Abstract:Recent advances in numerical simulation methods based on physical models and their combination with machine learning have improved the accuracy of weather forecasts. However, the accuracy decreases in complex terrains such as mountainous regions because these methods usually use grids of several kilometers square and simple machine learning models. While deep learning has also made significant progress in recent years, its direct application is difficult to utilize the physical knowledge used in the simulation. This paper proposes a method that uses machine learning to interpolate future weather in mountainous regions using forecast data from surrounding plains and past observed data to improve weather forecasts in mountainous regions. We focus on mountainous regions in Japan and predict temperature and precipitation mainly using LightGBM as a machine learning model. Despite the use of a small dataset, through feature engineering and model tuning, our method partially achieves improvements in the RMSE with significantly less training time.

What problem does this paper attempt to address?

The paper aims to address the issue of low accuracy in weather forecasting in mountainous regions. Specifically, existing Numerical Weather Prediction (NWP) methods experience a decline in forecasting accuracy in complex terrains such as mountainous areas. This is because these methods typically use grid cells that are several kilometers in size and combine simple machine learning models, which do not perform well in complex terrains like mountains. To solve this problem, the authors propose a new method that uses machine learning to "interpolate" future weather in mountainous regions. The method is implemented through the following steps: 1. **Data Source**: The study focuses on predicting the temperature at the summit of Mount Fuji and the precipitation in the Hakone area in the Kanto region of Japan. It utilizes observational data provided by the Japan Meteorological Agency, including meteorological observation data and forecast data from surrounding plain areas. 2. **Model Selection**: LightGBM is used as the machine learning model for prediction. This model can utilize current observational data and forecast data from surrounding plain areas, as well as past historical observational data, to improve the accuracy of weather forecasts in mountainous regions. 3. **Feature Engineering**: Feature engineering was conducted, including the use of past observational data, future forecast data (only from surrounding areas), and time series features (such as differencing and moving averages). 4. **Loss Function**: In addition to Mean Squared Error (MSE), a linear combination of binary cross-entropy and MSE was also tried as the loss function to improve prediction bias in non-rainfall situations. Experimental results show that for temperature prediction at Mount Fuji, the proposed method significantly outperforms existing weather forecasting services; for precipitation prediction in the Hakone area, it also shows good performance at certain times. Additionally, the study explores the impact of feature extraction on training time and overfitting, as well as the effect of different loss function choices on prediction results. Overall, the method can improve the accuracy of weather forecasts in mountainous regions to a certain extent.

Interpolation of mountain weather forecasts by machine learning

Improving prediction of mountain snowfall in the southwestern United States using machine learning methods

A Spatial Interpolation Method for Meteorological Data Based on a Hybrid Kriging and Machine Learning Approach

Seasonal Forecast of Winter Precipitation over China Using Machine Learning Models

Forecast of Winter Precipitation Type Based on Machine Learning Method

Predicting weather forecast uncertainty with machine learning

A Learning Based Approach for Uncertainty Analysis in Numerical Weather Prediction Models

Improving forecasts of precipitation extremes over northern and central Italy using machine learning

Medium- and Long-Term Precipitation Forecasting Method Based on Data Augmentation and Machine Learning Algorithms

Innovative Machine Learning Applications in Weather Forecasting and Long-Term Climate Analysis

Short-term precipitation prediction using deep learning

MetMamba: Regional Weather Forecasting with Spatial-Temporal Mamba Model

Improved forecasting via physics-guided machine learning as exemplified using "21·7" extreme rainfall event in Henan

Deep learning-based method for weather forecasting: A case study in Itoshima

Applications of machine learning to predict seasonal precipitation for East Africa

Research on Global Climate Change Prediction based on Machine Learning Model

Assessing a machine learning-based downscaling framework for obtaining 1km daily precipitation from GPM data

Simulating the climate driven runoff in data-scarce mountains by machine learning and downscaling reanalysis data

Prediction of municipality-level winter wheat yield based on meteorological data using machine learning in Hokkaido, Japan

Short-range forecasts of global precipitation using deep learning-augmented numerical weather prediction

Advances and prospects of deep learning for medium-range extreme weather forecasting