Artificial Intelligence to Advance Earth Observation: A review of models, recent trends, and pathways forward

Abstract

Earth observation (EO) is increasingly used for map-ping and monitoring processes occurring at thesurface of Earth. Data acquired by satellitesnowadays allow us to have a global view,consistent in time, of the state of our for-ests, oceans, and growing urban areas.However, such a wealth of data haslittle value without appropriateprocessing chains able to convertthe pixel values to informationuseful for decision makers.Recently, machine learning(ML) has seen fast advances—especially thanks to the rise ofdeep learning (DL) method-ologies—and is increasinglydeployed in EO image process-ing systems. The ever-growingmodels from computer vision(CV) and natural language pro-cessing (NLP) have inspired de-velopments in remote sensing, andnew approaches are constantly pro-posed in the field. However, despitetheir impressive results, the ever-growingmass of approaches and solutions makes itcomplicated to have a holistic overview and toknow the most promising approaches from the field.In this article, we aim to fill this knowledge gap andpropose to review the thriving ecosystem focusing ondeveloping artificial intelligence (AI) models for EO,its recent trends, and sketch potential pathways for fu-ture advances