An Open and Giant-Scale Dataset for Multi-Modal Local weather Change-aware Crop Yield Predictions
Authors: Fudong Lin, Kaleb Guillot, Summer Crawford, Yihe Zhang, Xu Yuan, Nian-Feng Tzeng
Summary: Exact crop yield predictions are of nationwide significance for guaranteeing meals safety and sustainable agricultural practices. Whereas AI-for-science approaches have exhibited promising achievements in fixing many scientific issues resembling drug discovery, precipitation nowcasting, and so forth., the event of deep studying fashions for predicting crop yields is consistently hindered by the shortage of an open and large-scale deep learning-ready dataset with a number of modalities to accommodate ample info. To treatment this, we introduce the CropNet dataset, the primary terabyte-sized, publicly accessible, and multi-modal dataset particularly focusing on local weather change-aware crop yield predictions for the contiguous United States (U.S.) continent on the county degree. Our CropNet dataset consists of three modalities of information, i.e., Sentinel-2 Imagery, WRF-HRRR Computed Dataset, and USDA Crop Dataset, for over 2200 U.S. counties spanning 6 years (2017–2022), anticipated to facilitate researchers in growing versatile deep studying fashions for well timed and exactly predicting crop yields on the county-level, by accounting for the consequences of each short-term rising season climate variations and long-term local weather change on crop yields. Apart from, we develop the CropNet bundle, providing three varieties of APIs, for facilitating researchers in downloading the CropNet knowledge on the fly over the time and area of curiosity, and flexibly constructing their deep studying fashions for correct crop yield predictions. Intensive experiments have been carried out on our CropNet dataset by way of using varied varieties of deep studying options, with the outcomes validating the final applicability and the efficacy of the CropNet dataset in local weather change-aware crop yield predictions
