Methods For Estimating Maize Biomass From Sentinel-1 Radar Data And Sentinel-2 Optical Remote Sensing Data

Maize is an important,globally cultivated food.Dynamic maize growth monitoring is essential for guiding maize production and resolving practical production problems.Above-ground biomass is a key biophysical metric for monitoring crop growth status,plays an important role in the strategies for irrigation,managing fertilizer application,disease control,and yield forecasting.Remote sensing data provides the advantages of wide coverage,timeliness,and non-destructiveness,providing strong data support for spatially continuous crop biomass estimation in comparison to the time-consuming and destructive sampling in field measurements.This study proposed approaches for retrieving maize biomass in Changchun and Songyuan,Jilin province,Northeastern China.We developed models from band reflectance,vegetation indices（VIs）,biophysical variables（BPVs）,SAR polarization indices and texture features by employing multi-temporal Sentinel-1 Synthetic Aperture Radar（SAR）and Sentinel-2 data to find methods to improve the accuracy of crop biomass estimation based on gaussian process regression（GPR）and random forest（RF）,and evaluated the capability of Sentinel-1 and Sentinel-2 to estimate maize total biomass,leaf biomass,and stem biomass based on multi-output random forest（MRF）and multi-output XGBoost（MXGB）.The main research contents and conclusions are as follows:（1）The Pearson’s correlation coefficients were calculated to assess the sensitivities of vegetation index and biophysical variables to maize biomass.Three variables were selected,which were most correlated to the biomass of three different growth stages.The new fusion index was constructed by integrating the most sensitive variables of different growth stages to substantially improve the accuracy of biomass estimation.The GPR model using a combination of ratio vegetation index（RVI）of June,normalized different infrared index（NDII）of July,and normalized difference vegetation index（NDVI）of August achieved a result with R²=0.83 and RMSE=0.39 kg/m²,much better than single VIs or their combination,or optimized features（R²of 0.31-0.77,RMSE of 0.47-0.87 kg/m²）.A BPV predictor,combined with leaf chlorophyll content（CAB）in June,canopy water content（CWC）in July,and fractional vegetation cover（FCOVER）in August,with RF,also yielded the highest accuracy（R²=0.85,RMSE=0.38kg/m²）compared to that of single BPVs or their combinations,or optimized subset.（2）The results showed that a predictor combined by vertical-horizontal polarization（VV）,vertical-horizontal polarization（VH）,and the difference of VH and VV（VH-VV）derived from Sentinel-1 images of June,July,and August,respectively,with GPR and RF,provided a more accurate estimation of biomass（R²=0.81-0.83,RMSE=0.40-0.41 kg/m²）than the models based on single SAR polarization indices or their combinations,or optimized features（R²=0.04-0.39,RMSE=0.84-1.08 kg/m²）.（3）Ratio texture index constructed from texture features MEA_VH+VV/MEA _VH–VVbased on GPR model achieved the highest retrieval accuracy,with an R² of 0.70 and a RMSE of 0.54 kg/m²,compared to that of single VIs,BPVs or SAR polarization indices（R²=0.02-0.65,RMSE=0.59-1.23 kg/m²）.The GPR model using a combination of HOM _VH+VV/SEM VH–VV of June,MEAVH+VV/DIS VH–VV of July,and HOM VH+VV/HOM VH-VV of August achieved a result with R²=0.87 and RMSE=0.34 kg/m²,much better than the integrated predictor of VIs,BPVs or SAR polarization indices.（4）Multi-output methods were proposed for the simultaneous estimation of leaf biomass,stem biomass and total biomass by combining Sentinel-1 and Sentinel-2 remote sensing data.The best predictive accuracy was achieved by MRF with B2,FCOVER,NDVI,RVI,MSR,VH+VV,VH×VV,MEA_VH+VV/MEA _VH–VV as predictors（R²=0.86,RMSE=0.07 kg/m²for leaf biomass;R²=0.73,RMSE=0.20kg/m²for stem biomass and R²=0.80,RMSE=0.43 kg/m²for total biomass）.The spatial and temporal distribution of biomass in the study area from June to August was mapped based on the model and statistically analyzed.The biomass statistics of each component in different growth periods were in high agreement with the results of field measurements and were consistent with the growth pattern of maize.