Scholarlink

<p>Accurate assessment of forest biomass is crucial for quantifying carbon stocks and informing climate change mitigation strategies. This study developed spatially explicit biomass and carbon maps by integrating extensive field measurements with remote sensing datasets. A total of 45 super plots (250&#xa0;m × 250&#xa0;m) and 180 nested quadrats (31.6&#xa0;m × 31.6&#xa0;m) were established for vegetation and soil sampling across north-eastern zone of Eastern Ghats, India. Field-derived biomass ranged from 1.799 to 361.697 Mg ha⁻¹, while carbon stocks varied from 12.744 to 206.259 Mg C ha⁻¹. Machine learning models such as Random Forest (RF), Gradient Boosting Machine (GBM), Artificial Neural Network (ANN), and Multiple Linear Regression (MLR) were trained using biomass and carbon stocks as dependent variables and soil attributes, topographic factors, bioclimatic variables, and remotely sensed products as predictors. All models performed strongly during calibration, with R² values of 0.94 (RF), 0.95 (GBM), 0.83 (ANN), and 0.95 (MLR), demonstrating their robustness for regional biomass estimation. Comparisons with global biomass datasets showed moderate agreement, with R² values ranging from 0.338 to 0.456, indicating structural differences between local and global products. This work offers strong ground data and models to enhance biomass estimation and refine remote sensing products.</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

Comparative analysis of machine learning biomass models against global remote sensing products using ground validation

  • 期刊论文

  • 发表日期：

    2026年1月6日

查看全文
  • Prakash Paraseth,
  • Arpan Khanal,
  • Kakoli Banerjee

摘要

Accurate assessment of forest biomass is crucial for quantifying carbon stocks and informing climate change mitigation strategies. This study developed spatially explicit biomass and carbon maps by integrating extensive field measurements with remote sensing datasets. A total of 45 super plots (250 m × 250 m) and 180 nested quadrats (31.6 m × 31.6 m) were established for vegetation and soil sampling across north-eastern zone of Eastern Ghats, India. Field-derived biomass ranged from 1.799 to 361.697 Mg ha⁻¹, while carbon stocks varied from 12.744 to 206.259 Mg C ha⁻¹. Machine learning models such as Random Forest (RF), Gradient Boosting Machine (GBM), Artificial Neural Network (ANN), and Multiple Linear Regression (MLR) were trained using biomass and carbon stocks as dependent variables and soil attributes, topographic factors, bioclimatic variables, and remotely sensed products as predictors. All models performed strongly during calibration, with R² values of 0.94 (RF), 0.95 (GBM), 0.83 (ANN), and 0.95 (MLR), demonstrating their robustness for regional biomass estimation. Comparisons with global biomass datasets showed moderate agreement, with R² values ranging from 0.338 to 0.456, indicating structural differences between local and global products. This work offers strong ground data and models to enhance biomass estimation and refine remote sensing products.