downslope Winds,” Americal Meteorological Society, 1985. [152] A. Est´ evez, C. Mu˜ noz-Pina, and R. Morales, “Wind Farms in Oaxaca, Mexico: Complex Contracting and Spatial Effects,” Fourth Annual Conference on Transforming Development Through Inclusive Green Growth, no. September, pp. 6–7, 2016. [153] D. Wood, S. LOZANO, and O. Romero-Hernandez, “Wind energy potential in Mexico’s Northern Border States,” Comprehensive Renewable Energy, vol. 2, no. May, pp. 73–92, 2012. [154] Q. Hern´ andez-Escobedo, F. Manzano-Agugliaro, and A. Zapata-Sierra, “The wind power of Mexico,” Renewable and Sustainable Energy Reviews, vol. 14, no. 9, pp. 2830–2840, 2010. [155] D. R. Durran, “Another look at downslope windstorms. Part I: The development of analogs to supercritical flow in an infinitely deep, continuously stratified fluid,” Journal of the Atmospheric Sciences, vol. 43, no. 21, pp. 2527–2543, 1986. [156] P. F. Sheridan and S. B. Vosper, “A flow regime diagram for forecasting lee waves, rotors and downslope winds,” Meteorological Applications, vol. 13, no. 2, pp. 179–195, 2006. [157] R. F. Hertenstein and J. P. Kuettner, “Rotor types associated with steep lee topography: Influence of the wind profile,” Tellus, Series A: Dynamic Meteorology and Oceanography, vol. 57, no. 2, pp. 117–135, 2005. 134 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ [158] R. B. Smith, “Low Froude Number Flow Past Three-Dimensional Obstacles. Part I: Baroclinically Generated Lee Vortices,” Journal of the Atmospheric Sciences, vol. 46, no. 8, pp. 3611–3613, 1989. [159] V. Grubiˇ si´ c, S. Serafin, L. Strauss, S. J. Haimov, J. R. French, and L. D. Oolman, “Wave-induced boundary-layer separation in the lee of the Medicine Bow Mountains. Part II: Numerical modeling,” Journal of the Atmospheric Sciences, p. 150904104933002, 2015. [160] A. K. Pokharel, M. L. Kaplan, and S. Fiedler, “Subtropical Dust Storms and Downslope Wind Events,” Journal of Geophysical Research: Atmospheres, vol. 122, no. 19, pp. 10 191–10 205, 2017. [161] H. ´ Ag´ ustsson and H. ´ Olafsson, “Simulations of Observed Lee Waves and Rotor Turbulence,” Monthly Weather Review, vol. 142, no. 2, pp. 832–849, 2014. [162] K. Jung-Hoon and I.-U. Chung, “Study on Mechanisms and Orographic Effect for the Springtime Downslope Windstorm over the Yeongdong Region Jung-Hoon,” Atmosphere, vol. 16, no. 2, pp. 67–83, 2006. [163] M. T. Prtenjak and D. Belusic, “Formation of reversed lee flow over the north-eastern Adriatic during bora,” Geofizika, vol. 26, no. 2, pp. 145–155, 2009. [164] M. D. Priestley, J. G. Pinto, H. F. Dacre, and L. C. Shaffrey, “The role of cyclone clustering during the stormy winter of 2013/2014,” Weather, vol. 72, no. 7, pp. 187–192, 2017. [165] Y. Cao and R. G. Fovell, “Downslope Windstorms of San Diego County. Part I: A Case Study,” Monthly Weather Review, vol. 144, no. 2, pp. 529–552, 2016. [166] G. Ganbat, J. M. Seo, J. Y. Han, and J. J. Baik, “A theoretical study of the interactions of urban breeze circulation with mountain slope winds,” Theoretical and Applied Climatology, vol. 121, no. 3-4, pp. 545–555, 2015. [167] R. N. Hoffman and S. M. Leidner, “An Introduction to the Near–Real–Time QuikSCAT Data,” Weather and Forecasting, vol. 20, no. 4, pp. 476–493, 2005. [168] J. G. Powers, J. B. Klemp, W. C. Skamarock, C. A. Davis, J. Dudhia, D. O. Gill, J. L. Coen, D. J. Gochis, R. Ahmadov, S. E. Peckham, G. A. Grell, J. Michalakes, S. Trahan, S. G. Benjamin, C. R. Alexander, G. J. Dimego, W. Wang, C. S. Schwartz, G. S. Romine, Z. Liu, C. Snyder, F. Chen, M. J. Barlage, W. Yu, and M. G. Duda, “The weather research and forecasting model: Overview, system efforts, and future directions,” Bulletin of the American Meteorological Society, vol. 98, no. 8, pp. 1717–1737, 2017. [169] P. Markowski and Y. Richardson, Mesoscale meteorology in midlatitudes, 2010, vol. 137. [170] S. Bontemps, P. Defourny, J. Radoux, E. Van Bogaert, C. Lamarche, F. Achard, P. Mayaux, M. Boettcher, C. Brockmann, G. Kirches, M. Z¨ ulkhe, V. Kalogirou, F. Seifert, and O. Arino, “Consistent global land cover maps for climate modelling communities: current achievements of the ESA’ and cover CCI,” ESA Living Planet Symposium 2013, vol. 2013, no. December, pp. 9–13, 2013. 