A Decade of Deep Learning in Remote Sensing: A Systematic Review of Reviews

Since 2014, Deep Learning (DL) has emerged as a powerful technology for analyzing Remote Sensing (RS) data, enabling advancements in land cover classification, object detection, change detection, and other RS tasks.With the rapid growth of research in this area, nearly ten thousand articles have been published with hundreds of review articles.This review paper aims to map the evolution of DL applications in the RS research domain, chronicling the major achievements from their widespread adoption from 2016 to March 2026.Instead of targeting a specific topic, we performed a systematic review of peer-reviewed ''review'' articles (namely, ''review of reviews'').Following the scoping review framework, we identified and summarized 448 review articles across four major scientific databases: Web of Science, Scopus, IEEE Xplore, and arXiv.The included reviews cover convolutional neural networks, vision transformers, generative adversarial networks and hybrid architectures, as well as transfer learning, selfsupervised learning and Foundation Models (FMs).These are applied to a range of applications, including agriculture, urban analysis, disaster monitoring, climate science, marine environments and hyperspectral imaging.A bibliometric analysis reveals that the annual growth rate for review publications is 74.9%, with agriculture and environmental monitoring being the most prevalent application domains.The review shows that, while the field has progressed from convolutional network-dominated pipelines towards transformerbased and FM architectures, persistent challenges in data scarcity, domain generalization, benchmark fragmentation and computational efficiency continue to constrain operational deployment.This article provides a comprehensive overview of achievements in DL applications for the RS domain, drawing on over a decade of research to offer evidence-based recommendations for future research.