Interpolation of mountain weather forecasts by machine learning
Recent advancements in numerical simulation methods based on physical models have enhanced the accuracy of weather forecasts. However, the precision diminishes in complex terrains like mountainous regions due to the several kilometers square grid used in numerical simulations. While statistical machine learning has also significantly advanced, its direct application is difficult to utilize physics knowledge. This paper proposes a method that employs machine learning to “interpolate” future weather in mountainous regions using current observed data and forecast data from surrounding plains. Generally, weather prediction relies on numerical simulations, so this approach can be considered a hybrid method that indirectly merges numerical simulation and machine learning. The use of binary cross-entropy in precipitation prediction is also examined.
READ FULL TEXT