In the lecture we will very briefly illustrate the general idea behind deep learning as a universal function approximator. To illustrate this universality, we will present several deep learning algorithms that we are developing to address very different operational challenges in oceanography and hydrology, ranging from storm surge forecasting, meteorological tsunami modeling and ocean gravity wave emulations to hydrological discharge predictions and sparse satellite data reconstructions. Each of presented deep algorithms turned out to be beneficial, but for different reasons. This leads us to the heuristic aspect of such predictions: do these deep algorithms function merely as numerically cheap, if opaque, replacements of classical numerical models or can they offer a complementary route to results that are beyond classical geophysical  models?

How to cite: Licer, M.: A deep role of deep learning in environmental research, forecasting and applications, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-736, https://doi.org/10.5194/ems2025-736, 2025.