Weather records from recent years performed better than analogue years when merging with real-time weather measurements for dynamic within-season predictions of rainfed maize yield

Abstract

Listen

Translate article

Within-season crop yield prediction with a dynamic crop model can provide valuable references for field management practices and regional food security. However, weather ensembles containing the unknown future weather conditions occurring after prediction dates are essential for such predictions using crop models. Two strategies were established for selecting analogue weather years as the target growing season based on a five-year maize experiment conducted at eight sites in the Loess Plateau of China. The first strategy tried weather data from different lengths of years ahead the planting year. The second strategy used the k-nearest neighbor (k-NN) algorithm to select analogue weather according to different combinations of weather variables with daily or accumulative values. The results showed that satisfactory predictions could be obtained after maize tasseling (about 50 d prior to maturity). The mean absolute relative error (ARE) and coefficient of variation (CV) of the daily yield predictions after tasseling were 6.6% and 5.7%, respectively, in 2010 at the Yulin site. In the leading-year strategy, the most reliable predictions were obtained by the weather data from the 10 years ahead of planting, with an overall average ARE of 11.7%. In the k-NN strategy, the most reliable predictions were obtained by using the analogue weather selected with only accumulative precipitation, with an overall average ARE of 11.5%. Additionally, both of the two optimal strategies improved the original predictions in most cases. However, the k-NN strategy was more likely to generate worse predictions in the early part of the growing season. Generally, it was more convenient to use the weather data of 10 leading years before the planting year to represent the unknown weather data after the prediction dates. This strategy provided reliable prediction accuracy without complex programming and requirement for long-term weather records.