S4A-CyL: A Sentinel-2 Time Series Dataset for Deep Learning in Agriculture (2020-2024)

It is specifically designed to support the development and validation of deep learning models for agricultural applications, with a primary focus on crop type classification. The dataset integrates dense Sentinel-2 L2A time series with meticulously processed agricultural parcel geometries and harmonized crop type labels derived from the Spanish Land Parcel Identification System (SIGPAC).

A key feature is the implementation of a persistent parcel identification system, ensuring the temporal traceability of agricultural plots across the five-year period. The data is structured in a format inspired by the Sen4AgriNet project, with individual NetCDF files for each spatial patch. Each patch is a self-contained unit that includes the multi-spectral Sentinel-2 time series alongside the corresponding parcel ID and crop type reference layers.