Detection of climate change signals using precipitation and temperature time series by a hybrid deep learning framework

Climate change is one of the most extreme challenges of the twenty-first century. Precipitation (pr) and temperature variability are key indicators of climate change detection. Whereas hybrid deep learning (DL) models have been widely applied, their integration with region-specific datasets remains limited. This study uses a CNN-BiLSTM-TCN-attention model to identify the signals of climate change in Thailand based on the 1993–2024 station measurements and CMIP6-GCM simulations. Historical temperature analysis revealed statistically significant warming trends: coastal stations experienced T_asmin increases of up to 0.0673 °C/year and T_asmax of up to 0.0838 °C/year, urban areas 0.0401–0.0733 °C/year, and high-altitude sites 0.02–0.03 °C/year. Precipitation trends were spatially heterogeneous, with increases at Khlong Yai (4.764 mm/year) and Samut Prakan (4.303 mm/year), but declines at Aranyaprathet (− 0.688 mm/year) and Kampaeng Phet (− 0.402 mm/year). After bias correction, the framework achieved high performance (R² = 0.9807 for T_asmin, 0.9782 for T_asmax, 0.9034 for precipitation) and low error metrics (MSE = 0.0461). Future projections under SSP3-7.0 and SSP5-8.5 indicate widespread emergence of temperature signals, with median detection around 2019 and warming rates of + 0.5 °C/decade; by 2100, T_asmin and T_asmax are projected to rise robustly (+ 0.50 to + 0.72 °C/decade), while precipitation anomalies remain weakly negative (− 0.03 to − 0.08%/decade) with high variability but non-emergent trends. This study’s findings show the importance of applying Thailand-specific observational and model data to effectively detect and quantify local climate change signals, which can be used to inform local adaptation planning.