Scholarlink

<p>Effective water resources management is germane to food security and climate resilience, particularly in semi-arid regions such as northern Nigeria. Limited studies have integrated time series (TS) and machine learning (ML) for estimating agricultural water footprints (AWF) and the agricultural water scarcity index (AWSI). This study presents a novel framework that combines advanced TS modelling techniques with six ML models. The ML (SVM, ANN, RF, KNN, LSTM, and GBM) models were applied singly and in ensembles to estimate key water-related indicators: blue (TWF<sub>b</sub>) and green (TWF<sub>g</sub>) water footprints, AWF, AWSI, and crop evapotranspiration (ET<sub>c</sub>). These were assessed for eight major crops in their respective highest-producing states in Nigeria. Among the TS approaches, the TBATS model (Trigonometric, Box-Cox transformation, ARMA errors, Trend and Seasonal components) demonstrated superior forecasting performance for most climatic variables across locations, with Nash–Sutcliffe Efficiency (NSE) values ranging from 0.523 to 0.923. LSTM architectures consistently outperformed other ML models in estimating AWF components, AWSI, and ET<sub>c</sub>, achieving NSE values between 0.994 and 0.998. However, the ensemble, Combined ML model (CML 11) integrating ANN, GBM, SVM, and LSTM emerged as the most robust and generalisable across variables and locations, since the NSE ranged between 0.691 and 0.994. SHAP analysis identified crop coefficient, maximum air temperature and radiation as the most influential variable across all target parameters. This study offers a valuable decision-support system for sustainable agricultural water management and climate-smart agriculture in Nigeria and comparable agro-ecological regions, globally.</p> Graphical abstract <p></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 and time series integration for agricultural water footprints and scarcity index estimation in Northern Nigeria

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Blessing Funmbi Sasanya,
  • Oludare Owolabi

摘要

Effective water resources management is germane to food security and climate resilience, particularly in semi-arid regions such as northern Nigeria. Limited studies have integrated time series (TS) and machine learning (ML) for estimating agricultural water footprints (AWF) and the agricultural water scarcity index (AWSI). This study presents a novel framework that combines advanced TS modelling techniques with six ML models. The ML (SVM, ANN, RF, KNN, LSTM, and GBM) models were applied singly and in ensembles to estimate key water-related indicators: blue (TWFb) and green (TWFg) water footprints, AWF, AWSI, and crop evapotranspiration (ETc). These were assessed for eight major crops in their respective highest-producing states in Nigeria. Among the TS approaches, the TBATS model (Trigonometric, Box-Cox transformation, ARMA errors, Trend and Seasonal components) demonstrated superior forecasting performance for most climatic variables across locations, with Nash–Sutcliffe Efficiency (NSE) values ranging from 0.523 to 0.923. LSTM architectures consistently outperformed other ML models in estimating AWF components, AWSI, and ETc, achieving NSE values between 0.994 and 0.998. However, the ensemble, Combined ML model (CML 11) integrating ANN, GBM, SVM, and LSTM emerged as the most robust and generalisable across variables and locations, since the NSE ranged between 0.691 and 0.994. SHAP analysis identified crop coefficient, maximum air temperature and radiation as the most influential variable across all target parameters. This study offers a valuable decision-support system for sustainable agricultural water management and climate-smart agriculture in Nigeria and comparable agro-ecological regions, globally.

Graphical abstract