Sentinel-1 SAR Backscatter Response to Agricultural Drought in The Netherlands

: Drought is a major natural hazard that impacts agriculture, the environment, and socio-economic conditions. In 2018 and 2019, Europe experienced a severe drought due to below average precipitation and high temperatures. Drought stress affects the moisture content and structure of agricultural crops and can result in lower yields. Synthetic Aperture Radar (SAR) observations are sensitive to the dielectric and geometric characteristics of crops and underlying soils. This study uses data from ESA’s Sentinel-1 SAR satellite to investigate the inﬂuence of drought stress on major arable crops of the Netherlands, its regional variability and the impact of water management decisions on crop development. Sentinel-1 VV, VH and VH/VV backscatter data are used to quantify the variability in the spatio-temporal dynamics of agricultural crop parcels in response to drought. Results show that VV and VH backscatter values are 1 to 2 dB lower for crop parcels during the 2018 drought compared to values in 2017. In addition, the growth season indicated by the cross-ratio (CR, VH/VV) for maize and onion is shorter during the drought year. Differences due to irrigation restrictions are observed in backscatter response from maize parcels. Lower CR values in 2019 indicate the impact of drought on the start of the growing season. Results demonstrate that Sentinel-1 can detect changes in the seasonal cycle of arable crops in response to agricultural drought. the effects of drought on crops. Here, we investigated the inﬂuence of inter-annual variability in the Sentinel-1 SAR signals over different crops due to the drought in 2018 and 2019 in the Netherlands. The inter-annual variation shows that VH and VV values are lower in the drought period compared to the normal year. The seasonal cycle for CR is found


Introduction
The threat to global food security is expected to increase as a result of climate change and the rising world population [1,2]. Monitoring agricultural crops to predict yield, predict change in water use efficiency and detect the impact of extreme weather events is vital for ensuring long-term food security. Drought is an abiotic stress that arises from a soil moisture deficit when the availability of precipitation does not meet evaporative demand, and can lead to a reduction in crop yields [3][4][5]. A reduction in crop yield can be caused by the shorter life-cycle of the crops, and/or the adaptation of the stomatal control and reduction of leaf area in order to cope with drought stress [6]. Future climate change is expected to increase the frequency of severe meteorological [7][8][9] and agricultural droughts [4,10]. Large-scale, continuous monitoring is essential to detect, monitor and mitigate the effects of drought on agricultural production.
Drought indices from meteorological station-based weather information are effective for drought assessment [11][12][13]. However, the global distribution of meteorological stations is highly variable in space and time, limiting their potential use for drought monitoring [14][15][16]. Remote sensing satellite observations provide continuous, global coverage and can provide insight into droughts at a range of resolutions [17][18][19]. Many studies have used spectral indices from the visible to near infra-red part of the spectrum (e.g., [20][21][22][23][24]). However, their reliability is limited by cloud cover. Microwave remote sensing offers continuous coverage without being hindered by atmospheric conditions. Synthetic Aperture Radar (SAR) data have been widely used in vegetation monitoring studies due to microwave sensitivity to dielectric and geometric changes during crop development [25,26]. However, the use of several previous and existing SAR missions viz. RADARSAT-1/2, ALOS-1/2 and TerraSAR-X was limited due to spatial coverage and coarse temporal resolution. With the launch of the Sentinel-1 SAR constellation [27], it is possible to have 12 days (6 days in Europe) of temporal resolution data across the globe and high temporal resolution over the Europe (<2 days depending on the latitude) by combining different acquisitions in interferometric wide (IW) swath mode. Additionally, Sentinel-1 SAR archive and real-time data accessibility in cloud computing servers, such as Google Earth Engine (GEE), allows for large scale monitoring without the constraints of data handling and processing [28].
The potential of Sentinel-1 SAR data have been shown in many studies for crop growth monitoring [29][30][31], phenology tracking [32] and biophysical parameter estimation [33]. The high temporal resolution, the fact that acquisitions are unaffected by cloud cover, and the sensitivity of backscatter to vegetation water content and geometry means that Sentinel-1 offers unique potential to capture the drought response of crops. However, limited research has been conducted to understand the potential of Sentinel-1 SAR signal response to the agricultural drought events [34,35].
Two consecutive drought summers occurred across central Europe in 2018 and 2019 [36][37][38][39]. Hari et al. [9] found that the consecutive droughts of 2018-2019 were the most severe in the last 250 years. These droughts had a severe impact in the Netherlands [40][41][42], resulting in a significant drop in yield for most crops [43]. The long term average precipitation during the summer growing season (April to September) is 438 mm. However, in 2018 this was just 304 mm [40]. The drought in 2018 caused a peak in agricultural water usage in the Netherlands and a lowering of the groundwater table, particularly in the southeast [42]. Water deficit influences crop water content and structure. Droughts have an adverse effect on the overall yield, which is caused by a reduction in leaf area, adaptation of the stomatal control and the shorter life-cycle of the crops to cope with the water deficit [6]. Since the SAR response depends on the geometrical and dielectric properties, it is expected that the water deficit caused by the drought will be observable in the SAR response.
Here, we compare the seasonal cycle of Sentinel-1 SAR observables across the Netherlands from several crops during drought-affected years to those of a normal year. Differences in the Sentinel-1 observables will be discussed with respect to rainfall, soil moisture and regional crop yield estimates during the drought-affected and normal years. We also examine the influence of Sentinel-1 viewing geometry and acquisition time. We investigate whether spatial variability in the drought response due to crop and soil type, irrigation water availability can be detected in the SAR observalbes (VV, VH backscatter and VH/VV ratio).

