Scholarlink

<p>Soil depth is a key physical attribute, plays a crucial role in determining the choice of crops grown, nutrient and water retention as well as controlling the various hydrological and erosion processes. However, there is lack of detailed spatial soil depth information in India, especially in hilly mountainous terrains of Himalayan region, which is limiting our various efforts towards sustainable management of soil resources in this fragile ecosystem. The study was attempted for spatial prediction and mapping of soil depth using various remote sensing derived environmental covariates employing random forest (RF) regression-based machine learning modelling approach in the Tehri Garhwal district of Uttarakhand. Soil depth observations were collected during a series of field surveys from 421 georeferenced locations belonging to different strata representing combinations of diverse land use/land cover (LULC), geology as well as elevation zones. Environmental covariates such as climatic/bioclimatic variables, terrain parameters, different spectral indices, LULC as well as geological information obtained from various sources such as terrain analysis, Google Earth Engine etc. were used for development of spatial prediction model using RF regression. Boruta algorithm was applied for feature selection prior to model development. RF model was trained and tested with 10-fold cross-validation strategy, using the depth observations from sampling locations and values of covariates, identified using feature selection procedure. Validated RF model could predict soil depth with considerable accuracy (R<sup>2</sup>: 0.76, RMSE: 7.01). Terrain Ruggedness Index (TRI), clay index, Multi Resolution Valley Bottom Flatness (MRVBF), Precipitation of Driest Month (bio14) and Temperature Annual Range (bio7) were found to be the most important variables accounting for soil depth variability. Topographic parameters (60%) form the majority in the top 10 important covariates, exhibiting the crucial role played by topography in governing depth distribution under hilly mountainous Himalayan terrain. The validated model could map soil depth effectively and the values varied from 22.5 to 72&#xa0;cm (mean:42.75&#xa0;cm and SD: 5&#xa0;cm), across the study area. The spatial uncertainty estimation was carried out in the form of 90% prediction intervals, which clearly showed the regions with varying prediction uncertainties. The study illustrated the use and potential of suitable environmental covariates and machine learning technique for depth distribution mapping in hilly and mountainous terrains. Such digital soil mapping-based activities may prove beneficial in our efforts towards effective resource management as well as comprehensive land use planning in various fragile mountain ecosystems.</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

Soil depth mapping in the hilly and mountainous landscape of North-West Himalayan region using machine learning technique

  • 期刊论文

  • 发表日期：

    2025年8月11日

查看全文
  • Justin George Kalambukattu,
  • Suresh Kumar,
  • Bappa Das,
  • Trisha Roy

摘要

Soil depth is a key physical attribute, plays a crucial role in determining the choice of crops grown, nutrient and water retention as well as controlling the various hydrological and erosion processes. However, there is lack of detailed spatial soil depth information in India, especially in hilly mountainous terrains of Himalayan region, which is limiting our various efforts towards sustainable management of soil resources in this fragile ecosystem. The study was attempted for spatial prediction and mapping of soil depth using various remote sensing derived environmental covariates employing random forest (RF) regression-based machine learning modelling approach in the Tehri Garhwal district of Uttarakhand. Soil depth observations were collected during a series of field surveys from 421 georeferenced locations belonging to different strata representing combinations of diverse land use/land cover (LULC), geology as well as elevation zones. Environmental covariates such as climatic/bioclimatic variables, terrain parameters, different spectral indices, LULC as well as geological information obtained from various sources such as terrain analysis, Google Earth Engine etc. were used for development of spatial prediction model using RF regression. Boruta algorithm was applied for feature selection prior to model development. RF model was trained and tested with 10-fold cross-validation strategy, using the depth observations from sampling locations and values of covariates, identified using feature selection procedure. Validated RF model could predict soil depth with considerable accuracy (R2: 0.76, RMSE: 7.01). Terrain Ruggedness Index (TRI), clay index, Multi Resolution Valley Bottom Flatness (MRVBF), Precipitation of Driest Month (bio14) and Temperature Annual Range (bio7) were found to be the most important variables accounting for soil depth variability. Topographic parameters (60%) form the majority in the top 10 important covariates, exhibiting the crucial role played by topography in governing depth distribution under hilly mountainous Himalayan terrain. The validated model could map soil depth effectively and the values varied from 22.5 to 72 cm (mean:42.75 cm and SD: 5 cm), across the study area. The spatial uncertainty estimation was carried out in the form of 90% prediction intervals, which clearly showed the regions with varying prediction uncertainties. The study illustrated the use and potential of suitable environmental covariates and machine learning technique for depth distribution mapping in hilly and mountainous terrains. Such digital soil mapping-based activities may prove beneficial in our efforts towards effective resource management as well as comprehensive land use planning in various fragile mountain ecosystems.