A data-driven approach for assessing future resilience of Trasimeno Lake in central Italy under CMIP6 climate scenarios

Study focus After depicting trends currently detectable by historical water level records, an LSTM (Long Short-Term Memory) modeling framework is exploited to produce forecasts up to 2100 of water level fluctuations on the basis of input rainfall and temperature data. Future resilience of the Trasimeno Lake under four representative CMIP6 (Coupled Model Intercomparison Project - Phase 6) scenarios is discussed. Moreover, the impacts on water level forecasts of uncertainties associated with future projections of precipitation and temperature data are quantified. New hydrological insights for the region The lower performance thresholds achieved for (Root Mean Square Error), , (Pearson correlation coefficient), (Mean Absolute Percentage Error), and (Mean Percentage Error) are 0.30 m, 0.04 m, 0.72, 9.76%, and 1.70%, respectively. The optimized model is subsequently applied to assess future lake-level dynamics under the SSP1–2.6, SSP2–4.5, SSP3–7.0, and SSP5–8.5 scenarios. Projections indicate that water level decline rates may increase by approximately a factor of 2 (SSP2–4.5), 4 (SSP3–7.0), and 7 (SSP5–8.5) compared with the historical trend of −0.008 m/year observed over 1961–2024. Across these scenarios, navigability is expected to become an increasingly recurrent concern, while hydrological conditions comparable to those recorded during the severe 1958 drought are projected to occur by the end of the century. Forecast results are presented together with a quantitative evaluation of the associated uncertainties.