OpenNovelty.org

GeoFAR: Geography-Informed Frequency-Aware Super-Resolution for Climate Data

ICLR 2026 Conference SubmissionAnonymous Authors
OpenReview Score: 5.5Download Report PDF
climate downscalingimage super-resolutionimplicit neural representationearth observationenvironmental science
Abstract:

Super-resolving climate data is crucial for fine-grained decision-making in various domains, ranging from agriculture to environmental conservation. However, existing super-resolution approaches struggle to generate the high-frequency spatial information present in climate data, especially over regions showing complex terrain variability. A key obstacle lies in a frequency bias existing in both deep neural networks (DNNs) and climate data: DNNs exhibit such bias by overfitting to low-frequency information, which is further exacerbated by the prevalence of low-frequency components in climate data (e.g., plains, oceans). As a consequence, geography-dependent high-frequency details are hard to reconstruct from coarse climate inputs with DNNs. To improve the fidelity of climate super-resolution (SR), we introduce GeoFAR: by explicitly encoding climatic patterns at different frequencies, while learning implicit geographical neural representations (i.e., related to location and elevation), our approach provides frequency-aware and geography-informed representations for climate SR, thereby reconstructing fine-grained climate information at high resolution. Experiments show that GeoFAR is a model-agnostic approach that can mitigate high-frequency prediction errors in both deterministic and generative SR models, demonstrating state-of-the-art performance across various spatial resolutions, atmospheric variables, and downscaling ratios. Datasets and code will be released.

Disclaimer
This report is AI-GENERATED using Large Language Models and WisPaper (A scholar search engine). It analyzes academic papers' tasks and contributions against retrieved prior work. While this system identifies POTENTIAL overlaps and novel directions, ITS COVERAGE IS NOT EXHAUSTIVE AND JUDGMENTS ARE APPROXIMATE. These results are intended to assist human reviewers and SHOULD NOT be relied upon as a definitive verdict on novelty.
NOTE that some papers exist in multiple, slightly different versions (e.g., with different titles or URLs). The system may retrieve several versions of the same underlying work. The current automated pipeline does not reliably align or distinguish these cases, so human reviewers will need to disambiguate them manually.
If you have any questions, please contact: mingzhang23@m.fudan.edu.cn

Overview

Overall Novelty Assessment

The paper introduces GeoFAR, a climate super-resolution method that combines frequency-aware representation learning with geography-informed neural representations to reconstruct high-frequency spatial details in climate data. It resides in the Specialized Deep Learning Techniques leaf, which contains four papers exploring novel training paradigms beyond standard GAN or residual architectures. This leaf represents a relatively sparse research direction within the broader Deep Learning-Based Super-Resolution Methods branch, suggesting the paper targets an emerging niche rather than a crowded subfield.

The taxonomy reveals that most deep learning climate super-resolution work concentrates in GAN-based and residual/convolutional architectures (four papers each), while the Specialized Techniques leaf explores alternative paradigms such as self-supervised learning and pixel recursive models. GeoFAR's emphasis on explicit frequency encoding and implicit geographical representations distinguishes it from neighboring leaves, which focus on adversarial training or standard convolutional designs. The taxonomy's scope and exclude notes clarify that hybrid statistical-physical methods belong elsewhere, positioning GeoFAR firmly within pure deep learning innovation.

Among twenty-three candidates examined via semantic search and citation expansion, none clearly refute any of the three core contributions. The GeoFAR approach examined ten candidates with zero refutable overlaps; frequency-aware representation learning with FCK examined three candidates, also with zero refutations; and Geo-INR for geography-informed learning examined ten candidates, again finding no clear prior work. These statistics suggest that within the limited search scope, the combination of frequency-aware and geography-informed representations appears relatively unexplored, though the analysis does not claim exhaustive coverage of all possible prior art.

Based on the limited literature search, GeoFAR appears to occupy a distinct position by explicitly addressing frequency bias and geographical variability in climate super-resolution. The absence of refutable candidates among twenty-three examined suggests novelty within the sampled literature, though a broader search might reveal additional related work. The taxonomy context indicates this work extends the Specialized Deep Learning Techniques direction into domain-specific feature engineering, complementing rather than duplicating existing GAN or residual approaches.

