Nowcasting from space: Impact of tropical cyclones on Fiji's agriculture

Noy, Ilan; Blanc, Elodie; Pundit, Madhavi; Uher, Tomas Working Paper Nowcasting from space: Impact of tropical cyclones on Fiji's agriculture ADB Economics Working Paper Series, No. 676 Provided in Cooperation with: Asian Development Bank (ADB), Manila Suggested Citation: Noy, Ilan; Blanc, Elodie; Pundit, Madhavi; Uher, Tomas (2023) : Nowcasting from space: Impact of tropical cyclones on Fiji's agriculture, ADB Economics Working Paper Series, No. 676, Asian Development Bank (ADB), Manila, https://doi.org/10.22617/WPS230007-2 This Version is available at: https://hdl.handle.net/10419/298122 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/3.0/igo/ ASIAN DEVELOPMENT BANK ASIAN DEVELOPMENT BANK 6 ADB Avenue, Mandaluyong City 1550 Metro Manila, Philippines www.adb.org Nowcasting from Space Tropical Cyclones Impact on Fiji’s Agriculture Satellite imagery can be a source of easily available, fast, affordable, and accurate data for assessing disaster impacts. This study investigates the feasibility of using remote sensing data for post-disaster damage assessment. It focuses on Fiji, its agriculture sector, and the tropical cyclones that wreaked havoc on the country in recent years. Findings of the study show that remote sensing data, when combined with pre-event socioeconomic and demographic information, can better explain the identified changes in the vegetation index, thereby improving both nowcasting and post-disaster damage assessments of tropical cyclones on agriculture. About the Asian Development Bank ADB is committed to achieving a prosperous, inclusive, resilient, and sustainable Asia and the Pacific, while sustaining its efforts to eradicate extreme poverty. Established in 1966, it is owned by 68 members —49 from the region. Its main instruments for helping its developing member countries are policy dialogue, loans, equity investments, guarantees, grants, and technical assistance. ADB ECONOMICS WORKING PAPER SERIES NO. 676 January 2023 NOWCASTING FROM SPACE IMPACT OF TROPICAL CYCLONES ON FIJI’S AGRICULTURE Ilan Noy, Elodie Blanc, Madhavi Pundit, and Tomas Uher ASIAN DEVELOPMENT BANK ADB Economics Working Paper Series Ilan Noy, Elodie Blanc, Madhavi Pundit, and Tomas Uher No. 676 | January 2023 Ilan Noy ([email protected]) is a professor at School of Economics and Finance, Victoria University of Wellington, New Zealand. Elodie Blanc (elodie. [email protected]) is a research fellow at Motu Economic and Public Policy Research, Wellington, New Zealand and a research scientist at the MIT Joint Program on the Science and Policy of Global Change, USA. Madhavi Pundit ([email protected]g) is a senior economist at the Economic Research and Regional Cooperation Department, Asian Development Bank. Tomas Uher ([email protected]) is a researcher at the School of Economics and Finance, Victoria University of Wellington. Nowcasting from Space: Impact of Tropical Cyclones on Fiji’s Agriculture The ADB Economics Working Paper Series presents research in progress to elicit comments and encourage debate on development issues in Asia and the Pacific. The views expressed are those of the authors and do not necessarily reflect the views and policies of ADB or its Board of Governors or the governments they represent. Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) © 2022 Asian Development Bank 6 ADB Avenue, Mandaluyong City, 1550 Metro Manila, Philippines Tel +63 2 8632 4444; Fax +63 2 8636 2444 www.adb.org Some rights reserved. Published in 2023. ISSN 2313-6537 (print), 2313-6545 (electronic) Publication Stock No. WPS230007-2 DOI: http://dx.doi.org/10.22617/WPS230007-2 The views expressed in this publication are those of the authors and do not necessarily reflect the views and policies ofthe Asian Development Bank (ADB) or its Board of Governors or the governments they represent. ADB does not guarantee the accuracy of the data included in this publication and accepts no responsibility for any consequence of their use. The mention of specific companies or products of manufacturers does not imply that they are endorsed or recommended by ADB in preference to others of a similar nature that are not mentioned. By making any designation of or reference to a particular territory or geographic area, or by using the term “country” inthis publication, ADB does not intend to make any judgments as to the legal or other status of any territory or area. This work is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) https://creativecommons.org/licenses/by/3.0/igo/. By using the content of this publication, you agree to be bound bytheterms of this license. For attribution, translations, adaptations, and permissions, please read the provisions andterms of use at https://www.adb.org/terms-use#openaccess. This CC license does not apply to non-ADB copyright materials in this publication. If the material is attributed toanother source, please contact the copyright owner or publisher of that source for permission to reproduce it. ADB cannot be held liable for any claims that arise as a result of your use of the material. Please contact [email protected] if you have questions or comments with respect to content, or if you wish toobtain copyright permission for your intended use that does not fall within these terms, or for permission to use theADB logo. Corrigenda to ADB publications may be found at http://www.adb.org/publications/corrigenda. Notes: In this publication, “$” refers to United States dollars, unless otherwise stated. ADB recognizes “China” as the People’s Republic of China and “USA” as the United States. The ADB Economics Working Paper Series presents data, information, and/or findings from ongoing research and studies to encourage exchange of ideas and to elicit comment and feedback about development issues in Asia and the Pacific. Since papers in this series are intended for quick and easy dissemination, the content may or may not be fully edited and may later be modified for final publication. ABSTRACT The standard approach to ‘nowcast’ disaster impacts, which relies on risk models, does not typically account for the compounding impact of various hazard phenomena (e.g., wind and rainfall associated with tropical storms). The alternative, traditionally, has been a team of experts sent to the affected areas to conduct a ground survey, but this is time-consuming, difficult, and costly. Satellite imagery may provide an easily available and accurate data source to gauge disasters’ specific impacts, which is both cheap, fast, and can account for compound and cascading effects. If accurate enough, it can potentially replace components of ground surveys altogether. An approach that has been calibrated with remote sensing imagery can also be used as a component in a nowcasting tool, to assess the impact of a cyclone, based only on its known trajectory, and even before post-event satellite imagery is available. We use one example to investigate the feasibility of this approach for nowcasting, and for post-disaster damage assessment. We focus on Fiji and on its agriculture sector, and on tropical cyclones (TCs). We link remote sensing data with available household surveys and the agricultural census data to obtain an improved assessment of TC impacts. We show that remote sensing data, when combined with pre-event socioeconomic and demographic data, can be used for both nowcasting and post-disaster damage assessments.1 Keywords: satellite, cyclone, damage, impact, disaster, nowcasting