基于Sentinel-2影像和随机森林方法的风蚀季翻耕地识别OACHSSCD

In arid and semi-arid regions,farmland ploughing during the wind erosion season exposes the soil surface and significantly intensifies soil wind erosion.Accurately identifying the spatiotemporal distribution of ploughed land is therefore essential for improving regional wind erosion assessments.Taking Kangbao County-located in the core area of the Beijing-Tianjin Sand Source Control Project and the Capital Two Zones Construction-as the study area,this research integrates field survey samples with Sentinel-2 imagery to analyze spectral index characteristics of ploughed land and residue-covered land during the wind erosion season.Classification thresholds are derived to construct a time-series high-confidence sample set,and a Random Forest(RF)model is used to identify ploughed land and analyze its spatiotemporal dynamics and impacts on soil wind erosion.The results reveal that:1)quantifying the contribution of each band in identification of land types by using the Maximal Information Criterion(MIC),confirming a Logistic index that composes of bands B6,B7,B8 and B8A-as well as NDTI,RI(11,12),and BSI,all of them combined together can effectively overcome the weak spectral separability between ploughed and residue-covered land.2)Using Otsu's method to determine classification thresholds,and applying a multi-index consistency strategy to build a high-confidence sample library,which greatly enhances sample purity and representativeness.3)Based on time-series high-confidence samples,the RF model achieves multi-temporal ploughed-land identification with an overall accuracy above 0.96 and a Kappa coefficient above 0.92,demonstrating high robustness and generalization ability even under limited field samples and strong spatiotemporal variability.