Assessment of the Hail Damage on Agricultural Crops from North of Baragan Plain, Romania, Using Remote Sensing Data


 Hail is one of the dangerous meteorological phenomena facing society. The present study aims to analyze the hail event from 20 July 2020, which affected the villages of Urleasca, Traian, Silistraru and Căldăruşa from the Traian commune, Baragan Plain. The analysis was performed on agricultural lands, using satellite images in the optical domain: Sentinel-2A, Landsat-8, Terra MODIS, as well as the satellite product in the radar domain: Soil Water Index (SWI), and weather radar data. Based on Sentinel-2A images, a threshold of 0.05 of the Normalized Difference Vegetation Index (NDVI) difference was established between the two moments of time analyzed (14 and 21 July), thus it was found that about 4000 ha were affected. The results show that the intensity of the hail damage was directly proportional to the Land Surface Temperature (LST) difference values in Landsat-8, from 15 and 31 July. Thus, the LST difference values higher than 12° C were in the areas where NDVI suffered a decrease of 0.4-0.5. The overlap of the hail mask extracted from NDVI with the SWI difference situation at a depth of 2 cm from 14 and 21 July confirms that the phenomenon recorded especially in the west of the analyzed area, highlighted by the large values (greater than 55 dBZ) of weather radar reflectivity as well, indicating medium–large hail size. This research also reveals that satellite data is useful for cross validation of surface-based weather reports and weather radar derived products.


25
In the context of climate change, the risk of weather disasters, such as hailstorms, threaten human well-being by  The hail phenomenon can be predicted and analyzed, in terms of its effects, based on weather radar information and 48 satellite images. The attempts to identify the agricultural areas affected by hailstorm, using photogrammetric means (color and infrared photograms), have been done since 1969, analyzing the relationship between real vegetation 50 damage and that determined remotely by photogrammetric analysis (Towery et al. 1975).
Multispectral imagery from remote sensing satellites represent an effective and unique instrument for rapid and 52 precise investigation of natural hazards and it has the potential to provide valuable assistance during damage surveys 53 after severe hailstorm events (Jedlovec et al. 2006).

54
The effects of hail have been highlighted over time by many specialists who have used various sources of satellite 55 data, for example: GOES-8 satellite data in the case of (Klimowski et al. 1998

60
Hail cannot always be recorded comprehensively by ground-based weather stations, if it occurs outside the spatial 61 extent monitored in-situ. Thus, the weather radars are the only measuring systems providing a large area under 62 constant surveillance and high temporal resolution (Kunz and Kugel 2015). Radar reflectivity can be used to assess 63 and monitor the intensity of precipitation and to infer the precipitation type. The occurrence of hail on the ground 64 that causes severe damage is very likely when the radar reflectivity exceeds a threshold of 55 dBZ (Sandu et al.

73
The defoliation of crops resulted as a combined action of wind and hail contributed to the greatest reduction, almost 74 to a total compromise, whereas strong winds alone may have only toppled the crops and allowed them to survive and 75 recover on their own later, leading to smaller Normalized Difference Vegetation Index (NDVI ) changes (Changnon

87
The rest of article is structured as follows: section 2 presents a short overview of the weather event, while the third 88 section refers to the data and methods used to analyze the hail event. The fourth section contains the results and 89 discussions with reference to the hail damage and in the last part are the conclusions which emphasize the 90 importance of using satellite data for hail evaluation.

93
The weather event for which the analysis was performed is the hailstorm that occurred on 20 July 2020 in the area of 94 Traian Administrative Territorial Units (ATU), located in the South-East of Romania (Baragan Plain). The study area

95
( Fig. 1) (Sandu et al. 2010). In 20 July 2020, between 10:00 and 19:00 UTC, National Meteorological Administration 100 (NMA) issued an orange code warning for most areas of the country (western and southern Moldova, northern 101 Dobrogea, northern and northeastern Muntenia and eastern and southeastern Transylvania). Accentuated atmospheric 102 instability with heavy rain, lightnings, strong winds and hailstorms were forecasted. In that day, the amount of 103 precipitation was expected to reach 30-40 l/m 2 and locally 50-60 l/m 2 .

