HyLake v1.0 is a hybrid lake model that hard-couples a process-based one-dimensional surface energy balance and vertical thermal structure backbone with a Bayesian-Optimized Bidirectional LSTM (BO-BiLSTM) surrogate, which replaces the implicit Euler solver for lake surface temperature. The model was first developed and validated against FLake at five eddy-covariance sites in Lake Taihu (China), and tested for transferability at the ungauged Lake Chaohu, showing improved skill in surface temperature, latent heat flux, and sensible heat flux under both in-situ and ERA5-based forcing (He and Yang, 2025, Geosci. Model Dev., https://doi.org/10.5194/gmd-18-9257-2025).
Source code and data are openly available at https://doi.org/10.5281/zenodo.15289113.
HyLake v1.0 is the first version contributed to ISIMIP, used for ISIMIP4b runs. Ongoing development extends the surrogate-training base to ~900 globally distributed lakes for historical thermal-structure simulation, with future projections planned under CMIP6-based GCM forcing. This work will inform per-lake calibration for the Local Lakes sub-sector and provide the foundation for uncalibrated runs in the Global Lakes sub-sector.
This entry covers HyLake's contribution to the Local Lakes sub-sector with per-lake calibration on selected well-studied lakes.
