| As one of the staple grain crops in developing countries,paddy rice plays an important role in alleviating the pressure of population growth on food demand and ensuring regional and global food security.Thus,accurate and timely acquisition of high-resolution paddy rice spatio-temporal distribution is critical for creating smart agriculture,ensuring national food security,adapting and optimizing agricultural production structure,and rationally utilizing water resources.In this study,we develop a high-resolution remote sensing extraction model of paddy rice planting range based on the decision tree,random forest,support vector machine,and classification and regression trees algorithms by taking advantage of the Google Earth Engine cloud computing platform and collaborating with Sentinel-1 SAR and Sentinel-2 MSI data on the basis of establishing a priori knowledge of spectral and polarization characteristics of paddy rice growing period.In order to achieve large-scale and high-precision paddy rice extraction,the time window and feature combination of image synthesis are optimized during the construction of the machine learning extraction model.Meanwhile,to assess paddy rice mapping accuracy,quantitative analysis the uncertainty factors of paddy rice mapping in terms of the classification algorithm,hybrid pixel,and scale effect.Finally,the paddy rice spatio-temporal distribution characteristics in Hunan Province were analyzed based on the high-resolution paddy rice dataset during 2017～2021.The main conclusions are as follows:(1)The overall accuracy and Kappa coefficient of the paddy rice planting range decision tree model built with Sentinel-1 SAR and Sentinel-2 MSI time-series data based on prior knowledge of spectral and polarization characteristics of the paddy rice growing period were93.97 percent and 0.908,respectively.On the whole,it meets the criteria of high-precision paddy rice planting range extraction.We also found that the paddy rice extraction accuracy based on EVI spectral feature approach was better than using the VH polarization feature method in the decision tree model.Due to the influence of cloud coverage,however,it is impossible to completely extract the paddy rice planting area using the spectral feature method.(2)In order to achieve the application requirements of large-range extraction and high-precision paddy rice mapping,the time window size of Sentinel-2 MSI and Sentinel-1SAR image synthesis were rate determined.Compared with other feature combination schemes,the fused spectral,index,polarization,and topographic features have the highest accuracy,the overall accuracy,and the Kappa coefficient reach 96.70% and 0.945,respectively.Furthermore,the paddy rice extraction performance of different classification algorithms were compared based on the optimal time window image synthesis and feature combination.The results show that random forest outperforms support vector machine,decision tree model,and classification and regression trees.(3)Quantitative analysis the uncertainty factors of paddy rice planting range estimation and their impact on the paddy rice estimation accuracy in terms of classification algorithm and hybrid pixel effect showed that the developed decision tree model and random forest classification algorithm show stronger ability,and the rice extraction accuracy is higher than support vector machine and classification regression tree algorithm.The average area accuracy of remote sensing estimating paddy rice planting range rises with the extension of spatial range at the same spatial resolution.The average area accuracy reduces as the spatial resolution lowers within the same spatial range.The more intensive the rice cultivation,the higher the average regional precision.The better the spatial resolution,the higher the average area accuracy under the same intensively planted rice planting.(4)The single and double cropping paddy rice are mixed structural features.The double cropping paddy rice is concentrated in Yueyang,Changde,and Yiyang cities,while the single cropping paddy rice is sparsely distributed relatively.Although the inter-annual variation of city-level paddy rice-planted acreage is small,the overall trend of double cropping paddy rice acreage is decreasing during 2017～2021.Moreover,the paddy rice distribution is significantly affected by topography and temperature.It is mainly distributed in the area with the elevation below 200 m,slope less than 6°,and annual average temperature greater than 17 ℃ in Hunan Province.Double cropping paddy rice tends to be grown at lower elevations,on gently sloping terrain,and in areas with higher average annual temperatures. |
