Speaker
Description
Introduction
Growing different crops in a repeating sequence on the same field - frequently referred to as crop rotation - has various benefits over monoculture. These include improved control of weeds and soil-borne pests and diseases, enhanced resource use efficiency, and an increase in crop yield overall. Agronomic theory behind crop rotational planning is well-established, and respective planning tools are abundantly available. Software like ROTOR, ROTAT, and Fruchtfolge have been designed to generate best-suited crop sequences for specific fields. Although these optimization approaches promise mathematical clarity with logically convincing rule sets and best-practice assumptions, real-world sequential cropping may not follow those agronomic rules exclusively. In addition to environmental growing conditions, farmers' preferences, market prices, and agro-political decisions are expected to play a decisive role in driving operational cropping decision (Jänicke et al., 2022; Notz et al., 2023; Stein and Steinmann, 2018). A comprehensive analysis of the drivers determining sequential cropping and consequently rotational patterns is however missing so far.
Material and Methods
Therefore, we conducted a data-driven analysis to reveal practically relevant information using large observational datasets over Central Europe. Based on field-level data from the Integrated Administration and Control System (IACS) for Germany (eight federal states included), Austria, and the Czech Republic, we collected observations of individual crop sequences over the past 18 years. Across the three countries, our study area covered more than five million and 17 different crop types (excluding grassland and perennial crops). We used an unsupervised K-means clustering approach via Principal Component Analysis (PCA) of soil and climate characteristics at 1 km resolution to create clusters of fields with similar growing conditions. We used these cluster and a field-level cropping history from the past five years, market price developments, agronomic best-practice rules, subsidy payments, county-level animal density data, and information on cropping decisions of neighbouring fields to train a random forest (RF) machine learning model to detect and predict the crop type grown at each field for the upcoming year. All the above predictors were defined dynamic in time such that e.g. changes in climatic conditions would result in changes of the resulting cluster or changes in individual cropping decisions affected neighbouring fields. RF training included spatio-temporal cross-validation and forward feature selection.
Results and Discussion
Preliminary results revealed the importance of the individual cropping history (with regional differences), agricultural subsidies such as the Diversification of Crop production promoting grain legumes through the Common Agricultural Policy of the EU, and prices of high-revenue products, especially for oilseed crops. When designing economic and agronomic scenarios for adapting crop production to a changing climate at a higher level, these results point to the importance of economic incentives to move away from habitual sequential patterns, while agronomic rules for best-practice crop rotations play a secondary role.
In the next step, we will use the trained RF to project crop rotational patterns until 2070 across the German grid at 5 km resolution. This outlook will be based on climatic projections from the German core ensemble of the German Weather Service (DWD) to inform the process-based crop model MONICA (Nendel, 2014) to simulate long-term climate change outlooks considering the projected rotational management for each simulation unit. To advance from a one-crop-covers-all approach will allow for a more realistic representation of the agricultural landscape when simulating crop production, greenhouse gas emissions, or carbon sequestration potentials over large areas and timespans.
References
Jänicke, C. et al., 2022. Field-level land-use data reveal heterogeneous crop sequences with distinct regional differences in Germany. European Journal of Agronomy, 141, Article 126632. https://doi.org/10.1016/j.eja.2022.126632
Nendel, C., 2014. MONICA: A Simulation Model for Nitrogen and Carbon Dynamics in Agro-Ecosystems. In L. Mueller, A. Saparov, & G. Lischeid (Eds.), Novel Measurement and Assessment Tools for Monitoring and Management of Land and Water Resources in Agricultural Landscapes of Central Asia (pp. 389-405). Springer International Publishing. https://doi.org/10.1007/978-3-319-01017-5_23
Notz, I. et al., 2023. Transition to legume-supported farming in Europe through redesigning cropping systems. Agronomy for Sustainable Development, 43(1), 12. https://doi.org/10.1007/s13593-022-00861-w
Stein, S. and Steinmann, H. H., 2018. Identifying crop rotation practice by the typification of crop sequence patterns for arable farming systems – A case study from Central Europe. European Journal of Agronomy, 92, 30-40. https://doi.org/10.1016/j.eja.2017.09.010
| Keywords | Crop rotation; Machine learning; Agricultural decision making; Crop modelling |
|---|
