Scholarlink

<p>Ladakh represents one of the most fragile agroecological systems of the Indian Trans-Himalaya, where agriculture is constrained by high altitude, severe aridity, short growing season, glacially regulated water availability, shallow soils and rapid socio-economic change. Although remote sensing and GIS have widely been applied to land evaluation in arid and mountainous environments, their integration with machine learning for land capability assessment in data-sparse cold desert landscapes remains insufficiently developed. This study addresses this gap by developing an integrated geospatial-machine learning framework for land capability classification and agroecological zonation in Ladakh. Sentinel-2 and Landsat imagery, Cartosat/ASTER digital elevation data, field-based soil observations, SoilGrids information, IMD climatic data, RUSLE-based erosion modelling, Analytic Hierarchy Process weighting, Random Forest classification and K-means clustering were combined to generate spatially explicit land capability and agroecological suitability maps. The Random Forest model was configured and validated through stratified training-validation partitioning, cross-validation and multiple performance metrics. Results show that approximately 62.1% of the region falls under Class VIII, indicating barren, glaciated, permafrost-affected or otherwise unsuitable terrain. Only about 8.9% of the area belongs to Classes I-IV and is potentially suitable for crop-based or marginal agricultural use, mainly along irrigated valleys and terraces. Agroecological zonation identified high-potential, moderate-potential and low-potential zones occupying 8%, 12% and 80% of the area, respectively. Slope, water accessibility, soil organic carbon and NDVI emerged as the most influential predictors of suitability. The findings demonstrate that sustainable agricultural planning in Ladakh must prioritize protection of limited valley agriculture, climate-resilient irrigation, soil conservation, regulated grazing and integration of indigenous knowledge with precision geospatial decision support. The proposed framework offers a transferable model for similar high-altitude cold desert environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Geospatial and machine learning techniques for land capability assessment and agroecological zoning in the Trans-Himalayan Cold Desert of Ladakh, India

  • 期刊论文

  • 发表日期：

    2026年7月30日

查看全文
  • Mahesh Kumar Gaur,
  • Rajesh Kumar Goyal,
  • Neelratan Singh

摘要

Ladakh represents one of the most fragile agroecological systems of the Indian Trans-Himalaya, where agriculture is constrained by high altitude, severe aridity, short growing season, glacially regulated water availability, shallow soils and rapid socio-economic change. Although remote sensing and GIS have widely been applied to land evaluation in arid and mountainous environments, their integration with machine learning for land capability assessment in data-sparse cold desert landscapes remains insufficiently developed. This study addresses this gap by developing an integrated geospatial-machine learning framework for land capability classification and agroecological zonation in Ladakh. Sentinel-2 and Landsat imagery, Cartosat/ASTER digital elevation data, field-based soil observations, SoilGrids information, IMD climatic data, RUSLE-based erosion modelling, Analytic Hierarchy Process weighting, Random Forest classification and K-means clustering were combined to generate spatially explicit land capability and agroecological suitability maps. The Random Forest model was configured and validated through stratified training-validation partitioning, cross-validation and multiple performance metrics. Results show that approximately 62.1% of the region falls under Class VIII, indicating barren, glaciated, permafrost-affected or otherwise unsuitable terrain. Only about 8.9% of the area belongs to Classes I-IV and is potentially suitable for crop-based or marginal agricultural use, mainly along irrigated valleys and terraces. Agroecological zonation identified high-potential, moderate-potential and low-potential zones occupying 8%, 12% and 80% of the area, respectively. Slope, water accessibility, soil organic carbon and NDVI emerged as the most influential predictors of suitability. The findings demonstrate that sustainable agricultural planning in Ladakh must prioritize protection of limited valley agriculture, climate-resilient irrigation, soil conservation, regulated grazing and integration of indigenous knowledge with precision geospatial decision support. The proposed framework offers a transferable model for similar high-altitude cold desert environments.