Potential of remote sensing data-crop model assimilation and seasonal weather forecasts for early-season crop yield forecasting over a large area

Abstract

Regional crop yield forecasting before harvest is critical for managing climate risk, optimizing agronomic management, and making food trade policy. The advantage of remote sensing data-crop model assimilation for yield estimates has been well recognized, however its potential for early-season crop yield forecasting has not yet been investigated. In this study, combining a crop model, remote sensing leaf area index (LAI) assimilation and weather forecasts, we conducted yield forecasting for winter wheat in the central North China Plain during 2008–2015. Sequential forecasting was conducted to assess the yield forecasting potential and the effects of LAI assimilation and multiple weather forecasts with different lead times. Results showed that forecasting skill increased with shortening of lead time. Assimilating remote sensing LAI into crop model was valuable and critical to improve yield forecasting skills. The uncertainties from weather forecasts could weaken the forecasting skills; and using historical weather observations performed better than using weather forecasts outputted by climate models. In general, winter wheat yield in the central North China Plain could be reliably forecasted with a lead time from five weeks (mean MAPE < 10%, ROC score > 0.8, ACC> 0.65) to two months (mean MAPE <12%, ROC score> 0.75, ACC > 0.55) before harvest. The study highlights that current available data can provide fair yield forecasting; nevertheless the remote sensing LAI data and weather forecasts need further improvements to improve yield forecasting skills and provide valuable suggestions for stakeholders to respond to the forecasts timely.