In-season Corn Yield Prediction Using Satellite-derived Solar-Induced Chlorophyll Fluorescence and Machine Learning Algorithms data

This data is for the U.S. Corn Belt (US-CB), a major agricultural region characterized by intensive corn cultivation. The data includes 210 counties consistently dominated by corn production, selected based on their suitability for sub-pixel SIF disaggregation and inclusion in the USDA’s annual yield reporting. The data include five growing seasons: 2015, 2016, 2018, 2019, and 2020. The year 2017 was excluded from the data due to incomplete satellite data caused by sensor malfunction.
Annual corn yield records were obtained from the U.S. Department of Agriculture’s National Agricultural Statistics Service (USDA-NASS). County-level yield data, measured in tons per hectare, are generated using a combination of farmer-reported surveys, field assessments, and calibrated simulation models. This integration ensures their reliability as ground-truth references for both training and validating predictive models. Their consistent spatial resolution and broad temporal span across the U.S. Corn Belt make them a solid basis for model development and performance testing.
Standard satellite-derived SIF products are provided at coarse spatial resolutions, complicating their use in heterogeneous agricultural landscapes. To overcome this limitation, I applied a sub-pixel extraction method developed by Kira and Sun (2020), which uses high-resolution land cover and crop maps to isolate the contribution of corn within each SIF pixel. This technique allowed for the derivation of corn-specific SIF values at the county level, minimizing contamination from other crops or land covers.

本数据集面向美国玉米带（U.S. Corn Belt）——这一以集约化玉米种植为典型特征的核心农业产区。本数据集涵盖210个以玉米种植为主导的县域，筛选依据为其适配亚像素日光诱导叶绿素荧光（SIF，Solar-Induced Chlorophyll Fluorescence）分解流程，且纳入美国农业部年度产量报告体系。数据集覆盖2015、2016、2018、2019及2020年共5个生长季，因传感器故障导致卫星数据不全，2017年被排除在外。
年度玉米产量记录取自美国农业部国家农业统计服务中心（USDA-NASS）。县域尺度的产量数据以吨/公顷为单位，通过农户上报调查、田间评估与校准后的模拟模型相结合的方式生成。该整合方案确保了数据作为训练与验证预测模型的地面真值参考的可靠性。其统一的空间分辨率与覆盖美国玉米带的广泛时间跨度，为模型开发与性能测试提供了坚实基础。
标准卫星反演的日光诱导叶绿素荧光（SIF）产品多为粗空间分辨率，这一特性使其在异质性农业景观中的应用受到限制。为克服该局限，本研究采用Kira与Sun（2020）提出的亚像素提取方法，该方法借助高分辨率土地覆盖与作物分布图，分离出每个SIF像素内玉米的荧光贡献。通过该技术，可获取县域尺度的玉米特异性SIF值，最大程度降低其他作物或土地覆盖类型带来的信号污染。