Seasonal flooded rice area extent dataset during dry seasons 2016 to 2019 at the Telangana state scale, South-India.

Indian agriculture largely depends on the timely and spatially variable availability of water resources which are replenished during the monsoon season. In the state of Telangana, a significant portion of the available water is utilized for flooded rice cultivation, both in surface water-fed command areas and in groundwater-dependent regions. The spatial extent of seasonal rice cultivation varies annually in response to water availability that is a key indicator of how farmers adapt to regional and global environmental and socio-economic changes. In this study, we present seasonal land use maps for the dry season (Rabi) from 2016 to 2019, derived using the (IOTA²) processing chain [1]. IOTA² is an open-source software that combines temporal interpolation and classification of multispectral Sentinel-2 time series to map land cover dynamics. Sentinel-2 Level 2A data-processed using the Multi-Temporal Cloud Screening and Atmospheric Correction Software (MAJA)-were used to generate 10-day composite reflectance time series for each season over the entire Telangana state. A Random Forest classifier was trained on interpolated spectral time series using ground-truth data collected by the authors during dedicated field campaigns conducted between January and March of each year from 2016 to 2019. Ground observations were labelled into nine land use classes: rice, vegetables, maize (when applicable), orchards, natural bush, bare ground, urban, water, and unharvested dry-season cotton (when applicable). For each season, the ground-truth dataset was randomly split into training and validation sets eight times to generate eight classification outputs, from which average precision, recall, and F-score values were calculated. The dataset associated with this paper includes four seasonal raster maps, each encoding, for every pixel, the number of times (from 0 to 8) it was classified as rice during the eight classification runs. These rice extent confidence maps serve as an empirical measure of spatial classification uncertainty and inter-annual variability. The ground-truth polygon dataset used for classification and validation is also provided. Together, these datasets support the monitoring of seasonal rice dynamics and can serve as a reference for agricultural and hydrological studies in South Asia or training data for deep learning approaches for extension in space and time of those maps. Such a compilation can be used to support decisions on crop or cropping pattern changes in response to climate change, as well as to inform government policy-making.