A two-stage deep model for geophysical data reconstruction

Zupančič Muc, Matjaž (author), Kristan, Matej (mentor), Ličer, Matjaž (comentor)

Sea surface temperature (SST) is critical for weather forecasting and climate modeling, however remotely sensed SST data often suffer from incomplete coverage due to cloud obstruction and limited satellite swath width. While deep learning approaches have shown promise in reconstructing missing data, existing methods struggle to accurately recover fine-grained details, which, however are crucial for many down-stream geophysical processing and prediction problems. We propose CRITER (Coarse Reconstruction with Iterative Refinement network), a novel two-stage approach comprising: (i) a transformer-based Coarse Reconstruction Module (CRM) that estimates low-frequency SST components by leveraging global spatio-temporal correlations in available observations, and (ii) an Iterative Refinement Module (IRM) for recovering high-frequency details absent from the initial CRM estimate. Extensive experiments across Mediterranean, Adriatic, and Atlantic sea datasets reveal CRITER's superior performance over the state-of-the-art DINCAE2 model. CRITER achieves substantial reconstruction error reductions in both missing and observed regions: $20\%$ and $89\%$ for the Mediterranean, $44\%$ and $80\%$ for the Adriatic, and $1\%$ and $88\%$ for the Atlantic dataset, respectively. These results mark a significant advancement in the field of sparse geophysical data reconstruction.

• Language: English
• Keywords: reconstruction of geophysical data, image inpainting, computer vision
• Work type: Master's thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2024
• PID: 20.500.12556/RUL-160890
• Publication date in RUL: 05.09.2024

Secondary language

• Language: Slovenian
• Title: Dvostopenjski globoki model za rekonstrukcijo geofizikalnih podatkov

Abstract

Površinska temperatura morja (angl. sea surface temperature, SST) je ključnega pomena za napovedovanje vremena in podnebno modeliranje. Vendar so meritve SST, pridobljene z daljinskim zaznavanjem, pogosto pomanjkljive zaradi omejene širine satelitskih pasov in prisotnosti oblakov. Čeprav so pristopi globokega učenja pokazali obetavne rezultate pri rekonstrukciji manjkajočih vrednosti, obstoječe metode visoko frekvenčnih podrobnosti ne uspejo rekonstruirati, le te pa so ključne za kasnejše geofizikalne analize in napovedne modele. V nalogi predlagamo metodo CRITER (Coarse Reconstruction with ITerative Refinement), nov dvostopenjski pristop, ki vključuje (i) modul za grobo rekonstrukcijo (CRM) na osnovi transformerske nevronske mreže, (ii) modul za iterativno izpopolnjevanje (IRM), namejen obnavljanju visokofrekvenčnih podrobnosti, ki manjkajo v prvotni rekonstrukciji CRM. Eksperimenti na podatkovnih bazah Sredozemskega, Jadranskega in Atlantskega morja kažejo, da CRITER znatno presega najuspešnejšo metodo DINCAE2. CRITER doseže zmanjšanje rekonstrukcijske napake na izbrisanih in na neizbrisanih regijah za $20\%$ in $89\%$ na Sredozemlju, $44\%$ in $80\%$ na Jadranu ter $1\%$ in $88\%$ na Atlantiku, kar predstavlja pomemben napredek na področju rekonstrukcije redkih geofizikalnih podatkov.

• Keywords: rekonstrukcija geofizikalnih podatkov, vrisovanje manjkajočih vrednosti v slikah, računalniški vid