Study Area
This study is focused on the impact of the 2018 and 2019 droughts in the Netherlands. The study areas, shown in Figure 1, were selected to account for different soil types, irrigation practices and the spatial distribution of drought in the Netherlands during 2018 and 2019, as well as the availability of in situ information. The general study area comprises three areas indicated in blue. Three factors were considered in the definition of the areas of interest. First, the yield data are available at province level. Second, water management decisions are made by water boards. These are regional governing bodies responsible for the management of surface water in the environment. Their boundaries do not coincide with province boundaries. Third, we need to consider the location of areas of interest within the swaths of the Sentinel-1 relative orbits. In Figure 1, the Vechstromen-Noord area consists of the northern part of the Vechstromen water board, which includes parts of the Drenthe and Overijssel provinces. Vechtstromen-Noord is observed in Sentinel-1 relative orbits (ROs) 37, 139, 88 and 15. Scheldestromen consists of the part of the Scheldestromen water board that falls within the Zeeland province. Common crop parcels within RO 37,110, 88 and 161 swaths were selected from Scheldestromen. Finally, the Flevopolder is an island of reclaimed land, and is part of the Zuiderzeeland water board. It is an area characterized by highly productive, intensive agriculture and has featured in many SAR studies [29,44,45]. The difference in soil texture and water management among the areas is expected to lead to a variation in drought response. The clay content of soil in Flevopolder is relatively high. In contrast, the soil in Vechtstromen-Noord has a high sand content. Crops in Zeeland (including the Scheldestromen area) suffered yield loss due the lack of precipitation and the ban on the open water irrigation.
To investigate the potential influence of water management (e.g., irrigation availability), we compared data from four municipalities of the Vechtstromen waterboard, highlighted in red in the inset of Figure 1. In Coevorden and Hardenberg, irrigation is only available from groundwater during droughts. In contrast, farmers in Tubbergen and Hof van Twente have access to both surface (open) water and groundwater for irrigation during the drought period [46].

