Modeling spatio-temporal crop allocation patterns by a stochastic decision tree method, considering agronomic driving factors

Evaluating the environmental impacts of agricultural practices increasingly involves the use of spatially distributed simulation models that account for crop allocations across fields as an input factor. Our objective was to develop a model for spatio-temporal allocation of crops to a field pattern that was able to account for agronomic and spatial driving factors including crop production objectives, spatial distribution of the crops around farmsteads, and preferential allocation of crops on soil waterlogging classes. We developed a model based on stochastic decision trees (SDTs) to integrate farm type and field characteristics (area, distance to farmstead, waterlogging, and current crop) in the spatio-temporal allocation process without prior expert knowledge, and we compared the model to a reference model based on first-order Markov chains or transition matrices. A case study comparing both models was performed in the Naizin catchment (Western France), where crop allocation to fields was known for the period 1993–2006. The SDTs built had a general structure similar to transition matrices. SDTs and transition matrices exhibited similar performances in predicting crop transitions in time and in allocating crops to the proper soil waterlogging class. However, SDTs proved to better reproduce the spatial distribution of crops around the farmsteads. SDTs provide an integrated way to analyze and simulate crop allocation processes within a single integrated framework. The ease of constructing decision trees suggests potential couplings of SDT to various landscape-scale ecological models requiring a detailed description of the land use mosaic as input data.

Research highlights
► Decision trees can be used to model and simulate crop successions in space and in time.
► In a landscape where crops are highly organized in space and crop rotations are followed without deviation, we observed that the performance of decision trees were similar to that of first-order Markov chains.
► Decisions trees provide a way to model and simulate spatio-temporal allocation of crops within a single integrated framework with a low computational cost.