135 Chapter 8. Bibliography [171] H. H. Shin and S.-Y. Hong, “Representation of the Subgrid-Scale Turbulent Transport in Convective Boundary Layers at Gray-Zone Resolutions,” Monthly Weather Review, vol. 143, no. 1, pp. 250–271, 2015. [172] C. Zhang, Y. Wang, and K. Hamilton, “Improved Representation of Boundary Layer Clouds over the Southeast Pacific in ARW-WRF Using a Modified Tiedtke Cumulus Parameterization Scheme *,” Monthly Weather Review, vol. 139, no. 11, pp. 3489–3513, 2011. [173] M. J. Iacono, J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins, “Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models,” Journal of Geophysical Research Atmospheres, vol. 113, no. 13, pp. 2–9, 2008. [174] M. Tewari, F. Chen, W. Wang, J. Dudhia, M. A. LeMone, K. Mitchell, M. Ek, G. Gayno, J. Wegiel, and R. H. Cuenca, “IMPLEMENTATION AND VERIFICATION OF THE UNIFIED NOAH LAND SURFACE MODEL IN THE WRF MODEL,” Bulletin of the American Meteorological Society, pp. 2165–2170, 2004. [175] T. L. Clark and W. R. Peltier, “Critical level reflection and the resonant growth of nonlinear mountain waves,” pp. 3122–3134, 1984. [176] L. Armi and G. J. Mayr, “The descending stratified flow and internal hydraulic jump in the lee of the Sierras,” Journal of Applied Meteorology and Climatology, vol. 50, no. 10, pp. 1955–2011, 2011. [177] J. Szmyd, “Influence of lee waves and rotors on the near-surface flow and pressure fields in the northern foreland of the Tatra Mountains,” Meteorological Applications, vol. 23, no. 2, pp. 209–221, 2016. [178] C. M. Appendini, J. Hern´ andez-Lasheras, R. Meza-Padilla, and J. A. Kurczyn, “Effect of climate change on wind waves generated by anticyclonic cold front intrusions in the Gulf of Mexico,” Climate Dynamics, vol. 0, no. 0, pp. 1–17, 2018. [179] X. Hong, M. Peng, S. Wang, and Q. Wang, “Simulating and understanding the gap outflow and oceanic response over the Gulf of Tehuantepec during GOTEX,” Dynamics of Atmospheres and Oceans, vol. 82, pp. 1–19, 2018. [180] O. A. Jaramillo and M. A. Borja, “Wind speed analysis in La Ventosa, Mexico: A bimodal probability distribution case,” Renewable Energy, vol. 29, no. 10, pp. 1613–1630, 2004. [181] R. Romero-Centeno, J. Zavala-Hidalgo, A. Gallegos, and J. J. O’Brien, “Isthmus of tehuantepec wind climatology and ENSO signal,” Journal of Climate, vol. 16, no. 15, pp. 2628–2639, 2003. [182] M. J. Brennan, H. D. Cobb, and R. D. Knabb, “Observations of Gulf of Tehuantepec Gap Wind Events from QuikSCAT: An Updated Event Climatology and Operational Model Evaluation,” Weather and Forecasting, vol. 25, no. 2, pp. 646–658, 2010. 136 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ [183] J. P. McCreary, H. S. Lee, and D. B. Enfield, “The response of the coastal ocean to strong offshore winds: With application to circulations in the Gulfs of Tehuantepec and Papagayo,” Journal of Marine Research, vol. 47, no. 1, pp. 81–109, 1989. [184] D. R. Durran, “Lee Waves and Mountain Waves,” Encyclopedia of Atmospheric Sciences, pp. 1161–1169, 2013. [185] W. J. Steenburgh, D. M. Schultz, and B. A. Colle, “The Structure and Evolution of Gap Outflow over the Gulf of Tehuantepec, Mexico,” Monthly Weather Review, vol. 126, no. 10, pp. 2673–2691, 1998. [186] R. S. Arthur, J. D. Mirocha, K. A. Lundquist, and R. L. Street, “Using a canopy model framework to improve large-eddy simulations of the neutral atmospheric boundary layer in the Weather Research and Forecasting model,” Monthly Weather Review, pp. MWR–D–18–0204.1, 2018. [187] P. Doubrawa, A. Montornes, R. J. Barthelmie, S. C. Pryor, G. Giroux, and