Scholarlink

<p>Accurate estimation of the actual crop evapotranspiration (AETc) is essential for optimizing water management in rice cultivation, particularly in middle Indo-Gangetic Plains (MIGP), characterized by the intensive agriculture and variable water availability. Conventional approaches such as weighing lysimeters and eddy covariance systems, while effective, are often economically and logistically impractical for widespread adoption in developing countries. To address this limitation, a field experiment was conducted using a locally constructed cost-effective, non-weighing paddy lysimeter to estimate AETc, quantify water balance components, and determine stage-wise crop coefficient (Kc) values for transplanted puddled rice (TPR). In addition, a suite of machine learning (ML) models-including Random Forest (RF), Gradient Boosting (GB), Linear Regression (LR), Artificial Neural Networks (ANN), Support Vector Machines (SVM), and k-Nearest Neighbours (kNN)—was evaluated for their performance in predicting AETc based on the weather variables. Major Water losses primarily due to percolation and evapotranspiration accounted for 35.4% and 49.8% of total applied water in the first year, and 25.8% and 51. 5% during second year, respectively, indicating limited water retention within the root zone in TPR conditions. The regional average Kc values (1.10 ± 0.19, 1.20 ± 0.23, 1.15 ± 0.17, and 0.88 ± 0.12 for initial, development, mid-season, and end-season stages, respectively) closely aligned with the FAO-standard values, suggesting their broader applicability in similar agro-ecological regions. Among the ML models tested, ANN demonstrated superior predictive performance, with the minimal error metrics (MAE: 0.04–0.08&#xa0;mm/day, RMSE: 0.02–0.06&#xa0;mm/day) and excellent goodness-of-fit indices (R²: 0.99; NSE: 0.99–1.0). Gradient Boosting (GB) followed closely R² (0.97–0.99), while kNN showed the poorest performance, R² (0.37–0.53). Integrating ML-based data driven approaches with non-weighing paddy lysimeter data enhances water use efficiency, helps in precise irrigation scheduling, and fostering sustainable agriculture in resource-limited conditions of the Middle Indo-Gangetic Plains of South Asia and similar agro-ecotypes of the globe.</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

Estimation of actual evapotranspiration and stage-wise crop coefficients for transplanted rice using a modified non-weighing paddy lysimeter and their prediction on machine ensemble approach in the middle Indo-Gangetic plains of South Asia

  • 期刊论文

  • 发表日期：

    2025年10月13日

查看全文
  • Arti Kumari,
  • Ashutosh Upadhyaya,
  • Pawan Jeet,
  • Rakesh Kumar,
  • Kirti Saurabh,
  • Ved Prakash,
  • Anup Das,
  • P. K. Sundaram,
  • A. K. Singh,
  • Akram Ahmed,
  • Amit Kumar

摘要

Accurate estimation of the actual crop evapotranspiration (AETc) is essential for optimizing water management in rice cultivation, particularly in middle Indo-Gangetic Plains (MIGP), characterized by the intensive agriculture and variable water availability. Conventional approaches such as weighing lysimeters and eddy covariance systems, while effective, are often economically and logistically impractical for widespread adoption in developing countries. To address this limitation, a field experiment was conducted using a locally constructed cost-effective, non-weighing paddy lysimeter to estimate AETc, quantify water balance components, and determine stage-wise crop coefficient (Kc) values for transplanted puddled rice (TPR). In addition, a suite of machine learning (ML) models-including Random Forest (RF), Gradient Boosting (GB), Linear Regression (LR), Artificial Neural Networks (ANN), Support Vector Machines (SVM), and k-Nearest Neighbours (kNN)—was evaluated for their performance in predicting AETc based on the weather variables. Major Water losses primarily due to percolation and evapotranspiration accounted for 35.4% and 49.8% of total applied water in the first year, and 25.8% and 51. 5% during second year, respectively, indicating limited water retention within the root zone in TPR conditions. The regional average Kc values (1.10 ± 0.19, 1.20 ± 0.23, 1.15 ± 0.17, and 0.88 ± 0.12 for initial, development, mid-season, and end-season stages, respectively) closely aligned with the FAO-standard values, suggesting their broader applicability in similar agro-ecological regions. Among the ML models tested, ANN demonstrated superior predictive performance, with the minimal error metrics (MAE: 0.04–0.08 mm/day, RMSE: 0.02–0.06 mm/day) and excellent goodness-of-fit indices (R²: 0.99; NSE: 0.99–1.0). Gradient Boosting (GB) followed closely R² (0.97–0.99), while kNN showed the poorest performance, R² (0.37–0.53). Integrating ML-based data driven approaches with non-weighing paddy lysimeter data enhances water use efficiency, helps in precise irrigation scheduling, and fostering sustainable agriculture in resource-limited conditions of the Middle Indo-Gangetic Plains of South Asia and similar agro-ecotypes of the globe.