B04
Probabilistic land use
Prof. Dr. Thomas Heckelei
University of Bonn | +49 228 73-2331 |
Dr. Hugo Storm
University of Bonn | +49 228 73-60828 / +49 157 75745561 |
Summary
Agricultural land use practices, such as crop choice and irrigation decisions, influence the temporal and spatial distribution of water and energy fluxes. In B04 we generate crop and irrigation maps that cover Europe at a 1x1 km resolution from 1990 - 2020 and that help model those fluxes. Importantly, the generated maps are consistent with existing knowledge, for example on regional crop production quantities. We apply Bayesian statistical approaches and Machine Learning to incorporate various types of information sources, such as soil and climate maps, survey data, economic statistics, and remote sensing data. The maps are probabilistic in nature, meaning that they transparently reflect data and model uncertainty.
Contribution to the CRC
The probabilistic maps generated by B04 provide ex-post crop and irrigation information for Europe at an unprecedented temporal and spatial resolution. The methodology developed by B04 additionally supports spatial disaggregation of simulation results for agricultural production from economic models in D01, it also allows to generate spatially explicit crop maps for the future. Other projects within the CRC use the maps as inputs to model water and energy fluxes. By transparently communicating the inherent uncertainty, B04 makes it possible to use those maps in an adequate and responsible manner.
Approach
In B04, we are following two – currently distinct – approaches to generate the probabilistic crop and irrigation maps: On the one hand, we are using existing soil and climate maps, together with survey-based crop observations, to understand the relationship between the local environmental and economic conditions and cropping decisions. On the other hand, we extract abundant crop cover information from satellite imagery. In further steps, we intend to combine those approaches using Bayesian probabilistic modeling. The maps are intended to reflect all relevant existing knowledge on the spatial distribution of crop cover and irrigation, meaning for example that consistency between the generated maps and regional statistics is guaranteed.
Main results in 2022
To test our developed methodology, we generated a crop map with a 1x1 km resolution for North-Rhine Westphalia (Germany). Only nine different explanatory variables related to natural agronomic conditions were used to provide crop production information that usually is available only at the province level (i.e., NUTS 1). We used IACS data that is publicly available only for selected years to validate our estimation. For visualization, the maps display for each 1x1 km cell only the dominant crop, that is the crop with the largest share in 2020.