104
Precipitation cumulated up to 25 l/m 2 between 06:00 UTC 20 July 2020 -06:00 UTC 21 July 2020 (Fig. 2), being 105 significantly higher compared to the previous period (14-20 July), when the amount of precipitation was below 5 the analyzed area being thus prone to agricultural drought.   The current study relies on remote sensing data, in order to assess the hail damage on agricultural crops on 20 July 116 2020 in the Traian ATU (Baragan Plain). Multispectral data from the Sentinel-2 satellite, the Land Surface 117 Temperature (LST) product obtained from the Landsat 8 satellite data, SAR images obtained from the Sentinel-1 118 satellite and the LST and reflectance products obtained from the Terra MODIS satellite were used (Table 1).

119
Auxiliary data, such as meteorological and Soil Water Index (SWI) products were used to identify the condition of 120 soil moisture before and after the hailstorm. Also, weather radar images were used to identify the hailstorm and Land validation of the results was done based on the statements and photos published in the online newspaper. According to the mayor of the Traian commune, and the statements of the locals, over 1000 ha of agricultural crops 125 (corn, soybeans, sunflowers, but also vegetables from gardens) have been affected by hailstorm. The hail lasted 10-126 20 minutes, the fields being covered with a layer of ice up to 10 cm thickness, this further damaging crops (some of 127 them being irrigated -example Căldărușa area - Fig. 3 c) that were already affected by drought (Radio Romania 128 Antena Satelor 2020). Also, according mass-media, it was stated the hail had a diameter of up to 4-5 cm, strongly 129 affecting the agricultural crops, whose damage was estimated at about € 608895 (Valsan 2020; Ferma 2020). It

141
The locals' gardens were also affected by the fury of nature, the medium-large hail accumulating in a significant 142 layer, as evidenced by the photos posted in the media and social media (Fig. 4).      The difference in vegetation between the two images is observed even in the case of the natural color combination, 178 with a spatial resolution of 10 m (Fig.6). Thus, the hail effects at the area of interest (Traian ATU) were detected 179 from the Sentinel-2 L2A multispectral images (level 2 processing), by comparing the vegetation situation before and

199
We computed ∆NDVI approach (∆NDVI= NDVI pre-event -NDVI post-event ) to distinguish changes in the vegetation

236
Although VHI was designed to identify the types of drought (VHI values below 40,

259
The difference of NDVI between the two images (14 July and 21 July) is significant especially for the agricultural 260 areas (between 0.8 and 0.9 in Fig. 7a, to below 0.6 in Fig. 7b) in the west of Traian ATU. The change detection of 261 NDVI, more exactly the subjective NDVI hail threshold of 0.05 showed that the most the most significant damage 262 were in the west of the Traian commune, especially around the Căldăruşa and Urleasca, were the NDVI recorded 263 generally a decrease higher than 0.25 and locally up to 0.5 (Fig.7).

264
From the NDVI hail threshold of 0.05 (Fig.8 a) it resulted that approximately 4000 ha were affected by hail (Table   265 3), mostly the arable land (3550 ha) being damaged. Also, the hail affected vegetable gardens and fruit trees inside the rural green area included in the area (126.78 ha) (Fig.8 b, c)   the hail event, the west of the Traian ATU being seriously affected (Fig.9 a-b), especially for and around the

296
Also, we tried to analyze the damages for the type of crops (eg. corn, sunflower, soybean etc.) identified on the 297 satellite imagery, considering also the photos taken from the field and the declaration of the authorities and locals.

298
One first result was that the correlation between NDVI and LAI has a Pearson coefficient of 0.99, meaning that both 299 identified the hail damage in the same manner, both indices decreasing with the defoliation of vegetation after 20

305
continued to suffer in August, but with a lower decreasing trend compared to previous time series (Fig. 12).

306
To analyze the damage for agricultural crops we decided to remove the harvested areas from the hail mask based on 307 the 0.05 NDVI threshold. To identify the type of crop we used the photointerpretation based on the Sentinel-2 308 images, from May to August (Fig. 13). The harvested crops (eg. winter wheat, rape) were observed with few weeks 309 before the hail event, so these areas were excluded from a damage assessment. In the case of pastures, vineyards, 310 fruit trees and berry plantations we used the LPIS 2020 to analyze the effects of hail. Thereby, from the 3922 ha of 311 agricultural lands extracted from NDVI hail mask, only 3142 ha were considered for the unharvested crops (Table 4).

