Evaluation of Landsat-8 TIRS data recalibrations and land surface temperature split-window algorithms over a homogeneous crop area with different phenological land covers.

ISPRS Journal of Photogrammetry and Remote Sensing 174 (2021) 237–253 Available online 4 March 2021 0924-2716/© 2021 The Authors. Published by Elsevier B.V. on behalf of International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Evaluation of Landsat-8 TIRS data recalibrations and land surface temperature split-window algorithms over a homogeneous crop area with different phenological land covers Raquel Nicl` os a , * , Jesús Puchades a , C´ esar Coll a , María J. Barber` a a , Lluís P´ erez-Planells a , Jos´ e A. Valiente b , Juan M. S´ anchez c a Department of Earth Physics and Thermodynamics, Faculty of Physics, University of Valencia, 50, Dr. Moliner, E-46100 Burjassot, Spain b Instituto Universitario Centro de Estudios Ambientales del Mediterr´ aneo–CEAM-UMH, 14 Charles Darwin, E-46980 Paterna, Spain c Regional Development Institute, University of Castilla-La Mancha, Campus Universitario s/n, E-02071 Albacete, Spain ARTICLE INFO Keywords: Landsat-8 TIRS Calibration LST algorithms Validation In-situ measurements Thermal-infrared ABSTRACT Successive re-calibrations were implemented in Landsat-8 TIRS data since launch. This paper evaluates the performances of both: (1) these re-calibrations, up to the last calibration update announced for TIRS data in the next Landsat Collection 2; and (2) single-channel (SC) corrections and split-window (SW) algorithms to retrieve land surface temperature (LST) from TIRS data. A robust and accurate multi-year (2014–2019) set of reference ground data were used, which included thermal infrared (TIR) radiance measurements taken along transects in a uniform and thermally homogeneous rice paddy area, but also emissivity measurements for the different ground covers at the site through the year. The calibration results showed significant biases at the site for data after the 2014 reprocessing, but negligible biases and root-mean-square differences (RMSDs) <1.5 K were obtained when using the current TIRS data in Collection 1 (i.e., data after the 2017 reprocessing). The last announced calibration update mainly introduced differences in biases, improving slightly the results for band 10 and presenting a calibration response difference between bands. The SC corrections showed negligible LST biases for both bands and the lowest RMSDs (<1.6 K) when using the band 10 data in the current Collection 1, and the bias disappeared for this band after applying the calibration update. Three of the seventeen different SW equations evaluated in the paper showed negligible biases and LST RMSDs lower than or equal to 0.8 K. These three SW algorithms are mainly recommended for users of the current TIRS data in Collection 1; one of them being that proposed to generate a SW LST product in the future Collection 3. Finally, bias differences around 1.4 K were shown in the results of the SW algorithms after applying the calibration update announced for Collection 2. 1. Introduction Land surface temperature (LST) is an Essential Climate Variable (ECV) for the monitoring of the Earth climate system (Bojinski et al., 2014; Guillevic et al., 2018), as defined by the Global Climate Observing System (GCOS, GCOS-200) and the Climate Change Initiative (CCI) programme of the European Space Agency (ESA). LST is required for studies in meteorology, climatology and hydrology (Weng, 2009; Ghent et al., 2010; Anderson et al., 2012; Trigo et al., 2015; Mokhtari et al., 2019). NASA and the United States Geological Survey (USGS) launched the Landsat-8 (L8) on February 11, 2013 to keep on providing Earth surface data at high spatial resolution like the previous missions of the Landsat series. The L8 Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) started the data acquisition in April 2013. The L8 TIRS is the first Landsat sensor with two Thermal Infrared (TIR) bands, 10 (10.60–11.19 µm) and 11 (11.50–12.51 µm), at a nominal spatial resolution of 100 m. These two TIR bands allow the use of split-window (SW) algorithms to obtain LST from atmospheric correction through the differential atmospheric absorption between the two bands. In November 2013, the L8 Calibration Team warned of the need to reprocess the TIRS data due to discrepancies observed when comparing * Corresponding author. E-mail addresses: [email protected] (R. Nicl` os), [email protected] (J. Puchades), [email protected] (C. Coll), [email protected] (M.J. Barber` a), lluis. [email protected] (L. P´ erez-Planells), [email protected] (J.A. Valiente), [email protected] (J.M. S´ anchez). Contents lists available at ScienceDirect ISPRS Journal of Photogrammetry and Remote Sensing journal homepage: www.elsevier.com/locate/isprsjprs https://doi.org/10.1016/j.isprsjprs.2021.02.005 Received 8 August 2020; Received in revised form 27 January 2021; Accepted 2 February 2021 ISPRS Journal of Photogrammetry and Remote Sensing 174 (2021) 237–253 238 the L8 TIRS data with equivalent top-of-atmosphere (TOA) data retrieved from in-situ data measurements acquired at reference water sites. They proposed subtracting 0.29 ±0.12 Wm -2 sr -1 µm −1 (2.1 ±0.8 K at 300 K) and 0.5 ±0.2 Wm -2 sr -1 µm −1 (4.4 ±1.8 K at 300 K) from radiances (brightness temperatures) measured by TIRS band 10 and 11, respectively. The USGS Earth Resource Observation Science (EROS) processing system implemented the bias error correction on February 3, 2014 (thus hereafter called 2014 reprocessing), but the residual variability for the recalibrated data was still larger than desired for the TIRS bands (Barsi et al., 2014). The research conducted after these first calibration activities indicated that the calibration errors were dependent on the scene due to radiation incursion from outside the Instantaneous Field of View (IFOV) of the TIRS, and thus they were inherent to each TIRS scene. Thus, the calibration based on local bias assessments could be just valid for particular subsets of data. The out-of-field stray light effect was shown to be larger for band 11 than for band 10, which made the use of band 11 data not recommended where absolute calibration was required (Barsi et al., 2014). Additionally, users were advised not to apply SW atmospheric and emissivity correction techniques, given the large uncertainties of the TIRS band 11 values. Montanaro et al. (2014a; 2015) proposed an algorithm for modeling and correcting the radiance introduced from outside the IFOV to reduce the calibration observed discrepancies to attain the required absolute error limit of 2% for TIRS. The out-of-field radiance produced, additionally to a banding observed in uniform and homogeneous areas, differences between the detected signals and the modeled TOA radiances from target temperatures up to 4–5% in band 10 and up to 8–9% in band 11 (Gerace and Montanaro, 2017). Maps of the out-of-field radiation locations were made for each TIRS detector, which were used in the correction algorithm. It was known as “TIRS-on-TIRS” algorithm, since the correction is estimated with TIRS data within the acquisition swath to avoid using external data (Montanaro et al., 2015). The algorithm implementation in the TIRS data was done in 2017 (hereafter called 2017 reprocessing), as a result of the successive improvements implemented. Gerace and Montanaro (2017) showed evaluation results when using independent scenes of EOS-Terra MODIS as reference data. They showed that the magnitude of the banding artifact was reduced by half on average over the uncorrected product and that the absolute radiometric error was reduced to approximately 0.5% in both TIRS bands on average, i.e., below the 2% requirement. Guo et al. (2020) observed that the relationship between the difference between TIRS TOA brightness temperatures at bands 10 and 11 and in-situ LSTs was closer to the atmospheric differential absorption theory after the straylight correction. They found this fact implied that 2007-corrected data were more proper to obtain suitable results when using SW algorithms than before the correction. In December 2019, the L8 Calibration Team announced a TIRS absolute radiometric calibration update in preparation for the upcoming Landsat Collection 2 processing. According to them, once the