Image processing and AI techniques for climate change detection using remote sensing: a comprehensive review

Climate change is accelerating spatially complex transformations in land, water, coasts, cryosphere, and ecosystems, creating a critical need for reliable, scalable, and timely monitoring based on Earth observation imagery. Conventional approaches that rely on sparse in-situ measurements, manual image interpretation, and simple spectral indices or thresholding often fail to capture subtle, heterogeneous, and multiscale changes, and they do not scale to today's multi-sensor, multi-temporal satellite archives. This review synthesizes image processing and AI techniques applied to optical, SAR, thermal, and hyperspectral remote sensing for climate change detection, covering classical change detection methods, machine learning classifiers, deep learning architectures (including Siamese and segmentation networks), spatio-temporal models for satellite image time series, and multi-sensor fusion, across application domains such as land use/land cover (LULC) and deforestation, hydrology and flooding, coastal and mangrove dynamics, cryospheric change, urban heat, ecosystems, and natural hazards. In addition, we analyze how these methods are evaluated using common performance metrics–Overall Accuracy (OA), precision, recall, F1-score, Intersection over Union (IoU), Kappa coefficient, and error measures such as RMSE–and discuss key challenges related to data quality and annotation, domain shift and generalization, computational and operational constraints, interpretability, and integration with climate and impact models. The distinctive contribution of this review is a unified method-application taxonomy that explicitly links algorithm families to specific climate monitoring tasks, a systematic comparison of reported performance metrics that clarifies trade-offs between techniques under different data and class-imbalance conditions, and a practical decision framework to guide researchers and practitioners in selecting appropriate image processing and AI approaches for given sensors, regions, and operational requirements, while outlining promising future directions such as foundation models, standardized benchmarks, and interoperable climate decision-support systems. Across the reviewed literature, deep learning approaches consistently demonstrate higher accuracy (e.g., improved IoU and F1-scores) in complex and heterogeneous environments, while classical methods remain effective for large-scale and data-scarce applications. However, significant gaps persist in model generalization across regions, availability of labeled datasets, and integration of multi-sensor time-series data.