Akademik Veri Yönetim Sistemi
Journal of Cleaner Production, cilt.523, 2025 (SCI-Expanded)
Water security in urban watersheds is increasingly becoming challenging as climate and land use changes intensify the hydrological cycle, shifting streamflow patterns and causing uncertainties in availability of water resources. In this research, using the advantages of both deep learning and physically based modeling, we apply the Soil and Water Assessment Tool (SWAT) coupled with Long Short Term Memory (LSTM) or LSTM-Attention to explore the relative attribution of changes in streamflow to land use change (LUC) and climate change in the Village Creek Watershed in Alabama, USA. SWAT was used to represent the physical processes of infiltration, sediment transport, and evapotranspiration, and the deep learning models were used to establish a connection between streamflow and statistically downscaled rainfall and temperature time series under present and future climates. The future precipitation levels for the period 2043–2055 are expected to rise by as much as 21 %, and minimum temperatures will decrease by −0.33 °C during winter seasons in the study area, leading to increased frequency of flooding and changed water resource availability. Land use change practice created an unexpected effect of streamflow reduction, by −6 % during colder months, in the study basin even though the effect was less compared to other processes. When changes in climate and land use are combined in simulation models, their combined effects, an increase of 12 %, drive more extreme changes in streamflow than from their individual impacts, +6 % for climate and −1 % for land use. The results showed the necessity for integrated management strategies because of complex non-linear responses to concurrent environmental changes. Comprehensive assessment across multiple gauging stations and performance metrics confirm the reliability of these models in representing both baseflow and high-flow conditions. In particular, the performance of Kling-Gupta efficiency (KGE) for LSTM-Attention remained above 0.90. Furthermore, the synergy between physically explicit and deep learning models uncovers emergent hydrological patterns, reinforcing the necessity of anticipatory planning for climate resilience.