Andreas Dombos | Searcharxiv

arXiv2026

D-SHIFT: Transferring High Spatial Information from GRACE Monthly TWSA Mascon to Daily Products Using Generative Adversarial Networks

The Gravity Recovery and Climate Experiment (GRACE) and GRACE Follow-On missions provide monthly terrestrial water storage anomaly (TWSA) estimates for monitoring large-scale water storage change. The monthly temporal resolution of official products limits the analysis of high-frequency hydrological events, while existing daily GRACE products often have reduced spatial resolution due to sparse groundtrack coverage and required smoothing and regularization. This study introduces D-SHIFT (Daily Spatial High-Resolution Inference via Feature Transformation), a deep learning-based framework for generating daily, high-resolution TWSA fields from daily spherical harmonic coefficient (SHC) solutions. The model is trained in the monthly domain by using low-resolution daily solutions and other auxiliary features as inputs, while targeting on monthly mascon products. The model is then applied to daily SHC inputs to generate products with similar spatial resolution of monthly products. Monthly validation against mascon products gives a global mean root mean square error of about 2.3cm, with good correlation and explained variance agreement. Daily analyses show that D-SHIFT produces spatially coherent day-to-day fields and improves basin-scale trend and seasonality estimates compared with low-resolution SHC. The basin-area double-difference analysis indicates that these gains are most relevant for spatially localized signals affected by smoothing and leakage. In Greenland, D-SHIFT better reproduces coastal mass-loss patterns and gives a basin-mean trend of -10.5cm/yr, close to the CSR Monthly value of -12.0cm/yr.

physics.geo-ph↗

arXiv2025

Solving Inverse Problems with FLAIR

Flow-based latent generative models such as Stable Diffusion 3 are able to generate images with remarkable quality, even enabling photorealistic text-to-image generation. Their impressive performance suggests that these models should also constitute powerful priors for inverse imaging problems, but that approach has not yet led to comparable fidelity. There are several key obstacles: (i) the data likelihood term is usually intractable; (ii) learned generative models cannot be directly conditioned on the distorted observations, leading to conflicting objectives between data likelihood and prior; and (iii) the reconstructions can deviate from the observed data. We present FLAIR, a novel, training-free variational framework that leverages flow-based generative models as prior for inverse problems. To that end, we introduce a variational objective for flow matching that is agnostic to the type of degradation, and combine it with deterministic trajectory adjustments to guide the prior towards regions which are more likely under the posterior. To enforce exact consistency with the observed data, we decouple the optimization of the data fidelity and regularization terms. Moreover, we introduce a time-dependent calibration scheme in which the strength of the regularization is modulated according to off-line accuracy estimates. Results on standard imaging benchmarks demonstrate that FLAIR consistently outperforms existing diffusion- and flow-based methods in terms of reconstruction quality and sample diversity. Our code is available at https://inverseflair.github.io/.

cs.CV↗