Supplementary Materials for "Mapping the Spatial Distribution of Photovoltaic Power Plants in Northwest China Using Remote Sensing and Machine Learning"

The supplementary materials include all data and figures supporting the methodology and results of the study Tables and Excel files: Mutual information result.xlsx, Pearson correlation results.xlsx, and The 360 feature names after screening.xlsx provide detailed statistical analysis and feature selection results, including mutual information, Pearson correlation coefficients, and the final screened feature set used in the study. CSV file: Variance calculation result.csv contains variance analysis for all candidate features. Text files: RELIEF-F Fine Sample Interval Results.txt and RFECV 10-fold cross-validation -Results.txt document detailed results of the feature selection procedures and model cross-validation performance. Shapefile datasets: npvcaiyangdian.* and pvcaiyangdian.* include sampling point locations and attribute data for PV and non-PV samples used in classification, containing .shp, .shx, .dbf, .prj, .sbn, .sbx, and .shp.xml files. Figures: Fig. S1: Location of the study area and long-term annual average values for the potential PV power generation period from 2007 to 2018 (data sourced from Global Solar Atlas, accessed on December 8, 2024). Fig. S2: (a) Spatial coverage frequency of Sentinel-2 images with cloud cover <15%; (b) Spatial distribution of PV and non-PV sample points. Fig. S3: Workflow for classifying PV power plants. Fig. S4: Comparative analysis of classification accuracy and running time between Relief-F and RFECV feature selection methods. Fig. S5: Pearson correlation coefficients of the top 20 preferred features (- indicates top ten in absolute ranking with PV_label). Fig. S6: Mutual information of the top 20 preferred features (- indicates top ten in ranking with PV_label). Fig. S7: SHAP-based feature importance analysis of the RF classifier (a: SHAP summary plot; b: mean absolute SHAP values for each feature, Class 0: non-PV, Class 1: PV). Fig. S8: Classification results of PV power plants. Fig. S9: Spatial distribution characteristics and statistical analysis of PV power plants (a: kernel density of PV stations in 2023; b: comparison of PV area among regions; c: correlation between PV area and newly added & cumulative grid-connected capacity). Fig. S10: Mapping result of PV power plants. Fig. S11: SHAP waterfall explanation comparison between PV and non-PV samples (a,b: PV samples; c,d: non-PV samples). Fig. S12: Illustration of main sources of classification errors. Fig. S13: Location map of detected PV power plants, main power lines, and NPP-VIRS (2023 mean). These supplementary files enable reproducibility and transparency of the PV mapping workflow, feature selection, and classification analyses.