From field to globe: upscaling of crop growth modelling

Recently, the scale of interest for application of crop growth models has extended to the region or even globe with time frames of 50-100 years. The application at larger scales of a crop growth model originally developed for a small scale without any adaptation might lead to errors and inaccuracies. Moreover, application of crop growth models at large scales usually gives problems with respect to missing data.

Knowledge about the required level of modelling detail to accurately represent crop growth processes in crop growth models to be applied at large scales is scarce. In this thesis we analysed simulated potential yields, which resulted from models which apply different levels of detail to represent important crop growth processes. Our results indicated that, after location-specific calibration, models in which the same processes were represented with different levels of detail may perform similarly. Model performance was in general best for models which represented leaf area dynamics with the lowest level of detail. Additionally, the results indicated that the use of a different description of light interception significantly changes model outcomes. Especially the representation of leaf senescence was found to be critical for model performance.

Global crop growth models are often used with monthly weather data, while crop growth models were originally developed for daily weather data. We examined the effects of replacing daily weather data with monthly data. Results showed that using monthly weather data may result in higher simulated amounts of biomass. In addition, we found increasing detail in a modelling approach to give higher sensitivity to aggregation of input data.

Next, we investigated the impact of the use of spatially aggregated sowing dates and temperatures on the simulated phenology of winter wheat in Germany. We found simulated winter wheat phenology in Germany to be rather similar using either non-aggregated input data or aggregated input data with a 100 km × 100 km resolution.

Generation or simulation of input data for crop growth models is often neces­sary if the model is applied at large scales. We simulated sowing dates of several rainfed crops by assuming farmers to sow either when temperature exceeds a crop-specific threshold or at the onset of the wet season. For a large part of the globe our methodology is capable of simulating reasonable sowing dates. To simulate the end of the cropping period (i.e. harvesting dates) we developed simple algorithms to generate unknown crop- and location-specific phenological parameters. In the main cropping regions of wheat the simulated lengths corresponded well with the observations; our methodology worked less well for maize (over- and underestimations of 0.5 to 1.5 month). Importantly, our evaluation of possible consequences for simulated yields related to uncertainties in simulated sowing and harvesting dates showed that simulated yields are rather similar using either simulated or observed sowing and harvesting dates (a maximum difference of 20%), indicating the applicability of our methodology in crop productivity assessments.

The thesis concludes with a discussion on a proposed structure of a global crop growth model which is expected to simulate reasonable potential yields at the global scale if only monthly aggregates of climate data at a 0.5° × 0.5° grid are available. The proposed model consists of a forcing function, defined in terms of sigmoidal and quadratic functions to represent light interception, combined with the radiation use efficiency approach, and phenology determining the allocation of biomass to the organs of the crop. Within the model sowing dates and pheno­logical cultivar characteristics are simulated. Based on the proposed model the thesis finally derives directions for future research to further enhance global crop growth modelling.