remote sensing

Article Evaluating the Impact of the 2020 Derecho on Corn and Soybean Fields Using Synthetic Aperture Radar

Mehdi Hosseini 1,*, Hannah R. Kerner 1 , Ritvik Sahajpal 1 , Estefania Puricelli 1, Yu-Hsiang Lu 1, Afolarin Fahd Lawal 1, Michael L. Humber 1, Mary Mitkish 1, Seth Meyer 2 and Inbal Becker-Reshef 1 1 NASA-Harvest, Department of Geographical Sciences, University of Maryland, College Park, MD 20740, USA; [email protected] (H.R.K.); [email protected] (R.S.); [email protected] (E.P.); [email protected] (Y.-H.L.); [email protected] (A.F.L.); [email protected] (M.L.H.); [email protected] (M.M.); [email protected] (I.B.-R.) 2 Food and Agricultural Policy Research Institute, University of Missouri, Columbia, MO 65211, USA; [email protected] * Correspondence: [email protected]; Tel.: +1-314-20-205-6263

 Received: 27 October 2020; Accepted: 25 November 2020; Published: 26 November 2020 

Abstract: On 10 August 2020, a series of intense and fast-moving windstorms known as a derecho caused widespread damage across Iowa’s (the top US corn-producing state) agricultural regions. This severe event bent and flattened crops over approximately one-third of the state. Immediate evaluation of the disaster’s impact on agricultural lands, including maps of crop damage, was critical to enabling a rapid response by government agencies, insurance companies, and the agricultural supply chain. Given the very large area impacted by the disaster, satellite imagery stands out as the most efficient means of estimating the disaster impact. In this study, we used time-series of Sentinel-1 data to detect the impacted fields. We developed an in- crop type map using Harmonized Landsat and Sentinel-2 data to assess the impact on important commodity crops. We intersected a SAR-based damage map with an in-season crop type map to create damaged area maps for corn and soybean fields. In total, we identified 2.59 million acres as damaged by the derecho, consisting of 1.99 million acres of corn and 0.6 million acres of soybean fields. Also, we categorized the impacted fields to three classes of mild impacts, medium impacts and high impacts. In total, 1.087 million acres of corn and 0.206 million acres of soybean were categorized as high impacted fields.

Keywords: SAR; derecho; crop damage; corn; soybean

1. Introduction Iowa is the biggest producer of corn and the second-largest producer of soybeans in the United States. In 2019, about 13.5 million acres of corn and 9.2 million acres of soybeans were planted in Iowa according to the United States Department of Agriculture’s National Agricultural Statistics Service (USDA-NASS) [1]. This amounted to roughly one-sixth of the total corn production and one-seventh of soybean production nationwide (USDA-NASS) [1]. USDA and Iowa Department of Agriculture and Land Stewardship collaborated to estimate the impacts of the derecho over Iowa agricultural fields. Their initial official numbers were huge as they crossed the track with counties and the area planted in each. They estimated that 8.2 million acres of corn and 5.6 million acres of soybeans may have been impacted by the storm and from these 3.57 million acres of corn and 2.5 million acres of soybeans were hardest hit by the derecho (https://iowaagriculture.gov/news/updated-derecho-impact- estimates-08142020). Although, as it is discussed further in this paper, these numbers were significantly

Remote Sens. 2020, 12, 3878; doi:10.3390/rs12233878 www.mdpi.com/journal/remotesensing Remote Sens. 2020, 12, 3878 2 of 16 reduced in the USDA report 2-3 months after the storm. In this study, it is demonstrated that using Synthetic Aperture Radar (SAR) satellite data, we were able to give a much more detailed analysis and estimate in a short period of time (about 2 weeks) after the storm that the satellite data were available for the whole region. One of the missions of the NASA-Harvest consortium at the University of Maryland is to rapidly assess the satellite data after the natural hazards such as the derecho happen and provide information to the government, insurance companies and farmers for the rapid response actions. The European Space Agency Sentinel-1A and 1B satellites are C-band SAR satellites which provide -free observations of the Earth’s surface since radar waves penetrate . In the Interferometric Wide (IW) swath mode, Sentinel-1A and 1B have a swath width of 250 km and spatial resolution of 5 m by 20 m [2]. This wide swath width is important for monitoring crop lands over large areas. Each of these satellites has a revisit time of 12 days, and together the Sentinel-1 satellites can have a revisit rate as frequent as 6 days. SAR signals are highly sensitive to the geometry and structure of crops. This sensitivity combined with the cloud-free nature of SAR made Sentinel-1 an ideal source of data for assessing crop damage resulting from the Iowa derecho. Thanks to the increasing number of SAR-equipped satellites and the open license of Sentinel-1 data, SAR data have been widely used in recent years for agricultural applications including crop monitoring. For example, Alonso-González et al., 2015 used time-series of polarimetric SAR data for crop growth monitoring [3]. They used elements of Pauli decomposition and demonstrated that using those elements it is possible to detect the changes on types of backscatter between different acquisitions and use them for crop growth monitoring. Lopez-Sanchez et al., 2012 [4] used polarimetric interferometric SAR (PolInSAR) data to study crop height variations. They used airborne L-band SAR data and demonstrated that even this low frequency SAR data could be used to derive crop height for rape and corn. Kontgis et al., 2017 [5] used time-series of Sentinel-1 VV and VH intensities to map rice paddy extent. They demonstrated that using those time-series and machine learning algorithms, it is feasible to map rice paddy extent with overall accuracy of 96.3%. Velso et al., 2017 [6] used integration of Sentinel-1 and 2 to study crop growth monitoring of five globally important crops including wheat, rapeseed, corn, soybean and sunflower. They demonstrated the high potential of SAR data with or without optical data for crop growth monitoring. RADARSAT-2 C-band data and RapidEye optical data have been used and compared by Hosseini et al., 2019 for estimation of corn biomass, with similar accuracies from both satellites for dry biomass estimation [7]. They also demonstrated that while optical satellite data outperformed SAR data in the early to mid-growing season, SAR provided more accurate estimates in the late growing season when optical signals are saturated. Mandal et al., 2019 [8] used both Sentinel-1 and RADARSAT-2 data for corn leaf area index (LAI) estimation in Canada and Poland, and showed that machine learning algorithms could be used to predict corn LAI from SAR with high accuracy (R2 of ~0.8 and mean absolute error of 0.5). Hosseini et al., 2015 [9] showed similar results for corn and soybean LAI estimates using a semi-empirical water cloud model. Bell et al., 2019 [10] studied the potential of using Sentinel-1 and Moderate Resolution Imaging Spectroradiometer (MODIS) data for monitoring vegetation damage resulting from in June 2016 in Iowa. Their work showed that both Normalized Difference Vegetation Index (NDVI) derived from MODIS and the VV and VH intensities from SAR are changed significantly for damaged fields compared to unaffected fields. Surek and Nador, 2015 [11] used polarimetric RADARSAT-2 data for monitoring damage to sunflower and corn fields from drought and in Hungary, observing that the SAR signals dropped by 1.5–2.5 dB after a drought and that NDVI decreased between 5–20 percent after a storm. Finally, Chauhan et al., 2020 [12] showed that VH backscatter, NIR and red-edge are highly sensitive to lodging in Wheat in a study conducted in Italy. Optical satellite data have also been used widely for crop damage monitoring. For example, Silleos et al., 2002 [13] used NDVI derived from SPOT images for damage monitoring over agricultural fields. They applied Maximum Likelihood classification using NDVI to detect the damaged crop area at the field level, and correctly detected 73% of the total damaged crop area. In another study, Young et al., 2004 [14] used NDVI and Modified Soil Adjusted Vegetation Index (MSAVI) from Landsat-TM images Remote Sens. 2020, 12, 3878 3 of 16 forRemote crop Sens. hail 2020 damage, 12, x FOR monitoring. PEER REVIEW They detected 90–95% of the damaged sites using NDVI and3 of 83–89% 16 using MSAVI. The previous studies showed high potential of optical satellite data for crop damage damaged sites using NDVI and 83–89% using MSAVI. The previous studies showed high potential monitoring. However, in rapid response situations, access to data in a short time-frame before and of optical satellite data for crop damage monitoring. However, in rapid response situations, access to after an event is critical. In many situations, including the 2020 Iowa derecho, limits the data in a short time-frame before and after an event is critical. In many situations, including the 2020 available optical data. Iowa derecho, cloud cover limits the available optical data. In this study, we used only SAR data for crop damage assessment due to cloudy weather conditions In this study, we used only SAR data for crop damage assessment due to cloudy weather beforeconditions and after before the and derecho. after the The derecho. main objective The main of objective this work of was this towork assess was the to damageassess the caused damage by the 2020caused Iowa by derechothe 2020 Iowa using derecho remote us sensinging remote data. sensing We have data. evaluated We have evaluated the potential the potential of using of SAR using data forSAR rapidly data for assessing rapidly crop assessing damage crop resulting damage fromresult theing 2020from Iowathe 2020 derecho Iowa whenderecho little when to no little in-situ to no data arein-situ available. data are We available. produced We damaged produced area damaged and damage area and severity damage maps severity based maps on based SAR dataon SAR before data and afterbefore the and derecho after the and derecho estimated and the estimated damaged the area dama ofged corn area and of soybean corn and crops. soybean Our crops. results Our demonstrate results thatdemonstrate SAR time-series that SAR change time-series monitoring change is amonitoring promising is means a promising for rapidly means detecting for rapidly changes detecting over crop fieldschanges that over are relatedcrop fields to derecho that are orrela otherted to storm derecho impacts. or other storm impacts.

