Regional Weather Variable Predictions by Machine Learning with Near-Surface Observational and Atmospheric Numerical Data

TL;DR

This work tackles the challenge of producing accurate, fine-grained regional weather forecasts by fusing high-frequency near-surface observations with coarse-grained atmospheric numerics. The authors introduce MiMa, an encoder–decoder Transformer framework with two encoders (micro and macro) and a decoder, yielding modelets that predict a single weather parameter at a given location; they also extend this to Re-MiMa for ungauged locations via regional transfer learning using elevations. Across Kentucky Mesonet stations and four weather parameters, MiMa outperforms baselines, and Re-MiMa achieves region-wide accuracy at ungauged sites, demonstrating a practical pathway to high-resolution regional nowcasting. The approach has strong implications for real-time decision-making in sectors like transportation and emergency management, and code/data availability supports replication and regional adaptation of the method.

Abstract

Accurate and timely regional weather prediction is vital for sectors dependent on weather-related decisions. Traditional prediction methods, based on atmospheric equations, often struggle with coarse temporal resolutions and inaccuracies. This paper presents a novel machine learning (ML) model, called MiMa (short for Micro-Macro), that integrates both near-surface observational data from Kentucky Mesonet stations (collected every five minutes, known as Micro data) and hourly atmospheric numerical outputs (termed as Macro data) for fine-resolution weather forecasting. The MiMa model employs an encoder-decoder transformer structure, with two encoders for processing multivariate data from both datasets and a decoder for forecasting weather variables over short time horizons. Each instance of the MiMa model, called a modelet, predicts the values of a specific weather parameter at an individual Mesonet station. The approach is extended with Re-MiMa modelets, which are designed to predict weather variables at ungauged locations by training on multivariate data from a few representative stations in a region, tagged with their elevations. Re-MiMa (short for Regional-MiMa) can provide highly accurate predictions across an entire region, even in areas without observational stations. Experimental results show that MiMa significantly outperforms current models, with Re-MiMa offering precise short-term forecasts for ungauged locations, marking a significant advancement in weather forecasting accuracy and applicability.

Figures (10)

• Figure 1: Kentucky Mesonet weather observational stations denoted by yellow circles, with those stations chosen for MiMa model evaluation and pointed by red line segments tagged with their latitudes, longitudes, and elevations.
• Figure 2: Overview of the MiMa (short for Micro-Macro) model inputted with data from both an individual station and WRF-HRRR modeling computation to yield the weather variable predictions.
• Figure 3: Structure of the Micro model, with the hidden state $\textbf{H}_{t}$ obtained by inputting $\textbf{X}_{\text{micro}}$ to an encoder and the output $\textbf{O}_{t+1}$ obtained by inputting $\textbf{H}_{t}$ plus $\textbf{Y}_{0}$ to a decoder. Output $\textbf{O}_{t+1}$ is then passed to a fully connected layer which generates the predicted parameter value $\textbf{Y}_{t+1}$ via a fully connected network.
• Figure 4: Structure of MiMa model, with its Micro Encoder and its Decoder identical to those depicted in Fig. \ref{['fig:MiED']} and with the hidden states of the Micro and the Macro Encoders concatenated as the Decoder's input.
• Figure 6: Ensemble temperature prediction plot of MiMa modelet for Station FARM.