P. Casso, “Effect of Wind Turbine Wakes on the Performance of a Real Case WRF-LES Simulation,” Journal of Physics: Conference Series, vol. 854, no. 1, 2017. [188] C. Xu, J. Yang, C. Li, W. Shen, Y. Zheng, and D. Liu, “A Research on Wind Farm Micro-sitting Optimization in Complex Terrain,” International Conference on aerodynamics of Offshore Wind Energy Systems and wakes, pp. 669–679, 2013. [189] A. C. Fitch, J. B. Olson, J. K. Lundquist, J. Dudhia, A. K. Gupta, J. Michalakes, and I. Barstad, “Local and Mesoscale Impacts of Wind Farms as Parameterized in a Mesoscale NWP Model,” Monthly Weather Review, vol. 140, no. 9, pp. 3017–3038, 2012. [190] A. H. Schrotenboer, M. A. uit het Broek, B. Jargalsaikhan, and K. J. Roodbergen, “Coordinating technician allocation and maintenance routing for offshore wind farms,” Computers and Operations Research, vol. 98, pp. 185–197, 2018. [191] GWEC, “GWEC Global Wind 2017 report: Offshore Wind,” 2018. [192] A. Montorn` es, “WRF LES, the microscale awakens,” Boulder, Colorado, pp. 1323–1330, 2002. [193] H. Xu, Y. Wang, and M. Wang, “The Performance of a Scale-Aware Nonlocal PBL Scheme for the Subkilometer Simulation of a Deep CBL over the Taklimakan Desert,” Advances in Meteorology, vol. 2018, 2018. [194] A. Montorn` es, P. Casso, B. Kosovic, and G. Lizcano, “WRF-LES in the real world: Towards a seamless modeling chain for wind industry applications,” 17th Annual WRF Users’ Workshop, 2016. [195] Wind Europe, “Wind in power 2017. Annual combined onshore and offshore wind energy statistics,” Wind Europe, Tech. Rep., 2018. [196] D. Mu˜ noz-Esparza, B. Kosovi´ c, C. Garc´ ıa-S´ anchez, and J. van Beeck, “Nesting Turbulence in an Offshore Convective Boundary Layer Using Large-Eddy Simulations,” Boundary-Layer Meteorology, vol. 151, no. 3, pp. 453–478, 2014. 137 Chapter 8. Bibliography Fitch, espec´ ıfico para parques e´ olicos, se simula un per´ ıodo de un a˜ no con una configuraci´ on de previsi´ on diaria de funcionamiento. En el segundo cap´ ıtulo, damos un paso adelante en el campo de la previsi´ on a corto plazo combinando la predicci´ on del del primer estudio con una herramienta estad´ ıstica de post-proceso (filtro Kalman-Bayesian). Con esta t´ ecnica, llegamos a la escala temporal de nowcasting, con interesantes aplicaciones en la industria e´ olica. La segunda parte, presentada en el cap´ ıtulo tres, se centra en el efecto estela de los parques e´ olicos en su entorno. Introducimos el concepto Annual Wake, que es el recurso e´ olico medio anual perdido en la zona debido a la estela del parque e´ olico. Probamos esta herramienta en el mismo parque e´ olico de la primera parte y en otro situado al este de China. La tercera parte aborda la cuesti´ on del estudio de los vientos extremos. En concreto, se centra en la investigaci´ on de los diferentes fen´ omenos de las mountain lee waves resultantes de la compleja interacci´ on entre las condiciones meteorol´ ogicas a gran escala y los forzamientos orogr´ aficos locales en un grupo de parques e´ olicos en M´ exico. Finalmente, en la cuarta y ´ ultima parte, en el cap´ ıtulo cinco, se analiza la microescala turbulenta en un ´ area de parque e´ olico en el sur de China usando simulaciones WRF-LES. A continuaci´ on se presenta un resumen de cada uno de los cap´ ıtulos de la tesis. Cap´ ıtulo 2: Predicci´ on de recurso e´ olico Los modelos meteorol´ ogicos regionales se est´ an convirtiendo en una herramienta cada vez m´ as usada para la predicci´ on del recurso e´ olico debido a su capacidad para simular la din´ amica de los flujos de viento locales que afectan a la producci´ on de parques e´ olicos. Este estudio se centra en la predicci´ on y validaci´ on de la producci´ on de un parque e´ olico real sobre terreno complejo mediante simulaciones del modelo WRF (Weather Research and Forecasting) a alta resoluci´ on. El parque e´ olico se encuentra en Galicia, en el noroeste