2.2. Data Data

2.1.2.1. Study Study AreaArea OurOur studystudy area is is the the state state of of Iowa, Iowa, United United States, States, which which was hit was hardest hit hardest by the byderecho. the derecho. We Wegenerated generated damage damage statistics statistics for the for whole the state whole as statewell as as for well each as county for each separately. county Figure separately. 1 shows Figure all 1 shows99 counties all 99 of counties Iowa. of Iowa.

FigureFigure 1.1. The state of Iowa, Iowa, the the sampling sampling field field locations, locations, and and the the counties counties of Iowa, of Iowa, USA USA

2.2.2.2. Ground Ground TruthTruth DataData WeWe obtained obtained aa limitedlimited number of of ground ground truth truth observations observations of ofcrop crop type type and and derecho derecho damage damage intensityintensity from from geo-tagged geo-tagged imagesimages provided provided by by Iowa Iowa farmers. farmers. These These 14 14 fields fields are are distributed distributed in Boone in Boone countycounty (5 (5 fields), fields), MarshallMarshall county (7 (7 fields), fields), and and Hardin Hardin county county (2 (2fields) fields) (Figure (Figure 1).1 For). For each each image, image, wewe extracted extracted the the locationlocation (latitude and and longitude) longitude) and and heading heading information information from from the theimage image metadata. metadata. SinceSince most most imagesimages of fields fields were were taken taken from from the the road road or oredge edge of the of thefield, field, we moved we moved points points ~60 m ~60 in m inthe the heading heading direction direction to toensure ensure points points would would lie lieinside inside the the field. field. Figure Figure 2 shows2 shows the the catastrophic catastrophic impactsimpacts of of the the derecho derecho onon threethree corn fields. fields.

Remote Sens. 2020, 12, 3878 4 of 16 Remote Sens. 2020, 12, x FOR PEER REVIEW 4 of 16

Figure 2. Derecho impacts on corn fieldsfields in Iowa. Ph Photooto credit: Lance Lillibridge (Iowa farmer). 2.3. Satellite Data 2.3. Satellite Data We used dual polarization Sentinel-1 ground range detected (GRD) products in this study. We used dual polarization Sentinel-1 ground range detected (GRD) products in this study. To To generate change maps, we processed 10 scenes from 29 July 2020 to 22 August 2020 to cover the generate change maps, we processed 10 scenes from 29 July 2020 to 22 August 2020 to cover the whole whole study area before and after the storm. We processed all data using the European Space Agency study area before and after the storm. We processed all data using the European Space Agency (ESA) (ESA) Sentinel-1 Toolbox (SNAP). First, we used the orbital file to update the geometry information of Sentinel-1 Toolbox (SNAP). First, we used the orbital file to update the geometry information of the the satellite in the image header. We applied the radiometric conversion to convert the pixel values to satellite in the image header. We applied the radiometric conversion to convert the pixel values to backscatter coefficients in the decibel unit. Next, we applied a 5 5 Gamma Map speckle filter to backscatter coefficients in the decibel unit. Next, we applied a 5×5× Gamma Map speckle filter to minimize the effects of speckle noise. Finally, we performed terrain correction using the Range-Doppler minimize the effects of speckle noise. Finally, we performed terrain correction using the Range- Terrain Correction algorithm available in SNAP. We used the Shuttle Radar Topography Mission Doppler Terrain Correction algorithm available in SNAP. We used the Shuttle Radar Topography (SRTM) digital elevation model (DEM) with 1 arc-second resolution with bilinear interpolation for Mission (SRTM) digital elevation model (DEM) with 1 arc-second resolution with bilinear ortho-rectification and resampled the spatial resolution to 30 m. interpolation for ortho-rectification and resampled the spatial resolution to 30 m. We also used and compared the time-series of Sentinel-1 images from 10 July to 27 August for both We also used and compared the time-series of Sentinel-1 images from 10 July to 27 August for 2019 and 2020. As explained in the methodology section, the objective of this analysis was to find the both 2019 and 2020. As explained in the methodology section, the objective of this analysis was to sensitivity of each Sentinel-1 polarization band to derecho impacts on crops and to find the threshold find the sensitivity of each Sentinel-1 polarization band to derecho impacts on crops and to find the for applying the change detection algorithm. In addition to the SAR observations processed using threshold for applying the change detection algorithm. In addition to the SAR observations processed SNAP, we used the Sentinel-1 Radiometric Terrain Corrected (RTC) Data on Amazon Web Services using SNAP, we used the Sentinel-1 Radiometric Terrain Corrected (RTC) Data on Amazon Web (AWS), which was released by Indigo Ag during our study (https://sentinel-s1-rtc-indigo-docs.s3-us- Services (AWS), which was released by Indigo Ag during our study (https://sentinel-s1-rtc-indigo- west-2.amazonaws.com/index.html). These data are corrected from speckle noise and the data are docs.s3-us-west-2.amazonaws.com/index.html). These data are corrected from speckle noise and the re-tiled to Sentinel-2 tile grid. To avoid the time-consuming preprocessing of Sentinel-1 data, we used data are re-tiled to Sentinel-2 tile grid. To avoid the time-consuming preprocessing of Sentinel-1 data, this freely accessible AWS dataset to extract time-series of VV and VH intensities of Sentinel-1 for July we used this freely accessible AWS dataset to extract time-series of VV and VH intensities of Sentinel- and August of 2019 and 2020 over the study area. 1 for July and August of 2019 and 2020 over the study area. In addition to SAR data from Sentinel-1, we used optical observations from the Harmonized In addition to SAR data from Sentinel-1, we used optical observations from the Harmonized Landsat and Sentinel-2 (HLS) dataset [15] as inputs for the in-season crop type classification method. Landsat and Sentinel-2 (HLS) dataset [15] as inputs for the in-season crop type classification method. The HLS dataset provides both Landsat-8 and Sentinel-2 observations at 30 m resolution and aligned The HLS dataset provides both Landsat-8 and Sentinel-2 observations at 30 m resolution and aligned band-passes, re-gridded to the Military Gridded Reference System (MGRS) used by Sentinel-2. We used band-passes, re-gridded to the Military Gridded Reference System (MGRS) used by Sentinel-2. We the cloud mask provided by HLS to remove pixels with clouds and filled in missing data using used the cloud mask provided by HLS to remove pixels with clouds and filled in missing data using linear interpolation. linear interpolation. 3. Methodology 3. Methodology 3.1. Damaged Area Detection 3.1. Damaged Area Detection The damaged area was detected based on a threshold value on Sentinel-1 cross-polarization (VH)The intensity damaged variations area was between detected images based of on before a thre theshold storm value and on after Sentinel the storm.-1 cross-polarization We found that VH(VH) intensity intensity was variations more sensitive between thanimages VV of to before the derecho the storm damage and after by analysingthe storm. time-seriesWe found that of both VH intensity was more sensitive than VV to the derecho damage by analysing time-series of both polarization bands. For each Sentinel-1 path and frame, the first available images before and after the