Crop, Soil Moisture and Meteorological Data
In the Netherlands, the Basisregistratie Gewaspercelen (BRP), containing parcel-level crop information, is publicly available via the Publieke Dienstverlening Op de Kaart (PDOK) [47] web server and is updated annually. The BRP database contains the boundary geometry of agricultural parcels along with the name of the cultivated crop, as well as the geometry and area of the parcel. Major summer crops, namely silage maize, potatoes (for consumption), sugar beets, summer barley and onions are considered in this study. The number of parcels for each crop type in the three study areas is presented in Table 1. For the comparison between the municipalities with and without access to surface water, only maize and potato parcels were considered because the differences in the number of crop parcels between the municipalities are too large to ensure a fair and reliable comparison. Annual crop yield information is openly available via Statistics Netherlands (https: //opendata.cbs.nl/#/CBS/en/dataset/7100eng/table, accessed on 22 May 2020) [43]. Here, we used the normalized yield (i.e., the total gross yield per harvested hectare, 1000 Kg/Ha) to compare results in the study areas. The normalized gross yield and inter-annual yield differences per province are expressed in 1000 kg/ha for studied crops. Figure 2 shows the normalized gross yield for 2017 per province and inter-annual yield deviation between drought and "normal" years, i.e., 2018-2017 and 2019-2017, respectively. Since the Vechtstromen-Noord study area is located in two provinces, the yield was interpreted by taking the average over Drenthe and Overijssel province data.
VanderSat's daily volumetric surface soil moisture (SM) product (V1.0) is used in this study [48]. The SM products are available at 100 × 100 m resolution on a daily basis from the combined use of Advanced Microwave Scanning Radiometer (AMSR)-2 and Soil Moisture Active Passive (SMAP) microwave satellite sensors retrieved using the Land Parameter Retrieval Model (LPRM) [49][50][51] . In this study, spatially aggregated volumetric SM data at municipality level were used. The surface soil moisture of the Coevorden, Goes and Zeewolde municipality areas were used for the Vechtstromen-Noord, Scheldestromen and Flevopolder study areas, respectively. Surface soil moisture data from Coevorden, Hardenberg, Tubbergen and Hof van Twente municipalities were used observe the impact of water management decisions. Data discontinuity was observed in the SM products in mid-2019 due to a gap in the SMAP data record.
Meteorological data were acquired from the Royal Netherlands Meteorological Institute (KNMI). KNMI provide daily precipitation data, measured at 48 weather stations throughout the Netherlands. The daily precipitation data acquired by the weather stations in Hoogeveen, Vlissingen and Lelystad were used for the Vechtstromen-Noord, Scheldestromen and Flevopolder study areas, respectively. Data from the Hoogeveen and Twente weather stations were used in the Vechtstromen-Noord area in the analysis related to water management.

Sentinel-1 SAR Data
Sentinel-1 SAR GRD data products in interferometric wide (IW) swath mode, available through GEE cloud based services, are used in this study. The Sentinel-1 SAR system consists of two satellites, S1A and S1B, which operate in the C-band (5.45 GHz) with a revisit time of 6 days together. With the combination of different viewing geometries (or relative orbits (RO)) in ascending and descending passes for data acquisition, it is possible to have Sentinel-1 data with a temporal resolution of less than 2 days across Europe. Most of the Netherlands is covered by six different ROs, namely 37, 110, 139, 15, 88 and 161 ( Table 2). The swaths of the descending and ascending orbits are shown in Figure 3, respectively.   In this study, parcel-level spatially-averaged Sentinel-1 VV and VH backscatter (σ 0 ) and cross-ratio (VH/VV) values were obtained from the Agricultural SandboxNL database [52,53]. This database was generated from GEE-hosted Sentinel-1 SAR GRD products. GEE-hosted Sentinel-1 data are already corrected for basic SAR pre-processing steps, namely (1) Metadata update with restituted orbit file, (2) Border noise removal after 2018 onward, (3) Thermal noise removal between the sub-swaths, and (4) Radiometric calibration and Terrain correction [28]. The crop parcel-level boundaries in Agricultural SandboxNL were obtained from the annual BRP data [47]. It consists of parcel boundaries, main cultivated crop type information and attributes such as unique OBJECTID and parcel geometry. Parcel-level backscatter values (VV, VH and CR) in the Agricultural SandboxNL database can be queried by crop types, administrative boundaries (province, municipality or user-defined), Sentinel-1 acquisition dates and ROs. Table 3 summarizes the various input datasets, their sources and scales used in this study.