JEL codes: Q54, Q1, C8 The authors would like to thank the Fiji Ministry of Agriculture and Bureau of Statistics for data inputs; Priscille Villanueva, Homer Pagkalinawan, and Paolo Magnata for their valuable research assistance; Erik Aelbers, Isoa Wainiqolo, and staff of Fiji Resident Mission for their support and feedback; and participants in the Economic Research and Regional Cooperation Department seminar for comments and discussions. The grant fund for the study was received from the Japan Fund for Prosperous and Resilient Asia and the Pacific financed by the Government of Japan through the Asian Development Bank. I. INTRODUCTION In the last decades, satellite-based Earth observation data have been increasingly used for various applications in different fields. One of these is disaster and emergency management. Emergency responders and disaster risk managers have started to explore the use of satellite imagery as a reliable and immediate source for disaster-impact information (Voigt et al. 2016). Recent technological advances in remote sensing have led to a massive increase in temporal, spatial, and spectral resolutions of these images. Furthermore, we have also seen improvements in data availability and accessibility, and advances in the processing methods that are required to effectively interpret these large amounts of data (Notti et al. 2018). These developments have led to an increasing application of satellite imagery for more effective disaster mitigation, preparedness, response, and recovery. One such disaster management application that has advanced due to improvements in remote sensing is post-disaster damage assessment. Traditionally, in a post-disaster scenario, a team of experts is sent to the area to conduct a ground survey and assess damages—this is sometimes connected to a commissioned Post-Disaster Needs Assessment (PDNA) report. This process is time-consuming and costly, and possibly difficult because often the access to the affected area is difficult, dangerous, or restricted. Alternatively, governments and, especially, public and private insurance companies use risk models to quantify these damages. Risk models are frequently inaccurate and are generally unable to account for compound or cascading events,1 as these are much more difficult to quantitatively model. In contrast, satellite imagery can function as an easily available, timely, and accurate data source to gauge the damage severity and the specific impacts caused by a disaster event (even a compound or cascading one). Indeed, one can also develop, with the assistance of remote sensing data, nowcasting tools to forecast the impact of the event when it is already happening (or immediately thereafter). While these can serve as effective complements to ground surveys, particularly for immediate assessment, if accurate enough, satellite imagery can even potentially replace ground survey efforts altogether. Satellite data are currently used for post-disaster damage assessment of various types of disasters caused by natural hazards—tropical storms, floods, earthquakes, landslides, and tsunamis. When calibrated well, these data are selected appropriately in terms of their spectral, temporal, and spatial resolutions to obtain the desired information. However, these data are rarely matched with traditional socioeconomic information (e.g., census) to better gauge the determinants of impacts and improve our ability to nowcast them. In this study, we use one example to investigate the feasibility of using remote sensing data for post-disaster damage assessment. We focus on the islands of Fiji, on its agriculture sector, and on tropical cyclones (the main disaster-inducing hazard in Fiji, and in much of the other Pacific island countries). The main crop grown in Fiji is sugarcane, representing 60% of the area harvested averaged over the period 2016–2020 (FAO). We consider the effect of four recent tropical cyclones (TCs) on agriculture since reliable satellite pictures of high resolution and frequency are only available for recent events. They are • TC Winston (15 February 2016), • TCs Josie (2 April 2018) and Keni (10 April 2018), and 1 Compound events refer to multiple disasters happening concurrently or consecutively, thereby reinforcing each other to create a more complex condition that amplifies the impact of individual disasters; whereas, cascading events happen when one hazard triggers another hazard in a series (IGES 2022). 2 • TC Harold (7–8 April 2020). As a Category 5 cyclone (the most extreme category), TC Winston was one of the strongest tropical storms in recorded history and the strongest cyclone ever to make landfall in the Southern Hemisphere, with estimated wind speeds of up to 230 kilometers (km) per hour. After making landfall in Viti Levu, Fiji in February 2016, TC Winston caused massive destruction, damaged 32,000 houses, and left approximately 350,000 Fijians in need of humanitarian aid. The total economic losses were estimated at $1.38 billion, approximately one-third of Fiji’s gross domestic product (Reliefweb 2016a, 2016b; Di Liberto 2016; WFP 2017; UNICEF). In 2018, Category 1 TC Josie affected the central and western parts of Fiji and caused heavy flooding, mainly on the main island Viti Levu. The Category 3 TC Keni passed close to Viti Levu just 1 week later, causing repeated flooding. The two cyclones affected approximately 78,000 Fijians, with 12,000 people seeking emergency shelter at evacuation centres. The economic losses caused by these two events were estimated at $3 million (Reliefweb 2018, RNZ 2018). The Category 5 TC Harold caused destruction in the Pacific island countries of, Fiji, Solomon Islands, Tonga, and Vanuatu in April 2020. It impacted the southern part of Fiji as a Category 4 event and affected more than 180,000 people, displaced around 10,000 people, and destroyed or damaged almost 4,000 homes. The estimated economic losses associated with the event were estimated to exceed $40 million (DFAT, Reliefweb 2020). Estimates of the value of mortality and aggregated figures of economic damage these events inflicted (for all assets and sectors) at the regional level are available indirectly from the Fijian government through the United Nations Office for Disaster Risk Reduction (in DesInventar Sendai, https://www.desinventar.net/), and from a nongovernment source, Emergency Events Database (EM-DAT, https://www.emdat.be/). These are, of course, not specifically attempting to count losses in agriculture; and, indeed, in Tables 1 and 2, we show that these regional-level data from EM-DAT and DesInventar are not very tightly correlated with the remote sensing data we use to identify spatially detailed information on the damage the cyclones wrought to agriculture. Our purpose, in this study, is to examine the possibility of combining remote sensing data with socioeconomic sources of information (household surveys and the census) to nowcast the damage from tropical cyclones on agriculture. We focus on Fiji and extract the change in the Enhanced Vegetation Index (details below) around four tropical cyclones that occurred in Fiji in recent years. We then aggregated these data spatially to the district level and match them with data we extract from the 2013-14 Household Income and Expenditure Survey (HIES) and the 2020 Fiji Agriculture Census (FAC). By matching remote-sensing observations with socioeconomic, demographic, and agronomic data, we are able to better explain the identified changes in the vegetation index, and thus improve the nowcasting of impacts of the tropical cyclones on agriculture. In other words, we show that, in the case of Fiji’s recent tropical cyclones, the remote sensing measurement we employed (the Enhanced Vegetation Index), when combined with pre-cyclone socioeconomic and demographic data, can be used for nowcasting damages (during the event and immediately thereafter). We conclude by describing, how, together with additional socioeconomic data, remote sensing can also be very useful for nowcasting post-disaster indirect losses (rather than the direct losses). 