de Espa˜ na, en una regi´ on compleja y con gran recurso e´ olico. Utilizando la parametrizaci´ on Fitch, espec´ ıfica para parques e´ olicos, se simula un per´ ıodo de un a˜ no a partir sistema de predicci´ on operativo diario. Se obtienen predicciones de producci´ on energ´ etica y viento, y se comparan con datos reales obtenidos en cada turbina proporcionados por la empresa que explota el parque. Los resultados muestran que el WRF realiza una buena predicciones energ´ ıa e´ olica para este tipo de parques e´ olicos, debido a su precisa representaci´ on del comportamiento de la capa l´ ımite planetaria de la regi´ on y al buen desempe˜ no del esquema Fitch en estas condiciones. El mejor error medio anual (MAE) obtenido es de 1,87 m/s para la velocidad del viento y 14,75% para la energ´ ıa e´ olica. Comparando experimentos con y sin el esquema de Fitch, estimamos las p´ erdidas del recurso e´ olico en la zona debido el efecto estela de los aerogeneradores. La estela media anual o la huella ambiental de la instalaci´ on se extiende varios kil´ ometros en la direcci´ on suroeste-noreste de los vientos dominantes, con p´ erdidas de recursos del 0,5% incluso a 17 km de las turbinas. Cap´ ıtulo 3: Predicci´ on de viento a corto plazo (nowcasting) Los modelos meteorol´ ogicos regionales a microescala se han convertido en una herramienta fundamental para la predicci´ on de la producci´ on de parques e´ olicos debido a su capacidad para resolver la din´ amica de los flujos locales. La alta demanda de herramientas de predicci´ on fiables en la industria energ´ etica es la motivaci´ on para el desarrollo de un sistema integrado que combina el modelo atmosf´ erico WRF con una optimizaci´ on obtenida por el conjunci´ on de 144 MIGUEL ´ ANGEL PR´ OSPER FERN ´ ANDEZ un filtro Kalman y un modelo Bayesiano. Este estudio se centra en el desarrollo y validaci´ on de este sistema combinado en un parque e´ olico en Galicia (NW Espa˜ na). Se simula un per´ ıodo de un a˜ no a una resoluci´ on horizontal de 333 m, configuraci´ on de la predicci´ on operativa diaria. El filtro Kalman-Bayesiano es probado tanto directamente en la velocidad del viento como en sus componentes U-V (zonal y meridional) en per´ ıodos de nowcasting de entre 10 minutos a 6 horas. Los resultados son prometedores, los principales ´ ındices de error estad´ ıstico mejoran significativamente en un horizonte de previsi´ on de 6 horas y a´ un m´ as en los casos con un horizonte m´ as corto. El error medio anual (MAE) para el horizonte de previsi´ on 1h es de 1,03 m/s para la velocidad del viento y 12,16opara la direcci´ on del viento. Adem´ as, el ´ exito de la utilizaci´ on del sistema integrado en los diferentes casos analizados demuestra la utilidad potencial que esta herramienta puede tener para una variedad de aplicaciones en parques e´ olicos operativos y el mercado energ´ etico. Cap´ ıtulo 4: Estela anual El efecto estela local en las condiciones de flujo alrededor de los parques e´ olicos afecta significativamente la producci´ on de energ´ ıa e´ olica. En este cap´ ıtulo se realizan simulaciones con el modelo WRF utilizando el esquema de Fitch para un parque e´ olico en la provincia china de Jiangsu, frente al mar Amarillo. Se simula un per´ ıodo de un a˜ no con una resoluci´ on horizontal de 333 m, obtiendo las predicciones de producci´ on energ´ etica y viento y se comparan con los datos reales proporcionados por la empresa desarrolladora del parque. Los resultados muestran que el WRF realiza buenas predicciones de energ´ ıa e´ olica para este tipo de parques, debido a una buena representaci´ on del comportamiento de la capa l´ ımite planetaria de la regi´ on y el buen desempe˜ no del esquema Fitch bajo estas condiciones. Se observan efectos de estela significativos a varios kil´ ometros a favor del viento desde el parque, especialmente en las direcciones predominantes del viento (sureste-este). Estos