2010-2019年基于Modis的宁夏县级尺度枸杞估产数据集

针对现有遥感估产数据集过度依赖于实测数据和人工经验的问题，基于宁夏16县Modis高光谱影像、年际枸杞统计产量数据和枸杞种植区域矢量图，构建了一种多波段、多时相融合的县级尺度遥感影像估产数据集。实验结果表明，通过卷积神经网络自动提取本文数据集的影像特征，估产模型在MRE、RMSE和R2上别分达到了14.52%、859.23吨和0.83，验证了该数据集的可用性，为宁夏县级尺度枸杞年际产量预测提供重要数据支持，对地区农业可持续发展和科学研究具有重要意义。

To address the issue that existing remote sensing yield estimation datasets overly rely on field-measured data and manual prior experience, this study developed a county-scale remote sensing yield estimation dataset with multi-band and multi-temporal fusion, using MODIS hyperspectral imagery from 16 counties in Ningxia, interannual statistical wolfberry yield data, and vector maps of wolfberry planting areas. Experimental results demonstrate that when convolutional neural networks (CNNs) are employed to automatically extract image features from this dataset, the yield estimation model achieved a mean relative error (MRE) of 14.52%, a root mean square error (RMSE) of 859.23 tons, and a coefficient of determination (R²) of 0.83, which validates the usability of the proposed dataset. This dataset provides critical data support for interannual wolfberry yield prediction at the county scale in Ningxia, and holds great significance for regional sustainable agricultural development and scientific research.