Remote Sens. 2020, 12, 3878 5 of 16 polarization bands. For each Sentinel-1 path and frame, the first available images before and after the storm were used for change detection. Pixels that had higher variations than the threshold value were detected as impacted pixels. In the next step, the impacted pixels were intersected with an in-season crop type map to determine the damage estimates for corn and soybean. We explain each of these steps in the following sections.

3.2. Threshold Estimation Prior work demonstrated that cross-polarization has higher sensitivity to crops than the co-polarizations [16–18]. In this study, both co- and cross-polarizations were analyzed to further verify their sensitivities to storm impacts on crops. To find the right thresholds to capture changes in VV and VH intensities caused by the derecho, we studied the time series of both polarization bands for 2019 and 2020. We randomly selected 50 corn fields and 50 soybean fields over the affected region, then selected 3 sites in different parts of each field (resulting in 150 corn and 150 soybean sites total). Our observation demonstrated that using more fields and sites, does not change the time-series of VV and VH intensities significantly. The 50 corn fields and the 50 soybean fields were distributed from East-Iowa to the West-Iowa to have good spatial distributions. Also, we made sure that none of the fields that we had in-situ data for included in those 100 fields. We extracted the VV and VH intensities at each site using an average 5 by 5 window. We used the in-season 2020 crop type map that was generated in this study and the 2019 Cropland Data Layer (CDL) produced by the USDA [19] to identify locations of corn and soybean fields. While finding a robust threshold for detecting impacted fields would require a high number of in-situ reference points, we determined approximate thresholds for each polarization based on analysis of the time series.

3.3. In-Season Crop Type Mapping Since the derecho happened in August 2020, an in-season crop type map was needed to estimate the extent of the derecho damage on corn and soybeans. The Cropland Data Layers (CDL) produced by the USDA each year [19] are not available until the year following the map year, so the 2020 CDL will not be available until 2021. We used the method described in [20] to produce an in-season crop type map using satellite observations available between January and August 2020. This method uses satellite observations from the Harmonized Sentinel-2 and Landsat (HLS) and Sentinel-1 datasets and normalizes inputs by growth stage to minimize inter-annual domain shift. Inter-annual domain shift occurs when the distribution of values in the satellite inputs to the model is shifted significantly in the test data compared to the training data, e.g., due to delayed planting and crop growth. We used satellite inputs from the greenup and peak growth stages with a random forest classifier to produce mid-season predictions of corn, soybean, or other for each 30 m pixel. Satellite data for 2017–2019 were used for the training of the algorithm. The reported accuracy in [20] for mid-season (between July and September) crop type classification using the growth stage-normalized random forest was 76.5% for 2019 in a study area in Illinois. The other class encompasses other crops, urban areas, forests, water, and any other land cover class that is not corn or soybean. The mid-season crop type maps used in this study can be accessed through the NASA Harvest Data Portal at https://data.harvestportal.org/dataset/iowa-in-season-crop-type-map-2020.

3.4. Derecho Impact on Crops The process was done using ESRI’s ArcGIS software. Figure3 shows a flowchart of the analysis. First, we generated a difference map using the images of before and after the storm. Then, we filtered out pixels that have change values lower than the threshold as non-impacted pixels. Next, we converted the change raster map to vector polygon format and it was reprojected to the crop type map coordinate system. It is assumed that the impacted area by the storm cannot be a very small island polygon (i.e., a signal polygon that is not connected with any other polygon and with an area less than 10 pixels). Therefore, those small island polygons were removed from the analysis. The study area was covered Remote Sens. 2020, 12, x FOR PEER REVIEW 6 of 16

3.4. Derecho Impact on Crops The process was done using ESRI’s ArcGIS software. Figure 3 shows a flowchart of the analysis. First, we generated a difference map using the images of before and after the storm. Then, we filtered out pixels that have change values lower than the threshold as non-impacted pixels. Next, we converted the change raster map to vector polygon format and it was reprojected to the crop type map coordinate system. It is assumed that the impacted area by the storm cannot be a very small island polygon (i.e., a signal polygon that is not connected with any other polygon and with an area Remoteless than Sens. 102020 pixels)., 12, 3878 Therefore, those small island polygons were removed from the analysis.6 ofThe 16 study area was covered by five Sentinel-1 images. We generated the change map for all five images andby five then Sentinel-1 merged images. them. We We masked generated out the changes change mapthat forwere all outside five images of the and study then mergedarea. Finally, them. Wewe intersectedmasked out the changes crop damage that were map outside with of the the in-seaso study area.n crop Finally, type map we intersected to produce the the crop damage damage map map for witheach thecrop in-season type. crop type map to produce the damage map for each crop type.

