OCQT-PTS: A Multi-Region Rice Phenology Time-Series Dataset

OCQT-PTS is a multi-region rice phenology time-series dataset designed for eight-class rice cropping-system classification and cross-region generalization evaluation. The public release contains 636,719 samples divided into seven partitions: 36,729 real training samples, 8,333 quality-screened Core-A phenology-transferred samples (PTS), 33,867 spatial-block validation samples, and four independent external test sets from Huangmei (329,890 samples), Liushi (7,680 samples), Dawu (159,391 samples), and Maonan (60,829 samples). Each observation contains a 72-step Sentinel-2 time series with ten spectral bands (B2, B3, B4, B5, B6, B7, B8, B8A, B11, and B12), an original day-of-year sequence, a valid-time mask, an eight-class class identifier, and a three-channel [early, middle, late] label. The released weighted training protocol includes all 8,333 Core-A PTS samples. Real training samples use a sample weight of 1.0, while Core-A PTS samples use a sample weight of 0.25. Validation and external test partitions are unweighted. PTS samples are intended only for training augmentation and must not be included in validation or external testing. The archive includes field definitions, a data card, partition-level class and region statistics, integrity manifests, a release verification script, and a script for reconstructing the combined weighted Real+PTS training set. Archive SHA-256:d5da135786d663cfec5d9a853508699c10b905c43f74795cce4e00184de00ca3