A METHOD FOR AUTOMATED EVALUATION OF WATER EROSION OF SOIL IN VINEYARDS USING SPECTRAL INDICES OF SATELLITE IMAGERY

Purpose: to develop an automated method for assessing water erosion of vineyard soils based on spectral indices of satellite images for the optimization of soil conservation measures. Materials and methods. Four models: RUSLE, ORUSCAL, an index model, and field measurements were used to assess soil loss. The index model is implemented in the Google Earth Engine environment and includes the calculation of normalized spectral indices NDVI, BSI, and the empirical erosion index (ELI). Automated detection of erosion forms was performed using machine learning (Random Forest) based on a training set of 256 images and 1200 marked plots. Soil loss (t/ha/year) is determined on a scale from 10 to 74, normalized by the average values of field measurements. The planting pattern (along/across the slope) is implemented through a reduction coefficient (0.85). Results. Key factors of water erosion: slope topography (slopes of 5–10°), soil characteristics (humus content below 2 %), and climatic conditions (heavy rainfall) were identified. An adapted methodology incorporating a dynamic C-factor and agronomic features (planting pattern, bush age) consideration was developed. The following measures were proposed to reduce erosion: placing rows across the slope (> 5°), terracing, and monitoring using Sentinel-2. The index model demonstrated the highest accuracy (28.8 t/ha/year), while the classical RUSLE (44.5 t/ha/year) and ORUSCAL (59.3 t/ha/year) showed overestimated values due to limitations in the calculation of the C-factor. The integration of satellite data and machine learning made it possible to automate the recognition of erosion types (linear, sheet, no erosion) with an accuracy of 89 %. Conclusions. The study confirms the effectiveness of combining remote sensing with adapted models, which reduces predicted soil loss by 30–50 %. This method reduces labor and financial costs for field research and the prompt identification of areas with a high erosion risk.