3 II. LITERATURE REVIEW A. Optical Satellite Imagery for Disaster Damage Assessment For disaster damage assessment, the most frequently used satellite sensors are optical, Synthetic Aperture Radar (SAR), and Light Detection and Ranging (LiDAR) sensors. Most optical satellite sensors gather surface reflectance data from the visible electromagnetic spectrum, as well as emissivity data from the infrared wavelengths to produce images. Optical imagery is relatively easier to interpret than SAR and LiDAR imagery, as the resulting imagery typically appears as a standard colored or black and white photograph and corresponds to how humans naturally view the world. The combination of visible and infrared wavelengths is particularly useful for the detection of water surfaces and vegetation-covered areas and is therefore well suited for flood mapping or estimating storm- or flood-induced vegetation impacts. The applicability of optical satellite data for storm and flood damage assessment is partially limited because of its reliance on relatively cloud-free weather conditions for gathering the data and the fact that high precipitation is typically correlated with heavy cloud cover during the days surrounding the event (Rahman 2019). Furthermore, many sources of high-resolution satellite data (such as GeoEye, RapidEye, QuickBird, or WorldView) are not publicly available and their data can be very costly to acquire. Despite these limitations, researchers have repeatedly shown the efficacy of optical satellite imagery for immediate storm and flood damage assessments. For rapid post-disaster impact assessment that can effectively assist emergency responders, there are high demands on both the temporal and spatial resolutions of the required satellite imagery. Imagery needs to be available quickly (Hodgson et al. 2010) and the spatial resolution requirements for reliable visual interpretation are also high. Where experts need to conduct damage assessment manually, and if accuracy is important, they require a high resolution at 1.5 meters per pixel (Battersby, Hodgson, and Wang 2012). Therefore, it is often necessary to obtain a combination of satellite sources to gather as much relevant data as is possible during the first few days after the disaster. Furthermore, satellite images may not provide direct insights of disaster-inflicted damages because of the vertical viewing angles of satellite sensors. To assess disaster impacts, researchers often require the use of proxies to estimate specific damages. For example, a building’s photographic texture or rubble and debris piles can be used to identify buildings that were damaged, or a building’s shadow can be used to approximate building collapse (Ghaffarian, Kerle, and Filatova 2018). Because of these difficulties, many damage assessment approaches also combine the remote sensing imagery with ancillary data such as satellite-derived Digital Elevation Models or Digital Terrain Models, soil or land use data, and other datasets that can be combined to derive more accurate disaster-related information. Many of these data types are freely accessible online from the United States Geological Survey, national sources, or other international organizations.2 It is less common, in this literature, to combine these data with spatial economic and demographic data, such as those data that are collected in standard household living standards surveys, or the decadal census. 2 See GeoSpatial Data Cloud (http://www.gscloud.cn/home#page1/1) and the United States Geological Survey (https://www.usgs.gov/). 4 B. General Damage Assessments Flood and storm damage assessments based on satellite imagery vary in their use of specific remote sensing sources and the corresponding spatial and temporal resolutions of the imagery. Often, high spatial, temporal, and spectral resolution imagery would be most efficient, but not always available. There can also be trade-offs between the spatial and the temporal resolution of the data (Giraldo-Osorio and García-Galiano 2012, Fayne et al. 2017). With respect to the spatial resolution, damage assessments are conducted on a range of available resolutions, from low (>100 meters per pixel) to moderate (5–100 meters per pixel) and high resolutions (<5 meters per pixel). Summarized below are general hazard assessments of floods and storms, while the next subsection focuses on specific damage types such as vegetation, agriculture, or buildings and infrastructure. Storms With respect to general damage estimations of storms, and specifically tropical storms, researchers used moderate-resolution (Al-Amin Hoque et al. 2015; Al-Amin Hoque et al. 2016; Al-Amin Hoque et al. 2017; Phiri, Simwanda, and Nyirenda 2021) or high-resolution imagery (Barnes, Fritz, and Yoo 2007; Mas et al. 2015; Doshi, Basu, and Pang 2018). Studies by Al-Amin Hoque et al. (2015), Al-Amin Hoque et al. (2016), and Al-Amin Hoque et al. (2017) used moderateresolution SPOT-5 imagery to analyze impacts of the 2007 Tropical Cyclone Sidr in Bangladesh. Phiri, Simwanda, and Nyirenda (2021) estimated the damages for the 2019 Cyclone Idai in Mozambique using Sentinel-2 images. Barnes, Fritz, and Yoo (2007) used high-resolution IKONOS imagery to estimate local damages of the 2005 Hurricane Katrina in the United States; and Mas et al. (2015) studied the 2013 Typhoon Haiyan in the Philippines using Google Earth imagery. Doshi, Basu, and Pang (2018) utilized freely available high-resolution satellite imagery datasets to analyze the 2017 Hurricane Harvey in the United States. Floods Flood mapping and impact assessments on a large scale were predominantly conducted on publicly available low-resolution optical sensors such as the Visible Infrared Imaging Radiometer Suite (Li et al. 2018, Goldberg et al. 2018) or, more frequently, the Moderate Resolution Imaging Spectroradiometer (MODIS) (Brakenridge and Anderson 2006; Irimescu et al. 2009; Sun, Yu, and Goldberg 2011; Sun et al. 2012; Haq et al. 2012; Zhang et al. 2012; Li et al. 2013; Senthilnath et al. 2013; Kwak, Park, and Fukami 2014; Nigro et al. 2014; Memon et al. 2015; Arvind et al. 2016; Coltin et al. 2016; Ban et al. 2017; Lin et al. 2017; Lin et al. 2019). Several older studies applied data from the predecessor to MODIS, the advanced very high-resolution radiometer (Barton and Bathols 1989; Ali, Quadir, and Huh 1989; Sheng, Gong, and Xiao 2001; Jain et al. 2006). Benefits of these satellite sensors are high spatial coverage and temporal resolution, along with free availability. However, the low spatial resolution limits these studies to less detailed or accurate impact estimations. Freely available Landsat data were utilized in flood mapping and damage assessments by Gianinetto, Villa, and Lechi (2006); Ma et al. (2011); Li, Xu, and Chen (2016); Hutanu, Urzica, and Enea (2018); Sivanpillai et al. (2021); and Musiyam et al. (2020). Kordelas et al. (2018) used