Evaluation of downscaled DEMETER multi-model ensemble seasonal hindcasts in a northern Italy location by means of a model of wheat growth and soil water balance

Abstract

In this paper we explore the new possibilities for early crop yield assessment at the local scale arising from the availability of dynamic crop growth models and of downscaled multi-model ensemble seasonal forecasts.We compare the use of the latter with other methods, based on crop growth models driven by observed climatic data only. The soil water balance model developed and used at ARPA Emilia-Romagna (CRITERIA) was integrated with crop growth routines from the model WOFOST 7.1. Some validation runs were first carried out and we verified with independent field data that the new integrated model satisfactorily simulated above-ground biomass and leaf area index. The model was then used to test the feasibility of using downscaled multi-model ensemble seasonal hindcasts, coming from the DEMETER European research project, in order to obtain early (i.e. 90, 60 and 30 d before harvest) yield assessments for winter wheat in northern Italy. For comparison, similar runs with climatology instead of hindcasts were also carried out. For the same purpose, we also produced six simple linear regression models of final crop yields on within season (end of March, April and May) storage organs and above-ground biomass values. Median yields obtained using downscaled DEMETER hindcasts always outperformed the simple regression models and were substantially equivalent to the climatology runs, with the exception of the June experiment, where the downscaled seasonal hindcasts were clearly better than all other methods in reproducing the winter wheat yields simulated with observed weather data. The crop growth model output dispersion was almost always significantly lower than the dispersion of the downscaled ensemble seasonal hindcast used as input for crop simulations.