stray-light correction was in place, it became possible to work on reducing remaining radiometric calibration errors, as those that they observed from vicarious calibration results using instrumented buoys in various water bodies. The new calibration update proposes two different corrections: one from launch to March 2, 2015 (with TIRS instrument operating on side-A electronics); and another one after March 2, 2015 (with TIRS operating on side-B electronics, switched over from nominal side-A electronics for correcting the anomaly observed in TIRS data operation since November 2014 (England, 2019)). The corrections use linear gain and bias adjustments. The notice on the U.S. Geological Survey website (usgs.gov/land-resources/nli/landsat/landsat-8-oli-andtirs-calibration-notices) provides a table that lists the estimated change in TIRS bands response (in K) as a result of the calibration update for a range of temperatures that is being corrected and for both operational modes of the instrument (sides A and B). It also shows figures of the estimated change both in radiance and temperature per band. The first objective of the paper is the evaluation of the abovementioned recalibrations implemented in the TIRS data in 2014 and 2017, and also when using the last calibration update proposed in 2019, not yet evaluated with independent reference ground data over land sites. The second objective of the paper is to check different methods for atmospheric and emissivity corrections of TIRS brightness temperatures for LST retrievals. We evaluated single-channel (SC) corrections and five different SW algorithms. Different sets of regression coefficients for these SW algorithms were estimated in previous papers (Du et al., 2015; Guo et al., 2020; Gerace et al., 2020; Jimenez-Mu˜ noz et al., 2014; Meng et al., 2019; Rozenstein et al.,2014; Yu et al., 2014), and thus finally this paper validates a set of seventeen SW equations against ground data. Previous studies (Guo et al., 2020; Gerace et al., 2020; Meng et al., 2019; Yu et al., 2014) evaluated either the change in TIRS TOA radiances after the stray-light correction in the 2017 reprocessing or the results of specific SW algorithms using Surface Radiation Budget Monitoring (SURFRAD) data as reference. The SURFRAD stations measure upwelling and downwelling irradiances using pyrgeometers in the spectral range from 3.5 to 50 µm, from which estimates of LSTs can be derived. However, according to Guillevic et al. (2012), the SURFRAD instrumental uncertainty alone give a LST uncertainty of around 1 K. Krishnan et al. (2020) reported average differences up to 2 K between LST derived from TIR radiometers and those obtained from pyrgeometer data for grassland sites, depending on the heterogeneity of the site and season, and even larger for an asphalt sample. Gerace et al. (2020) indicated that standard deviations (SDs) of approximately 2 K were observed for the differences between LSTs retrieved from spaceborne TIR instruments and LSTs estimated from SURFRAD data. They concluded that given that this 2 K residual error was consistently observed and that the pyrgeometers are sensitive from 4 to 50 μ m, this fact could indicate that reflected solar radiation might be contributing the measurements and that the uncertainty in the required broadband emissivity was potentially a limiting factor for applications requiring high accuracy. Guo et al. (2020) indicated that, besides SURFRAD sites, it was necessary to validate using more robust and homogeneous ground-measured datasets to help clarify the effect of the recalibrations on LST retrieval. We evaluated all the TIRS recalibrations, and also LSTs retrieved with SC corrections and a complete set of SW algorithms, using accurate radiance data measured by hand-held radiometers with narrow-bands in the TIR, comparable to those of L8 TIRS, instead of pyrgeometer data. These TIR radiance measurements were acquired along transects over a uniform and thermally-homogeneous crop area, with very different phenological land covers, concurrently with the registration of TIRS images in the years 2014 and 2016–2019. Additionally, emissivities for these land covers were measured at the site, and not assumed or estimated from a mean emissivity climatology or threshold methods like in previous papers (Gerace et al. 2020; Yu et al., 2014; Rozenstein et al., 2014; Meng et al., 2019; Guo et al., 2020). To ensure the high quality and accuracy of the data used for both evaluations and of the results, we also did a comprehensive uncertainty analysis and applied a test based on the R-based validation method (Wan and Li, 2008) to assess the quality of the atmospheric profiles used in the radiative transfer calculations. Therefore, the main strengths and novelties of this paper can be summarized as: (1) Multi-year evaluation of L8 TIRS calibrations for very different ground covers and atmospheric conditions through the year. (2) Evaluation of the successive re-calibration procedures implemented for L8 TIRS data since launch and up to the present, i.e., including the last calibration update announced for the Landsat Collection 2. (3) Ground LST measurements, used as reference data, acquired along transects by multiband TIR radiometers (instead of pyrgeometers at fix stations) over a thermally homogeneous site at the spatial scales of both ground radiometric measurements and the L8 TIRS pixel. (4) In situ measurements of narrow-band TIR emissivities for the different covers of the site. (5) Comprehensive uncertainty analysis of R. Nicl` os et al. ISPRS Journal of Photogrammetry and Remote Sensing 174 (2021) 237–253 239 the ground-measured and Landsat-derived LSTs used in the comparison, taking into account all possible sources of error involved in the measurements. (6) Evaluation of a complete set of SC and SW methods for atmospheric and emissivity corrections of L8 TIRS brightness temperatures for LST retrievals. (7) Application of a test in order to assess the quality of the atmospheric profiles used in the required radiative transfer calculations. The following section describes the site and the acquisition of ground data. Section 3 explains the required atmospheric radiative transfer simulations, and the SC corrections. Section 4 describes the evaluated SW techniques and the sets of coefficients. Section 5 includes the results and discussion, together with the uncertainty analysis and the atmospheric profile quality test. The calibration results are shown and discussed in 5.1; the results of the evaluation of the LSTs retrieved using SC corrections are included in 5.2; and the results obtained for the different SW techniques are described and discussed in 5.3. Finally, section 6 summarizes the main conclusions of the paper. 2. Experimental site and ground data The experimental site is a uniform and thermally-homogeneous rice paddy area located near the city of Valencia, previously used in calibration/validation (CAL/VAL) studies for other satellite sensors, and called the Valencia Test Site (Coll et al., 2005, 2012; Nicl` os et al., 2018). This area is covered by L8 scenes acquired at paths 198 and 199 (Fig. 1). The rice paddy has land cover changes along the year due to the phenology of the rice crop, with nearly full vegetation cover (fraction of vegetation cover close to one) in July-September, flooded soils (water surfaces) in December-January and also in May-June, and bare soils in February-April, with a trend from wet to dry soils (Fig. 2). The soil at the site is a silty clay loam soil (textural classification: 14% sand, 50% silt and 35% clay; mineral classification: 19.3% quartz, 3.5% feldspar, 6% filosilicate, 8.9% hematite, 62.3% calcite; organic matter: 4.5% (Nicl` os et al., 2018)). The land cover variations, with three totally different but uniform land covers at the same site, make it interesting for calibrations and validations under very different surface conditions and temperatures along the year (Nicl` os et al., 2018). In previous papers, we analyzed the thermal homogeneity of the site at different scales to evaluate possible dissimilarities between the footprint of satellite pixels and the areas covered by in-situ measurements (Coll et al., 2005, 2012; Nicl` os et al., 2018). These analyses used data acquired over the site (with full vegetation cover) by