Scholarlink

<p>River water quality in monsoon-influenced, industrialized basins reflect interactions between seasonal hydrology, land-use change, and anthropogenic pressures. However, forecasting approaches rarely integrate temporal modeling with driver attribution. This study examined monthly water quality dynamics at ten sites along a tropical river system between 2014 and 2024 using principal component-weighted indices. Classical seasonal autoregressive integrated moving average (SARIMA) models were benchmarked against hybrid neural architectures including ARIMA combined with temporal convolutional networks, ARIMA integrated with long short-term memory layers, and bidirectional gated recurrent units to identify site-specific forecast skill. Pooled regressions were used to test associations between non-vegetated land cover and water quality, while distributed-lag frameworks estimated rainfall and evapotranspiration effects across zero to six-month lags. Hybrid models achieved superior performance at reaches exhibiting abrupt, event-driven variability, whereas SARIMA matched or exceeded alternatives where strong seasonal persistence dominated. Non-vegetated cover showed a robust negative association with water quality, declining approximately 0.5 index units per 3 percentage point increase, respectively. Rainfall and evapotranspiration exhibited small, site-dependent cumulative effects, indicating localized dilution versus mobilization regimes. These findings support reach-specific management prioritizing riparian restoration where land-use pressures are rising and flow-conditioned interventions where storm pulses drive quality swings, offering a practical framework for operational forecasting and targeted restoration in complex river systems.</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

Hybrid SARIMA-Neural Network Frameworks for River Water Quality Forecasting: Integrating Land-Use and Hydrometeorological Drivers Across Spatial Scales

  • 期刊论文

  • 发表日期：

    2026年5月25日

查看全文
  • Sanghamitra Sanyal,
  • Sanchari Sarkar,
  • Moitreyee Chakrabarty

摘要

River water quality in monsoon-influenced, industrialized basins reflect interactions between seasonal hydrology, land-use change, and anthropogenic pressures. However, forecasting approaches rarely integrate temporal modeling with driver attribution. This study examined monthly water quality dynamics at ten sites along a tropical river system between 2014 and 2024 using principal component-weighted indices. Classical seasonal autoregressive integrated moving average (SARIMA) models were benchmarked against hybrid neural architectures including ARIMA combined with temporal convolutional networks, ARIMA integrated with long short-term memory layers, and bidirectional gated recurrent units to identify site-specific forecast skill. Pooled regressions were used to test associations between non-vegetated land cover and water quality, while distributed-lag frameworks estimated rainfall and evapotranspiration effects across zero to six-month lags. Hybrid models achieved superior performance at reaches exhibiting abrupt, event-driven variability, whereas SARIMA matched or exceeded alternatives where strong seasonal persistence dominated. Non-vegetated cover showed a robust negative association with water quality, declining approximately 0.5 index units per 3 percentage point increase, respectively. Rainfall and evapotranspiration exhibited small, site-dependent cumulative effects, indicating localized dilution versus mobilization regimes. These findings support reach-specific management prioritizing riparian restoration where land-use pressures are rising and flow-conditioned interventions where storm pulses drive quality swings, offering a practical framework for operational forecasting and targeted restoration in complex river systems.