Figure 3. Flowchart of the processing steps to generate the derecho impacted fields map. Figure 3. Flowchart of the processing steps to generate the derecho impacted fields map. 4. Results 4. Results 4.1. In-Season Crop Type Map 4.1. In-Season Crop Type Map Figure4 shows the in-season crop type map produced using HLS observations available between JanuaryFigure and 4 August shows 2020 the usingin-season the method crop type described map inproduced Kerner etusing al., 2020 HLS [20 observations]. Ten of the 14available ground betweenreference January points wereand August correctly 2020 classified using the as corn,method though described we emphasize in Kerner thatet al., this 2020 sample [20]. Ten size of is toothe 14small ground to provide reference a representative points were assessmentcorrectly classifi of theed crop as corn, type though classification we emphasize accuracy that across this the sample entire sizemap is of too Iowa. small In to future provide work, a repr we planesentative to develop assessment crowd-sourcing of the crop capabilities type classification to increase accuracy the number across of theground entire reference map of pointsIowa. weIn future can collect work, for we evaluating plan to develop in-season crowd-sourcing crop type classification capabilities products, to increase and willthe number evaluate of the ground map accuracyreferencecompared points we tocan the collec 2020t for Cropland evaluating Data in-season Layer (CDL) cropwhen type classification it is released products,in 2021. To and provide will evaluate additional the insights map accuracy about the compared in-season to cropthe 2020 type mapCropland accuracy, Data weLayer compared (CDL) whenthe pixel-wise it is released classifications in 2021. to To the provide 2018 CDL. additional Of the pixels insights classified about as the “other” in-season in our crop in-season type map 2020 accuracy,map, 30% we were compared classified the as grasslandpixel-wise/ pastureclassifications in the 2018to the CDL, 2018 17% CDL. as forest,Of the 15% pixels as corn,classified 14% as “other”developed, in our 11% in-season as soybean, 2020 4% map, as wetlands,30% were classified 4% as hay, as 3% grassland/pasture as water, and 2% in as the alfalfa 2018 (roundingCDL, 17% as to forest,nearest 15% percent; as corn, classes 14% constituting as developed,< 1% 11% of as “other” soybean, pixels 4% not as wetlands, discussed). 4% Since as hay, the total3% as planted water, areaand 2%of corn as alfalfa and soybeans (rounding was to approximatelynearest percent; the classes same inconstituting 2018 and 2020< 1% [ 21of, 22“other”], this pixels analysis not suggests discussed). that ~26%Since ofthe pixels total classifiedplanted area as “other” of corn in and our soybeans in-season was 2020 approximately map should have the beensame classifiedin 2018 and as either 2020 [21,22],corn or soybeans.this analysis Of thesuggests pixels that classified ~26% asof cornpixels or classified soybean in as our “other” in-season in our 2020 in-season map, only 2020< 0.13% map wereshould classified have been as somethingclassified as other eith thaner corn corn or orsoybeans. soybean Of in the pixels 2018 CDL. classified as corn or soybean in

RemoteRemote Sens. 2020Sens., 202012, x, FOR12, x PEERFOR PEER REVIEW REVIEW 7 of 167 of 16 our in-seasonour in-season 2020 2020 map, map, only only < 0.13% < 0.13% were were classified classified as something as something other other than than corn corn or soybean or soybean in the in the Remote Sens. 2020, 12, 3878 7 of 16 20182018 CDL. CDL.

FigureFigureFigure 4. In-season 4.4. In-season crop croptype typetypemap mapmapproduced producedproduced using usingusing HLS HLSHLobservationsS observationsobservations available availableavailable in August inin AugustAugust 2020. 2020.2020. 4.2. Time-Series Analysis 4.2. Time-Series4.2. Time-Series Analysis Analysis As discussed in the Methodology section, we assessed the sensitivity of VV and VH time series As discussedAs discussed in the in Methodology the Methodology section, section, we asse we ssedasse ssedthe sensitivity the sensitivity of VV of andVV VHand timeVH time series series to crop damage impacts from the storm. We plotted the mean intensities for 150 corn sites and the to cropto crop damage damage impacts impacts from from the storm.the storm. We plotteWe plotted thed meanthe mean intensities intensities for 150 for corn150 corn sites sitesand andthe the 150 soybean sites from 10 July 2020 to 27 August 2020 for both polarizations and for 2019 and 2020 in 150 soybean150 soybean sites sitesfrom from 10 July 10 July2020 2020 to 27 to August 27 August 2020 2 for020 both for both polarizations polarizations and forand 2019 for 2019 and 2020and 2020 in in Figure5. FigureFigure 5. 5.

Figure 5. Cont. Remote Sens. 2020, 12, x FOR PEER REVIEW 8 of 16 Remote Sens. 2020, 12, 3878 8 of 16

Figure 5. Time-series of VV and VH intensities for corn and soybean. Figure 5. Time-series of VV and VH intensities for corn and soybean. The average intensities were relatively stable in July for both VV and VH for corn in 2019. The average intensities were relatively stable in July for both VV and VH for corn in 2019. Then, Then, in August, because of the crop growth, backscatter intensities increased for both polarizations. in August, because of the crop growth, backscatter intensities increased for both polarizations. From From 3 August to 15 August 2019, the increase was 0.8 dB for VV and 0.32 dB for VH. The average 3 August to 15 August 2019, the increase was 0.8 dB for VV and 0.32 dB for VH. The average intensities for corn fields dropped more in mid-July in 2020 compared to 2019, probably due to crop intensities for corn fields dropped more in mid-July in 2020 compared to 2019, probably due to crop and soil water content variations and/or drought conditions. The decreases were higher for VV than and soil water content variations and/or drought conditions. The decreases were higher for VV than VH (1.48 dB for VV versus 0.84 dB for VH) as this polarization is more sensitive to soil moisture [23]. VH (1.48 dB for VV versus 0.84 dB for VH) as this polarization is more sensitive to soil moisture [23]. Then, in late July 2020, the intensities were stable until the storm happened on August 10th which Then, in late July 2020, the intensities were stable until the storm happened on August 10th which made significant increases for both VV and VH intensities (1.45 dB for VV versus 1.35 dB for VH). If we made significant increases for both VV and VH intensities (1.45 dB for VV versus 1.35 dB for VH). If compare the plots for corn-2020 with those of corn-2019, it demonstrates that the difference in changes we compare the plots for corn-2020 with those of corn-2019, it demonstrates that the difference in between the two years (from 3 August to 15 August) are 0.65 dB for VV and 1.03 dB for VH which changes between the two years (from 3 August to 15 August) are 0.65 dB for VV and 1.03 dB for VH shows higher sensitivity of VH polarization compared to VV polarization to the storm impacts on corn which shows higher sensitivity of VH polarization compared to VV polarization to the storm impacts crop. The strong derecho winds caused corn stalks to bend and in some cases snap, which caused on corn crop. The strong derecho winds caused corn stalks to bend and in some cases snap, which many corn crops to dry out and/or die in the weeks following the storm. The drier conditions for corn caused many corn crops to dry out and/or die in the weeks following the storm. The drier conditions due to the derecho and the ongoing drought were observed in the time series from 15–27 August and for corn due to the derecho and the ongoing drought were observed in the time series from 15–27 show 0.83 dB drop in VV intensity and 0.7 dB drop in VH intensity. August and show 0.83 dB drop in VV intensity and 0.7 dB drop in VH intensity. For soybean in 2019, the intensities dropped by 1.13 dB for VV and 0.29 dB for VH from 10 July to For soybean in 2019, the intensities dropped by 1.13 dB for VV and 0.29 dB for VH from 10 July to 22 July, likely due to variations in soil and crop water content. From 22 July to 15 August, the intensities 22 July, likely due to variations in soil and crop water content. From 22 July to 15 August, the intensities did not change significantly and then in mid to late August, because of the crop growth, small increases did not change significantly and then in mid to late August, because of the crop growth, small increases in both VV (0.4 dB) and VH (0.47 dB) intensities were observed. The plots for soybean in 2020 show in both VV (0.4 dB) and VH (0.47 dB) intensities were observed. The plots for soybean in 2020 show relatively stable intensities for July. However, like corn, the intensities for soybean increased by 1.02 dB relatively stable intensities for July. However, like corn, the intensities for soybean increased by 1.02 dB for VV and 0.7 dB for VH after the storm happened. Since soybean has lower biomass compared for VV and 0.7 dB for VH after the storm happened. Since soybean has lower biomass compared to to corn, the soil has higher contribution to the scattered signal for a soybean field than a corn field, corn, the soil has higher contribution to the scattered signal for a soybean field than a corn field, especially at the time of the growing season when corn is more than 1 m tall and the soil effects on especially at the time of the growing season when corn is more than 1 m tall and the soil effects on the the backscattered signal are reduced. This can explain the higher increase in VV than VH after the backscattered signal are reduced. This can explain the higher increase in VV than VH after the storm storm for soybean. Since soybeans have a more flexible structure and grow lower to the ground than for soybean. Since soybeans have a more flexible structure and grow lower to the ground than corn, corn, most soybean crops affected by the derecho continued to grow from 15–27 August and so the most soybean crops affected by the derecho continued to grow from 15–27 August and so the intensities intensities continued to increase (0.34 dB for VV and 1.13 dB for VH). continued to increase (0.34 dB for VV and 1.13 dB for VH).