resultados muestran que este m´ etodo puede proporcionar informaci´ on valiosa para el an´ alisis de posibles ubicaciones futuras de parques e´ olicos. Cap´ ıtulo 5: Estudio de eventos extremos Los Tehuantepecers o Tehuanos son vientos extremos producidos en el Istmo de Tehuantepec, que soplan hacia el sur a trav´ es del paso de Chivela, la brecha monta˜ nosa a trav´ es del Istmo, desde el Golfo de M´ exico hasta el Oc´ eano Pac´ ıfico. Son el resultado de la compleja interacci´ on entre las condiciones meteorol´ ogicas a gran escala y las forzamientos orogr´ aficos locales alrededor del paso de Chivela y ocurren principalmente en los meses de invierno, debido a las advecciones de aire que se producen tras los frentes fr´ ıos que llegan hasta el sur. Pueden generar eventos extremos localizados, como downslope windstorms y saltos hidr´ aulicos, fuertes flujos turbulentos que tienen un efecto directo en el lado Pac´ ıfico del istmo y en el Golfo de Tehuantepec. Este estudio se centra en la investigaci´ on de estos fen´ omenos mediante simulaciones de modelos WRF (Weather Research and Forecasting) de alta resoluci´ on horizontal y vertical. En particular, empleamos una configuraci´ on de cuatro dominios anidados, con resoluci´ on horizontal de hasta 444 m en el dominio m´ as interior y 70 niveles verticales h´ ıbridos-sigma, 8 de los cuales son que se encuentran dentro de los primeros 200 m sobre el nivel del suelo. Seleccionamos un per´ ıodo de 36 horas en diciembre de 2013, que es cual se observaron condiciones favorables para la fenomenolog´ ıa local estudiada. El 145 Chapter 8. Bibliography experimento de alta resoluci´ on del WRF revela una clara estructura compleja en el fuerte flujo de viento del Tehuano. Dependiendo del n´ umero de Froude en lo alto de la barrera topogr´ afica, las condiciones de tormentas de viento (flujo supercr´ ıtico) o los saltos hidr´ aulicos (flujo cr´ ıtico) se desarrollan simult´ aneamente en diferentes lugares al este del paso de Chivela con diferentes alturas de terreno. La comparaci´ on con las observaciones sugiere que el modelo representa con precisi´ on el intenso viento de bajada espacialmente heterog´ eneo y la formaci´ on de mountain wave clouds durante varias horas, con bajos errores en la velocidad y viento, direcci´ on del viento, y temperatura. Cap´ ıtulo 6: An´ alisis de fen´ omenos a microescala El estudio de los fen´ omenos de flujo de viento a microescala ha sido siempre un punto de inter´ es dentro de la industria e´ olica. Hay varios eventos e´ olicos bien conocidos relacionados con ´ el que los dise˜ nadores de turbinas y los gerentes de operaciones deben tener en cuenta, tales como rachas de viento altas, los cambios extremos de direcci´ on, r´ afagas de viento coherentes con cambios de direcci´ on o el cizallamiento extremo del viento. WRF Large Eddy Simulations (LES) es una herramienta de modelizaci´ on num´ erica que se est´ a utilizando cada vez m´ as en este campo. Basado en una simulaci´ on WRF de muy alta resoluci´ on y una desactivaci´ on del esquema de capa l´ ımite planetaria, esta configuraci´ on WRF puede ser utilizada para simular flujos de viento turbulentos en numerosas aplicaciones en el sector e´ olico. En este estudio, evaluamos el las simulaciones LES en un terreno complejo en el ´ area de un parque e´ olico en el sur de China. Para ello, simulamos un periodo de 10 d´ ıas dividido en ejecuciones diarias utilizando una configuraci´ on de acoplamiento mesoescala-microescala que termina en dos dominios LES y alcanza una resoluci´ on horizontal de 98 m en el m´ as interior. Para resolver la inestabilidad de las simulaciones, aplicamos una t´ ecnica de alisamiento zonal que suaviza la orograf´ ıa cerca de puntos num´ ericamente inestables. Los resultados obtenidos se validan utilizando datos observacionales de tres estaciones meteorol´ ogicas en diferentes puntos de la zona. 146