Developing products for use in drought monitoring: improved crop yield and production forecasts (multi-crop modelsapproach)

Given the variety of crop growth models and their relative strengths and weaknesses, ensembles of crop models are being progressively developed for yield/production forecasting purposes, quantification of the impact on yield of changes due to climate or crop management. Such approach helps improve accuracy and consistency in simulating growth dynamics under various environmental conditions. The aim of the project is to develop an operational integrated seasonal climate-crop modelling system for yield and production forecasting of major crops in Queensland (wheat, sorghum, sugarcane and cotton). A focused literature review was conducted in order to select potential crop models for the multi-model Ensemble. The main criteria used include: (1) the popularity of the model (at national and international levels), (2) its performance as reported in published studies, (3) the model structure and easiness of use within the project time frame (e.g., implementation, data collection and calibration, etc.), (4) the availability of model updates, and (5) a crop model which considers the various aspects of climate change as drivers (including rainfall, atmospheric CO2, temperature and ozone). The models APSIM (Agricultural Production Systems sIMulator) and DSSAT (Decision Support System for Agrotechnology Transfer) were then selected after review. We also conducted a calibration/parameterisation of the selected models and performed the growth and yield simulation for wheat, sorghum, sugarcane and cotton. Different combinations of factors (crop management practices) were used to evaluate the ability of these models to give reasonable results in such various conditions at 10 to 19 selected simulation sites across the agricultural landscape in Queensland. Because the simulation configurations represent a range of crop management practices, soil types, crop varieties, etc., the aim was not to predict 'one single' yield value. Rather, the objective was to configure the models so that it can be used for impact analysis and/or be coupled to other forecasting systems. Comparing the simulated yields (from each model or the mean simulated value from the Ensemble) to the observed yield (available at regional scale) the overall trend in year to year variability is satisfactorily captured for cotton across the different sets of configurations, whereas for the other crops, there is generally a trend to yield overestimation or underestimation depending on the site and year. A statistical assessment of model performance revealed RMSE values as low as 0.27 t ha-1 and 0.40 t ha-1 for cotton and sorghum, respectively, in APSIM simulations; the respective values in DSSAT being 0.17 t ha-1 and 0.37 t ha-1, and 0.24 t ha-1 and 0.45 t ha-1, for the Ensemble. The spread of RMSE variation however was lesser for wheat and cotton compared to sorghum, regardless of the model. Similar pattern is observed with MAE values. Regarding sugarcane the variations of RMSE and MAE in DSSAT simulations and the Ensemble were the lowest. A modelling framework was thus developed through the project with interesting potentialities for decision support system tools. This framework as well as the simulated variables have served in the sister projects DCAP USQ 6 ('Enhanced multi-peril crop insurance'), DCAP USQ 14 ('Crop production modelling under climate change and regional adaptation'), and DCAP USQ 15 ('Value of seasonal climate forecasts').