China is the world’s largest consumer and importer of soybeans, with rising domestic demand driving the expansion of production and cultivation areas since 2000. Knowing the spatial distribution of soybean, including the trends and interannual variability, is essential for yield estimation, agricultural planning, and ensuring national food security. However, a high-precision, long-term, national-scale spatial dataset for soybean cultivation areas of China remains unavailable. To address this gap, we developed the China Soybean Area (ChinaSoyA30m) dataset at a 30-m resolution—covering the years 2000–2022 at the national scale, using Landsat imagery and a combined phenology-based and machine learning approach. We analyzed time-series phenological characteristics of major crops across nine major agricultural regions in China and automatically generated annual training samples for supervised classifiers through a GWCCI-derived unsupervised method. To enhance the accuracy of these samples, we applied gap statistics, K-means clustering, and spectral angle mapping techniques to minimize noise and improve classification reliability. The supervised classification was conducted using a multi-random forest fusion strategy on the Google Earth Engine (GEE) platform, leveraging dense Landsat-5/7/8/9 data to generate yearly soybean maps. Our ChinaSoyA30m dataset showed strong correlation with official statistics at provincial, prefectural, and county levels, with R2 values of 0.95, 0.89, and 0.80, respectively.