Landsat-5 TM and EOS ASTER and showed SDs<0.5 K for an area of 16 km 2 . They also used TIR radiance measurements along transects covering areas of 250 m by 250 m and showed SDs<0.6 K. The latter was repeated by Nicl` os et al. (2018) for the three land cover conditions in the site, and SDs<0.5 K, 0.4 K, 0.9 K were obtained for full vegetation cover, flooded soil, and bare soil, respectively, which proved that the site was adequately thermally homogeneous, even for the bare soil conditions. In this paper, unlike previous studies (Guo et al., 2020, Gerace et al., 2020, Meng et al., 2019, Yu et al. 2014) that use pyrgeometer data of SURFRAD stations for TIRS data validations over land sites, ground TIR radiances were measured along transects in our experimental area using hand-held Cimel Electronique multiband radiometers model CE-312 (Legrand et al., 2000, Brogniez et al., 2003, Coll et al., 2019). These ground measurements were acquired, concurrently with L8 overpasses, following the methodology detailed described in Coll et al. (2005) for a total of 25 cloud-free days from 2014 to 2019. The CE-312 radiometers measured the surface radiance within a spectral band i, L surf,i , which depends on the surface emissivity, ε i , as follows: Lsurf,i= ε iBi(T)+(1− ε i)Li a,hem↓(1) where Bi(T)is the channel Planck’s function for a temperature T (here T being the LST). Li a,hem↓ is the atmospheric downwelling irradiance divided by π (Nicl` os et al., 2018). Li a,hem↓ was measured using an Infragold Reflectance Target (IRT-94–100) made by Labsphere (Guillevic et al., 2018), which is a highly diffuse gold panel with a reflectivity close to 0.92 in the 8 – 14 µm region. Emissivity measurements were taken for the different land covers at the site using the Temperature-Emissivity Separation method (TES) (Gillespie et al., 1998; Hulley & Hook, 2009) applied to the ground data measured by the CE-312 radiometers and also the Box Method (Nicl` os et al., 2018). The TES method was used for the bare soils at the site, since this method works better for surfaces with significant emissivity spectral variations (Coll et al., 2005; Nicl` os et al., 2018). This method requires Lsurf,i measured over a specific sample with the different bands of the CE312 radiometer, the Li a,hem↓ in each band and an initial ε i guess to retrieve the maximum-minimum emissivity difference (MMD), i.e., the emissivity spectral contrast, among the CE-312 bands. Then, an empirical relationship between the MMD and the minimum ε i is used to rescale these band ε i to the actual values. The empirical relationship was obtained by integrating the spectra contained in the ASTER spectral library (Baldridge et al., 2009) with the response functions of the CE-312 bands. Using the TES method, ε i values for the bare soil in the site were previously measured for different soil moisture contents, and an increase in ε i with the soil moisture was observed (Nicl` os et al., 2018). Fig. 1. L8 scenes (false color composites of L8 OLI bands; RGB 654) acquired by the paths 198 and 199 (left and right figures, respectively) over the rice paddy site. R. Nicl` os et al. ISPRS Journal of Photogrammetry and Remote Sensing 174 (2021) 237–253 240 The Box Method was used for the full vegetation cover, with a neargrey body spectrum (Coll et al., 2005; Nicl` os et al., 2018). This method uses a thermally insulated box with inner walls made of highly reflecting polished aluminum and two interchangeable top lids. One lid is made of the same aluminum (ideally with ε =0) and the other lid is made of highly emitting material (corrugated aluminum painted in Parson’s black, ideally ε =1). The latter can be heated to about 330 K and has a thermostat. Both lids have a small aperture for the radiometer to observe the radiance coming from the bottom of the box, which can be open to the sample or closed with another aluminum lid without aperture. The sample ε i measurement requires a series of radiance measurements for the system of sample and box lids in different configurations: 1) aluminum top lid – sample at bottom, 2) hot top lid – sample at bottom, 3) hot top lid – aluminum lid at bottom, and 4) aluminum top lid – aluminum lid at bottom. Further details on this procedure and the correction required for a not ideal box are given in Rubio et al. (2003). For water surfaces, ε i values were retrieved with eq. (1) from CE-312 L surf,i measurements and T values measured by contact probes (Nicl` os et al., 2014). The ε i values obtained from the measurements for the different land cover conditions at the site were around 0.990, 0.985, and 0.970 for the CE-312 band centered in 10.9 µm for flooded soils (water), full vegetation cover, and dry bare soils, respectively. Uncertainties ranging from 0.004 to 0.006 were obtained for the ε i values measured over flooded soils and full vegetation cover, and from 0.004 to 0.011 for ε i over bare soils. The uncertainty values were in agreement with those obtained for the TES method over other surfaces in Coll et al. (2019). These emissivity measurements acquired in field conditions were used to correct the ground-measured TIR radiances. Finally, the uncertainties in ground LSTs were estimated taking into account three error sources added together in quadrature (Coll et al. 2005, 2019). The error sources are: (i) the calibration uncertainty of the ground radiometers, estimated from 0.12 K to 0.15 K in Coll et al. (2019); (ii) the spatial variability within the test site taken as the SD of all the LST measured in the transects within five minutes of each overpass (lower than the maximum values mentioned above for the different covers); and (iii) the uncertainty resulting from the propagation of the uncertainties in the input variables used to retrieve the LSTs through eq. (1); the emissivity one producing the major effect in terms of LST (from 0.4 K to 0.6 K) (Coll et al., 2019). We obtained LSTs from 282 K to 316 K for the whole period, with uncertainties ranging from 0.6 to 0.9 K (with an average LST uncertainty of 0.67 ±0.08 K). 3. Atmospheric radiative transfer and single-channel correction The evaluation of the successive recalibrations consisted of comparing L8 TIRS TOA radiances / brightness temperatures with those simulated from ground LSTs, and the evaluation of the LSTs retrieved from L8 TIRS data consisted of comparing them with ground LSTs, as summarized in the evaluation flow chart shown in Fig. 3. To obtain TOA radiances, L i , from ground-measured LSTs, or inversely LSTs from L i , the radiative transfer equation (RTE) is used. The RTE is written as follows: Li= τ iLsurf,i+Li a↑(2) where Lsurf,i is dependent on T and ε i as described in eq. (1), and τ i and Li a↑ are the atmospheric transmittance and upwelling radiance, respectively. A single-channel (SC) correction can be applied to retrieve LSTs from L i inverting eq. (2) for Lsurf,i and then eq. (1) for Bi(T). The error introduced by using eqs. (1) - (2) with band magnitudes instead of the spectral integration in terms of radiance was evaluated in previous papers (Coll et al., 2012), and errors lower than 0.1 K were shown for satellite TIR bandwidths lower than 1 µm. TOA brightness temperatures, T i , are obtained from L i =B i (T i ) in eq. (2). In addition to ε i values (measured for each land cover at the site as described in section 2), the atmospheric τ i, Li a↑ and Li a,hem↓ variables should be estimated for each atmospheric condition. The thermal window band (10–12.5 μ m) is affected by absorption and emission from atmospheric water vapor, whose effect in terms of T i (and also in LST retrieval from them) depends on the vertical profiles of air humidity and temperature (Coll et al., 2012). Spatial and temporal interpolations of the National Centers for Environmental Prediction (NCEP) atmospheric profiles sampled on a 1◦x 1◦grid and generated every six hours were used as input data to a radiative transfer code, the MODerate resolution atmospheric TRANsmission (MODTRAN) 5 (Berk et al., 2006; Guillevic et al., 2018), to simulate the required τ i, Li a↑ and Li a,hem↓. A sensitivity analysis was performed to assess the uncertainty in the LSTs retrieved with the SC method, i.e., inverting eq. (1)-(2). We have followed the procedure proposed by Ghent et al. (2019) to define the different uncertainties affecting