A Physics-Aware Neural Network Approach for Flow Data Reconstruction From Satellite Observations

Abstract

An accurate assessment of physical transport requires high-resolution and high-quality velocity information. In satellite-based wind retrievals, the accuracy is impaired due to noise while the maximal observable resolution is bounded by the sensors. The reconstruction of a continuous velocity field is important to assess transport characteristics and it is very challenging. A major difficulty is ambiguity, since the lack of visible clouds results in missing information and multiple velocity fields will explain the same sparse observations. It is, therefore, necessary to regularize the reconstruction, which would typically be done by hand-crafting priors on the smoothness of the signal or on the divergence of the resulting flow. However, the regularizers can smooth the solution excessively and will not guarantee that possible solutions are truly physically realizable. In this paper, we demonstrate that data recovery can be learned by a neural network from numerical simulations of physically realizable fluid flows, which can be seen as a data-driven regularization. We show that the learning-based reconstruction is especially powerful in handling large areas of missing or occluded data, outperforming traditional models for data recovery. We quantitatively evaluate our method on numerically-simulated flows, and additionally apply it to a Guadalupe Island case study—a real-world flow data set retrieved from satellite imagery of stratocumulus clouds.

Adapting a deep convolutional RNN model with imbalanced regression loss for improved spatio-temporal forecasting of extreme wind speed events in the short to medium range

“…It may also be worthwhile to consider other atmospheric variables such as the convective available potential energy (CAPE) and deep-layer wind shear (DLS) due to their strong correlation with severe convective storm activity such as the occurrence of thunderstorms and supercells (see, for example, Rädler et al, 2015;Tsonevsky et al, 2018;Chavas and Dawson II, 2021). Another possible extension would be to implement categorical scores directly in the loss function (see, for example, Lagerquist and Ebert-Uphoff, 2022) or even combine the ConvLSTM with a socalled physics-aware loss function (see, for example, Schweri et al, 2021;Cuomo et al, 2022).…”

Neural Flow Map Reconstruction

“…Closely related to superresolution are works that aim to reconstruct the vector field from a sparse set of scattered data. Prior works [SFT∗22, EMY∗20, YWS∗21] achieve flow field reconstruction by directly optimizing for the velocity vectors. Han et al [HTZ∗19] and Gu et al [GHCW21] reconstruct vector fields from a set of representative streamlines.…”

Unsteady 2D Flow Reconstruction around Arbitrary Shapes via Conformal Mapping aided Deep Neural Networks

“…Various statistical tools have been developed for FR [1,2], however substantial strides were made recently with the application of deep learning (DL) to the field [3,4,5]. Applications outside typical wind tunnel-like scenarios, such as reconstruction of atmospheric flows based on satellite imagery [6], have also been developed.…”