312
It is necessary for the unharvested crops to be evaluated and monitored in order to make future connections between 313 the effects of hail, the decrease of agricultural productions and the compensation of the insured farmers.

327
After the hailstorm, on 21 July 2020, the NDVI and LAI values significantly decreased for the corn and sunflower, 328 these crops continued to suffer more than the soybean in the next period. So, if on 14 July, the NDVI for corn had a 329 value of 0.78, on 21 July decreased to 0.74, registering a loss of 15.58%. On 29 July, the corn affected by hail had a 330 NDVI value of 0.54 and a loss of 17.37% compared with the 21 July (Fig. 14). Compared with the NDVI results, the 331 LAI pointed out a higher increase of the loses (in some cases even over to half) suffered by the crops for the entire 332 selected interval (Fig.15). The LAI difference between 14 and 21 July marks a loss of 33.65% in the case of corn,

333
whereas the NDVI registered a loss of 16.35%. For the sunflower the LAI had a loss of 29.54%, whereas the NDVI 334 only 13.86%. Regarding the soybean the least sensitive to hail, the LAI had a loss of 15.58% and the NDVI 335 registered 9.37%. For 21-29 July difference, the LAI continued to have high loses for corn (25.36%) and sunflower 336 (18.52%), whereas for the next time difference (29 July-08 August), the soybean registered higher losses (16.38%),

337
than the other two crops.

338
In the case of the 8-18 August difference, the LAI soybean reached the highest loss of -30.58% during the time series 339 analyzed, whereas the NDVI marks a loss of 15.15%. The evolution of LAI and NDVI losses for the soybean during 340 the interval selected highlights the idea that soybean was less affected and the process of biophysical degradation 341 was not so significant as in the case of corn and sunflower. As a conclusion, it can be said that sensitivity of LAI it 342 seems to be a good indicator for estimating the losses of vegetation condition for different crops.

4.4.WHI 373
The areas most affected by the hail have registered the lowest values, suggesting an extreme vegetative stress for the 374 crops from the west of the Traian commune, crops that had optimal conditions for growth and high productivity 375 before the hail event. Thus, for the period 28 July-04 August, the VHI values around the Căldăruşa village indicated 376 an extreme agricultural drought, followed by the severe and moderate drought (Fig. 18). The hail effects on VHI are noted just in the case of synthesis 28 July-04 August because in the previous synthesis 381 (20-27 July 2020), the maximum of NDVI, respectively the mean LST necessary for Terra MODIS products are 382 recorded around the time 09:30 UTC, long before the hail occurred (12 UTC).

383
During the 21 days of recorded and processed data, the VHI surprised very well the remarkable decreasing trend of 384 vegetation affected by the hail in the Traian commune. So, if the mean of VHI for the NDVI hail mask was 72 in the 385 case of synthesis 201 (20-27 July), for the next synthesis (28 July-04 August) the VHI decreased to 41 (Fig. 19).  In the case of Sentinel-1 data, we tried to identify a homogeneous hail mask through Normalized Ratio Procedure 395 between Bands (NRPB) and the difference between bands in dual polarization (eg. VH) and bands in single 396 polarization (eg. VV) but we could not find out satisfactory results. Therefore, we selected and processed to linear 397 gamma only the band VV which is vertically in dual polarization (Fig. 21).

398
On 21 July, after the hail occurred the maximum of VV (0.35 dB) was recorded, then on the next available radar data 399 (25 July), the affected surfaces registered a significant decrease, reaching the 0.26 dB. Further, the VV values were 400 oscillated around 0.20 dB (Fig. 22).

