Scholarlink

<p>Water-induced soil erosion is a critical driver of land degradation in the Eastern Himalayas, threatening agricultural productivity and regional stability. Previous assessments in Northeast India often rely on coarse global datasets or resource-intensive field approaches that lack the spatial granularity required for district-level management in complex terrains such as Aizawl. This study aimed to quantify the modelled average annual soil loss and delineate high-risk erosion zones using high-resolution remote sensing data. We employed a hybrid approach that integrated the Revised Universal Soil Loss Equation (RUSLE) with machine learning algorithms, specifically Random Forest (RF) and Support Vector (SV). Parallelized pixel-based operations for rainfall (R), Soil Erodibility (K), cover management (C), and conservation practice (P) factors were conducted in Google Earth Engine (GEE), whereas Slope length and steepness factor (LS) factor was computed in ArcGIS 10.8. The findings show Rainfall erosivity ranged from 850 to 1108&#xa0;MJ mm ha<sup>− 1</sup> h<sup>− 1</sup> yr<sup>− 1</sup> and extreme topographic relief with LS values up to 83. The model results indicated that RF performed with higher accuracy (R<sup>2</sup> = 0.93) than SV (R<sup>2</sup> = 0.44) in estimating soil erosion. The estimated mean annual soil loss was 151.15 t ha<sup>− 1</sup> yr<sup>− 1</sup> using the RUSLE model and 149.64 t ha<sup>− 1</sup> yr<sup>− 1</sup> using RF. A significant positive correlation was observed between soil loss and the LS (<i>r</i> = 0.34, <i>p</i> &lt; 0.001), the P factors (<i>r</i> = 0.34, <i>p</i> &lt; 0.001) and the R factor (<i>r</i> = 0.27, <i>p</i> &lt; 0.001), while soil loss was weakly and negatively correlated with K factor (<i>r</i> = – 0.03, <i>p</i> &lt; 0.001). Furthermore, the probability zonation map shows that more than 52% of the district falls under a slight erosion risk zone, while 45% falls under a moderate zone; the remains is shared between high-to very-high erosion-prone areas, primarily in barren lands, built-up areas, riverbanks, and steep terrains. These findings highlight a critical “erosion paradox” in which high forest cover masks localized degradation hotspots. These patterns identify critical hotspots where extreme topography overrides forest protection, allowing managers to prioritize high-risk riverbanks and steep slopes for targeted agroforestry and landslide prevention efforts. This study recommends the urgent need for LS-based zoning and the adoption of multi-strata agroforestry systems to mitigate sediment yields and landslides.</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

Hybrid machine learning-RUSLE approach for soil erosion assessment in Aizawl district, Mizoram, Northeast India

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月15日

查看全文
  • Ahmed Abdallah Adam,
  • Chhakchhuak Lalruatkimi,
  • Khumanthem Babina Devi,
  • Nistha Khanna,
  • Gyati Yam,
  • Faisal Ismail Musa,
  • Ashish Paul,
  • Om Prakash Tripathi

摘要

Water-induced soil erosion is a critical driver of land degradation in the Eastern Himalayas, threatening agricultural productivity and regional stability. Previous assessments in Northeast India often rely on coarse global datasets or resource-intensive field approaches that lack the spatial granularity required for district-level management in complex terrains such as Aizawl. This study aimed to quantify the modelled average annual soil loss and delineate high-risk erosion zones using high-resolution remote sensing data. We employed a hybrid approach that integrated the Revised Universal Soil Loss Equation (RUSLE) with machine learning algorithms, specifically Random Forest (RF) and Support Vector (SV). Parallelized pixel-based operations for rainfall (R), Soil Erodibility (K), cover management (C), and conservation practice (P) factors were conducted in Google Earth Engine (GEE), whereas Slope length and steepness factor (LS) factor was computed in ArcGIS 10.8. The findings show Rainfall erosivity ranged from 850 to 1108 MJ mm ha− 1 h− 1 yr− 1 and extreme topographic relief with LS values up to 83. The model results indicated that RF performed with higher accuracy (R2 = 0.93) than SV (R2 = 0.44) in estimating soil erosion. The estimated mean annual soil loss was 151.15 t ha− 1 yr− 1 using the RUSLE model and 149.64 t ha− 1 yr− 1 using RF. A significant positive correlation was observed between soil loss and the LS (r = 0.34, p < 0.001), the P factors (r = 0.34, p < 0.001) and the R factor (r = 0.27, p < 0.001), while soil loss was weakly and negatively correlated with K factor (r = – 0.03, p < 0.001). Furthermore, the probability zonation map shows that more than 52% of the district falls under a slight erosion risk zone, while 45% falls under a moderate zone; the remains is shared between high-to very-high erosion-prone areas, primarily in barren lands, built-up areas, riverbanks, and steep terrains. These findings highlight a critical “erosion paradox” in which high forest cover masks localized degradation hotspots. These patterns identify critical hotspots where extreme topography overrides forest protection, allowing managers to prioritize high-risk riverbanks and steep slopes for targeted agroforestry and landslide prevention efforts. This study recommends the urgent need for LS-based zoning and the adoption of multi-strata agroforestry systems to mitigate sediment yields and landslides.