Space-scale exploration of the poor reliability of deep learning models: the case of the remote sensing of rooftop photovoltaic systems

• DOI: 10.1017/eds.2025.13
• Publication Date: 2025-04-10T06:59:38Z
• Authors: Gabriel Kasmi; Laurent Dubus; Yves-Marie Saint-...; Philippe Blanc

Abstract Photovoltaic (PV) energy grows rapidly and is crucial for the decarbonization of electric systems. However, centralized registries recording technical characteristics rooftop PV systems are often missing, making it difficult to monitor this growth accurately. The lack monitoring could threaten integration into grid. To avoid situation, remote sensing using deep learning has emerged as a promising solution. existing techniques not reliable enough be used by public authorities or transmission system operators (TSOs) construct up-to-date statistics on fleet. reliability comes from models being sensitive distribution shifts. This work comprehensively evaluates shifts’ effects classification accuracy trained detect panels overhead imagery. We benchmark isolate sources shifts introduce novel methodology that leverages explainable artificial intelligence (XAI) decomposition input image model’s decision regarding scales understand how affect models. Finally, based our analysis, we data augmentation technique designed improve robustness classifiers under varying acquisition conditions. Our proposed approach outperforms competing methods can close gap with more demanding unsupervised domain adaptation methods. discuss practical recommendations mapping imagery