Taxonomy

Core-task Taxonomy Papers
50
3
Claimed Contributions
23
Contribution Candidate Papers Compared
0
Refutable Paper

Research Landscape Overview

Core task: super-resolution for climate data. The field addresses the challenge of enhancing coarse-resolution climate model outputs or observations to finer spatial scales, a critical step for regional impact assessments and decision-making. The taxonomy reveals several complementary branches: Deep Learning-Based Super-Resolution Methods leverage neural architectures—including specialized techniques such as generative models and self-supervised approaches—to learn complex mappings from low- to high-resolution climate fields. Statistical Downscaling Methods encompass traditional regression-based and stochastic techniques that have long been used to relate large-scale predictors to local climate variables. Integrated Frameworks and Validation focus on benchmarking, intercomparison projects, and hybrid pipelines that combine multiple downscaling strategies. High-Resolution Climate Datasets provide the reference products and reanalysis outputs essential for training and evaluation. Domain-Specific Applications tailor downscaling to particular variables (precipitation, temperature, wind) or sectors (hydrology, agriculture, infrastructure), reflecting the diversity of end-user needs. Recent work highlights a tension between physically interpretable statistical methods and data-driven deep learning approaches that can capture fine-scale spatial patterns but may lack explicit physical constraints. For instance, Climate Downscaling[1] and Downscaling Appraisal[5] illustrate traditional statistical frameworks, while Adversarial Climate[2] and Stochastic GAN[33] demonstrate how generative models can produce realistic high-resolution fields. Within the Specialized Deep Learning Techniques cluster, GeoFAR[0] sits alongside Self-supervised Downscaling[42] and Pixel Recursive[45], emphasizing novel training paradigms and architectural innovations. Compared to these neighbors, GeoFAR[0] appears to focus on integrating domain-aware priors or feature representations, distinguishing it from purely self-supervised or autoregressive strategies. Transferability Evaluation[46] further underscores ongoing questions about how well these models generalize across regions and climate regimes, a key concern as the community seeks robust, scalable solutions for diverse applications.

Claimed Contributions

GeoFAR approach for climate super-resolution

The authors propose GeoFAR, a model-agnostic approach that combines frequency-aware representations (encoding both low and high-frequency climatic patterns) with geography-informed implicit neural representations (jointly encoding location and elevation) to improve the fidelity of climate super-resolution.

10 retrieved papers
Frequency-aware representation learning with FCK

The authors introduce a Frequency-aware Convolution Kernel (FCK) that uses DCT bases to explicitly encode different frequency components into separate channels, mitigating the low-frequency bias in both neural networks and climate data by enhancing high-frequency information.

3 retrieved papers
Geo-INR for geography-informed learning

The authors develop Geo-INR, an implicit neural representation that goes beyond location-only encoding by jointly encoding both spherical location (latitude and longitude) and terrain information (elevation and slope) on a 3D geographic manifold to inform climate super-resolution.

10 retrieved papers

Core Task Comparisons

Comparisons with papers in the same taxonomy category

Contribution Analysis

Detailed comparisons for each claimed contribution

Contribution

GeoFAR approach for climate super-resolution

The authors propose GeoFAR, a model-agnostic approach that combines frequency-aware representations (encoding both low and high-frequency climatic patterns) with geography-informed implicit neural representations (jointly encoding location and elevation) to improve the fidelity of climate super-resolution.

Contribution

Frequency-aware representation learning with FCK

The authors introduce a Frequency-aware Convolution Kernel (FCK) that uses DCT bases to explicitly encode different frequency components into separate channels, mitigating the low-frequency bias in both neural networks and climate data by enhancing high-frequency information.

Contribution

Geo-INR for geography-informed learning

The authors develop Geo-INR, an implicit neural representation that goes beyond location-only encoding by jointly encoding both spherical location (latitude and longitude) and terrain information (elevation and slope) on a 3D geographic manifold to inform climate super-resolution.

Table of Contents