Sentinel-2 data, while Feng et al. (2015) used HJ-CCD data. Du et al. (2021) developed a flood mapping method for data-sparse regions, combining Landsat, Google Earth Engine, and satellitebased soil moisture data. 11 For the district, rather than the grid-cell level, average district level EVIdiff and EVIch values are calculated as the average of the grid-level EVI values for all grid cells within each district, based on Fiji’s administrative division. C. Socioeconomic and Demographic Data The socioeconomic and demographic data used for the linear regressions were gathered from two primary sources: 2013-14 Household Income and Expenditure Survey (HIES) and 2020 Fiji Agriculture Census (FAC). The HIES contains household-level data for a sample of 6,020 households and is aimed to be representative per district. Household income values were averaged per district using weights included in the original HIES dataset to arrive at average household income values at the district (Tikina Covata) level. The respective districts were assigned to households based on the household ID numbers. FAC is organized at a subdistrict (Tikina Vou) level. The data from a total of over 71,000 households capture detailed economic and demographic information relating to the agriculture sector in rural and peri-urban areas, where most agricultural activities are concentrated. Grouping subdistricts into respective districts was done based on the administrative associations as defined in the House of Chiefs (n.d.) list, and a combination of regional maps in FAC documentation. Average district values were calculated using a weighted average with the number of farmers, agricultural households, or their members used as weights for the respective indicators. With respect to the variables indicating the ratio of agricultural land used for specific crops, only the crop types that covered at least 1% of the total area planted were selected. Similarly, for the land tenure variables, only the land tenure types that accounted for at least 5% of the total farmland area were selected. Besides the data from the HIES and FAC, a “cyclone distance” variable was added to the regression model to account for the effect of cyclone proximity on vegetation damage. The value of this variable ranges from 1 (closest to the cyclone path) to 4 (furthest from the cyclone path). More specifically, distance from the cyclone was accounted for by establishing four zones, with borders at 50 km, 100 km, and 200 km from the cyclone path and with the value of 1 indicating that the majority (>50%) of the district area is closer than 50 km from the cyclone path. The values were calculated using cyclone trajectory maps from the International Best Track Archive for Climate Stewardship (IBTrACS; NOAA) and Reliefweb (2016a). For the combined cyclone events of Josie and Keni, the lower value of cyclone distance was selected from the two (implicitly assuming no cumulative damage). A complete list of the independent variables used in the regression models is presented in Table 3 in the Appendix3, along with their description and sources. From the list of variables that were extracted from the HIES and FAC, selected variables were removed from variable pairs that had a correlation coefficient higher than 0.7. The variables removed were (i) no savings account due to lack of access, (ii) number of females per agricultural household, (iii) ratio of agricultural land 3 The Appendix is available online at http://dx.doi.org/10.22617/WPS230007-2. 12 used for growing coconuts, (iv) native lease land ownership, (v) average age, and (vi) imputed rent and wages.4 D. Regression Models A linear regression model was applied to identify potential relationships between demographic and socioeconomic factors on one hand, and cyclone-induced vegetation damage on the other, which is approximated by the EVI measures described earlier. In principle, our aim is to identify the correlates of significant damage from cyclones. We are not necessarily aiming to identify any causal mechanism from these independent control variables on the dependent variables in our models (the variants of the EVI measure; see below). We estimate the following equation: 𝐸𝐸𝐸𝐸𝐸𝐸𝑐𝑐𝑐𝑐 𝑥𝑥=𝛼𝛼+𝛽𝛽1𝐻𝐻𝐸𝐸𝐸𝐸𝐻𝐻𝑐𝑐+𝛽𝛽2𝐹𝐹𝐹𝐹𝐹𝐹𝑝𝑝𝑐𝑐⊂𝑐𝑐 +𝛽𝛽3𝐷𝐷𝐸𝐸𝐻𝐻𝐷𝐷𝑐𝑐𝑐𝑐 +𝜀𝜀𝑐𝑐𝑐𝑐 (4) Whereby 𝐸𝐸𝐸𝐸𝐸𝐸𝑐𝑐𝑐𝑐 𝑥𝑥, the dependent variable, is calculated as variations of the EVI around the time of the respective cyclone (c) in district (d). Variants include (i) the absolute EVI change, EVIdiff (eq. 2); (ii) the relative EVI change, EVIch (eq. 3); (iii) the same as (i) but including the condition that the EVIdiff<0; and (iv) the same as (ii) but including the condition that EVIch>0. The latter two variations of the independent variable are used as an attempt to limit the effect of other potential influencing factors that may have caused the EVI values in certain districts to be positive despite the cyclone occurrence. As for the independent variables (described in detail in section III-C, 𝐻𝐻𝐸𝐸𝐸𝐸𝐻𝐻𝑐𝑐 is the vector of districtlevel variables available from the Fiji household survey. 𝐹𝐹𝐹𝐹𝐹𝐹𝑝𝑝𝑐𝑐⊂𝑐𝑐 is the vector of subdistrict-level variables available from the agricultural census, aggregated to the district level (to match the level of aggregation in the household survey data).5 𝐷𝐷𝐸𝐸𝐻𝐻𝐷𝐷𝑐𝑐𝑐𝑐 is the measured distance from the cyclone path, identified for each cyclone, for each district. The error term (𝜀𝜀𝑐𝑐𝑐𝑐) is assumed to be independent and identically distributed. For each of the four independent variables, we estimate a set of regressions for four distinct samples: all cyclones together (Harold, Winston, Josie, and Keni); Winston, Josie, and Keni; and all events except Harold. TC Harold is the only cyclone within the dataset that is associated with a positive average EVI value (when combining both Sentinel and MODIS data), which would, on its own, suggest that the overall condition of Fiji’s vegetation measured by EVI improved after TC Harold’s occurrence. This further suggests that there might have been some measurement errors or other factors that affected the outcome associated with TC Harold. We are unable to explain this, and we therefore also estimated a sample that excluded only the observations on TC Harold. 4 With respect to outliers, for the indicator “Average Income from Sale of Crops,” two extreme values were removed for subdistricts Muaira and Vaturova as these were more than 20 times higher than the average. Therefore, the average crop income values for the respective districts did not account for these two subdistricts. 5 We note that the FAC data was collected after the TCs hit Fiji. Ideally, we would have had access to a similar census done before cyclones. However, the census is not run frequently (the previous one was in 2009), and we were not granted access to it. In addition, land use is a slowly moving variable, and it is unlikely that the TCs themselves have led to such rapid land use changes to make the 2020 census irrelevant. We therefore use the 2020 census, in spite of its timing. 