Remote Sens. 2020, 12, x FOR PEER REVIEW 9 of 16

To further compare VV and VH polarizations for crop damage monitoring, we plotted the VV vs. VH intensity at each timestep for the time series of corn and soybean fields in 2019 and 2020 in Figure 6. The correlation of determination (R2) between VV and VH was higher for corn (0.93) than soybean (0.86) in 2020. This suggests that the choice of polarization for detecting derecho-impacted Remotefields Sens.will 2020have, 12 a, 3878greater effect for soybean fields than corn fields. The correlations for 2019 are much9 of 16 lower for both crops but still the correlation for corn (R2 of 0.57) is higher than the correlation for soybean (R2 of 0.48). The lower correlations for 2019 are because of different crop growth stages. ActuallyTo further, in 2019, compare there was VV anda significant VH polarizations delay (~3 forweeks) crop on damage planting monitoring, date for much we plotted of the the crop VV due vs. VHto the intensity flood. at each timestep for the time series of corn and soybean fields in 2019 and 2020 in Figure6. The correlation of determination (R2) between VV and VH was higher for corn (0.93) than soybean (0.86) in 2020. This suggests that the choice of polarization for detecting derecho-impacted fields will have a greater effect for soybean fields than corn fields. The correlations for 2019 are much lower for both crops but still the correlation for corn (R2 of 0.57) is higher than the correlation for soybean (R2 of 0.48). The lower correlations for 2019 are because of different crop growth stages. Actually, in 2019, there was a significant delay (~3 weeks) on planting date for much of the crop due to the flood.

Figure 6. Correlations between VV and VH intensities for corn and soybean fields in 2019 and 2020.

Given the higher sensitivity of VH polarization compared to VV polarization to crop structure variations especially for corn (Figure5), the VH polarization was used for the change detection operation and detecting the impacted fields. Also, the analysis of the time series plots and our visual inspectionsFigure of6. Correlations the SAR images between before VV andand VH after intensities the storm for demonstrated corn and soybean that fields the VHin 2019 intensity and 2020. changes more than 1.5 dB within a 12-day time-frame between the two Sentinel-1 acquisitions before and after the stormGiven are the related higher to sensitivity the impacts of ofVH the polarization storm on crops. compared Lower to threshold VV polarization values doto crop not e ffstructureectively suppressvariations changes especially that for are corn due to (Figure natural 5), variations the VH orpolarization crop growth, was rather used than for thedue change to storm detection impacts. operation and detecting the impacted fields. Also, the analysis of the time series plots and our visual 4.3.inspections Impacted of Area the SAR images before and after the storm demonstrated that the VH intensity changes more than 1.5 dB within a 12-day time-frame between the two Sentinel-1 acquisitions before and after We used the 1.5 dB threshold for VH intensity changes to detect regions impacted by the derecho. In total, 2.59 million acres were impacted, including 1.99 million acres of corn and 0.60 million acres of soybean (Figures7 and8). All 14 fields for which we had in-situ data were detected in the damage map. We observed that the storm did not have uniform impacts on all fields. Figure8 shows a zoomed-in inset of the affected region. While many fields in this inset were detected as storm-impacted fields, other fields were not as severely impacted. The severity of the storm impacts is discussed further in the next section. Remote Sens. 2020, 12, 3878 10 of 16 Remote Sens. 2020, 12, x FOR PEER REVIEW 10 of 16 Remote Sens. 2020, 12, x FOR PEER REVIEW 10 of 16

Figure 7. Impacted fields by Derecho for (a) corn and (b) soybean fields. Figure 7. Impacted fields fields by Derecho for (a) corn and (b) soybean fields.fields.

