Advances in land surface forecasting: a comparison of LSTM, gradient boosting, and feed-forward neural networks as prognostic state emulators in a case study with ecLand

The most useful weather prediction for the public is near the surface. The processes that are most relevant for near-surface weather prediction are also those that are most interactive and exhibit positive feedback or have key roles in energy partitioning. Land surface models (LSMs) consider these processes together with surface heterogeneity and, when coupled with an atmospheric model, provide boundary and initial conditions. They forecast water, carbon, and energy fluxes, which are an integral component of coupled atmospheric models. This numerical parametrization of atmospheric boundaries is computationally expensive, and statistical surrogate models are increasingly used to accelerate experimental research. We evaluated the efficiency of three surrogate models in simulating land surface processes for speeding up experimental research. Specifically, we compared the performance of a long short-term memory (LSTM) encoder-decoder network, extreme gradient boosting, and a feed-forward neural network within a physics-informed multi-objective framework. This framework emulates key prognostic states of the Integrated Forecasting System (IFS) land surface scheme of the European Centre for Medium-Range Weather Forecasts (ECMWF), ecLand, across continental and global scales. Our findings indicate that, while all models on average demonstrate high accuracy over the forecast period, the LSTM network excels in continental long-range predictions when carefully tuned, extreme gradient boosting (XGB) scores consistently high across tasks, and the multilayer perceptron (MLP) provides an excellent implementation time-accuracy trade-off. While their reliability is context-dependent, the runtime reductions achieved by the emulators in comparison to the full numerical models are significant, offering a faster alternative for conducting experiments on land surfaces.