the satellite LST retrieval. We have considered four uncertainty sources. (i) Uncertainties in radiative transfer simulations including possible errors in the atmospheric profiles and the radiative transfer model used (Hook et al., 2004; Coll et al., 2012), obtaining values of 0.6 K and 1.1 K for TIRS bands 10 and 11, respectively for variations of 10% in relative humidity and 1 K in air temperature at each atmospheric profile level. (ii) Uncertainties in the measured emissivity values, that propagate through the inverse of eq. (1). (iii) Uncertainties in TOA T i that contribute to the LST uncertainty by propagation in eq. (2). The T i uncertainty was taken as the noiseequivalent-error (NEΔT) specified for the TIRS bands, i.e., 0.4 K at 300 K for both bands (Irons et al., 2012; Meng et al., 2019). Although the actual TIRS performance was shown to be closer to 0.05 K by Montanaro et al. (2014b), this value of 0.4 K was considered for the sensitivity analysis to be consistent with those in previous studies (Jimenez-Mu˜ noz Fig. 2. Views of the different land covers at the site: full vegetation cover (10 July 2018; left), bare soil (27 March 2018; middle), and flooded soil (2 June 2016; right). False color composites of L8 OLI (RGB 654). R. Nicl` os et al. ISPRS Journal of Photogrammetry and Remote Sensing 174 (2021) 237–253 241 et al., 2014; Du et al., 2015; Meng et al., 2019). For the evaluations, we considered the average LST value for an array of 17 ×17 TIRS pixels resampled to 30-m centered on the test site, that corresponds to around 500 ×500 m 2 or 5 ×5 native 100-m pixels. The array size was selected as a compromise between having sufficient number of pixels and preserving the thermal homogeneity of the selected area. Since the NEΔT is purely random, its uncertainty contribution to the averaged LST must by divided by  N √(Ghent et al., 2019), with N =17 ×17 being the number of samples (SD of the mean). Note that error sources (i) and (ii) are not random but locally correlated, and thus the uncertainties are not divided by  N √(Ghent et al., 2019). (iv) Finally, a fourth error contribution was taken as the SD of the calculated LSTs over the 17 ×17 pixel array to account for the LST spatial variability. The four error contributions were added together in quadrature to obtain the total SC LST uncertainty, which ranged from 0.6 to 0.8 K for band 10, and from 1.2 to 1.3 K for band 11, and with average uncertainties of 0.69 ±0.05 K and 1.24 ± 0.03 K for bands 10 and 11, respectively. Guo et al. (2020) used the SC algorithm proposed by Jimenez-Mu˜ noz et al. (2009; 2014) to retrieve LSTs from TOA L i and T i in L8 TIRS band 10 and 11 and evaluate the stray-light correction effect in terms of LSTs. This SC algorithm uses parametric atmospheric terms as function of the total water vapor content in the atmosphere (W), which implies a simplification for users and has been used for multiple applications (e.g., for estimating evapotranspiration using L8 data (Mokhtari et al., 2019; Olivera-Guerra et al., 2017)). However, it involves higher uncertainties, which are not proper for calibration studies. Yu et al. (2014) compared the results obtained with the RTE (i.e., inverting eq (1)-(2)) and those when using this SC algorithm, concluding that the LSTs estimated from the former was more accurate than the latter. Jim´ enez-Mu˜ noz et al. (2014) quantified the effect of the uncertainty in W, as one of the main input variables in their SC algorithm (together with emissivity), in terms of the retrieved LSTs and estimated differences up to 1 K for band 10, which implies much larger LST uncertainties than those above estimated for the RTE, since LST uncertainties are due to both input data uncertainties and those from the SC algorithm regression. They also tested the algorithm results for band 10 with an independent simulated database and obtained a RMSD of 2 K (1.5 K for W <3 cm). 4. Split-window algorithms for L8 TIRS Several split-window (SW) algorithms were proposed or adjusted for the L8 TIRS spectral bands in the literature (Jimenez-Mu˜ noz et al., 2014; Yu et al. 2014; Rozenstein et al. 2014; Du et al., 2015; Meng et al. 2019; Guo et al. 2020; Gerace et al. 2020). In this section, we describe five SW algorithms with different estimates of the algorithm regression coefficients, leading a total of seventeen different SW equations, which will be later validated. Jimenez-Mu˜ noz et al. (2014) followed the expression (hereafter referred to as JM): T=T10 +c0+c1(T10 −T11)+c2(T10 −T11)2+(c3+c4W)(1− ε )+(c5 +c6W)Δ ε (3) where ε and Δ ε are the average and the difference of ε i for the TIRS bands 10 and 11, respectively, and T, T 10 and T 11 are the LST and TOA T i for the TIRS bands. W is again the total water vapor content in the atmosphere. The c j coefficients were obtained by regression of the synthetic TOA T i simulated with the MODTRAN 4 radiative transfer code (Beck et al., 1999) using the Global Atmospheric Profiles from Reanalysis Information (GAPRI). The GAPRI database is a compilation of selected atmospheric profiles at a global scale derived from ERA-Interim reanalysis data during 2011 (Mattar et al., 2015). Similarly, Meng et al. (2019) estimated the c j coefficients in eq. (3) using the MODTRAN 5 code and the GAPRI database. In this case, the coefficients were estimated per W subranges, and also globally for any W between 0 cm and 7 cm. Table 1 shows the c j in eq. (3) obtained by Jimenez-Mu˜ noz et al. (2014) (JM) and Meng et al. (2019) (JM_M) for the subranges of W that cover the actual atmospheric conditions in our site. The algorithm regression uncertainties, δ alg , provided by the corresponding papers and defined as standard errors of estimate (Jimenez-Mu˜ noz et al., 2014) or root-mean-square errors of the predicted T (Meng et al., 2019), are also shown in the last column of Table 1. Thus, δ alg is the uncertainty obtained with algorithm regression, which minimizes the sum of squared deviations of predictions from actual scores on the simulation database, and takes into account the variance of the residuals. Meng et al. (2019) also derived the coefficients in other two SW algorithms for the TIRS bands using MODTRAN 5 and the GAPRI database. They calculated the c j in the generalized SW algorithm designed by Wan and Dozier (1996) and refined by Wan (2014) by adding an additional term related to the quadratic difference between both TOA T i . This algorithm is given as follows: T=c0+(c1+c2 1− ε ε +c3 Δ ε ε 2)T10 +T11 2+(c4+c5 1− ε ε +c6 Δ ε ε 2)T10 −T11 2+c7(T10 −T11)2(4) The algorithm given in eq. (4) will be hereafter referred to as W&D, and W&D_M when the c j are those derived by Meng et al. (2019) for TIRS. Du et al. (2015) previously also derived the c j in eq. (4) for TIRS (hereafter W&D_D) using MODTRAN 5 and the Thermodynamic Initial Ground measured emissivity Ground surface and atmospheric TIR radiance measurements Ground LST NCEP atmospheric profiles MODTRAN radiative transfer modelAtmospheric transmittances and radiances Reference TOA radiances and brightness temperatures L8 TIRS TOA radiances and brightness temperatures (17x17 pixels & 1 pixel) Single-channel LST Split-window LST ETRETR NCEP W Ground emissivities EVALUATION OF L8 TIRS LST RETRIEVALS EVALUATION OF L8 TIRS CALIBRATIONS Selection of optimal mathchups (Rbased method) Fig. 3. Flow chart of the procedure followed for evaluating both the successive L8 TIRS recalibrations and the LSTs retrieved from L8 TIRS data. R. Nicl` os et al. ISPRS Journal of Photogrammetry and Remote Sensing 174 (2021) 237–253 242 Guess Retrieval (TIGR) atmospheric profile database. Guo et al. (2020) refined the coefficients as Zheng et al. (2019) did for Sentinel-3A SLSTR, dividing the simulation data into temperature subranges. In this case, data were divided into several subranges in terms of T 10 , additionally to the W subranges. This version will be referred to as W&D_D_W_T. They also increased the LST minus bottom air temperature maximum difference in the simulations from 20 K (in Du et al. (2015)) to 35 K, considering that it was more probable for barren or desert surfaces. Gerace et al. (2020) derived the c j in eq. (4) again (hereafter as W&D_G). Their purpose was proposing a SW algorithm for a future operational SW LST product generated from L8 and L9 TIRS image data