13 From the full set of independent variables extracted from the HIES and FAC, the variables used in the final regression models were selected by the following method. Initially, all available independent variables were included in the regression model, and the least statistically significant variable was identified based on its p-value. Then, we rerun the regression, removing the least statistically significant independent variable. This process was repeated until all the independent variables were statistically significant at a 10% level of significance. This method was used to identify statistically significant variables separately for each of the total of 16 regression models reported below (four EVI categories and four cyclone-grouping categories). The regression models applied in the first table within each of the four EVI categories (Tables 4, 6, 8 and 10 in the Appendix) and within each cyclone type (table columns 1-4) contain only variables that were identified as statistically significant at a 10% level of significance for the regression of each specification. The second table within each category (Tables 5, 7, 9, 11 in the Appendix) shows a single regression model (a selection of independent variables) applied across all five types of cyclone events. Here, all variables that proved statistically significant for at least one cyclone event were included. The regression outputs for TC Harold by itself are in Table 12 in the Appendix,6 while Table 13 presents the summary statistics of all variables. E. Data and Method for Countrywide Analysis Data for the annual agricultural income of Fiji were obtained from the Reserve Bank of Fiji (2021), which contains Fiji's agricultural gross value added at constant basic prices of 2014. Ideally, we would have liked to use annual district level agricultural income, but this data was not available. The data used for the years 2015–2019 are from the Fiji Bureau of Statistics; whereas, the data for 2020 are based on the Macroeconomic Committee’s estimates as of July 2021. A list of variables used for countrywide analysis is presented in Table 14 in the Appendix. The country-level weighted averages of EVIdiff and EVIch were calculated from district-level EVIdiff and EVIch values, using household crop income values from FAC as weights, to account for varying relevance of the district for the countrywide level of agricultural income. Since the only available district-level crop income data were for the year 2020 (during which TC Harold occurred), these values were also used as weights to calculate the weighted EVI averages for the years 2016 and 2018 (the years of the other two cyclone events). In addition, since the national-level data only include a few observations, we could not pursue any statistical analysis. However, we discuss below the correlation between the change in agricultural income during the cyclone years, and the change in the EVI measures. 6 Regression results for TC Harold when using negative EVIdiff and EVIch values are not reported due to the small number of observations (28). 14 IV. REGRESSION RESULTS A. Absolute Enhanced Vegetation Index Change Tables 4–7 in the Appendix show the regression results when EVIdiff is the dependent variable. Many of the variables do not seem very robustly associated with the difference in the EVI experienced post-cyclone. One variable that repeatedly comes out as significant at 1% of statistical significance, and maybe not surprisingly, is the distance from the TC path.7 This is the only proxy we use for the intensity of the hazard, and the data seem to suggest that indeed a change in the EVI is therefore a satisfactory proxy for damage from TCs. The coefficient for the Distance variable is consistently positive (except for TC Harold), suggesting that an increased distance from the cyclone path was associated with higher EVIdiff values (i.e., less vegetation damage, a better outcome), which is consistent with our basic hypothesis. Four other variables seem to consistently be statistically significant for the full set of observations (i.e., all districts, rather than just those in which the EVI decreased). From the FAC measures that describe the crop composition of each district, the variables that stand out are Banana (negative) and Cassava (positive). The negative coefficient for banana suggests that a higher proportion of land cultivating banana, in an exposed district, is associated with more damage from the cyclone. Put differently, banana plants seem more vulnerable to the cyclone shock. The opposite is true for cassava; the more cassava a district has, the less damage from the cyclone it experiences (holding everything else constant).8 The total income and total transfer variables are also consistently statistically significant in Tables 4 and 5; although, Income is consistently positive and Total transfer is consistently negative. These results suggest that higher-income districts suffer less damage from tropical cyclones, other things being equal, and increased government transfers and remittances (from abroad) were associated with more damage to vegetation from tropical cyclones. Beyond these five consistently identified associations with TC vegetation damage (Distance, Banana, Cassava, Income, and Total transfer), the coefficients for the Dalo crop share, similar to Banana, are consistently negative and significant for All TCs, but were not significant in the regression model that includes the same independent variables across the four cyclone categories (Table 5). Next, we examine only those districts that experienced noticeable (remotely from space) damage from the TCs (the sample of observations with negative EVIdiff values). The results are shown in Tables 6 and 7 in the Appendix. The distance (from the cyclone path) is again consistently positive and statistically significant, except for the combined events of TCs Josie and Keni. This set of regressions includes a significantly smaller number of observations, so it might not be surprising that fewer variables are now statistically significant. Banana and Dalo still appear to be the crops 7 The only exception is in the case of TCs Josie and Keni, while using only negative EVI values. 8 We remind the readers that, by “damage” here, we refer to a decrease in the EVI. It therefore might also be that banana plants experience more (remotely) visible damage, rather than genuine economic damage that manifests in reduced income from these crops. Unfortunately, data on income, by district and/or year, from specific crops are not available. 15 that are most vulnerable to cyclone damage; although, for banana, these results are slightly less consistent than what we had in Tables 4 and 5 for the full population of districts in Fiji. Household size is now significant and consistently positive for All TCs and for TCs Josie and Keni in Table 6. When considering only observations with negative EVIdiff, the districts with bigger households were associated with less cyclone-induced vegetation damage. This variable is related to Household income, which is also positive and significant; hence, it is noteworthy that it is still statistically significant. Cassava is now statistically significant on at least 5% level of significance across all categories in Table 6. The variable measuring the irrigated farm area (as a share of the district’s total farm area), was significant and consistently negative for TC Winston and TCs Josie and Keni (in the regressions focusing on a specific TC event; Table 6, columns 2 and 3). This suggests that irrigated agricultural land may be less resistant to cyclone impacts, possibly because of its location or the crops using irrigation. The ratio of traditionally owned land is significant and consistently negative, except for TCs Josie and Keni, in the regression models including all variables (Table 7). This regression output also suggests that the more there is agricultural land under traditional ownership, the more