Satellite-derived vegetation indices are widely utilized in yield forecasting models; however, they can be heavily impacted by atmospheric conditions due to their reliance on visible and near-infrared portions of the electromagnetic spectrum. Given the importance of soil moisture (SM) for crop development, the objective of this study was to investigate the use of passive microwave-derived estimates of surface SM obtained by the SM Ocean Salinity Mission (SMOS) satellite for forecasting canola yields across the Canadian Prairies within Agriculture and Agri-Food Canada’s (AAFC) Canadian Crop Yield Forecaster (CCYF) model. Weekly SMOS SM observations were combined with climate variables and normalized difference vegetation index (NDVI) data derived from the Advanced Very High Resolution Radiometer (AVHRR) platform and used as an input for forecasting canola yields at the township-scale across the Canadian Prairies from 2010 to 2016. Top predictors were identified, and regression models were built using a robust least angle regression (RLARS) and leave-one-out cross-validation (LOOCV) scheme. SM was found to provide a better descriptor of canola stress than the more widely utilized NDVI, being selected as a predictor in 74.2 % of developed ecodistrict models over the 7-year period, compared to just 41.2 % for NDVI. The difference between model R² values (i.e. R²_diff) when SMOS SM predictors were included and excluded from the forecast, respectively, revealed varying degrees of model improvements; however, the majority of ecodistricts under study (53.3 %) showed improved model fit (i.e. R²_diff > 0) with observed canola yields when SMOS SM indices were included as potential predictors within the CCYF. Overall, greater improvements in the CCYF performance were observed in Manitoba and Saskatchewan where meteorological stations are more sparsely distributed. However, performance for both sets of model inputs was relatively low with R² values ranging from 0 to 0.74 (mean = 0.13) and from 0 to 0.52 (mean = 0.12) across the study area both when SM was included and excluded from the model, respectively. These findings suggest that while SMOS SM observations may provide a more effective indicator of canola yields, the CCYF’s performance at the township-scale, where interannual yield variability is often quite high, is limited by the short temporal satellite record.

在产量预报模型中广泛使用了卫星衍生的植被指数。但是，由于它们依赖于电磁波谱的可见和近红外部分，因此它们可能会受到大气条件的严重影响。考虑到土壤水分（SM）对作物生长的重要性，本研究的目的是调查利用SM海洋盐度任务（SMOS）卫星获得的无源微波估算的表面SM来预测加拿大全境的油菜籽产量加拿大农业和农业食品公司（AAFC）的加拿大作物单产预报员（CCYF）模型中的大草原。SMOS SM的每周观测值与气候变量以及从超高分辨率高分辨率辐射仪（AVHRR）平台获得的归一化差异植被指数（NDVI）数据相结合，用作预测2010年加拿大大草原地区城镇规模油菜籽产量的输入到2016年。确定了最佳预测变量，并使用鲁棒最小角度回归（RLARS）和留一法交叉验证（LOOCV）方案建立了回归模型。与使用更广泛的NDVI相比，SM被发现提供了更好的低芥酸菜籽胁迫特征，在7年的发展中，被选为74.2％的已开发生态区模型的预测指标，而NDVI仅为41.2％。型号R的区别 使用健壮的最小角度回归（RLARS）和留一法交叉验证（LOOCV）方案建立回归模型。与使用更广泛的NDVI相比，SM被发现提供了更好的低芥酸菜籽胁迫特征，在7年的发展中，被选为74.2％的已开发生态区模型的预测指标，而NDVI仅为41.2％。型号R的区别 使用健壮的最小角度回归（RLARS）和留一法交叉验证（LOOCV）方案建立回归模型。与使用更广泛的NDVI相比，SM被发现提供了更好的低芥酸菜籽胁迫特征，在7年的发展中，被选为74.2％的已开发生态区模型的预测指标，而NDVI仅为41.2％。型号R的区别SMOS SM预测变量分别包括在预测中和从预测中排除时的^2个值（即R ² _diff）显示了不同程度的模型改进；但是，当将SMOS SM指数作为CCYF中的潜在预测指标纳入考虑时，大多数正在研究的生态区（53.3％）表现出改善的模型拟合（即R ² _diff > 0），并具有观察到的油菜籽产量。总体而言，在曼尼托巴省和萨斯喀彻温省的气象站分布较稀疏的地方，CCYF性能得到了更大的改善。然而，在R ^2下，两组模型输入的性能都相对较低。当在模型中包括SM或从模型中排除SM时，整个研究区域的值分别在0到0.74（平均值= 0.13）和0到0.52（平均值= 0.12）之间。这些发现表明，尽管SMOS SM观测结果可以提供更有效的双低油菜籽产量指标，但CCYF在乡镇尺度上的表现（年际产量波动往往很高）受到短期卫星记录的限制。