A Self-Supervised Framework for Refined Reconstruction of Geophysical Fields via Domain Adaptation

Reconstructing fine-grained, detailed spatial structures from time-evolving coarse-scale geophysical fields has been a long-standing challenge. Current deep learning approaches addressing this issue generally require massive fine-scale fields as supervision, which is often unavailable due to limitations in existing observational systems and the scarcity of widespread high-precision sensors. Here, we present AdaptDeep, a self-supervised framework for refined reconstruction of geophysical fields via domain adaptation from the coarse-scale source domain to the fine-scale target domain. This method incorporates two pretext tasks, cropped field reconstruction and temporal augmentation-assisted contrastive learning, to leverage spatial and temporal correlations in the target domain. A global propagation structure is proposed in the feature extraction network to leverage bidirectional information for enhanced long-range dependencies and robustness against estimation errors. In experiments, AdaptDeep correctly identifies local, fine structures and significantly recovers 81.2% detailed information in sea surface temperature fields. In recent years, deep learning has greatly contributed to the field of meteorology and we perform an investigation into the use of convolutional neural networks, to tackle the issue of reconstructing precise spatial structures from coarse-scale geophysical fields that evolves over time. Right now, techniques using deep learning need highly detailed "ground truth" data to guide them, but this kind of data is often hard to get. This is particularly true in less economically developed areas where there's a lack of high-quality sensors. To address these concerns, we've developed AdaptDeep, which teaches itself how to reconstruct geophysical fields at a finer scale. It does this by mimicking the process of turning coarse-scale source data into finer-scale target data. The backbone of AdaptDeep is a neural network, which uses information from all directions and a global context to increase its understanding of long-range dependencies and reduce estimation errors. We put AdaptDeep to the test, and found that it could accurately pick out local, detailed structures, recovering an impressive 81.2% of detailed information in sea surface temperature fields. This shows that AdaptDeep could be a powerful new tool in helping us to understand our environment in more detail. A self-supervised framework is presented for refined reconstruction of geophysical fields without high-resolution fields as ground truth Two pretext tasks are designed and incorporated into the model to leverage spatial and temporal correlations in the target domain The feature extraction network employs a global propagation structure to exploit global information