A 10-m annual maize mapping dataset for Northeast China from 2017 to 2024

It is the first long-term, high-resolution maize-specific dataset for this region derived from AlphaEarth Foundations Satellite Embedding V1. The maps were produced on Google Earth Engine using random forest classifiers trained separately for five agro-climate zones across 119 grids.

High-confidence training samples for 2017–2022 were extracted by harmonizing six existing crop products, while samples for 2023–2024 were migrated across years using cosine similarity (θ = 0.90) of the 64-dimensional embedding features. The dataset is stored in GeoTIFF format, with pixel values of 1 indicating maize and 0 indicating non-maize, covering the spatial extent of 38°42′–53°55′N and 118°53′–135°09′E. Independent validation against ground-truth samples (2017–2019) yielded maize F1 scores of 0.915–0.936, comparison with municipal statistical yearbook data (2017–2024) showed R² of 0.78–0.90, and test-sample validation produced overall accuracies of 0.934–0.985.