sensitive to cyclone impacts is the district. The ratio of female farmers is now significant and consistently positive in TC Winston, in TCs Josie and Keni, and in TC Winston plus TCs Josie and Keni. These results, while indicative, are not robust enough to reach any firmer conclusions. B. Relative Enhanced Vegetation Index Change The regression models involving a relative EVI percentage change (EVIch) as the dependent variable (rather than the absolute EVI change) showed, to a certain extent, similar outcomes as the EVIdiff regressions. The most consistently significant variable is Distance and has positive coefficients. The Banana and Cassava variables, similar to the EVIdiff regressions, are significant except for the regressions for TCs Josie and Keni. The values of their coefficients also indicate that these crops are, again, associated with more vegetation damage in the case of bananas, and less damage for cassava. The Total transfer variable is significant for All TCs, for TCs Josie and Keni, and for TC Winston plus TCs Josie and Keni in Tables 8 and 9 in the Appendix. Similar to the EVIdiff regressions, the coefficient values here are consistently negative and pointing at the same conclusion—that increased government transfers and remittances were associated with larger vegetation damage. Noticeably, there are a few more variables that are statistically significant in some specifications, especially in the case of TC Winston (by far the strongest TC to hit Fiji in the last decade, and possibly the strongest TC ever recorded in the South Pacific). These include education, the yaqona crop, total income, and other income. However, since our aim is to identify a set of variables that will assist in nowcasting TC damage, we do not believe these can be consistently and reliably used. In the last set of regressions (Tables 10 and 11 in the Appendix), we again use the percentage change in EVI as the dependent variable but restrict the sample to those observations for which the change was negative (i.e., there was observed decrease in the EVI). As we have seen above, the results become less statistically robust since the sample decreases, but the explanatory power of the model is quite high. The main consistently and reliably positive coefficient is still the distance from the cyclone path. Interestingly, in the specification of only the selected independent variables 16 (Table 10), the Dalo variable is consistently negative and significant at a 1% level of significance, suggesting a higher vulnerability to cyclone damage. There is less consistency in any of the other variables, and we note that the regressions for TC Winston and for TCs Josie and Keni show the highest number of statistically significant associations with the EVI change variable. In Table 11, where we include all statistically significant variables that came up through the algorithm, we find very little that is statistically and robustly significant (except for the distance variable). 17 V. COUNTRYWIDE ANALYSIS When examining countrywide data, we can compare the EVI change (EVIdiff or EVIch) identified for each of the TCs with the change in Fiji’s annual agricultural income as measured by agricultural gross value added. As Table 15 in the Appendix shows, the year 2016, which is associated with the largest decrease (both absolute and relative) in EVI values, is also associated with the worst outcome in terms of agricultural income that decreased by 8.7% between 2015 and 2016. While the year 2018 still shows an overall negative EVI change related to TC Josie and TC Keni, Fiji’s annual agricultural income increased by 5.6%, which is the highest agricultural income increase among the three observed years. The year 2020 is the only year that shows an increase in EVI values despite TC Harold’s occurrence, by 0.002 and 0.44% in absolute and relative terms, respectively. In the same year, the country’s agricultural income increased by 3.0%. This pattern can, of course, be completely coincidental. But we do believe that the destruction wrought by TC Winston is easily identifiable from space and, in principle, so should other intense cyclones. Ideally, if we were granted access to the district-level estimations of agricultural income, from which the aggregate figure is possibly derived, we would have been able to better examine the reliability of our proposed indicator. Finally, we would like to suggest a possible algorithm for estimating the predicted change in agricultural income following a tropical cyclone. • Once the TC path is known (this information is posted immediately after the event), it is possible to provide a preliminary estimate of the district-level change in the EVI, based on the coefficients for the variables we identified: distance from the path of the cyclone, banana and cassava (as grown in each district), income, and transfers. While this information is not timesensitive, it allows one to estimate the likelihood that the cyclone will entail significant damage to agricultural production based on the basic parameters of the event. • Once remote sensing readings of the cyclone are available, one can try to identify affected districts more closely, and attempt to redirect assistance toward them (when that is relevant). This, together with the information about general vulnerability (e.g., the share of banana plantations in the district) can assist in disaster risk reduction planning. • With additional geolocated observations about local agricultural production, one could potentially design a tool that allows even more precise estimates of the economic impact on the agriculture sector more directly. We leave that for a time when the additional information is forthcoming. 18 VI. NEXT STEPS The aim of this project was to develop a tool that will enable nowcasting of disaster impacts. While there is an extensive literature that attempts to link hazard indicators (such as ground shaking) with remote sensing data, we attempt to model the agricultural economic damage from a tropical cyclone also using socioeconomic and demographic information. Typically, only remote sensing geospatial data are used. This project was hampered by the unavailability of socioeconomic and demographic information in sufficiently high spatial and temporal resolution, so the analysis had to be cross-sectional, and at the district level. In essence, the aim of that was to show a prototype tool that can be used with more detailed data (spatially and temporally) to nowcast disaster impacts, in general, and agricultural damages from tropical cyclones, in particular. This type of nowcasting is not yet done by disaster risk management agencies (multilateral or national), but we believe it holds significant promise in such cases as tropical cyclone impacts in the South Pacific. Using remote sensing high-spatial and high-frequency imaging that is now available, coupled with information from sociodemographic and economic data, should be superior to nowcasting modelling approaches that have been used in initiatives such as the Pacific Catastrophe Risk Assessment and Financing Initiative. 19 REFERENCES Adriano, B., N. Yokoya, J. Xia, H. Miura, W. Liu, M. Matsuoka, and S. Koshimura. 2021. Learning from Multimodal and Multitemporal Earth Observation Data for Building Damage Mapping. ISPRS Journal of Photogrammetry and Remote Sensing. 175. pp. 132–143. Ahmed, M. R., K. R. Rahaman, A. Kok, and Q. K. Hassan. 2017. Remote Sensing-Based Quantification of the Impact of Flash Flooding on the Rice Production: A Case Study Over Northeastern Bangladesh. Sensors. 