in the Collection 3 processing. They used 1393 TIGR atmospheric profiles (after filtering those where relative humidity were higher than 90%), 7 surface temperatures for each of them, 113 emissivity values obtained from the MODIS UCSB database (not using the man-made materials, unlike Du et al. (2015)), and MODTRAN simulations. Table 2 includes the coefficients c j in eq. (4) obtained by Du et al. (2015), Meng et al. (2019) and Guo et al. (2020) for the different subranges of W content, and temperatures for the latter, that cover the actual conditions in the validation site. The constant c j derived by Gerace et al. (2020) are also included in Table 2. The corresponding δ alg values (K), defined as the root-mean-square errors in these papers, are also shown. Meng et al. (2019) also derived the coefficients in the NOAA JPSS Enterprise algorithm for TIRS and W subranges (hereafter Ent_W) (Guillevic et al., 2018; Yu et al., 2017): T=c0+c1T10 +c2(T10 −T11)+c3 ε +c4 ε (T10 −T11)+c5Δ ε (5) Table 3 includes the c j in eq. (5), and the δ alg values, obtained for the subranges of W that cover the atmospheric conditions in the validation site. Qin et al. (2001) presented a SW technique for the AVHRR dependent on what they defined as two essential variables: ε i and τ i. Yu et al. (2014) proposed an algorithm for TIRS based on that of Qin et al. (2001). It relied on the determination of a parameter p i (with dimensions of temperature, in K) for the TIRS spectral bands using linear regression equations with the TIRS TOA T i : pi=ai+bi⋅Ti(6) where a i and b i are the regression coefficients. The following equations (7)-(13) shows the Yu et al. (2014) algorithm (hereafter referred to as Yu): T=T10 +B1⋅(T10 −T11)+B0(7) B0=C11⋅(1−A10⋅C10)⋅p10 −C10⋅(1−A11 −C11)⋅p11 C11⋅A10 −C10⋅A11 (8) B1=C10 C11⋅A10 −C10⋅A11 (9) A10 = ε 10⋅ τ 10 (10) A11 = ε 11⋅ τ 11 (11) C10 = (1− τ 10)⋅(1+(1− ε 10)⋅ τ 10)(12) C11 = (1− τ 11)⋅(1+(1− ε 11)⋅ τ 11)(13) Term B 0 in eq. (8) is explicitly dependent on p i (obtained with the regression coefficients a i and b i given in Table 4). Yu et al. (2014) also provided quadratic regressions between τ i and W for two standard atmospheres: a mid-latitude summer profile and a 1976 US Standard profile. Rozenstein et al. (2014) proposed another SW algorithm based on that of Qin et al. (2001) for TIRS bands. Table 4 shows the regression coefficients in eq. (6) provided by Rozenstein et al. (2014) for the ranges of temperature in the study database. However, the algorithm proposed by Rozenstein et al. (2014) do not use the p i parameter, but just the a i and b i coefficients. The following equations (14)-(23) provide the Rozenstein et al. (2014) algorithm (hereafter referred to as Ro): T=A0+A1⋅T10 −A2⋅T11 (14) A0=E1⋅b10 −E2⋅b11 (15) A1=1+A+E1⋅a10 (16) A2=A+E2⋅a11 (17) A=D10/E0(18) E1=D11⋅(1−C10 −D10)/E0(19) E2=D10⋅(1−C11 −D11)/E0(20) E0=D11⋅C10 −D10⋅C11 (21) Ci= ε i⋅ τ i(θ)(22) Table 1 Coefficients in the SW algorithm given in eq. (3) provided by Jimenez-Mu˜ noz et al. (2014) (JM) and Meng et al. (2019) (JM_M) for different W subranges. Reference W (cm) c 0 c 1 c 2 c 3 c 4 c 5 c 6 δ alg (K) Jimenez-Mu˜ noz et al. (2014) (JM) 0.0–6.0 −0.268 1.378 0.183 54.30 −2.238 −129.20 16.40 0.6 Meng et al. (2019) (JM_M) 0.0–2.5 −0.39 2.116 −0.045 64.386 −3.7 −147.522 21.065 0.431 2.0–3.5 −1.631 2.681 −0.054 67.827 −3.213 −204.953 41.441 0.503 0.0–7.0 −0.717 1.988 0.121 70.148 −7.006 −143.246 19.247 0.72 Table 2 Coefficients in the SW algorithm given in eq. (4) provided by Du et al. (2015), Meng et al. (2019), Guo et al. (2020) and Gerace et al. (2020). Reference W (cm) T 10 (K) c 0 c 1 c 2 c 3 c 4 c 5 c 6 c 7 δ alg (K) Du et al. (2015) (W&D_D) 0.0–2.5 – −2.78009 1.01408 0.15833 −0.34991 4.04487 3.55414 −8.88394 0.09152 0.34 2.0–3.5 – 11.00824 0.95995 0.17243 −0.28852 7.11492 0.42684 −6.62025 −0.06381 0.6 0.0–6.3 – −0.41165 1.00522 0.14543 −0.27297 4.06655 −6.92512 −18.27461 0.24468 0.87 Meng et al. (2019) (W&D_M) 0.0–2.5 – −1.56 1.007 0.162 −0.288 3.179 6.864 −11.209 0.165 0.44 2.0–3.5 – −0.099 0.998 0.148 −0.252 5.236 5.488 −5.455 0.02 0.57 0.0–7.0 – −2.64 1.012 0.142 −0.201 2.844 −0.569 −7.6 0.263 0.844 Guo et al. (2020) (W&D_D_W_T) 0.0–2.5 270–300 1.6214 0.9968 0.1739 −0.3965 4.3444 5.6164 12.8573 −0.1175 0.30 0.0–2.5 300–330 7.3937 0.9788 0.1917 −0.3384 3.0247 3.2533 −14.4977 0.1291 0.30 2.0–3.5 <300 24.913 0.911 0.174 −0.299 6.351 3.92 −5.582 −0.0640 0.54 2.0–3.5 ≥300 27.467 0.904 0.187 −0.349 5.675 2.842 −7.853 0.023 0.58 Gerace et al. (2020) ((W&D_G) – – 2.2925 0.9929 0.1545 −0.3122 3.7186 0.3502 −3.5889 0.1825 0.73 R. Nicl` os et al. ISPRS Journal of Photogrammetry and Remote Sensing 174 (2021) 237–253 243 Di= [1− τ i(θ)]⋅[1+(1− ε i)⋅ τ i(θ)] (23) The negative sign in eq. (15) was considered following the correction published by the authors. Rozenstein et al. (2014) also provided linear regressions between τ i and W for two standard atmospheres: a midlatitude summer profile and a 1976 US Standard profile, and recommend general users to use the regression coefficients for the second one. Estimates of W are required for the algorithms and they were obtained from the four nearest NCEP profiles by interpolating them spatially for the site and temporally for each L8 TIRS scene acquisition time. The ε i measured for each land cover condition at the site (described in section 2) were used as input values in the SW algorithms. Previous studies used pixel ε i estimates based mainly on the ASTER global emissivity dataset (ASTER-GED) (Hulley et al., 2015; García-Santos, et al., 2018; Gerace et al. 2020), which includes a worldwide mean emissivity climatology for the period 2000–2008, or threshold methods in terms of the Normalized-Difference Vegetation Index (Yu et al., 2014; Rozenstein et al., 2014; Meng et al., 2019; Guo et al., 2020) for validating some of the SW algorithms. Recently, an alternative method for estimating TIRS ε i using neural networks trained on OLI surface reflectances and the ECOSTRESS spectral library (Meerdink et al., 2019) was proposed by Vanhellemont (2020), although they did not evaluate the method in estimating input ε i in SW algorithm results. However, when possible, a good knowledge of the study site and in-situ ε i measurements of representative ground covers that concur with satellite overpasses are desired (Rozenstein et al., 2014), since land ε i are variable due to seasonal effects, land use variability, and changes in soil moisture content (Vanhellemont, 2020; Hongtao et al., 2019; Koley and Jeganathan, 2020), and they also depend on the site specific surface elements (e.g., vegetation species and soil types). 5. Results and discussion 5.1. Evaluation of TIRS calibrations This section shows the results of the evaluations of the successive TIRS calibrations, i.e., using data after the 2014 and 2017 reprocessings and with the changes in TIRS bands response introduced by the 2019 calibration update. The evaluation consisted of comparing L8 TIRS TOA L i and T i with those retrieved using the RTE (eq. (1)-(2)) from ground T and ε i (as shown in Fig. 3). First, and for comparison purposes, the calibration of the TIRS data acquired and downloaded in 2014 before the 2014 reprocessing (i.e., original or pre-correction data) was evaluated. The reference ground LSTs acquired this year covered a significant range (from around 285 K to 314 K) due to land cover changes at the site. The first row in Table 5 shows the statistical results of the TIRS minus ground data differences per band in terms of TOA L i and T i for the array of 17 ×17 pixels and only 1 pixel over the experimental site. Biases and SD of the differences, and root-mean-square differences (RMSD), are shown. Close results were obtained for 17 ×17 pixels and 1 pixel due to the thermal homogeneity of the experimental site, which was also proven in previous papers as described in Section 2 (Coll et al., 2005; Nicl` os et al., 2018). Fig. 4a shows the TOA T i simulated from ground LSTs against the TIRS TOA T i (average values for 17 ×17 pixels) for bands 10 and 11. The uncertainties in TIRS TOA L i and T i were calculated by adding in quadrature the SD of TIRS data in 17 ×17 pixels and the TIRS NEΔT of 0.4 K for both bands, which was divided by  N √, N =17 ×17 (Ghent et al., 2019). The resulting TOA T i uncertainties ranged between 0.1 and 0.4 K for both