Scholarlink

<p>Water scarcity and declining groundwater levels pose serious threats to agricultural sustainability in northwestern Bangladesh. Artificial rainwater harvesting (ARH)—the process of collecting, storing, and reusing surface runoff—offers a nature-based solution to improve water availability, especially in drought-prone areas. This study assessed ARH suitability and crop water demand using machine learning (ML) and geospatial techniques across the Rajshahi and Rangpur Divisions. To evaluate ARH suitability, seven ML models were used: random forest (RF), gradient boosting (GB), support vector regression (SVR), neural network (NN), eXtreme gradient boosting (XGBoost), K-nearest neighbors (KNN), and an ensemble model (EM). The EM showed the best performance (<i>R</i><sup>2</sup> = 0.99, RMSE = 0.07, MAE = 0.02), identifying over 99% of Rajshahi’s land as moderately to highly suitable. Rangpur showed limited suitability due to sandy soils, steeper slopes, and fragmented vegetation. Crop water demand estimated using RF, GB, SVR, elastic net (EN), decision tree (DT), and KNN, along with stacking models (SM, ESM), achieved high accuracy (SM,<i> R</i><sup>2</sup> = 1.000; ESM,<i> R</i><sup>2</sup> = 0.9999). Higher evapotranspiration (ET) values in Rajshahi (1.15&#xa0;mm/day) were linked to longer sunshine and warmer temperatures, compared to Rangpur (1.06&#xa0;mm/day). Cluster analysis based on NDVI, ET, and EM scores identified two ARH zones. Rajshahi formed a high-potential cluster (NDVI = 0.244; EM = 0.205; ET = 1.15&#xa0;mm/day), with 19,064.03 sq. km suitable area. Key districts included Rajshahi (95.62%), Naogaon (92.51%), and Pabna (90.84%). Rangpur formed the low-potential cluster (529.81 sq. km), including Kurigram, Nilphamari, and Lalmonirhat. Although model accuracy was high, limitations included the lack of ground-truth validation and exclusion of water infrastructure. This framework supports region-specific, climate-resilient ARH planning.</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

Optimizing Artificial Rainwater Harvesting in the Northwest Zone of Bangladesh: Integrated Machine Learning and a GIS-Based Multicriteria Approach

  • 期刊论文

  • 发表日期：

    2025年5月30日

查看全文
  • Md. Najmul Kabir,
  • Mirza Nazmul Hasan,
  • Tanha Tabassum Hridi,
  • Md. Mominul Islam,
  • Emon Barua,
  • Nidhy Mostan,
  • Shah Md Shahnewaz Sayem

摘要

Water scarcity and declining groundwater levels pose serious threats to agricultural sustainability in northwestern Bangladesh. Artificial rainwater harvesting (ARH)—the process of collecting, storing, and reusing surface runoff—offers a nature-based solution to improve water availability, especially in drought-prone areas. This study assessed ARH suitability and crop water demand using machine learning (ML) and geospatial techniques across the Rajshahi and Rangpur Divisions. To evaluate ARH suitability, seven ML models were used: random forest (RF), gradient boosting (GB), support vector regression (SVR), neural network (NN), eXtreme gradient boosting (XGBoost), K-nearest neighbors (KNN), and an ensemble model (EM). The EM showed the best performance (R2 = 0.99, RMSE = 0.07, MAE = 0.02), identifying over 99% of Rajshahi’s land as moderately to highly suitable. Rangpur showed limited suitability due to sandy soils, steeper slopes, and fragmented vegetation. Crop water demand estimated using RF, GB, SVR, elastic net (EN), decision tree (DT), and KNN, along with stacking models (SM, ESM), achieved high accuracy (SM, R2 = 1.000; ESM, R2 = 0.9999). Higher evapotranspiration (ET) values in Rajshahi (1.15 mm/day) were linked to longer sunshine and warmer temperatures, compared to Rangpur (1.06 mm/day). Cluster analysis based on NDVI, ET, and EM scores identified two ARH zones. Rajshahi formed a high-potential cluster (NDVI = 0.244; EM = 0.205; ET = 1.15 mm/day), with 19,064.03 sq. km suitable area. Key districts included Rajshahi (95.62%), Naogaon (92.51%), and Pabna (90.84%). Rangpur formed the low-potential cluster (529.81 sq. km), including Kurigram, Nilphamari, and Lalmonirhat. Although model accuracy was high, limitations included the lack of ground-truth validation and exclusion of water infrastructure. This framework supports region-specific, climate-resilient ARH planning.