Integrating ICESat-2 laser altimeter observations and hydrological modeling for enhanced prediction of climate-driven lake level change

Abstract

Predicting future lake levels under climate change is critical for advancing our understanding of hydrological processes in a changing environment. However, continuous and long-term prediction of lake levels is challenging due to discrepancies in multi-source data and the lack of integration of hydrological models and climate scenarios. Physical and statistical models have been used for lake levels prediction, however, physical models are difficult to calibrate and statistical models often fail to account for the effects of climate changes. In this study, a lake level prediction model was proposed by assimilating the hydrophysical model based on water balance and the ICESat-2 observations using Variational Bayesian Monte Carlo. The model can predict monthly lake levels combining CMIP6-SWAT climate-driven projections and ICESat-2 observations. Short-term validation over 24 months showed the R2 was 0.91 and the RMSE was 5 cm between the proposed model and the ICESat-2 observation. The accuracy is superior to both the hydrophysical model based on water balance and the Prophet time series model. Long-term validation from 1978 to 2021 showed the proposed model has the potential to enhance prediction accuracy within 2 to 3 years compared with the hydrophysical model based on water balance and it is suitable for predicting long-term lake level trends. Notably, it successfully predicted the significant turning point around 2005 where lake levels shift from decline to increase based on past data. The water levels of Lake Qinghai were predicted to rise at a rate of 3.7 cm per year by 2050 under the SSP2-4.5 scenario. The proposed assimilation model combines the strengths of hydrological modeling based on water balance (incorporating the effects of climate change) and the latest ICESat-2 lake level observations (incorporating the effects of recent historical lake levels), improving the accuracy of short-term lake levels and long-term lake level trends prediction. Moreover, as the model is based on satellite remote sensing observations, it has the potential to be applied to any lake globally.