Figure 8. DerechoDerecho impact impact map map over over corn, corn, soybean and the in-situ fields. fields. Figure 8. Derecho impact map over corn, soybean and the in-situ fields. We compared our results to two other studies published by private companies McKinsey and We compared our results to two other studies published by private companies McKinsey and Company, Inc.Inc. (https: (https://www.mckinsey.com/indus//www.mckinsey.com/industries/agriculturetries/agriculture/our/our-insights-insights/how-the-iowa-/how-the-iowa-derecho- Company, Inc. (https://www.mckinsey.com/industries/agriculture/our-insights/how-the-iowa- derecho-has-affected-2020-crops#),has-affected-2020-crops#), and Indigo Ag, Incand (https: //www.indigoag.comIndigo /hubfsAg,/Derecho_Report_ Inc derecho-has-affected-2020-crops#), and Indigo Ag, Inc (https://www.indigoag.com/hubfs/Derecho_ReporvFINAL.pdf?hsCtaTracking=6743af10-1790-4712-8d47-ad2af4fafd70%7Ce2e218a2-bdc5-4fdb-9e0c-t_vFINAL.pdf?hsCtaTracking=6743af10-1790- (https://www.indigoag.com/hubfs/Derecho_Report_vFINAL.pdf?hsCtaTracking=6743af10-1790- 4712-8d47-ad2af4fafd70%7Ce2e218a2-bdc5-4fdb-9e0c-f2037f212ef6).f2037f212ef6). Table1 shows the damaged area estimates from their Table studies 1 shows compared the damaged to ours. area 4712-8d47-ad2af4fafd70%7Ce2e218a2-bdc5-4fdb-9e0c-f2037f212ef6). Table 1 shows the damaged area estimatesMcKinsey from and their Company studies Inc. compared estimated to ours. that 3.1–3.8McKinsey million and totalCompany acres Inc. were estimated impacted, that of 3.1–3.8 which estimates from their studies compared to ours. McKinsey and Company Inc. estimated that 3.1–3.8 millionthey estimated total acres 10–0% were were impacted, soybean of which and 80–90% they estimated were corn. 10–0% This were translates soybean to and 0.31–0.76 80–90% million were million total acres were impacted, of which they estimated 10–0% were soybean and 80–90% were corn.acres This for soybean translates and to 2.48–3.42 0.31–0.76 million million acres acres for for corn. soybean The estimatesand 2.48–3.42 from million the Indigo acres Ag, for Inc. corn. report The corn. This translates to 0.31–0.76 million acres for soybean and 2.48–3.42 million acres for corn. The estimatesare 1.4 million from the acres Indigo for soybean Ag, Inc. andreport 2.1 are million 1.4 million acres foracres corn for (3.5soybean million and acres 2.1 million total). acres We also for estimates from the Indigo Ag, Inc. report are 1.4 million acres for soybean and 2.1 million acres for corncompared (3.5 million the forecasted acres total). harvested We also compared acres reported the forecasted by the USDA harvested in the acres August reported Crop by Production the USDA corn (3.5 million acres total). We also compared the forecasted harvested acres reported by the USDA inReport the( https:August//downloads.usda.library.cornell.edu Crop Production Report /(husda-esmisttps://downloads.usda.li/files/tm70mv177brary.cornell.edu/usda-/kw52jz21g/00000n572/ in the August Crop Production Report (https://downloads.usda.library.cornell.edu/usda- esmis/files/tm70mv177/kw52jz21g/00000n572/cropcrop0820.pdf) and the October Crop Production0820.pdf) Report (https:and //thedownloads.usda.library.cornell. October Crop Production esmis/files/tm70mv177/kw52jz21g/00000n572/crop0820.pdf) and the October Crop Production Reportedu/usda-esmis /files/tm70mv177/ng452781q/mk61s709f(https://downloads.usd/crop1020.pdf). Thea.library.cornell.edu/usda- difference in forecasted Report (https://downloads.usda.library.cornell.edu/usda- esmis/files/tm70mv177/ng452781q/mk61s709f/crop1020harvested acres between these reports likely indicates.pdf). how manyThe difference acres the in USDA forecasted estimated harvested to be esmis/files/tm70mv177/ng452781q/mk61s709f/crop1020.pdf). The difference in forecasted harvested acrescompletely between damaged these reports (thus unable likely toindicates be harvested) how ma byny the acres derecho. the USDA The USDA estimated reduced to be the completely harvested acres between these reports likely indicates how many acres the USDA estimated to be completely damagedacres forecast (thus by unable 0.85 million to be harvested) acres for corn by the in the derecho. October The report USDA compared reduced to the the harvested August report. acres damaged (thus unable to be harvested) by the derecho. The USDA reduced the harvested acres forecastHowever, by the 0.85 forecasted million acres harvested for corn acres in the for Octo soybeanber report remained compared constant to the between August these report. reports. However, forecast by 0.85 million acres for corn in the October report compared to the August report. However, the forecasted harvested acres for soybean remained constant between these reports. the forecasted harvested acres for soybean remained constant between these reports. Overall, our estimates lie in between those reported by USDA, McKinsey, and Indigo. Our Overall, our estimates lie in between those reported by USDA, McKinsey, and Indigo. Our estimated 0.6 million soybean acres is within the range reported by McKinsey but 0.8 million acres estimated 0.6 million soybean acres is within the range reported by McKinsey but 0.8 million acres

Remote Sens. 2020, 12, 3878 11 of 16

Table 1. Estimates from published studies on derecho impacts over Iowa corn and soybean fields compared to our study. All estimates are in million acres.

McKinsey and NASA-Harvest Indigo Ag Inc. Company Inc. (i.e., This Study) Total 3.1–3.8 3.5 2.59 Soybean 0.31–0.76 1.4 0.6 Corn 2.48–3.42 2.1 1.99

Overall, our estimates lie in between those reported by USDA, McKinsey, and Indigo. Our estimated 0.6 million soybean acres is within the range reported by McKinsey but 0.8 million acres lower than the 1.4 million acres reported by Indigo. Our estimated 1.99 is close to the Indigo estimate of 2.1 million acres but slightly lower than both the Indigo and McKinsey estimates. Our estimates for corn and soybean acres are both higher than the USDA reductions in forecasted harvested acres, which is expected since our estimates include varying degrees of damage while the USDA acres likely accounts only for completely destroyed (unharvestable) crops. The difference between our estimates and those from Indigo and McKinsey can be attributed to differences in methods, particularly that the Indigo and McKinsey studies used VV polarization for change detection rather than VH, which we used in this study. In Figure6, we demonstrated that there is a very high correlation between VV and VH intensities for corn and so using either polarization provides similar results for corn. However, the correlation between the two bands was lower for soybean and as such there could be a higher difference between the estimates using VV and VH for soybeans. We also evaluated the impacted areas at the county level (Figure9). Story county had the highest damage impacts with 0.174 million acres impacted (0.140 million acres corn and 0.034 million acres soybean), followed by Boone county (0.154 million acres in total, 0.132 million acres corn and 0.022 million acres soybean) and Marshall county (0.150 million acres in total, 0.126 million acres corn and 0.024 million acres soybean). Remote Sens.Remote2020 Sens., 12 2020, 3878, 12, x FOR PEER REVIEW 12 of 16 12 of 16

Figure 9. TotalTotal impacted areas areas for for each each county. county.

Remote Sens. 2020, 12, x; doi: FOR PEER REVIEW www.mdpi.com/journal/remotesensing