bands. Such small uncertainty values, which account mainly for the spatial variability of TIRS T i , show the thermal homogeneity of the test site at the satellite scale. The uncertainties in the TOA L i and T i simulated from ground LSTs were obtained by addition in quadrature of the three uncertainty components resulting from the propagation in eq. (1)-(2) of the input variables. They were: (i) the uncertainty in the ground-measured temperatures, (ii) the emissivity uncertainty for each surface in the experimental site, and (iii) uncertainties in atmospheric radiances and transmittances as obtained from a radiative transfer model, in a similar way as in section 3. According to this, uncertainties in simulated T i ranged between 0.7 and 0.9 K for band 10 and between 1.0 and 1.2 K for band 11. As shown in Fig. 4a and Table 5, TOA L i and T i originally provided by TIRS overestimated those simulated from ground LSTs at the experimental site, confirming the calibration problem mainly for band 11. However, the observed biases at the rice paddy site (in the first row of Table 5) were much lower than those proposed to be subtracted by the L8 calibration team in 2014 (0.29 ±0.12 Wm -2 sr -1 µm −1 (2.1 ±0.8 K) and 0.5 ±0.2 Wm -2 sr -1 µm −1 (4.4 ±1.8 K) for bands 10 and 11, respectively). Fig. 4b and the second row in Table 5 show the evaluation results when using TIRS data after the 2014 reprocessing downloaded for the same acquisition dates in 2014. Since the overestimation of the originally acquired TIRS data was lower than that observed by the L8 Calibration Team using reference data at water sites, the reprocessing applied in 2014 on the TIRS data was not proper for our experimental Table 3 Coefficients in the SW algorithm given in eq. (5) provided by Meng et al. (2019) for different W subranges (Ent_M). Reference W (cm) c 0 c 1 c 2 c 3 c 4 c 5 δ alg (K) Meng et al. (2019) (Ent_M) 0.0–2.5 54.95 1.01 1.557 −57.805 0.147 −103.52 0.481 2.0–3.5 50.035 1.006 5.377 −52.801 −3.16 −87.906 0.589 0.0–7.0 67.297 0.985 −6.916 −63.855 9.548 −90.919 1.075 Table 4 Regression coefficients in eq. (6) to obtain the parameter p i according to Yu et al. (2014) and Rozenstein et al. (2014), per ranges of temperature. Reference T i range (◦C) a 10 (K) b 10 a 11 (K) b 11 Yu et al. (2014) −10–20 −55.58 0.4087 −59.85 0.4442 20–50 −66.61 0.4464 −71.23 0.4831 Rozenstein et al. (2014) 10–40 −62.8065 0.4338 −67.1728 0.4694 Table 5 Statistical results, in terms of bias ±SD (RMSD), of the TIRS minus ground data differences per band in terms of TOA L i and T i when using the original TIRS data acquired before reprocessing, and those after the 2014 and 2017 reprocessings. Results are shown both for an array of 17 ×17 pixels (resampled to 30-m) centered on the test site and only 1 pixel at the site. L 10 (W m −2 sr -1 µm −1 ) T 10 (K) L 11 (W m −2 sr -1 µm −1 ) T 11 (K) Original data array 0.09 ±0.10 (0.13) 0.7 ±0.7 (1.0) 0.33 ±0.13 (0.35) 2.7 ±1.0 (2.9) pixel 0.08 ±0.10 (0.13) 0.6 ±0.7 (0.9) 0.32 ±0.12 (0.34) 2.7 ±1.0 (2.8) 2014 reprocessed data array −0.20 ± 0.10 (0.22) −1.4 ± 0.7 (1.6) −0.18 ± 0.13 (0.22) −1.6 ± 1.1 (2.0) pixel −0.21 ± 0.10 (0.24) −1.5 ± 0.7 (1.7) −0.19 ± 0.12 (0.23) −1.7 ± 1.1 (2.0) 2017 reprocessed data array 0.02 ±0.17 (0.17) 0.2 ±1.2 (1.2) 0.01 ±0.18 (0.18) 0.1 ±1.6 (1.6) pixel 0.00 ±0.17 (0.17) 0.1 ±1.2 (1.2) 0.01 ±0.18 (0.18) 0.1 ±1.5 (1.5) R. Nicl` os et al. ISPRS Journal of Photogrammetry and Remote Sensing 174 (2021) 237–253 244 site and yielded similar underestimations (of around −1.5 K) for both bands. Results for band 10, used for SC LST retrievals by users, were worse than before this reprocessing. Fig. 4c and the third row of Table 5 show the results when data after the reprocessing in 2017 were used. The 2014 scenes were again downloaded after this new reprocessing, but additionally we used scenes acquired concurrently with ground data measured along transects in 2016, 2017, 2018 and 2019 to extend the evaluation of the currently available L8 TIRS data. A total of 25 scenes acquired at the site for different seasonal land covers were used for it, covering ground LSTs ranging from 282 K to 316 K. The resulting TIRS TOA T i uncertainties ranged between 0.1 and 0.5 K for both bands and the complete period, and the uncertainties in TOA T i simulated from ground LSTs ranged again between 0.7 and 0.9 K for band 10 and between 1.0 and 1.2 K for band 11 for this dataset. Unlike in the 2014 reprocessing, when just bias subtracting was applied, the reprocessing in 2017 used the TIRS-on-TIRS algorithm that allowed modeling and eliminating the out-of-field radiance, thus using a custom correction for each scene and pixel. Results in Fig. 4c and Table 5 shows that this correction improved the results in the currently available TIRS data. Negligible bias and SDs (RMSDs) of 0.17 W m −2 sr -1 µm −1 and 0.18 W m −2 sr -1 µm −1 (1.2 K and 1.6 K) were obtained for bands 10 and 11, respectively. Fig. 5a-b show the 2017 calibration results for both TIRS bands by type of surface cover at the experimental site in order to analyze possible land cover dependences. The largest differences were mainly observed for the dry bare soil condition. This fact could be due to the lowest homogeneity observed in previous studies for this surface cover at the site (Nicl` os et al., 2018). However, these largest differences corresponds to the highest LSTs (>307 K) and they could be also a consequence of this fact. Results were also analyzed by scene, since the TIRS-on-TIRS algorithm used in the 2017 reprocessing should not be dependent on scene. Fig. 5c-d shows the results split in L8 paths 198 and 199. Path 198 covers a larger proportion of sea than path 199, as shown in Fig. 1. There are differences in the results between paths 198 and 199 of around 0.8 K in terms of SDs, but not in terms of bias. However, since the number of matchups are quite different for both paths, with just 8 matchups for the 198 and 17 matchups for the 199, the observed SD differences can be just consequence of this fact. For some cases in Figs. 4-5, the comparison between simulated and satellite T i shows differences larger than the combined uncertainty of both measurements. The largest source of error in the simulated T i is the radiative transfer calculations. In some cases, the atmospheric profiles (a) (b) (c) y = 1.00x - 0.77 R² = 0.99 y = 0.95x - 1.54 R² = 0.98 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45 T i simulated from ground LST (°C) L8 TIRS T i (°C) band 10 band 11 y = 0.99x + 1.76 R² = 0.99 y = 0.92x + 3.24 R² = 0.98 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45 T i simulated from ground LST (°C) L8 TIRS T i (°C) band 10 band 11 y = 1.05x - 1.21 R² = 0.97 y = 1.04x - 0.89 R² = 0.95 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45 T i simulated from ground LST (°C) L8 TIRS T i (°C) Band 10 Band 11 Fig. 4. Comparison of TOA T i simulated from ground LST and acquired by L8 TIRS to evaluate the successive recalibrations, using TIRS data: a) before reprocessing (original data), b) after the 2014 reprocessing, and c) after the 2017 reprocessing (currently available data in the Landsat Collection 1 L8 database). a) and b) with the available data of 2014, and c) with matchups from 2014 to 2019. The straight lines show the linear regression of the data, with their fitting equations and coefficient of determination (R 2 ). R. Nicl` os et al. ISPRS Journal of Photogrammetry and Remote Sensing 174 (2021) 237–253 245 used may differ from the actual atmospheric profile over the site at the overpass time more than we assumed in the sensitivity analysis. We can check the suitability of the atmospheric profiles by comparing the L8 TIRS observed T i differences, (T 10 -T 11 ) obs with the same T i difference as obtained from the simulation from ground data, (T 10 -T 11 ) sim . Then the difference δ(T 10 -T 11 ) =(T 10 -T 11 ) obs - (T 10 -T 11 ) sim can be analyzed; a concept that is related with the R-based validation method (Wan and Li, 2008; Coll et al., 2012). Note that in the two bands, atmospheric absorption is mostly due to the same mechanism (water vapor continuum), which is stronger in band 11, so T 10 -T 11 is usually positive and increases with W. Then if the atmospheric profile is close to the actual profile, the difference δ(T 10 -T 11 ) should be close