17 (10). 2347. Al-Amin Hoque, M., S. Phinn, C. Roelfsema, and I. Childs. 2015. Assessing Tropical Cyclone Damage Using Moderate Spatial Resolution Satellite Imagery: Cyclone Sidr, Bangladesh 2007. Paper prepared for the 36th Asian Conference on Remote Sensing: Fostering Resilient Growth in Asia. Manila. 24–28 October. _____. 2016. Assessing Tropical Cyclone Impacts Using Object-Based Moderate Spatial Resolution Image Analysis: A Case Study in Bangladesh. International Journal of Remote Sensing. 37 (22). pp. 5320–5343. _____. 2017. Tropical Cyclone Disaster Management Using Remote Sensing and Spatial Analysis: A Review. International Journal of Disaster Risk Reduction. 22. pp. 345–354. Albawi, S., T. A. Mohammed, and S. Al-Zawi. 2017. Understanding of a Convolutional Neural Network. 2017 International Conference on Engineering and Technology. pp. 1–6. Ali, A., D. A. Quadir, and O. K. Huh. 1989. Study of River Flood Hydrology in Bangladesh with AVHRR Data. International Journal of Remote Sensing. 10 (12). pp. 1873–1891. Arvind, C. S., A. Vanjare, S. N. Omkar, J. Senthilnath, V. Mani, and P. G. Diwakar. 2016. Flood Assessment Using Multi-Temporal MODIS Satellite Images. Procedia Computer Science. 89. pp. 575–586. Ayala-Silva, T. and Y. A. Twumasi. 2004. Hurricane Georges and Vegetation Change in Puerto Rico Using AVHRR Satellite Data. International Journal of Remote Sensing. 25 (9). pp. 1629–1640. Ban, H. J., Y. J. Kwon, H. Shin, H. S. Ryu, and S. Hong. 2017. Flood Monitoring Using Satellite- Based RGB Composite Imagery and Refractive Index Retrieval in Visible and Near- Infrared Bands. Remote Sensing. 9 (4). 313. Barnes, C. F., H. Fritz, and J. Yoo. 2007. Hurricane Disaster Assessments with Image-Driven Data Mining in High-Resolution Satellite Imagery. IEEE Transactions on Geoscience and Remote Sensing. 45 (6). pp. 1631–1640. Barton, I. J. and J. M. Bathols. 1989. Monitoring Floods with AVHRR. Remote Sensing of Environment. 30 (1). pp. 89–94. 20 Battersby, S. E., M. E. Hodgson, and J. Wang. 2012. Spatial Resolution Imagery Requirements for Identifying Structure Damage in a Hurricane Disaster. Photogrammetric Engineering and Remote Sensing. 78 (6). pp. 625–635. Bhowmik, A. K. and P. Cabral. 2011. Damage and Post-Cyclone Regeneration Assessment of the Sundarbans Botanic Biodiversity Caused by the Cyclone Sidr. Paper prepared for the 1st World Sustainability Forum. Electronic conference at www.sciforum.net. 1–30 November. _____. 2013. Cyclone Sidr Impacts on the Sundarbans Floristic Diversity. Earth Science Research. 2 (2). pp. 62–79. Blanc, E. and E. Strobl. 2016. Assessing the Impact of Typhoons on Rice Production in the Philippines. Journal of Applied Meteorology and Climatology. 55 (4). pp. 993–1007. Brakenridge, R. and E. Anderson. 2006. MODIS-Based Flood Detection, Mapping and Measurement: The Potential for Operational Hydrological Applications. In J. Marsalek, G. Stancalie, and G. Balint, eds. Transboundary Floods: Reducing Risks Through Flood Management. Berlin: Springer Dordrecht. pp. 1–12. Brun, J. and A. P. Barros. 2013. Vegetation Activity Monitoring as an Indicator of Eco- Hydrological Impacts of Extreme Events in the Southeastern USA. International Journal of Remote Sensing. 34 (2). pp. 519–544. Byun, Y., Y. Han, and T. Chae. 2015. Image Fusion-Based Change Detection for Flood Extent Extraction Using Bi-Temporal Very High-Resolution Satellite Images. Remote Sensing. 7 (8). pp. 10347–10363. Cao, Q. D. and Y. Choe. 2018. Deep Learning Based Damage Detection on Post-Hurricane Satellite Imagery. arXiv preprint arXiv:1807.01688. _____. 2020a. Building Damage Annotation on Post-Hurricane Satellite Imagery Based on Convolutional Neural Networks. Natural Hazards. 103 (3). pp. 3357–3376. _____. 2020b. Post-Hurricane Damage Assessment Using Satellite Imagery and Geolocation Features. arXiv preprint arXiv:2012.08624. Capellades, M. A., S. Reigber, and M. Kunze. 2009. Storm Damage Assessment Support Service in the US Corn Belt Using RapidEye Satellite Imagery. Paper prepared for the Remote Sensing for Agriculture, Ecosystems, and Hydrology XI Conference. Berlin. 1–3 September. In Proceedings of SPIE, the International Society for Optical Engineering. 2009. Vol. 7472. Charrua, A. B., R. Padmanaban, P. Cabral, S. Bandeira, and M. M. Romeiras. 2021. Impacts of the Tropical Cyclone Idai in Mozambique: A Multi-Temporal Landsat Satellite Imagery Analysis. Remote Sensing. 13 (2). 201. Chehata, N., C. Orny, S. Boukir, D. Guyon, and J. P. Wigneron. 2014. Object-Based Change Detection in Windstorm-Damaged Forest Using High-Resolution Multispectral Images. International Journal of Remote Sensing. 35 (13). pp. 4758–4777. 27 Phiri, D., M. Simwanda, and V. Nyirenda. 2021. Mapping the Impacts of Cyclone Idai in Mozambique Using Sentinel-2 and OBIA Approach. South African Geographical Journal. 103 (2). pp. 237–258. Radhika, S., Y. Tamura, and M. Matsui. 2015. Cyclone Damage Detection on Building Structures from Pre- and Post-Satellite Images Using Wavelet Based Pattern Recognition. Journal of Wind Engineering and Industrial Aerodynamics. 136. pp. 23–33. Radio New Zealand (RNZ). 2018. Cyclones Josie and Keni Cost Fiji $US3 Million. 1 May. https://www.rnz.co.nz/international/pacific-news/356391/cyclones-josie-and-keni-cost-fiji- us3-million. Rahman, M. S. 2019. Remote Sensing Based Rapid Assessment of Flood Crop Damage. Doctoral dissertation. George Mason University. Rahman, M. S. and L. Di. 2020. A Systematic Review on Case Studies of Remote-Sensing- Based Flood Crop Loss Assessment. Agriculture. 10 (4). 131. Rahman, M. S., L. Di, E. Yu, L. Lin, and Z. Yu. 2021. Remote Sensing Based Rapid Assessment of Flood Crop Damage Using Novel Disaster Vegetation Damage Index (DVDI). International Journal of Disaster Risk Science. 12 (1). pp. 90–110. Ramlal, B., D. Davis, and K. De Bellott. 2018. A Rapid Post-Hurricane Building Damage Assessment Methodology Using Satellite Imagery. West Indian Journal of Engineering. 41 (1). pp. 74–83. Ramsey, E. W. I., D. K. Chappell, and D. G. Baldwin. 1997. AVHRR Imagery Used to Identify Hurricane Damage in a Forested Wetland of Louisiana. Photogrammetric Engineering and Remote Sensing. 63 (3). pp. 293–297. Reliefweb. 2016a. Fiji: TC Winston Cyclone Path. United Nations Office for the Coordination of Humanitarian Affairs (OCHA) Services. https://reliefweb.int/map/fiji/fiji-tc-winston- cyclone-path-26-feb-2016 (accessed 4 November 2021). Reliefweb. 2016b. Tropical Cyclone Winston. OCHA Services. https://reliefweb.int/disaster/tc- 2016-000014-fji (accessed 29 November 2021). Reliefweb. 2018. Republic of Fiji: Tropical Cyclone Josie and Tropical Cyclone Keni Rapid Gender, Protection and Inclusion Analysis. OCHA Services. https://reliefweb.int/report/fiji/republic-fiji-tropical-cyclone-josie-and-tropical-cyclone-keni- rapid-gender-protection (accessed 29 November 2021). Reliefweb. 2020. Tropical Cyclone Harold and COVID-19: A Double Blow to the Pacific Islands. OCHA Services. https://reliefweb.int/report/fiji/tropical-cyclone-harold-and-covid-19- double-blow-pacific-islands (accessed 29 November 2021). Reserve Bank of Fiji. 2021. Gross Domestic Product by Industry at Constant Prices. https://www.rbf.gov.fj/wp-content/uploads/2021/10/5.2-Gross-Domestic-Product-by- Industry-at-Constant-Prices.xlsx (accessed 4 November 2021). 28 Rodgers, J. C., A. W. Murrah, and W. H. Cooke. 