Remote Sens. 2020, 12, x FOR PEER REVIEW 13 of 16

Remote Sens. 2020, 12, 3878 13 of 16 4.4. Damage Severity Estimates

4.4. DamageAsRemote discussed, Sens. Severity 2020, 12 ,the Estimatesx FOR storm PEER REVIEWdid not have similar impacts on all the fields. We categorized 13 of 16 the detected damage areas into three classes based on the amount of change observed in the Sentinel-1 As discussed, the storm did not have similar impacts on all the fields. We categorized the detected VH intensities4.4. Damage (Figure Severity 10).Estimates The range of VH intensity changes for the low-impact class is 1.5–2.5 dB damage areas into three classes based on the amount of change observed in the Sentinel-1 VH intensities and it is expected that this class will show mild impacts from the storm. We hypothesize that yields (Figure 10As). Thediscussed, range ofthe VH storm intensity did not changes have similar for the impacts low-impact on all class the isfields. 1.5–2.5 We dBcategorized and it is expectedthe for the fields in this class will not be affected by the derecho, but the harvesting speed and costs may that thisdetected class damage will show areas mild into impacts three classes from based the storm. on the Weamount hypothesize of change that observed yields in for the the Sentinel-1 fields in this be affected,VH intensities especially (Figure for 10).corn The fields. range This of VH class intensity includes changes 0.389 for million the low-impact acres of corn class and is 1.5–2.5 0.199 dBmillion class will not be affected by the derecho, but the harvesting speed and costs may be affected, especially acresand of soybean.it is expected The that second this class class will is theshow medium-impacted mild impacts from areasthe storm. which We covershypothesize fields that for yields which we for corn fields. This class includes 0.389 million acres of corn and 0.199 million acres of soybean. anticipatefor the the fields storm in this will class have will notnegative be affected impacts by the on derecho, yield and but theincludes harvesting the speedsignal and variations costs may in the The second class is the medium-impacted areas which covers fields for which we anticipate the storm rangebe of affected, 2.5–3.5 especiallydB. This classfor corn includes fields. This0.514 class millio includesn acres 0.389 of cornmillion and acres 0.195 of cornmillion and acres0.199 millionof soybean. will have negative impacts on yield and includes the signal variations in the range of 2.5–3.5 dB. This The rangeacres ofof soybean. VH intensity The second changes class for is the the lastmedium-impacted class includes areasvariations which of covers more fields than for 3.5 which dB. This we class class includes 0.514 million acres of corn and 0.195 million acres of soybean. The range of VH intensity is consideredanticipate tothe include storm willhighly have impacted negative areasimpacts in whichon yield crops and includeswere fully the damaged signal variations or had inthe the largest changesrange for of the2.5–3.5 last dB. class This includes class includes variations 0.514 millio of moren acres than of 3.5corn dB. and This 0.195 class million is considered acres of soybean. to include negative yield impacts. This class includes 1.087 million acres of corn and 0.206 million acres of highlyThe impacted range of VH areas intensity in which changes crops for were the last fully class damaged includes variations or had the of largestmore than negative 3.5 dB. This yield class impacts. soybean. This class is close to the USDA reduced area between their August and October reports (i.e., This classis considered includes to 1.087include million highly acres impacted of corn areas and in which 0.206 millioncrops were acres fully of damaged soybean. or This had classthe largest is close to 0.85 million acres for corn) and the difference goes into a reduction in yield but is harvested. Corn the USDAnegative reduced yield impacts. area between This class their includes August and1.087 October million reportsacres of (i.e.,corn 0.85and million0.206 million acres foracres corn) of and and soybeansoybean. fieldsThis class that is areclose highly to the impactedUSDA reduced by the area storm between (>3.5 their dB August change) and are October mapped reports in Figure (i.e., 11. the difference goes into a reduction in yield but is harvested. Corn and soybean fields that are highly 0.85 million acres for corn) and the difference goes into a reduction in yield but is harvested. Corn impacted by the storm (>3.5 dB change) are mapped in Figure 11. and soybean fields that are highly impacted by the storm (>3.5 dB change) are mapped in Figure 11.

Figure 10. Derecho damage severity estimates. FigureFigure 10. 10.Derecho Derecho damage severity severity estimates. estimates.

Remote Sens. 2020, 12, x; doi: FOR PEER REVIEW www.mdpi.com/journal/remotesensing Figure 11. Corn and soybean fields that are highly impacted by the storm. Remote Sens. 2020, 12, x; doi: FOR PEER REVIEW www.mdpi.com/journal/remotesensing

Remote Sens. 2020, 12, 3878 14 of 16

5. Summary and Conclusions In this study, we assessed the sensitivity of Sentinel-1 signals at VV and VH polarizations to the Iowa derecho storm impacts on corn and soybean crops. As is common in many rapid response scenarios, very limited amounts of in-situ data were available for calibrating and validating analyses. To determine an appropriate change detection threshold without in-situ data, we compared the time series of Sentinel-1 data from about 40 days before the storm to 20 days after to the same time period in 2019. Comparing to 2019 provided a baseline to assess the magnitude of change that could be caused by the derecho versus natural variations and growth in the crops. We observed that while both VV and VH polarizations were highly sensitive to the storm impacts, VH polarization was more sensitive to the crop damages, thus we used changes in VH intensity to evaluate the damage extent and severity. We detected damaged areas using a 1.5 dB threshold on changes in VH intensity before and after the storm. We generated an in-season crop type map using multispectral time series observations from the Harmonized Landsat and Sentinel-2 (HLS) dataset from 2017 to 2020. We intersected the impacted area map with the crop type map to detect damaged area estimates for each crop type. These analyses resulted in damaged area estimates of 1.99 million acres for corn and 0.6 million acres for soybean. The map showed that 42 counties were affected by the storm and 12 counties (Story, Boone, Marshall, Cedar, Jasper, Clinton, Greene, Tama, Poweshiek, Benton, Dallas and Carroll) had each more than 0.1 million acres of impacted fields in total corn and soybean acreages. We further analyzed the severity of the storm impacts by categorizing the impacted areas into three severity categories based on the SAR signal variations before and after the storm. These categories ranged from a low impact class with 0.389 million acres of corn and 0.199 million acres of soybean to a high impact class with 1.087 million acres of corn and 0.206 million acres of soybean. In this study, we demonstrated the feasibility of using satellite data, especially from SAR satellites, for monitoring the impacts of derechos and other destructive storms on agricultural lands. We showed that when optical satellite images are unsuitable due to high cloud coverage, no signal, or saturated signal, SAR satellites can provide timely, reliable and actionable information for rapid response scenarios. Given the rapid and easy accessibility to satellite data and their large swath widths and high revisit times, Earth observation is the most efficient way of studying natural disasters and their impacts on agricultural fields and other land cover. As events become more frequent and severe due to global climate change, increased capacity for satellite monitoring of natural disasters will be critical to provide the necessary maps and analyses for rapid response actions by governments, insurance companies, and farmers.

Author Contributions: Conceptualization, M.H.; methodology, M.H.; software, M.H., Y.-H.L., A.F.L.; validation, M.H.; formal analysis, M.H.; investigation, M.H.; resources, M.H., H.R.K., R.S., E.P., M.M., S.M., I.B.-R.; data curation, M.H., Y.-H.L., A.F.L., H.R.K., R.S., E.P., M.M., S.M.; writing—original draft preparation, M.H.; writing—review and editing, H.R.K., R.S., M.L.H., S.M., I.B.-R., E.P., M.M.; visualization, M.H., H.R.K., R.S.; supervision, M.H., I.B.-R.; project administration, M.H.; funding acquisition, I.B.-R. All authors have read and agreed to the published version of the manuscript. Funding: This work was funded by the NASA’s Harvest Consortium on Food Security and Agriculture (Grant #80NSSC18M0039). Acknowledgments: The authors would like to thank the Iowa farmers who contributed field photos and other valuable information during this study (Julius Schaaf, Arick Baker, Lance Lillibridge, Rod Pierce, and Doug Svendsen) as well as AgriTalk radio for helping reach farmers. Conflicts of Interest: The authors declare no conflict of interest.