to zero. However, if the profile used is overestimating (underestimating) the atmospheric effect, the value of (T 10 -T 11 ) sim will be larger (smaller) than (T 10 -T 11 ) obs and therefore δ(T 10 -T 11 ) <0 (>0). This can be seen in Fig. 6, where the differences between T 10 simulated from ground data and TIRS observed T 10 are plotted against δ(T 10 -T 11 ) for the whole dataset. The cases where δ(T 10 -T 11 ) is small (in absolute value) are associated with small absolute values of temperature differences, the opposite being also true. Thus, we can use the δ(T 10 -T 11 ) difference as a check of the atmospheric profile quality and consequently of the accuracy of the T i simulation. To do that, we can set a threshold on δ(T 10 -T 11 ) based on an uncertainty analysis applied to the temperature differences plotted in Fig. 6 using the results of the preceding uncertainty analyses for the individual temperatures (a) (b) (c) (d) 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45 T i simulated from ground LST (°C) L8 TIRS T i (°C) band 10 flooded soil (water) full vegetation cover dry bare soil wet bare soil 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45 T i simulated from ground LST (°C) L8 TIRS T i (°C) band 11 flooded soil full vegetation cover dry bare soil wet bare soil 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45 T i simulated from ground LST (ºC) L8 TIRS T i (ºC) band 10 199 198 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45 T i simulated from ground LST (ºC) L8 TIRS T i (ºC) band 11 199 198 Fig. 5. Comparison of TOA brightness temperatures (for bands 10 and 11) simulated from ground LST and acquired by L8 TIRS after the reprocessing in 2017 split by: (a)-(b) land cover at the site, and (c)-(d) acquisition path (198 and 199). y = 2.86x - 0.17 R² = 0.74 -3 -2 -1 0 1 2 3 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 simulated T10 - L8 TIRS T10 (K) (T10-T11) (K) Fig. 6. Differences between T i simulated from ground data and TIRS observed T i at band 10 against the differences between the observed and simulated T i difference for both bands, δ(T 10 -T 11 ). The straight line is the linear regression of the data, with the fitting equation and coefficient of determination (R 2 ) shown in the graph. Open symbols represent cases not satisfying the condition −0.5 K <δ(T 10 -T 11 ) <0.5 K. R. Nicl` os et al. ISPRS Journal of Photogrammetry and Remote Sensing 174 (2021) 237–253 252 W&D_G for our site. Therefore, the abovementioned three SW algorithms are recommended for the users of current TIRS data to retrieve LST; being the one proposed by Gerace et al. (2020), for an operational SW product in the future Collection 3, the only one without W as input variable and thus more simple to apply. Finally, the results when the calibration update proposed in December 2019 was applied to the TIRS data showed biases in TOA T i of −0.08 K and 0.5 K for bands 10 and 11, respectively, thus introducing a calibration response difference between both bands. The calibration update improved slightly the LSTs obtained for our site with the SC correction applied to band 10 data, for which the RMSD was the lowest (1.5 K; or 0.7 K after removing doubtful matchups). A positive difference was observed between the reported biases for the SW algorithms applied to 2017-reprocessed TIRS data and those biases obtained for the algorithms when using the TIRS dataset after the 2019 update correction. The differences in biases were from 1.3 K to 1.9 K for all the algorithms, with a difference of 1.4 K for the JM, W&D_G and Ent_M_W algorithms, and no significant differences were observed in SDs. These results could mean that the best SW algorithms shown for the current TIRS data in Collection 1 would likely provide a negative bias at our site when applied to TIRS data in Collection 2. In any case, a new evaluation would be advisable when the Collection 2 data will be finally available for users, i.e., after the operational application of the calibration update in the L8 TIRS data processing system. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements This study was funded by the Spanish Ministry of Economy and Competitiveness and the European Regional Development Fund (FEDER) through the project CGL2015-64268-R (MINECO/FEDER, UE). A researcher was hired for it with the support of a grant of the regional program for training of technicians for R&D&i (Youth Guarantee GJIDI2018-A-142). The validation campaigns in 2014 and 2016 were also funded by the Spanish Ministry of Economy and Competitiveness under the projects CGL2011-30433-C02-02 and CGL2013-46862-C2-1-P. References Anderson, M.C., Allen, R.G., Morse, A., Kustas, W.P., 2012. Use of Landsat thermal imagery in monitoring evapotranspiration and managing water resources. Rem. Sens. Environ. 122, 50–65. Baldridge, A.M., Hook, S.J., Grove, C.I., Rivera, G., 2009. The ASTER spectral library version 2.0. Remote Sens. Environ. 113, 711–715. https://doi.org/10.1016/j. rse.2008.11.007. Barsi, J.A., Schott, J.R., Hook, S.J., Raqueno, N.G., Markham, B.L., Radocinski, R.G., 2014. Landsat-8 Thermal Infrared Sensor (TIRS) Vicarious Radiometric Calibration. Remote Sens. 6 (11), 11607–11626. Beck, A., Anderson, G.P., Acharya, P.K., Chetwynd, J.H., Bernstein, L.S., Shettle, E.P., Matthew, M.W., Adler-Golden, S.M., 1999. MODTRAN4 User’s Manual. Air Force Res. Lab, Hanscom AFB, MA, USA. Berk, A., Anderson, G.P., Acharya, P.K., Bernstein, L.S., Muratov, L., Lee, J., Fox, M., Adler-Golden, S.M., Chetwynd, J.H., Hoke, M.L., Lockwood, R.B., Gardner, J.A., Cooley, T.W., Borel, C.C., Lewis, P.E., Shettle, E.P., 2006. MODTRAN5: 2006 Update. Proc. SPIE 6233 (62331F). Bojinski, S., Verstraete, M., Peterson, T., Richter, C., Simmons, A., Zemp, M., 2014. The Concept of Essential Climate Variables in Support of Climate Research, Applications, and Policy. Bull. Amer. Meteor. Soc. 95 (9), 1431–1443. Brogniez, G., Pietras, C., Legrand, M., Dubuisson, P., Haeffelin, M., 2003. A high accuracy multiwavelength radiometer for in situ measurements in the thermal infrared. Part II: Behavior in field experiments. J. Atmos. Oceanic Technol. 20, 1023–1033. Coll, C., Caselles, V., Galve, J.M., Valor, E., Nicl` os, R., S´ anchez, J.M., Rivas, R., 2005. Ground measurements for the validation of land surface temperatures derived from AATSR and MODIS data. Rem. Sens. Environ. 97, 288–300. Coll, C., Caselles, V., Valor, E., Nicl` os, R., 2012. Comparison between different sources of atmospheric profiles for land surface temperature retrieval from single channel thermal infrared data. Rem. Sens. Environ. 117, 199–210. Coll, C., Nicl` os, R., Puchades, J., García-Santos, V., Galve, J.M., P´ erez-Planells, L.l., Valor, E., Theocharous, E., 2019. Laboratory calibration and field measurement of land surface temperatura and emissivity using thermal infrared multiband radiometers. Int. J. Appl. Earth Obs. Geoinformation 78, 227–239. Du, C., Ren, H., Qin, Q., Meng, J., Zhao, S., 2015. A practical split-window algorithm for estimating land surface temperature from landsat 8 data. Remote Sens. 7, 647–665. England, M.N., 2019. Landsat 8: TIRS SSM Encoder Current Anomaly Resolution. In: 2019 IEEE Aerospace Conference, Big Sky, pp. 1–10. https://doi.org/10.1109/ AERO.2019.8741692. García-Santos, V., Cuxart, J., Martínez-Villagrasa, D., Jim´ enez, M., Sim´ o, G., 2018. Comparison of three methods for estimating land surface temperature from Landsat 8-TIRS sensor data. Remote Sens. 10, 1450. GCOS-200. The Global Observing System for Climate: Implementation Needs, 342PP. WMO, 2016. https://library.wmo.int/opac/doc_num.php?explnum_id=3417 (Accessed 3 August, 2020). Gerace, A., Montanaro, M., 2017. Derivation and validation of the stray light correction algorithm for the thermal infrared sensor onboard Landsat 8. Rem. Sens. Environ. 