Estimation Of County Rice Straw Resources In Hilly Areas Of Central Sichuan Based On Sentinel-1/2 Images

Rice is a vital food crop in the hilly regions of central Sichuan.Accurately estimating the quantity and spatial distribution of rice straw resources is crucial for developing effective strategies to utilize them efficiently.Currently,estimation methods rely on traditional statistics and grass-grain ratio coefficients.However,the delayed nature of statistical data and regional variations in average grass-grain ratio coefficients hinder obtaining timely and accurate information on the current season’s rice straw resources.Remote sensing technology,with its wide monitoring range and independence from natural conditions,holds promise as a primary approach for investigating straw resource quantities.However,current remote sensing techniques face challenges in directly obtaining regional rice straw resource data.Optical images,the primary data source,often suffer from interference caused by factors like cloud cover and rainfall,leading to missing information and impacting the implementation of remote sensing-based estimation of rice straw resources.To address this,this study focuses on Zhongjiang County and utilizes Sentinel-1/2 satellite images along with field measurements from representative rice fields to estimate the quantity of rice straw resources and understand their spatial distribution characteristics in typical hilly areas of central Sichuan.The research aims to provide technical support for estimating and comprehensively utilizing rice straw resources at the county scale in hilly regions.The key research findings are as follows:（1）The results of multi-scale segmentation have a direct impact on the classification outcome of the images.In this study,we employed the ESP2 plugin in combination with the RMNE scale evaluation function to determine the optimal segmentation scale of 30.The best index factor was used to determine the combination of band weights.By conducting qualitative and quantitative analysis,we established a shape factor of 0.1 and a compactness factor of 0.5.Additionally,we employed the Relif F-RF coupled feature selection method to reduce the initial set of 127 feature factors to just 18.（2）Based on three types of machine learning,the spatial distribution of rice planting area is classified and extracted.The OA of the support vector machine model reaches 89.32%,and the Kappa coefficient is 0.8478;the OA of the random forest classification model reaches 97.49%,and the Kappa coefficient is 0.9749;the OA of the decision tree classification model reaches 98.67%,Kappa coefficient is 0.9809.The decision tree classification model showed better classification performance among the three models.The spatial distribution area of rice extracted by the decision tree classification model was 26444.6ha,compared with the statistical data of rice planting area of 27821.5ha,the relative accuracy reached 95.05%.（3）Among the three regression models to construct the rice straw yield inversion model,the particle swarm optimization random forest regression model verification set R²reached 0.86,RMSE was 0.056 kg.m^-2,and MAE was 0.045 kg.m^-2;random forest regression The R²of the model verification set reached 0.75,the RMSE was 0.061 kg.m^-2,and the MAE was 0.049 kg.m^-2;the R²of the multiple linear stepwise regression model verification set reached 0.68,the RMSE was 0.113 kg.m^-2,and the MAE was 0.093 kg.m^-2,particle swarm optimization random forest regression model is the optimal production estimation model.Based on the spatial distribution of rice planting area and the optimal model of straw per unit yield,the total output of rice straw resources in the research area in 2021 is 138475.15 t,and the range of rice straw production per unit area is between 4.73 t.ha^-1and 7.62 t.ha^-1.The average yield per unit area of rice straw was 5.24 t.ha^-1.（4）The results of Moran index and spatial cold and hot spots show that the spatial distribution of rice straw resources shows significant spatial aggregation.Based on the analysis and determination of the results of geographic detectors,the main influencing factors that cause the differences in the spatial distribution of rice straw resources are the rice growth period The monthly average precipitation and temperature in May,June and July,and the interactive detection results show that altitude is a cofactor affecting the spatial distribution of rice straw resources.