Physics-informed LSTMs improve streamflow prediction when storage-discharge dynamics are temporally resolved
The results show that the value of physics-informed learning depends on whether the constrained physical process is resolved at the modelling timestep, and suggest that temporal resolution should be considered explicitly when designing and evaluating physics-informed hydrological models.
A. Ardid, D. Dempsey, C. Cattoën et al.
· 0 citations