|Title||Modelling regional land use: the quest for the appropriate method|
|Author(s)||Soltani Largani, A.|
|Source||University. Promotor(en): Tom Veldkamp, co-promotor(en): Martha Bakker; Jetse Stoorvogel. - S.l. : s.n. - ISBN 9789461736574 - 153|
Soil Geography and Landscape
|Publication type||Dissertation, internally prepared|
|Keyword(s)||landgebruik - modelleren - gewasopbrengst - aardappelen - wintertarwe - gewasgroeimodellen - land use - modeling - crop yield - potatoes - winter wheat - crop growth models|
|Categories||Land Use Systems / Plant Production Systems|
The demand for spatially-explicit predictions of regional crop-yield patterns is increasing. An approach to assess a priori and/or future ranges of alternative scenarios spatial yield patterns at the regional scale is the application of mechanistic crop growth simulation models (CGSMs) (e.g. Launary, 2002). However, two main problems emerge in the application of field-level CGSMs at regional scales. Firstly, the required input data on weather, soils, and management are often not available; and secondly, if they are, generally not at the required level of detail. There are two possible approaches to address the identified problems. One is replacing the CGSM by a metamodel (Kleijnen and Sargent, 2000). The second approach is a simple empirical model (e.g. Lobell et al., 2008).
The modelling-approach choices and performances are context dependent. The context conditions that determine the best approach are input data requirements, problem definition, study sub-objective, the scale at which output results are expected, model end-users, and utilization of the output. The selection of the modelling approaches can be considered as one of the most difficult, and often ignored, steps to model crop yield at the regional level. However, a structured, systematic way of modelling-approach selection is lacking. In order to address this issue this thesis aimed to develop a framework for recommendable practices to model regional patterns of crop yield.
We reviewed literature for existing approaches that have been used to overcome the problem of data availability for the application of CGSMs at the regional level. Then we used the review to formulate decision rules as to what approach to take under different circumstances. Which of the approaches should be used depends on the following questions: (i) do observations of the input variable allow to estimate semivariograms?; (ii) are there auxiliary data correlated to the target variable?; (iii) do the input variables exhibit spatial correlation?; and (iv) is there spatial correlation in the residuals of the regression that related auxiliary data to the target variable?. Summarized, the selection of possible approaches depends on the data availability, the spatial variability, the temporal variability, the correlations with other variables, the data acquisition methods, the expected accuracy from a particular approach used to describe spatial variability, and the sensitivity of the CGSM to the variable.
We also evaluated different procedures (interpolate first, calculate later; calculate first, interpolate later) to simulate regional patterns of crop yields in the Carchi province in Northern Ecuador with field-level CGSMs. We also examined scaling effects that arise from spatial variability in input data by using different supports. Results demonstrated that the order of calculation and interpolation was of major importance, while aggregation had a minor effect on the regional patterns of potato yield. From an uncertainty propagation and variability point of view it is in general preferable to calculate first before interpolation.
We compared and evaluated the performance of three different modelling approaches for their capacity to model regional patterns of crop yield for two different cases: potato yields in Carchi and wheat yields in Western Germany. Based on these findings, various criteria for selecting a modelling approach are defined including credibility, relevance, sensitivity, and user friendliness. The empirical model and the metamodel are very easy to use and transparent. However, their application domain is limited to the case study area. The application of the CGSM remains complex and the model functions as a black box. The strength of the CGSMs is that impacts over a wider range of conditions can be simulated, taking into account many factors in a way that would not be possible using empirical models and metamodels (Lobell and Burke, 2010). It can be concluded that the various modelling approaches each have their unique merit. Hence, the different modelling approaches are therefore complementary for the interpretation of the observed patterns. There is not a single optimal solution to modelling agricultural systems to model, e.g. regional yield patterns. Moreover, we analysed the effect of spatial aggregation on the performance of the modelling approaches. The results showed that aggregation of calculated data leads to less variability and increasing linear fits at higher aggregation levels. The spatial variability in the case study area determines how strong this effect is. Finally we conclude:Regional crop yield modelling is very sensitive to the choice of model-type and data used; This sensitivity is usually not specifically addressed and not properly and systematically documented in many studies; The outcomes of such modelling exercises cannot be properly used when the underlying decisions on model and data type and sensitivities are unknown; Without this crucial knowledge regional crop simulation models can be easily misused by non-specialists; Standard decision rules are proposed to document these choices in a standard format allowing cross comparisons of different approaches despite the often strong context dependency of the results.