Scholarlink

<p>Betel vine cultivation sustains rural economies across South and Southeast Asia but depends critically on fine-scale interactions among climate, soil, hydrology, and infrastructure. This study refines betel-vine land-suitability assessment through an ensemble machine-learning framework integrating Random Forest (RF), Logistic Regression (LR), and Extra Trees Classifier (ETC) with a harmonized suite of multi-source predictors. Environmental inputs, covering climate, hydrology, soil, topography, vegetation, land use, and settlement pressure, were pre-processed, normalized, and balanced using SMOTE. The models were trained on presence–absence data for existing cultivation zones, optimized by GridSearchCV, and validated using accuracy, F1-score, Matthews Correlation Coefficient (MCC), ROC-AUC, and log-loss metrics.A proxy validation using block-wise multi-year production, area, and yield data was further conducted to externally evaluate the ensemble model’s reliability at the administrative scale. Results demonstrated that ensemble learning improved overall calibration and spatial stability, with ETC achieving the highest discrimination (ROC-AUC = 0.96), followed by RF (0.95), the ensemble (0.94), and LR (0.84). The ensemble outperformed individual models in calibration and spatial coherence, providing more balanced probability distributions across suitability classes. The ensemble delineated ~ 42% of the deltaic plain as “very high suitability,” primarily within riparian belts and irrigated floodplains characterized by organic-rich alluvial soils, gentle slopes, and stable groundwater. Proxy validation confirmed strong monotonic and categorical agreement (ρ = 0.93; κ = 0.68) between modelled suitability and actual productivity, reinforcing the model’s external validity and practical applicability. Variable-importance analysis identified land-use/land-cover, rainfall, soil pH, and organic carbon as dominant predictors, confirming the synergy of moisture regime and fertility in shaping betel viability. The proposed probabilistic, multi-model framework establishes a transferable methodology for perennial crop suitability mapping in deltaic 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

Probabilistic Ensemble Machine Learning and Remote Sensing Integration for Betel Vine (Piper betle L.) Site-Suitability Assessment in Purba Medinipur District, India

  • 期刊论文

  • 发表日期：

    2026年6月24日

查看全文
  • Debabrata Chanda,
  • Sayanti Poddar,
  • Uttam Mukhopadhyay,
  • Swapan Talukdar,
  • Shahfahad,
  • Atiqur Rahman

摘要

Betel vine cultivation sustains rural economies across South and Southeast Asia but depends critically on fine-scale interactions among climate, soil, hydrology, and infrastructure. This study refines betel-vine land-suitability assessment through an ensemble machine-learning framework integrating Random Forest (RF), Logistic Regression (LR), and Extra Trees Classifier (ETC) with a harmonized suite of multi-source predictors. Environmental inputs, covering climate, hydrology, soil, topography, vegetation, land use, and settlement pressure, were pre-processed, normalized, and balanced using SMOTE. The models were trained on presence–absence data for existing cultivation zones, optimized by GridSearchCV, and validated using accuracy, F1-score, Matthews Correlation Coefficient (MCC), ROC-AUC, and log-loss metrics.A proxy validation using block-wise multi-year production, area, and yield data was further conducted to externally evaluate the ensemble model’s reliability at the administrative scale. Results demonstrated that ensemble learning improved overall calibration and spatial stability, with ETC achieving the highest discrimination (ROC-AUC = 0.96), followed by RF (0.95), the ensemble (0.94), and LR (0.84). The ensemble outperformed individual models in calibration and spatial coherence, providing more balanced probability distributions across suitability classes. The ensemble delineated ~ 42% of the deltaic plain as “very high suitability,” primarily within riparian belts and irrigated floodplains characterized by organic-rich alluvial soils, gentle slopes, and stable groundwater. Proxy validation confirmed strong monotonic and categorical agreement (ρ = 0.93; κ = 0.68) between modelled suitability and actual productivity, reinforcing the model’s external validity and practical applicability. Variable-importance analysis identified land-use/land-cover, rainfall, soil pH, and organic carbon as dominant predictors, confirming the synergy of moisture regime and fertility in shaping betel viability. The proposed probabilistic, multi-model framework establishes a transferable methodology for perennial crop suitability mapping in deltaic environments.