Scholarlink

<p>Groundwater sustainability is under growing threat from climate variability and intensive land use/land cover (LULC) changes, making precise recharge projections indispensable. This study focuses on Nakane Lake in Dhule, Maharashtra (India), recognizing it as an essential natural recharge zone and models its groundwater recharge dynamics for the period 2024–2030. By integrating remote-sensing data with hydro-climatic factors (i.e., rainfall, slope, soil type, LULC, temperature, evapotranspiration) and cutting-edge machine-learning (ML) techniques, we captured complex subsurface behavior with remarkable accuracy. Historical water demand patterns were clustered using fuzzy logic to account for real-world uncertainties, while an autoencoder revealed hidden nonlinear dependencies. Groundwater balance was assessed by incorporating various transfer rate constants (kb) to simulate distinct flow regimes. Hydroclimatic data, spatial parameters, and lake statistics were integrated with multivariate techniques to address uncertainty and identify recharge-prone areas. Among nine ML algorithms tested including Random Forest, XGBoost and Gradient Boosting, the autoencoder delivered the best performance, achieving an exceptional R² of 0.999. Results reveal a consistent upward trend in groundwater levels, peaking at 1.09&#xa0;mm in June, with an average rise of 0.57&#xa0;mm. These findings underscore the value of intelligent modeling for anticipating groundwater trends and offer actionable guidance for sustainable water-management strategies in climate-sensitive lake regions.</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

Lake recharge projections with autoencoders and hydro-climatic indicators

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • Sanchita Nawale,
  • Santosh Murlidhar Pingale,
  • Meenal Mategaonkar

摘要

Groundwater sustainability is under growing threat from climate variability and intensive land use/land cover (LULC) changes, making precise recharge projections indispensable. This study focuses on Nakane Lake in Dhule, Maharashtra (India), recognizing it as an essential natural recharge zone and models its groundwater recharge dynamics for the period 2024–2030. By integrating remote-sensing data with hydro-climatic factors (i.e., rainfall, slope, soil type, LULC, temperature, evapotranspiration) and cutting-edge machine-learning (ML) techniques, we captured complex subsurface behavior with remarkable accuracy. Historical water demand patterns were clustered using fuzzy logic to account for real-world uncertainties, while an autoencoder revealed hidden nonlinear dependencies. Groundwater balance was assessed by incorporating various transfer rate constants (kb) to simulate distinct flow regimes. Hydroclimatic data, spatial parameters, and lake statistics were integrated with multivariate techniques to address uncertainty and identify recharge-prone areas. Among nine ML algorithms tested including Random Forest, XGBoost and Gradient Boosting, the autoencoder delivered the best performance, achieving an exceptional R² of 0.999. Results reveal a consistent upward trend in groundwater levels, peaking at 1.09 mm in June, with an average rise of 0.57 mm. These findings underscore the value of intelligent modeling for anticipating groundwater trends and offer actionable guidance for sustainable water-management strategies in climate-sensitive lake regions.