Analysis
Sentinel-1 data for the "normal" year of 2017 were compared to the drought years of 2018 and 2019 to characterize the drought impacts on SAR signals. The VV, VH and CR values were averaged at parcel-level and aggregated per crop type over the studied regions in different relative orbits of the Sentinel-1. The inter-annual comparisons were made by observing the differences in VV, VH backscatter and CR values for 2018 and 2019 compared to 2017. The differences in resulting observations were supported by soil moisture, precipitation, yield, and any openly available ground information during the drought. This comparison will be presented for the Vechtstromen-Noord area because the highest drought impact in 2018 was reported in the north eastern part of the Netherlands [57]. Spatially averaged parcel-level VV, VH and CR values for maize and onion will be presented and discussed. Similar results for potato and sugar beet are provided in supplementary materials. The influence of viewing geometry is considered by comparing data from four relative orbits allowing for differences in incidence angle and between ascending/descending passes. Figure 2 showed that there were spatial variations in the impact of the droughts on crop yield. This is due to differences in hydrometeorological conditions, soil types and water management decisions during the drought periods. Data from the three study areas were compared to determine if these regional variations in yield correspond to discernible regional differences in the VV, VH and CR values of Sentinel-1 data during the season. Results are presented for maize and onion (potato and sugar beet are in supplementary materials). To ensure that the results are not influenced by differences in viewing geometry, the relative orbits were chosen to minimize the difference in incidence angle.
Finally, data from four municipalities were compared to investigate whether the impact of water use restrictions on the crop development are discernible in the Sentinel-1 data. The spatially-averaged parcel level VV, VH and CR values aggregated per crop type from Coevorden and Hardenberg were contrasted with those in the Tubbergen and Hof van Twente municipalities. Recall that the the first two only have access to groundwater during the droughts, while the last two also have access to surface water. Results will be presented for maize and potato parcels because the number of parcels are comparable.

Influence of Drought on the Seasonal Cycle of Sentinel-1 SAR Data over Agricultural Crops
Spatially-averaged, parcel-level Sentinel-1 data during the 2018 and 2019 drought years were compared to data in a "normal" year (2017). Figure 4 shows the time series of Sentinel-1 VV, VH backscatter and cross-ratio (CR, VH/VV) from maize parcels in Vechtstromen-Noord. Maize is typically planted between mid-April and the first week of May, with emergence occuring in mid-May. Silage maize usually ripens in the field and is harvested between mid-September and mid-October [29,58]. Before crop emergence, soil surface roughness and moisture dominate the backscatter response in VV and VH polarizations. Backscatter variations during this period are caused by the rainfall events apparent in the bottom two rows. After emergence, VV and VH backscatter increase rapidly due to soil-vegetation interaction and volume scattering as the crop grows. Figure 4m shows that the soil moisture in 2018 (~0.07 m 3 /m 3 ) was much lower than that in 2017 (~0.20 m 3 /m 3 ) between day 175 and day 220. During this period, the VV and VH backscatter values in 2018 are~2 dB lower than the values of 2017. Precipitation around day 220 draws the backscatter values closer to those observed in 2017. Note, however, that the 2018 values decrease more rapidly from DOY 250. The backscatter, but particularly the CR, suggest that the growing season of 2018 was~30 days shorter than that of 2017. This can be explained by the earlier ripening and harvesting of maize due the dry conditions [59]. The shorter 2018 season also led to a significant drop in yield of 9.1 % compared to 2017 (Figure 2a).

Maize
Low soil moisture occurred during the tasseling and flowering phase, which is the most drought sensitive stage according to Kurt Thelen [60]. This is due to a lack of synchronization between silking and shedding of the pollen during pollination. The pollen grains may not remain viable and silking might be delayed. If maize crops have tasselled and shed their pollen while blisters have not appeared yet, the crop will be barren [60]. Furthermore, no large cumulative precipitation increases are observed from DOY 70-105. Thus, it can be assumed that the maize crops were planted and germinated in relatively dry soil. Water stress during the vegetative phase typically reduces leaf and plant size Anami et al. [61]. This would explain the lower VH backscatter and CR values compared to other years. Furthermore, the CR ratio was lower than in 2017 throughout the season. The lower normalized yield (5.9%) for 2019 compared to 2017 infers the lower CR values during the maize growth cycle. The yield reduction is due to anomalies at key moments. The resultant changes to the timing and length of growing season should be considered along with backscatter values to find potential indicators of reduced yield.
Comparing the columns of Figure 4, it is clear that the features discussed above are discernible in all four ROs. The effect of incidence angle is visible in VV and VH backscatter at the crop emergence. The difference in VV backscatter is larger than in VH backscatter, resulting in in a high CR for larger incidence angles.

