Department of Plant Science, College of Agriculture and Life Sciences, Seoul National University, Seoul, Republic of Korea.
Research Institute of Agriculture and Life Sciences, Seoul National University, Seoul, Republic of Korea.
PLoS One. 2019 Feb 25;14(2):e0211874. doi: 10.1371/journal.pone.0211874. eCollection 2019.
Crop growth models and remote sensing are useful tools for predicting crop growth and yield, but each tool has inherent drawbacks when predicting crop growth and yield at a regional scale. To improve the accuracy and precision of regional corn yield predictions, a simple approach for assimilating Moderate Resolution Imaging Spectroradiometer (MODIS) products into a crop growth model was developed, and regional yield prediction performance was evaluated in a major corn-producing state, Illinois, USA. Corn growth and yield were simulated for each grid using the Crop Environment Resource Synthesis (CERES)-Maize model with minimum inputs comprising planting date, fertilizer amount, genetic coefficients, soil, and weather data. Planting date was estimated using a phenology model with a leaf area duration (LAD)-logistic function that describes the seasonal evolution of MODIS-derived leaf area index (LAI). Genetic coefficients of the corn cultivar were determined to be the genetic coefficients of the maturity group [included in Decision Support System for Agrotechnology Transfer (DSSAT) 4.6], which shows the minimum difference between the maximum LAI derived from the LAD-logistic function and that simulated by the CERES-Maize model. In addition, the daily water stress factors were estimated from the ratio between daily leaf area/weight growth rates estimated from the LAD-logistic function and that simulated by the CERES-Maize model under the rain-fed and auto-irrigation conditions. The additional assimilation of MODIS data-derived water stress factors and LAI under the auto-irrigation condition showed the highest prediction accuracy and precision for the yearly corn yield prediction (R2 is 0.78 and the root mean square error is 0.75 t ha-1). The present strategy for assimilating MODIS data into a crop growth model using minimum inputs was successful for predicting regional yields, and it should be examined for spatial portability to diverse agro-climatic and agro-technology regions.
作物生长模型和遥感是预测作物生长和产量的有用工具，但在预测区域尺度上的作物生长和产量时，每种工具都有其固有缺陷。为了提高区域玉米产量预测的准确性和精度，开发了一种将中分辨率成像光谱仪 (MODIS) 产品同化到作物生长模型中的简单方法，并在美国主要玉米产区伊利诺伊州评估了区域产量预测性能。使用作物环境资源综合模型 (CERES)-玉米模型对每个网格进行了玉米生长和产量模拟，模型输入最少包括播种日期、施肥量、遗传系数、土壤和天气数据。使用具有叶面积持续时间 (LAD)-逻辑函数的物候模型来估计播种日期，该函数描述了 MODIS 衍生叶面积指数 (LAI) 的季节演变。玉米品种的遗传系数确定为成熟组的遗传系数[包含在决策支持系统技术转让 (DSSAT) 4.6 中]，该系数显示了从 LAD-logistic 函数导出的最大 LAI 与 CERES-Maize 模型模拟的最大 LAI 之间的最小差异。此外，根据从 LAD-logistic 函数估计的每日叶面积/重量增长率与 CERES-Maize 模型模拟的增长率之比，估算了每日水分胁迫因子。在雨养和自动灌溉条件下，对额外同化 MODIS 数据衍生的水分胁迫因子和 LAI 进行了估算，对玉米年产量预测的精度最高（R2 为 0.78，均方根误差为 0.75 t ha-1）。使用最少输入将 MODIS 数据同化到作物生长模型中的策略对于预测区域产量是成功的，应该在不同的农业气候和农业技术区域中检验其空间可移植性。
