Spatial resolution effects on crop yield forecasts: An application to rainfed wheat yield in north Greece with CERES-Wheat

- An approach was developed for within-season yield predictions at two spatial levels.
- Yield predictions varied by the spatial scale, planting and prediction date.
- Yield predictions at the lower resolution spatial level were in most cases moderate.
- Yield predictions at the higher resolution spatial level worsened.

The present study aims at forecasting hard wheat (Triticum turgidum L. var. durum) yield at seven prediction dates (planting and six 30-day intervals after planting) prior to harvest in northern Greece. It is based on (a) reported crop yields at two relatively high spatial resolution regional levels (three NUTS2 (Nomenclature of Units for Territorial Statistics) and 16 NUTS3 regions) and (b) crop agroclimatic indicators simulated with CERES-Wheat, at four planting dates, for the years 1979-2006. Principal component analysis (PCA) was applied to explore major patterns of joint variability in a number of simulated crop agroclimatic indicators at the selected prediction dates during growing season. Stepwise regression and hindcast were employed for the selection of the modes identified by PCA as predictors in multivariate linear models used for yield forecast. Yield forecasting skill varied to a large extent by the spatial scale, planting date and timing of forecast. When the simulation results were aggregated to the larger spatial level (NUTS2), the cross-validated forecasting skill was rated as moderate in Central Macedonia (CM) (R2 = 43%) and Thrace (THR) (R2 = 35.9%) and as low in West Macedonia (WM) (R2 = 21.5%). Soil water availability to plants was the most important indicator. Except for THR, these forecasts were achieved three months before harvest in CM and four in WM. Compared with the NUTS2 level, yield predictions at the higher resolution spatial level (NUTS3) worsened in 11 and 12 out of 16 NUTS3 regions in terms of R2 and RMSE, respectively. The results demonstrate the potential of this approach and the suitability of CERES-Wheat for regional crop yield forecasting in northern Greece.