Onion
Onion is a bulb vegetable typically planted at the end of March and harvested in September [62]. The period between planting and emergence is relatively long for onions. Soil surface roughness and moisture conditions dominate the backscatter contribution during this period. Figure 5 shows that VH backscatter and CR values increase during the leaf development phase (after DOY 140). Similar observations were reported in previous studies [63,64]. VH backscatter and CR decrease as the leaves bend and dry out during the ripening phase of onions. However, VV backscatter keeps increasing due to the increased sensitivity to soil moisture.
The inter-annual differences in the SAR observables were negligible before the onset of drought stress in 2018. The drought period coincides with the stem elongation and bulb development phases of onions. In 2018, all SAR observables started to decrease around DOY 210. CR values in Figure 5i-l show that the 2018 onion growing season was~10 days shorter than the 2017 season. This is likely because the onions ripened faster and entered senescence earlier. Onions that experience drought stress from the start of the bulb formation stage have a reduced final bulb weight [65] and thus lower crop yield. This decrease is seen in the normalized gross yield data, shown in Figure 2b. The 2018 onion yield is 6.8% lower than the 2017 yield. Figure 5i-l shows that the 2019 CR values are~1 dB lower during the emergence and leaf compared to other years. Combined with limited precipitation for a short period in the early season, this led to onion crops being planted into a relatively dry soil and thus a decreased above ground biomass overall compared to other years. The 2019 CR values become similar to the 2017 CR values during ripening and harvest. When comparing the onion yield of 2019 with 2017, the yield was slightly (2.75%) increased (Figure 2b. Hence, the lower CR values before ripening did not translate into a lower yield for onion crops in 2019. The influence of the satellite overpass time can also be found in the Sentinel-1 backscatter from onion parcels in Figure 5. After VV and especially VH backscatter become relatively stable around DOY 160, VV and VH backscatter rapidly increase in the ascending orbits (Figure 5c   In Scheldestromen (Figure 6h), and to a lesser extent in Flevopolder (Figure 6i), the 2019 CR is mostly similar to 2018. A likely cause is the soil type in Vechtstromen-Noord, which is dominated by sand. In Scheldestromen and Flevopolder, the soil has a higher clay content.
Furthermore, it can be observed that lack of precipitation events from DOY 70 to 110 leads to the largest drop in surface soil moisture in Vechtstromen-Noord. Surface soil moisture levels during planting were lower in Vechtstromen-Noord than in other study areas. Early dry periods for maize crops can lead to an increase in largest root diameter and a decrease in leaf elongation [2]. This results in a decrease in VH and thus CR values. In addition, the cumulative precipitation shows a dry period in the mid-season for Vechtstromen-Noord. This period is shorter for Scheldestromen and non-existent for Flevopolder, which could explain the decreased CR values in 2019. Another distinct difference between the study areas is the rapid CR decrease due to harvest at the end of the cropping season. Compared to 2017, the CR in 2018 starts decreasing approximately 40, 25, and 10 days earlier in Vechtstromen-Noord, Scheldestromen and Flevopolder, respectively. This is consistent with the normalized gross yield that reduced by 9.1%, 5.4% and 6% compared to the 2017 values (Figure 2a).
For 2019, the CR decreases rapidly approximately 15 and 10 days earlier than in 2017 in Vechtstromen-Noord and Scheldestromen, respectively. In Flevopolder, no early decline is observed for the 2019 CR. The influence of incidence angle on CR values during emergence, can also be observed here, especially in 2019. During emergence, Vechtstromen-Noord has the highest CR values in all years followed by Flevopolder. Scheldestromen shows the low-est values during emergence. This follows the the incidence angles which are approximately 43°, 38°and 41°for Vechtstromen-Noord, Scheldestromen and Flevopolder, respectively.  Figure 7 shows the Sentinel-1 backscatter from onion parcels in the three different study areas. Onion is a bulb crop and its yield is directly associated with water intake [66], hence drought impacts and regional variabilities are more pronounced in onion compared to other crops. The effect of the 2018 drought are clearly visible in the VH backscatter and CR values for all the studied regions. The early drop in CR values is indicative of the early maturity of onions in 2018. However, among all the studied regions, onion VH backscatter and CR values are lower in Scheldestromen compared to Vechtstromen-Noord and Flevopolder. CR decreases earlier in 2018 in Scheldestromen compared to other areas. Yield data also suggest that the onion crop was hit hardest during the 2018 drought in Scheldestromen.