406
In the case of the SWI analysis, from the 8 levels at which the soil moisture is calculated, the depth of 2 cm most 407 clearly highlights the increase in humidity that occurred on 21 July, as a result of the precipitation generated by the 408 hailstorm on 20 July. It can be noticed that on 22 July, the humidity recorded at a depth of 2 cm decreased 409 substantially, below the first pre-hail day, so that from 23 July to be below the other depth levels, thus confirming the 410 infiltration of precipitation inside the soil (Fig. 23). To emphasize the temporal correlation of the NDVI index, respectively of the resulting hail mask, with SWI, the 414 humidity data at the depth of 2 cm from 14 and 21 July were analyzed based on the values difference. Thereby, the 415 increase of humidity predominated, the most significant increases (15-20%) being in the area of Urleasca and 416 Căldăruşa, while in the east of Traian commune were reported increases of up to 5%.

417
The overlap of the NDVI hail mask with the SWI situation confirms that the phenomenon occurred mainly in the 418 west of the analyzed area (Fig. 24). Regarding the 14-21 July difference for the SWI values converted to volumetric 419 units, soil moisture increases were higher south of the Căldăruşa and around Urleasca village (0.010-0.012), 420 compared to the rest of the affected territory (Fig. 25). Also, the evolution of converted SWI (Fig. 26)

447
Hail is a dangerous weather phenomenon, causing significant damage to agriculture. Modern weather radars allow 448 the identification of hail, while high-resolution multispectral satellite imagery have the ability to provide valuable 449 assistance in analyzing damage resulting from severe hail events.

450
The hail occurring in Traian commune was identified on weather radar data, from which the reflectivity values 2 images, the spatial resolution of 10 m representing a great advantage in detecting and analyzing the vegetation 453 affected by hail. It should also be emphasized that Sentinel-2 data was the only tool capable of identifying hail 454 damage to vegetation. The 0.05 threshold applied for ∆NDVI and ∆LAI emphasizes better the hail damage extent 455 compared to the simple positive difference ∆NDVI and ∆LAI. Also, comparing the results based on Otsu's method, 456 it turned out that the identification of the spatial extent of hail is considerably limited for LAI comparing to NDVI, 457 the latter overestimating the hail extent. According to the NDVI difference values (14-21 July 2020) higher than 458 0.05, it was found that about 4000 ha were affected by hail, the arable lands registering the most damages (3550 ha), 459 especially in the proximity of Căldărușa and Urleasca localities.

460
Given the results obtained by the two methods (∆NDVI and ∆LAI, followed by a visual subjective hail threshold, 461 respectively the Binary Thresholding Function) one can conclude that the hail affected especially the agricultural 462 areas from Urleasca and Căldăruşa and overall, the hail produced damages between 2000 and 4000 ha.

463
Considering that NDVI hail mask extent, based on the visual threshold, is more suitable than the LAI, we chose to 464 compare several indices evolution (LAI, LST, SWI) with the extend derived from 0.05 NDVI threshold, in order to 465 evaluate the hail damages. Another reason why we chose NDVI as hail mask for further several analyzes, instead of 466 LAI, is the fact that NDVI has been widely used for the hail analysis.

467
The evolution of LAI and NDVI losses for the soybean after the hail event highlights the idea that soybean was less 468 affected and the process of biophysical degradation was not so significant as in the case of corn and sunflower.

469
The differences between the LST values, from 15 to 31 July 2020, are higher than in the surrounding areas, with 470 temperature increases, generally between 8-12° C. The average LST for the affected area was 38.27° C on 31 July, 471 compared to 30.48° C on 15 July, 2020, thus marking a difference of + 7.79° C. At the same time, it was observed 472 that the intensity of hail damage was directly proportional to the LST values. Thus, the LST difference values higher 473 than 12° C were in the areas where the NDVI suffered a decrease of 0.4-0.5.

474
Based on the SWI difference at a depth of 2 cm from July 14 and 21, it was observed that the most significant 475 increases (15-20%) were in the area of Urleasca and Căldăruşa localities, while in the east of Traian commune was a 476 moderate increase, up to at 5%. The overlap of the hail mask extracted from NDVI with the SWI situation confirms 477 that the hail occurred especially in the west of the analyzed area, which is also highlighted by the weather radar data.

478
The use of images and remote sensing techniques allow the identification, analysis and evaluation of the hail 479 phenomenon, providing continuous information. Sentinel-1,2 satellite data and products distributed free through the