191, 246–257. Gerace, A., Kleynhans, T., Eon, R., Montanaro, M., 2020. Towards an Operational, SplitWindow-Derived Surface Temperature Product for the Thermal Infrared Sensors Onboard Landsat 8 and 9. Remote Sens. 12, 224. Ghent, D., Kaduk, J., Remedios, J., Ard¨ o, J., Balzter, H., 2010. Assimilation of land surface temperature into the land surface model JULES with an ensemble Kalman filter. J. Geophys. Res. 115 (D19), 112. Ghent, D., Veal, K., Trent, T., Dodd, E., Sembhi, H., Remedios, J., 2019. A new approach to defining uncertainties for MODIS land surface temperature. Remote Sens. 11, 1–24. Gillespie, A.R., Matsunaga, T., Rokugawa, S., Hook, S.J., 1998. Temperature and emissivity separation from Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) images. IEEE Trans. Geosci. Rem. Sens. 36, 1113–1125. Guillevic, P.C., Privette, J.L., Coudert, B., Palecki, M.A., Demarty, J., Ottl´ e, C., Augustine, J.A., 2012. Land Surface Temperature product validation using NOAA’s surface climate observation networks – scaling methodology for the Visible Infrared Imager Radiometer Suite (VIIRS). Rem. Sens. Environ. 124, 282–298. Guillevic, P., G¨ ottsche, F., Nickeson, J., Hulley, G., Ghent, D., Yu, Y., Trigo, I., Hook, S., Sobrino, J.A., Remedios, J., Rom´ an, M., Camacho, F., 2018. Land surface temperature product validation best practice protocol version 1.1. Best Pract. Satell. L. Prod. Valid. (p. 60) L. Prod. Valid. Subgr. doi, 58. https://doi.org/10.5067/doc/ ceoswgcv/lpv/lst.001. Guo, J., Ren, H., Zheng, Y., Lu, S., Dong, J., 2020. Evaluation of Land Surface Temperature Retrieval from Landsat 8/TIRS Images before and after Stray Light Correction Using the SURFRAD Dataset. Remote Sens. 12, 1023. Hongtao, J., Huanfeng, S., Xinghua, L., Chao, Z., Huiqin, L., Fangni, L., 2019. Extending the SMAP 9-km soil moisture product using a spatio-temporal fusion model. Rem. Sens. Environ. 231, 111224. Hook, S.J., Chander, G., Barsi, J.A., Alley, R.E., Abtahi, A., Palluconi, F.D., Markham, B. L., 2004. In-flight validation and recovery of water surface temperature with Landsat-5 thermal infrared data using an automated high-altitude lake validation site at Lake Tahoe. IEEE Trans. Geosci. Remote Sens. 42 (12), 2767–2776. Hulley, G.C., Hook, S.J., 2009. The North American ASTER Land Surface Emissivity Database (NAALSED). Version 2.0. Remote Sens. Environ. 113, 1967–1975. Hulley, G.C., Hook, S.J., Abbott, E., Malakar, N., Islam, T., Abrams, M., 2015. The ASTER Global Emissivity Dataset (ASTER GED): Mapping Earth’s emissivity at 100 meter spatial scale. Geophys. Res. Lett. 42 (19), 7966–7976. Irons, J.R., Dwyer, J.L., Barsi, J.A., 2012. The next Landsat satellite: The Landsat data continuity mission. Remote Sens. Environ. 122, 11–21. Jim´ enez-Mu˜ noz, J.-C., Crist´ obal, J., Sobrino, J.A., Soria, G., Ninyerola, M., Pons, X., 2009. Revision of the Single-Channel Algorithm for Land Surface Temperature Retrieval From Landsat Thermal-Infrared Data. IEEE Trans. Geosci. Remote Sens. 47, 339–349. Jim´ enez-Mu˜ noz, J.-C., Sobrino, J.A., Skokovi c, D., Mattar, C., Crist´ obal, J., 2014. Land Surface Temperature Retrieval Methods From Landsat-8 Thermal Infrared Sensor Data. IEEE Geosci. Remote Sens. Lett. 11, 1840–1843. Koley, S., Jeganathan, C., 2020. Estimation and evaluation of high spatial resolution surface soil moisture using multi-sensor multi-resolution approach. Geoderma 378, 114618. Krishnan, P., Meyers, T.P., Hook, S.J., Heuer, M., Senn, D., Dumas, E.J., 2020. Intercomparison of In Situ Sensors for Ground-Based Land Surface Temperature Measurements. Sensors 20, 5268. Legrand, M., Pietras, C., Brogniez, G., Haeffelin, M., Abuhassan, N.K., Sicard, M., 2000. A high-accuracy multiwavelength radiometer for in situ measurements in the thermal infrared. Part I: characterization of the instrument. J. Atmos. Ocean Techn. 17, 1203–1214. Malakar, N., Hulley, G., Hook, S., Laraby, K., Cook, M., Schott, J., 2018. An Operational Land Surface Temperature Product for Landsat Thermal Data: Methodology and Validation. IEEE Trans. Geosci. Remote Sens. 2018 (56), 5717–5735. Mattar, C., Dur´ an_Alarc´ on, C., Jim´ enez-Mu˜ noz, J.C., Santamaría-Artigas; A., OliveraGuerra, L. Sobrino, J.A., 2015. Global Atmospheric Profiles from Reanalysis Information (GAPRI): a new database for earth surface temperature retrieval. Int. J. Remote Sens. 36, 5045-5060. Meerdink, S.K., Hook, S.J., Roberts, D.A., Abbott, E.A., 2019. The ECOSTRESS spectral library version 1.0. Remote Sens. Environ. 230, 111196. R. Nicl` os et al. ISPRS Journal of Photogrammetry and Remote Sensing 174 (2021) 237–253 253 Meng, X., Cheng, J., Zhao, S., Liu, S., Yao, Y., 2019. Estimating Land Surface Temperature from Landsat-8 Data using the NOAA JPSS Enterprise Algorithm. Remote Sens. 11, 155. Mokhtari, A., Noory, H., Pourshakouri, F., Haghighatmehr, P., Afrasiabian, Y., Razavi, M., Fereydooni, F., Naeni, A.S., 2019. Calculating potential evapotranspiration and single crop coefficient based on energy balance equation using Landsat 8 and Sentinel-2. ISPRS J. Photogramm. Remote Sens. 154, 231–245. Montanaro, M., Gerace, A., Lunsford, A., Reuter, D., 2014. Stray Light Artifacts in Imagery from the Landsat 8 Thermal Infrared Sensor. Remote Sens. 11, 10435–10456. Montanaro, M., Levy, R., Markham, B., 2014b. On-Orbit Radiometric Performance of the Landsat 8 Thermal Infrared Sensor. Remote Sens. 6, 11753–11769.Montanaro, M., Gerace, A., Rohrbach, S, 2015. Toward an operational stray light correction for the Landsat 8 thermal infrared sensor. Appl. Opt. 54 (13), 3963–3978. Montanaro, M., Gerace, A., Rohrbach, S., 2015. Toward an operational stray light correction for the Landsat 8 Thermal Infrared Sensor. Appl. Opt. 54 (13), 3963–3978. Nicl` os, R., Do˜ na, C., Bisquert, M., Valor, E., 2014. Thermal-infrared spectral and angular characterization of crude oil and seawater emissivities for oil slick identification. IEEE Trans. Geosci. Remote Sens. 52, 5387–5395. Nicl` os, R., P´ erez-Planells, Ll, Valiente, J.A., Coll, C., Valor, E., 2018. Evaluation of the SNPP VIIRS Land Surface Temperature product using ground data acquired by an autonomous system at a rice paddy. ISPRS J. Photogramm. Remote Sens. 135, 1–12. Olivera-Guerra, L., Mattar, C., Merlin, O., Duran-Alarcon, C., Santamaria-Artigas, A., Fuster, R., 2017. An operational method for the disaggregation of land surface temperature to estimate actual evapotranspiration in the arid region of Chile. ISPRS J. Photogramm. Remote Sens. 128, 170–181. Qin, Z., Karnieli, A., Berliner, P., 2001. A mono-window algorithm for retrieving land surface temperature from Landsat TM data and its application to the Israel-Egypt border region. Int. J. Remote Sens. 22, 3719–3746. Rozenstein, O., Qin, Z., Derimian, Y., Karnieli, A., 2014. Derivation of land surface temperature for Landsat-8 TIRS using a split window algorithm. Sensors. 14, 5768–5780. Rubio, E., Caselles, V., Coll, C., Valour, E., Sospedra, F., 2003. Thermal-infrared emissivites of natural surfaces: Improvements on the experimental set-up and new measurements. Int. J. Remote Sens. 24, 5379–5390. https://doi.org/10.1080/ 0143116031000102412. Trigo, I.F., Boussetta, S., Viterbo, P., Balsamo, G., Beljaars, A., Sandu, I., 2015. Comparison of model land skin temperature with remotely sensed estimates and assessment of surface-atmosphere coupling. J. Geophys. Res. 120 https://doi.org/ 10.1002/2015JD023812. Vanhellemont, Q., 2020. Combined land surface emissivity and temperature estimation from Landsat 8 OLI and TIRS. ISPRS J. Photogramm. Remote Sens. 166, 390–402. Wan, Z., 2014. New refinements and validation of the collection-6 MODIS land-surface temperature/emissivity product. Remote Sens. Environ. 140, 36–45. Wan, Z., Dozier, J., 1996. A generalized split-window algorithm for retrieving landsurface temperature from space. IEEE Trans. Geosci. Remote Sens. 34, 892–905. Wan, Z., Li, Z.L., 2008. Radiance-based validation of the V5 MODIS land-surface temperature product. Int. J. Remote Sens. 29 (17–18), 5373–5395. Weng, Q., 2009. Thermal infrared remote sensing for urban climate and environmental studies: methods, applications, and trends. ISPRS J. Photogramm. Rem. Sens. 64 (4), 335–344. Yu, X., Guo, X., Wu, Z., 2014. Land surface temperature retrieval from Landsat 8 TIRS—Comparison between radiative transfer equation-based method, split window algorithm and single channel method. Remote Sens. 6, 9829–9852. Yu, Y., Liu, Y., Yu, P., Wang, H., 2017. Enterprise Algorithm Theoretical Basis Document for VIIRS Land Surface Temperature Production. NOAA, Silver Spring, MD, USA. Zheng, Y., Ren, H., Guo, J., Ghent, D., Tansey, K., Hu, X., Nie, J., Chen, S., 2019. Land Surface Temperature Retrieval from Sentinel-3A Sea and Land Surface Temperature Radiometer. Using a Split-Window Algorithm. Remote Sens. 11, 650. R. Nicl` os et al.