Onion
The normalized gross yield dropped by 6.81%, 27.1% and 17.66% compared to the 2017 drought in Vechtstromen-Noord, Scheldestromen and the Flevopolder, respectively (Figure 2b). Scheldestromen may have been impacted the heaviest due to the lack of available fresh water for irrigation because of the increased salt intrusion in the drought period [67]. Figure 7h shows that the rate of CR decrease is faster in 2018 compared to other areas, and the harvest period is earlier than in other areas. Furthermore, the 2019 CR data clearly display that the root zone soil moisture in Vechtstromen-Noord had not been restored yet [57]. The values during and after planting until DOY 190 are extremely low compared to 2017 and 2018. This is not observed for Scheldestromen since it is situated below sea level and the soil mostly consists of clay. Therefore, the 2019 CR does not show signs of drought early in the season for most crops. In addition, for 2019, CR values are slightly lower in Scheldestromen, while in Vechtstromen-Noord and Flevopolder they are close to the 2017 CR values. These changes are also evident in cumulative rainfall and surface soil moisture plots. After DOY 220, 2019 rainfall is similar to 2018 rainfall for Scheldestromen, whereas in other two areas these are higher than 2018. There is a reduction in onion yield in these two areas as well. However, the observations show that this does not affect the backscatter.  The 2018 VH/VV ratio of north (Hardenberg and Coevorden) and south (Tubbergen and Hof van Tewnte) municipalities in Vechtstromen starts deviating from 2017 after DOY 230 (Figure 8). It seems that the crop season was shorter for both north and south municipalities and harvesting occurred earlier in 2018. However, the CR in the Tubbergen and Hof van Tewnte municipalities decreases earlier and faster at the end of the season than the CR in the Hardenberg and Coevorden municipalities. A possible explanation is that a higher number of maize parcels in the south municipalities were impacted heavily by the drought due to the ban on open water irrigation, which resulted in an earlier ripening and senescence and thus an early harvest. Furthermore, after DOY 230, the standard deviation of CR in the south municipalities is significantly higher than the north municipalities. This could be because farmers in the south municipalities were only allowed to irrigate from groundwater. Not all farmers are willing to irrigate silage maize with groundwater due to the cost. This may lead to an increase in the backscatter variation across the parcels. In addition, the VH and VV backscatter is significantly lower in the south municipalities during the harvest period around DOY 260.
Coevorden and Hardenberg have very similar CR signatures. The same can be concluded for Tubbergen and Hof van Twente. This can also be seen in Figure 9 where CR values of maize parcels are mapped. The top-left map, which shows the CR ratio on DOY 230, shows regional variability between the north and the south municipalities. This difference increases with each subsequent SAR acquisition shown in Figure 9, i.e., DOY 236, 242 and 248.
In 2019, the differences between the north and south municipalities are negligible until harvest, especially in CR. The CR in the south municipalities starts to decrease around DOY 260, while in the north municipalities, the CR starts to decrease around DOY 275. A high deviation of CR is also visible in the early growth phase (before DOY 150) of maize in all the municipalities for 2019 compared to other years.

Potato
Potatoes are typically planted between the end of April and the beginning of May, and harvesting starts in September. Potato parcels are easily recognizable due to the deep ridge planting. As a result, the backscatter variation in potato prior emergence is larger due to the deep ridges compared to other crops [68]. Figure 10 shows that in 2017, the backscatter from potato parcels of all the four municipalities is similar. However, the VH backscatter and CR response of potato parcels in 2018 start to diverge around DOY 210 for Tubbergen and Hof van Twente municipalities. Interestingly, CR values are relatively higher for Hof van Twente and lower for Tubbergen than the other two north municipalities (Figure 10h). These differences can be attributed to water management decisions taken in individual areas as cumulative precipitation (Figure 8n) and surface soil moisture (Figure 8k) are comparable among municipalities. The difference in VV, VH and CR values among north and south municipalities is relatively low compared to maize. This is because potato has a higher economic value and is likely to be irrigated with groundwater. Although average CR differences are not high, the standard deviation differences are relatively high between the north and south municipalities. A possible cause is the relatively low number of parcels in Tubbergen, especially in Hof van Twente. The backscatter differences between the north and south municipalities in 2019 are similar but smaller than in 2018.

