A multi agro-climatic reference dataset for irrigated and non-irrigated fields: France and Kazakhstan

Accurate spatial information on irrigated area extents is essential for assessing agricultural water use and food production. While recent advances in optical and synthetic aperture radar (SAR) remote sensing have enabled cost-effective large-scale irrigation mapping through machine and deep learning approaches, these models still depend heavily on reliable in situ reference data. Major limitation in irrigation mapping thus resides in the scarcity and uneven global distribution of well-documented irrigated reference datasets. Here, we present a new geospatial reference dataset of irrigated and non-irrigated agricultural fields collected across two contrasting agro-climatic regions: France and Kazakhstan. The dataset was compiled through extensive field campaigns conducted in 2023 (France) and 2025 (France and Kazakhstan). It provides field-level labels of irrigation status, associated crop types, and irrigation practices. In addition, we provide ready-to-use auxiliary time series derived from Sentinel-1 SAR backscatter and Sentinel-2 vegetation indices to facilitate machine learning–based irrigation mapping. This dataset supports the development, validation, and intercomparison of remote sensing–based irrigation products as well as the spatio-temporal transferability of deep learning models across regions, crop types, and climatic conditions.