2009. The impact of Hurricane Katrina on the Coastal Vegetation of the Weeks Bay Reserve, Alabama from NDVI Data. Estuaries and Coasts. 32 (3). pp. 496–507. Rossi, E., J. Rogan, and L. Schneider. 2013. Mapping Forest Damage in Northern Nicaragua After Hurricane Felix (2007) Using MODIS Enhanced Vegetation Index Data. GIScience & Remote Sensing. 50 (4). pp. 385–399. Rudner, T. G., M. Rußwurm, J. Fil, R. Pelich, B. Bischke, V. Kopacková, and P. Bilinski. 2018. Rapid Computer Vision-Aided Disaster Response via Fusion of Multiresolution, Multisensor, and Multitemporal Satellite Imagery. In Proceedings of the First Workshop on AI for Social Good. Neural Information Processing Systems (NIPS-2018). Montreal, Quebec. pp. 3–8. Scarsi, A., W. J. Emery, G. Moser, F. Pacifici, and S. B. Serpico. 2014. An Automated Flood Detection Framework for Very High Spatial Resolution Imagery. Paper prepared for the 2014 IEEE Geoscience and Remote Sensing Symposium. Quebec City. 13–18 July. Senthilnath, J., S. N. Omkar, V. Mani, and P. G. Diwakar. 2013. Multi-Temporal Satellite Imagery for Flood Damage Assessment. Journal of the Indian Institute of Science. 93 (1). pp. 105–116. Sheng, Y., P. Gong, and Q. Xiao. 2001. Quantitative Dynamic Flood Monitoring with NOAA AVHRR. International Journal of Remote Sensing. 22 (9). pp. 1709–1724. Shrestha, R., L. Di, G. Y. Eugene, L. Kang, Y. Z. Shao, and Y. Q. Bai. 2017. Regression Model to Estimate Flood Impact on Corn Yield Using MODIS NDVI and USDA Cropland Data Layer. Journal of Integrative Agriculture. 16 (2). pp. 398–407. Shrestha, R., L. Di, G. Yu, Y. Shao, L. Kang, and B. Zhang. 2013. Detection of Flood and Its Impact on Crops Using NDVI-Corn Case. Paper prepared for the Agro-Geoinformatics 2013: The Second International Conference on Agro-Geoinformatics. Fairfax, VA. 12–16 August. Singh, Y. P., S. M. Veerabhadrappa, N. Nishant, and D. Chutia. 2019. Flood Induced Crop Loss Assessment Based on Satellite Derived Flood Parameters and Vegetation Indices in North-Eastern Region of India. Paper prepared for the National Symposium on Innovations in Geospatial Technology for Sustainable Development with Special Emphasis on NER. Shillong. 20–22 November. Sivanpillai, R., K. M. Jacobs, C. M. Mattilio, and E. V. Piskorski. 2021. Rapid Flood Inundation Mapping by Differencing Water Indices from Pre- and Post-Flood Landsat Images. Frontiers of Earth Science. 15. pp. 1–11. Son, N. T., C. F. Chen, C. R. Chen, and L. Y. Chang. 2013. Satellite-Based Investigation of Flood-Affected Rice Cultivation Areas in Chao Phraya River Delta, Thailand. ISPRS Journal of Photogrammetry and Remote Sensing. 86. pp. 77–88. 29 Sun, D., Y. Yu, and M. D. Goldberg. 2011. Deriving Water Fraction and Flood Maps from MODIS Images Using a Decision Tree Approach. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 4 (4). pp. 814–825. Sun, D., Y. Yu, R. Zhang, S. Li, and M. D. Goldberg. 2012. Towards Operational Automatic Flood Detection Using EOS/MODIS Data. Photogrammetric Engineering & Remote Sensing. 78 (6). pp. 637–646. United Nations Children's Fund (UNICEF). Cyclone Winston. https://www.unicef.org.au/ourwork/unicef-in-emergencies/cyclone-winston (accessed 29 November 2021). Van der Sande, C. J., S. M. De Jong, and A. P. J. De Roo. 2003. A Segmentation and Classification Approach of IKONOS-2 Imagery for Land Cover Mapping to Assist Flood Risk and Flood Damage Assessment. International Journal of Applied Earth Observation and Geoinformation. 4 (3). pp. 217–229. Voigt, S., F. Giulio-Tonolo, J. Lyons, J. Kučera, B. Jones, T. Schneiderhan, G. Platzeck, K. Kaku, M. Kumar Hazarika, L. Czaran, S. Li, W. Pedersen, G. Kadiri James, C. Proy, D. Macharia Muthike, J. Bequignon, and D. Guha-Sapir. 2016. Global Trends in Satellite- Based Emergency Mapping. Science. 353 (6296). pp. 247–252. Wang, F. and E. J. D’Sa. 2010. Potential of MODIS EVI in Identifying Hurricane Disturbance to Coastal Vegetation in the Northern Gulf of Mexico. Remote Sensing. 2 (1). pp. 1–18. Wang, F. and Y. J. Xu. 2010. Comparison of Remote Sensing Change Detection Techniques for Assessing Hurricane Damage to Forests. Environmental Monitoring and Assessment. 162 (1). pp. 311–326. Wang, M. and H. Xu. 2018. Remote Sensing-Based Assessment of Vegetation Damage by a Strong Typhoon (Meranti) in Xiamen Island, China. Natural Hazards. 93 (3). pp. 1231– 1249. Wang, W., J. J. Qu, X. Hao, Y. Liu, and J. A. Stanturf. 2010. Post-Hurricane Forest Damage Assessment Using Satellite Remote Sensing. Agricultural and Forest Meteorology. 150 (1). pp. 122–132. World Food Programme (WFP). 2017. Tropical Cyclone Winston: Fiji Government and World Food Programme Joint Emergency Response – Lessons Learned Workshop Report, April 2017. Suva, Fiji: WFP Pacific Office. Wheeler, B. J. and H. A. Karimi. 2020. Deep Learning-Enabled Semantic Inference of Individual Building Damage Magnitude from Satellite Images. Algorithms. 13 (8). 195. Yu, G., L. Di, B. Zhang, Y. Shao, R. Shrestha, and L. Kang. 2013. Remote-Sensing-Based Flood Damage Estimation Using Crop Condition Profiles. Paper prepared for the Agro- Geoinformatics 2013: The Second International Conference on Agro-Geoinformatics. Fairfax, VA. 12–16 August. 30 Zhang, B., L. Di, G. Yu, Y. Shao, R. Shrestha, and L. Kang. 2013a. A Web Service Based Application Serving Vegetation Condition Indices for Flood Crop Loss Assessment. Paper prepared for the Agro-Geoinformatics 2013: The Second International Conference on Agro-Geoinformatics. Fairfax, VA. 12–16 August. Zhang, R., D. Sun, Y. Yu, and M. D. Goldberg. 2012. Mapping Nighttime Flood from MODIS Observations Using Support Vector Machines. Photogrammetric Engineering & Remote Sensing. 78 (11). pp. 1151–1161. Zhang, X., Y. Wang, H. Jiang, and X. Wang. 2013b. Remote-Sensing Assessment of Forest Damage by Typhoon Saomai and its Related Factors at Landscape Scale. International Journal of Remote Sensing. 34 (21). pp. 7874–7886. Zhu, X. X., D. Tuia, L. Mou, G. S. Xia, L. Zhang, F. Xu, and F. Fraundorfer. 2017. Deep Learning in Remote Sensing: A Comprehensive Review and List of Resources. IEEE Geoscience and Remote Sensing Magazine. 5 (4). pp. 8–36. ASIAN DEVELOPMENT BANK ASIAN DEVELOPMENT BANK 6 ADB Avenue, Mandaluyong City 1550 Metro Manila, Philippines www.adb.org Nowcasting from Space Impact of Tropical Cyclones on Fiji’s Agriculture Satellite imagery can be a source of easily available, fast, affordable, and accurate data for assessing disaster impacts. This study investigates the feasibility of using remote sensing data for post-disaster damage assessment. It focuses on Fiji, its agriculture sector, and the tropical cyclones that wreaked havoc on the country in recent years. Findings of the study show that remote sensing data, when combined with pre-event socioeconomic and demographic information, can better explain the identified changes in the vegetation index, thereby improving both nowcasting and post-disaster damage assessments of tropical cyclones on agriculture. About the Asian Development Bank ADB is committed to achieving a prosperous, inclusive, resilient, and sustainable Asia and the Pacific, while sustaining its efforts to eradicate extreme poverty. Established in 1966, it is owned by 68 members —49 from the region. Its main instruments for helping its developing member countries are policy dialogue, loans, equity investments, guarantees, grants, and technical assistance. ADB ECONOMICS WORKING PAPER SERIES NO. 676 January 2023 NOWCASTING FROM SPACE IMPACT OF TROPICAL CYCLONES ON FIJI’S AGRICULTURE Ilan Noy, Elodie Blanc, Madhavi Pundit, and Tomas Uher