11:00 AM - 1:00 PM
[MSD41-P06] INTERPRETATION OF RADAR AND OPTICAL DATA FOR CROPS FIELD YIELD MONITORING
Keywords:agriculture, yield, remote sensing, radar
The yield and production forecasting of the main crops are a societal and economic challenge at global, national and farm levels. In the context of climate change, the uncertainty on the future of crop productions is a major issue. The evolution of optical and radar satellites imagery from moderate to very high resolution provides new tools to address yield variability at different spatial scales. The Moderate Resolution Imaging Spectroradiometer (MODIS), and Visible Infrared Imaging Radiometer Suite (VIIRS) tools have provided significant information in the crop yield status at the national and regional levels (Franch et al, 2019). But to improve local scale forecasting models, it is necessary to integrate high and very high-resolution data.
This study was focused on the yield variability study between fields by averaging yields data at field level from 258 fields of Corn, Soybean and Rice in Nebraska. We used Planet/Dove-Classic, Sentinel-2, Landsat 8 (through Harmonized Landsat Sentinel-2 - HLS) and radar data (HLS, Planet, UAVSAR, RADARSAT-2, Sentinel 1-SAR).
We observed a very variable performance between radar index and between crops. In term of spectral bands, NIR (~0.865 μm) was the most important to explain yield variability whatever the resolution and crop types (R2adj>0.5). In term of polarimetric indexes, the results showed that in general the results for rice are mostly stable and better than the other crops (R2adj ~ 0.4 on average). The best results are obtained for the S1 VHasc indices with R2adj=0.42 and for RS2 phase difference with R2adj=0.42. The results for corn are the worst with an R2adj <0.35 for all the indexes. The results for soybean are more variable and seem to be highly correlated with certain indexes such as Rs2 HV, RS2 Volume and RS2 Pauli HV with R2adj>0.4. For optical and radar satellite-derived features, we investigated the day of the years of the maximum of the correlation for corn, soybean, and rice. The maximum of correlation for optical features occurs in a short time range between DOY 155 and DOY 185 for corn and rice, and DOY 190 and DOY 211 for soybean. It shows a consistency between of the optical derived feature. On the contrary, the maximum of correlation for radar derives features occurs for very variable DOY from DOY 120 to DOY 225. A study of the time series parameters correlation show that the optical parameters are highly correlated, but the radar parameters show strong temporal decorrelation. We identified by this way, 6 interested radar parameters for rice, 4 for corn and 4 for soybean.
We also investigated crop varieties for rice and soybean (insufficient data for corn). The results of the linear model seem to be closely linked to the variety studied. For the best radar indexes (RadarSat2: HV-VV-HH), the soybeans variety, 46-D08, R2adj=0.79 and for the variety, AG-46X6, R2adj =0.64. As a previous study, we also found that performance of the satellite-derived models was better for corn, soybean, and rice fields with lower average yields. It means that multi-spectral and polarimetric data saturate over high yields and cannot capture yield variability (Skakun et al, 2021). These results suggest the importance of combining radar data and optical data to integrate other biophysical variables, such as soil moisture and evapotranspiration.
This study was focused on the yield variability study between fields by averaging yields data at field level from 258 fields of Corn, Soybean and Rice in Nebraska. We used Planet/Dove-Classic, Sentinel-2, Landsat 8 (through Harmonized Landsat Sentinel-2 - HLS) and radar data (HLS, Planet, UAVSAR, RADARSAT-2, Sentinel 1-SAR).
We observed a very variable performance between radar index and between crops. In term of spectral bands, NIR (~0.865 μm) was the most important to explain yield variability whatever the resolution and crop types (R2adj>0.5). In term of polarimetric indexes, the results showed that in general the results for rice are mostly stable and better than the other crops (R2adj ~ 0.4 on average). The best results are obtained for the S1 VHasc indices with R2adj=0.42 and for RS2 phase difference with R2adj=0.42. The results for corn are the worst with an R2adj <0.35 for all the indexes. The results for soybean are more variable and seem to be highly correlated with certain indexes such as Rs2 HV, RS2 Volume and RS2 Pauli HV with R2adj>0.4. For optical and radar satellite-derived features, we investigated the day of the years of the maximum of the correlation for corn, soybean, and rice. The maximum of correlation for optical features occurs in a short time range between DOY 155 and DOY 185 for corn and rice, and DOY 190 and DOY 211 for soybean. It shows a consistency between of the optical derived feature. On the contrary, the maximum of correlation for radar derives features occurs for very variable DOY from DOY 120 to DOY 225. A study of the time series parameters correlation show that the optical parameters are highly correlated, but the radar parameters show strong temporal decorrelation. We identified by this way, 6 interested radar parameters for rice, 4 for corn and 4 for soybean.
We also investigated crop varieties for rice and soybean (insufficient data for corn). The results of the linear model seem to be closely linked to the variety studied. For the best radar indexes (RadarSat2: HV-VV-HH), the soybeans variety, 46-D08, R2adj=0.79 and for the variety, AG-46X6, R2adj =0.64. As a previous study, we also found that performance of the satellite-derived models was better for corn, soybean, and rice fields with lower average yields. It means that multi-spectral and polarimetric data saturate over high yields and cannot capture yield variability (Skakun et al, 2021). These results suggest the importance of combining radar data and optical data to integrate other biophysical variables, such as soil moisture and evapotranspiration.