References

1. USDA-NASS. Iowa Ag News–2019 Crop Production. Available online: https://www.nass.usda.gov/ Statistics_by_State/Iowa/Publications/Crop_Report/2020/IA-Crop-Production-Annual-01-20.pdf (accessed on 20 September 2020). Remote Sens. 2020, 12, 3878 15 of 16

2. Dingle Robertson, L.; Davidson, A.; McNairn, H.; Hosseini, M.; Mitchell, S.; Abelleyra, D.; Verón, S.; Cosh, M.H. Synthetic Aperture Radar (SAR) Image Processing for Operational Space-Based Agriculture Mapping. Int. J. Remote Sens. 2020, 41, 7112–7144. [CrossRef] 3. Lopez-Sanchez, J.M.; Irena, H.; Ballester-Berman, J.D. First demonstration of agriculture height retrieval with PolInSAR airborne data. IEEE Geosci. Remote Sens. Lett. 2012, 9, 242–246. [CrossRef] 4. Alonso-González, A.; Jagdhuber, T.; Hajnsek, I. Agricultural monitoring with polarimetric SAR time series. In Proceedings of the 8th International Workshop on the Analysis of Multitemporal Remote Sensing Images (Multi-Temp), Annecy, France, 22–24 July 2015; IEEE: Piscataway, NJ, USA, 2015. 5. Kontgis, C.; Warren, M.S.; Skillman, S.W.; Chartrand, R.; Moody, D.I. Leveraging Sentinel-1 time-series data for mapping agricultural land cover and land use in the tropics. In Proceedings of the 9th International Workshop on the Analysis of Multi Temporal Remote Sensing Images (MultiTemp), Brugge, Belgium, 27–29 June 2017. 6. Veloso, A.; Mermoz, S.; Bouvet, A.; Le Toan, T.; Planells, M.; Dejoux, J.-F.; Ceschia, E. Understanding the temporal behavior of crops using Sentinel-1 and Sentinel-2-like data for agricultural applications. Remote Sens. Environ. 2017, 199, 415–426. [CrossRef] 7. Hosseini, M.; McNairn, H.; Mitchell, S.; Davidson, A.; Dingle Robertson, L. Synthetic Aperture Radar and Optical Satellite Data for Estimating the Biomass of Corn. Int. J. Earth Obs. Geoinf. 2019, 83, 101933. [CrossRef] 8. Mandal, D.; Hosseini, M.; McNairn, H.; Kumar, V.; Bhattacharya, A.; Rao, Y.S.; Mitchell, S.; Dingle Robertson, R.; Davidson, A.; Dabrowska-Zielinska, K. An investigation of inversion methodologies to retrieve the Leaf Area Index of corn from C-Band backscatter. Int. J. Earth Obs. Geoinf. 2019, 82, 101893. [CrossRef] 9. Hosseini, M.; McNairn, H.; Merzouki, A.; Pacheco, A. Estimation of Leaf Area Index (LAI) in corn and soybeans using multi-polarization C- and L-band radar data. Remote Sens. Environ. 2015, 170, 77–89. [CrossRef] 10. Bell, J.; Gebremichael, E.; Molthan, A.; Schultz, L.; Meyer, F.; Shrestha, S. Synthetic Aperture Radar and Optical Remote Sensing of Crop Damage Attributed to in the Central United States. In Proceedings of the IGARSS 2019, Yokohama, Japan, 28 July–2 August 2019; pp. 9938–9941. 11. Surek, G.; Nador, G. Monitoring of Damage in Sunflower and Maize Parcels Using Radar and Optical Time Series Data. J. Sens. 2015, 2015.[CrossRef] 12. Chauhan, S.; Darvishzadeh, R.; Lu, Y.; Boschetti, M.; Nelson, A. Understanding wheat lodging using multi-temporal Sentinel-1 and Sentinel-2 data. Remote Sens. Environ. 2020, 243, 111804. [CrossRef] 13. Silleos, N.; Perakis, K.; Petsanis, G. Assessment of crop damage using space remote sensing and GIS. Int. J. Remote Sens. 2002, 23, 417–428. [CrossRef] 14. Young, F.; Chandler, O.; Apan, A. Crop Hail Damage: Insurance Loss Assessment using Remote Sensing. In Proceedings of the Remote Sensing and Photogrammetry Society Conference, Aberdeen, UK, 7–10 September 2004. 15. Claverie, M.; Ju, J.; Masek, J.G.; Dungan, J.L.; Vermote, E.F.; Roger, J.C.; Skakun, S.V.; Justice, C. The Harmonized Landsat and Sentinel-2 surface reflectance data set. Remote Sens. Environ. 2018, 219, 145–161. [CrossRef] 16. McNairn, H.; Hochheim, K.; Rabe, N. Applying polarimetric radar imagery for mapping the productivity of wheat crops. Can. J. Remote Sens. 2004, 30, 517–524. [CrossRef] 17. Jiao, X.; McNairn, H.; Shang, J.; Pattey, E.; Liu, J.; Champagne, C. The sensitivity of RADARSAT-2 polarimetric SAR data to corn and soybean leaf area index. Can. J. Remote Sens. 2009, 37, 69–81. [CrossRef] 18. McNairn, H.; Shang, J. A Review of Multitemporal Synthetic Aperture Radar (SAR) for Crop Monitoring. In Multitemporal Remote Sensing: Methods and Applications; Ban, Y., Ed.; Springer: Berlin/Heidelberg, Germany, 2016; Chapter 15; pp. 317–340. 19. Boryan, C.; Yang, Z.; Mueller, R.; Craig, M. Monitoring US agriculture: The US Department of Agriculture, National Agricultural Statistics Service, Cropland Data Layer program. Geocarto Int. 2011, 26, 341–358. [CrossRef] Remote Sens. 2020, 12, 3878 16 of 16

20. Kerner, H.; Becker-Reshef, I.; Estefania, I.P.; Barker, B.; Sahajpal, R.; Skakun, S.; Gray, P.; Hosseini, M. Resilient In-Season Crop Type Classification in Multispectral Satellite Observations using Growth Stage Normalization. In Proceedings of the SIGKDD ACM Conference on Knowledge Discovery and Data Mining Workshops, San Diego, CA, USA, 23–27 August 2020. 21. USDA-NASS. Crop Production Report. 11 December 2018. Available online: https://downloads. usda.library.cornell.edu/usda-esmis/files/tm70mv177/r781wm151/vm40xw490/crop1218.pdf (accessed on 20 September 2020). 22. USDA-NASS. Crop Production Report. 9 October 2020. Available online: https://downloads. usda.library.cornell.edu/usda-esmis/files/tm70mv177/ng452781q/mk61s709f/crop1020.pdf (accessed on 20 September 2020). 23. Vreugdenhil, M.; Wagner, W.; Bauer-Marschallinger, B.; Pfeil, I.; Teubner, I.; Rudiger, C.; Strauss, P. Sensitivity of Sentinel-1 Backscatter to Vegetation Dynamics: An Austrian Case Study. Remote Sens. 2018, 10, 1396. [CrossRef]

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).