Discussion
The interannual variability in the VV, VH backscatter and CR values indicates that SAR signals are sensitive to the drought stress for agricultural crops. However, these changes vary depending on the crop and soil types, water management practices followed in the studied regions. The lower VV and VH backscatter during drought may be caused by dry soil, reduction of VWC and changes in leaf geometry. Furthermore, it was observed that the seasonal cycle of CR for maize and onion was shorter in 2018. However, CR trends for sugar beet and potato did not show a significant difference in 2018 compared to 2017. A possible reason for this is an almost similar drop in VH and VV polarization values in 2018 for sugar beet and potato. Additionally, CR values are relatively lower during the vegetative stages in 2019 compared to 2017 and 2018 for all the crop types.
Regional variability occurs due to many factors, including soil type, hydrometeorological conditions and irrigation policies. Generally, drought responses were most extreme in the Vechtstromen-Noord study area. This is to be expected since this area is most vulnerable to droughts due to its sandy soil. However, onion crops in 2018 were impacted the most in the Scheldestromen area according to yield data ( Figure 2). This clearly translated into lower VH backscatter and CR ratio values during and after the drought period.
Among the three SAR observables, CR was found to be a potential candidate to investigate the change of crop growth status due to agricultural drought in 2018 for vertical crops.The utility of CR for identifying key crop growth stages have been discussed in previous studies [31,69,70].

Conclusions
To date, there have been few studies focusing on the use of SAR to monitor the effects of drought on crops. Here, we investigated the influence of inter-annual variability in the Sentinel-1 SAR signals over different crops due to the drought in 2018 and 2019 in the Netherlands. The inter-annual variation shows that VH and VV values are lower in the drought period compared to the normal year. The seasonal cycle for CR is found to be shorter for drought years for both maize and onion. This is caused by faster crop development and thus an earlier harvest due to drier environmental conditions in 2018.
Overall, the influence of the incidence angle on the drought-induced inter-annual variability of Sentinel-1 backscatter time series is found to be limited. On the other hand, the overpass time significantly influences drought-induced inter-annual variability. During drought periods, the drop in VV and VH backscatter was larger in descending orbits (05:45 UTC).
Water management decisions play a crucial role during agricultural drought events. It is observed that for potato crop parcels where open irrigation is allowed, the spatialtemporal variations are similar in all municipalities. Whereas, for maize, a clear difference is observed in CR between the municipalities where irrigation is allowed and those where it is not.
The outcomes of this study revealed the spatial and seasonal variability of SAR signals during the drought years for different crops in the Netherlands where agricultural practices are intensive and regulated. Future research should be extended to understand the responses of different crop types in different agro-climatic regions such as drylands. The sensitivity of the SAR signals also suggests that apart from the backscatter response, other SAR derived observables, interferometric coherence and retrieved bio-geophysical parameters could also be sensitive to the agricultural drought. The use of longer wavelength SAR data may provide a better insight into underlying soil moisture conditions during the drought events. In addition, the synergistic use of optical satellite derived information and SAR observables could combine the sensitivity to spectral characteristics and vegetation water content of the crop canopy affected by the drought.
The research findings demonstrate the Sentinel-1 SAR data contains useful information for water boards, irrigation planners and crop-information services. These data should be included to improve preparedness for future water stress events. They can contribute to decision-making regarding crop parcel water requirements and resource allocation depending on the spatial severity of the drought and economic viability of the crops grown in that area. In the longer-term, monitoring of inter-annual variability in the start and duration of the growing season using SAR data could be included in yield prediction and real-time crop monitoring.

•
Although Sentinel-1 SAR ensures data availability through the crop growing season, it is not without limitations. Results presented here show the potential value of parcel-level Sentinel-1 backscatter. However, data need to be integrated or synergistically used with high-resolution optical satellite observations to understand within parcel heterogeneity. • This study also suggests that Sentinel-1 observation geometry (incidence angle and overpass time) influences the backscatter response from different crop types which is relevant for the interpretation of SAR signals. This warrants attention when data from multiple orbits are combined to create a dense time series for near-real monitoring of the crops. • Previous studies have reported the saturation of C-band SAR signal response to high biomass crops during the peak vegetative stage [71][72][73]. This could be a potential limitation of using C-band SAR data for seasonal drought observations. However, using a longer wavelength (L-band), SAR data partially overcomes penetration depth limitations. Planned future SAR missions such as NISAR, BIOMASS, ROSE-L will operate at a longer wavelength compared to C-band Sentinel-1 SAR.