Scholarlink

<p>Effective irrigation water management in sub-Saharan Africa faces major challenges, including water scarcity, climate variability, and long-term meteorological data. Accurate estimation of reference evapotranspiration (ETo) is critical for assessing crop water requirements, irrigation optimizing, and developing decision support systems for sustainable agriculture. The FAO Penman-Monteith (FAO PM) equation, the standard for estimating ETo, requires extensive weather station networks and comprehensive meteorological data, often unavailable in developing countries. Although empirical models with fewer inputs exist, they typically need calibration. Machine learning (ML) models offer promising alternatives, especially under data-limited conditions. This study evaluates five ML models, including Bootstrap Aggregating (Bagging), Linear Regression (LR), Multilayer Perceptron (MLP), Random Forest (RF), and M5P Pruning (M5P), for ETo estimation across six agro-ecological zones in Nigeria using 15 different data combinations. A total of 450 scenarios were assessed against the FAO PM model using standard statistical indices. Results show that ML model performance varies by location and input data, with notable differences between northern and southern regions. Under complete data conditions, all models showed high accuracy (NSE: 0.924–0.999; RMSE: 0.038–0.446&#xa0;mm/day during testing). In data-limited scenarios, models using solar radiation and relative humidity were most effective in the south, while relative humidity and wind speed yielded better results in the north. Overall, the ML models ranked as follows: Bagging &gt; M5P &gt; MLP &gt; RF &gt; LR. This study highlights the potential of ML models to estimate ETo reliably with minimal inputs, offering practical guidance for irrigation planning and sustainable agriculture in Nigeria.</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

Machine learning-based estimation of daily reference evapotranspiration across agro-ecological zones in Nigeria: comparative analysis and model ranking

  • 期刊论文

  • 发表日期：

    2025年11月26日

查看全文
  • Emeka Ndulue,
  • Ikenna Onyekwelu,
  • Kelechi Igwe,
  • Vintus Ogwo,
  • Okechukwu Michael

摘要

Effective irrigation water management in sub-Saharan Africa faces major challenges, including water scarcity, climate variability, and long-term meteorological data. Accurate estimation of reference evapotranspiration (ETo) is critical for assessing crop water requirements, irrigation optimizing, and developing decision support systems for sustainable agriculture. The FAO Penman-Monteith (FAO PM) equation, the standard for estimating ETo, requires extensive weather station networks and comprehensive meteorological data, often unavailable in developing countries. Although empirical models with fewer inputs exist, they typically need calibration. Machine learning (ML) models offer promising alternatives, especially under data-limited conditions. This study evaluates five ML models, including Bootstrap Aggregating (Bagging), Linear Regression (LR), Multilayer Perceptron (MLP), Random Forest (RF), and M5P Pruning (M5P), for ETo estimation across six agro-ecological zones in Nigeria using 15 different data combinations. A total of 450 scenarios were assessed against the FAO PM model using standard statistical indices. Results show that ML model performance varies by location and input data, with notable differences between northern and southern regions. Under complete data conditions, all models showed high accuracy (NSE: 0.924–0.999; RMSE: 0.038–0.446 mm/day during testing). In data-limited scenarios, models using solar radiation and relative humidity were most effective in the south, while relative humidity and wind speed yielded better results in the north. Overall, the ML models ranked as follows: Bagging > M5P > MLP > RF > LR. This study highlights the potential of ML models to estimate ETo reliably with minimal inputs, offering practical guidance for irrigation planning and sustainable agriculture in Nigeria.