Observation-Guided Meteorological Field Downscaling at Station Scale: A Benchmark and a New Method

TL;DR

This work tackles downscaling meteorological fields from coarse gridded data to arbitrary station locations, addressing sub-grid biases by incorporating multi-scale observational information. It introduces a new benchmark and dataset for station-scale downscaling and proposes HyperDS, a data-conditioned hypernetwork that fuses high-resolution satellite observations with scattered station data to produce a continuous, coordinate-based meteorological field via an MLP decoder. HyperDS outperforms SR-based baselines across multiple surface variables, with especially large gains in wind speed and surface pressure, demonstrating the value of explicit sub-grid, observation-guided modeling. The approach enables continuous-resolution modeling and practical station-scale predictions, and the authors plan to release data and code to support broader research and applications.

Abstract

Downscaling (DS) of meteorological variables involves obtaining high-resolution states from low-resolution meteorological fields and is an important task in weather forecasting. Previous methods based on deep learning treat downscaling as a super-resolution task in computer vision and utilize high-resolution gridded meteorological fields as supervision to improve resolution at specific grid scales. However, this approach has struggled to align with the continuous distribution characteristics of meteorological fields, leading to an inherent systematic bias between the downscaled results and the actual observations at meteorological stations. In this paper, we extend meteorological downscaling to arbitrary scattered station scales, establish a brand new benchmark and dataset, and retrieve meteorological states at any given station location from a coarse-resolution meteorological field. Inspired by data assimilation techniques, we integrate observational data into the downscaling process, providing multi-scale observational priors. Building on this foundation, we propose a new downscaling model based on hypernetwork architecture, namely HyperDS, which efficiently integrates different observational information into the model training, achieving continuous scale modeling of the meteorological field. Through extensive experiments, our proposed method outperforms other specially designed baseline models on multiple surface variables. Notably, the mean squared error (MSE) for wind speed and surface pressure improved by 67% and 19.5% compared to other methods. We will release the dataset and code subsequently.

Figures (6)

• Figure 1: The difference between the previous SR (super-resolution)-based downscaling pipeline with fixed grid-level scale sun2024deep (a), and the proposed observation-guided downscaling pipeline with arbitrary scatter station-level scale.
• Figure 2: The study area and scatter stations used in our paper. The red dots represent the training stations, the blue dots represent the validation stations and the yellow dots represent the test stations.
• Figure 3: The proposed HyperDS architecture. It mainly consists of three parts: a dual-branch feature encoder is used to extract semantic features from the input low-resolution meteorological field and high-resolution remote sensing images respectively; subsequently, the implicit retrieval network utilizes a cross-attention mechanism to implicitly fuse different feature information and align the remote sensing image features with meteorological field variables; and finally, the FC (fully connected)-based weight generator predicts the weight vector for the target network. The MLP (Multi-Layer Perceptron) decoder, the target network, learns the mapping from the sampled subgrid coordinates to the corresponding location state values. It is supervised at both the observation station scale and the high-resolution grid scale, allowing for the continuous modeling of the meteorological fields.
• Figure 4: Two variants of MLP decoders based on implicit neural representations with subgrid sampling.
• Figure 5: Visualization comparison of downscaling to station-scale using different methods, where the color of each station represents the magnitude of the normalized mean square error at that station, with lighter colors indicating larger errors, i.e. the darker the color of the site, the better the performance of the downscaling. The base map of each visual image is also the result of downscaling at the grid scale with $0.25^\circ$ spatial resolution of each model.