modelling and analysis
In times of climate change and increasingly limited resources, Austrian agriculture is caught between economic efficiency and ecologically sustainable production. In addition to general questions on forms of agricultural land management, changing framework conditions and agricultural policy measures are also of importance. In Austria, a wide range of measures under the ÖPUL agri-environmental program promote environmentally friendly farming. There are numerous interactions between the different economic and ecological sub-areas of the agricultural, environmental and food systems, which can result in conflicts of objectives but also synergy effects.
In addition to theoretical, conceptual considerations, quantitative methods can also be helpful in demonstrating these complex interrelationships. For example, mathematical models for the simulation of plant growth (crop models) depict biophysical processes such as biomass production, nutrient cycles, soil water balance and erosion. With the help of model scenarios and depending on the site conditions, relationships between plant yields and management measures as well as the environmental effects of cultivation can be analyzed. In addition to the complex and data-intensive crop models, there are a number of other quantitative methods with varying degrees of detail, data requirements and explanatory potential.
Objective
The aim of the project is to develop an analytical concept for the investigation of negative and positive environmental effects of agricultural land use and, conversely, the effects of current and expected future ecological framework conditions on agricultural land use. This concept should include a theoretical framework based on a systematic approach, a database consisting of relevant data and indicators as well as a selection of methods for quantitative analysis. Adapted to regional conditions and equipped with a certain flexibility, the analysis concept can be a valuable basis for decision-making for a variety of questions in the agri-environmental sector.
Status of the project
Data on site conditions for agricultural land use, such as natural conditions like topography, soil composition and meteorological data, as well as on management measures like crop rotation, irrigation, tillage and the use of fertilisers and pesticides, were taken from a wide variety of data sources (including the BML's geodata catalogue and soil maps) and collated on a site-specific basis. In addition, remote sensing data from the JRC on soil moisture are included in the analyses. Among other things, the yield data from the farms (FADN) are available for validation.
As part of the ‘Biophysical Processes’ project, some work steps were taken over by the IKT staff unit. This mainly involved input from the fields of remote sensing, spatial analyses and climate data: the available data portals for raw data, processed data and indices from satellite data were determined. A GIS analysis localised impact areas that are located in spatial proximity to the measurement networks of the climate stations, soil profiles and groundwater levels in order to be able to analyse the associated influences on the model accuracy. In a third step, national and international data sources of climate data were compared. Ultimately, the INCA data was determined to be the most suitable for the purpose of the project.
In the area of methods, research was carried out into the various modelling methods. Although the large data requirements of process-based crop models, which are sometimes difficult to fulfil, can theoretically lead to very detailed analyses, in practice they are often a source of uncertainty. In contrast, statistical and machine learning models used in this area are usually less data-intensive and more flexible in terms of data requirements, but are often less detailed and designed in the form of a black box, i.e. without modelling the causal relationships. Using the database, a process-based model was parameterised and calibrated and several machine learning models (Elastic Net, MARS, Random Forest, Extreme Gradient Boosted Trees) were tuned and trained. Three different aggregation levels were created to compare the performance of the models. The machine learning models were additionally trained on these aggregation levels with three different sets of explanatory variables (management, biophysical variables, all variables). The results of the machine learning models are supplemented and interpreted by explanatory models based on Shapley values (SHAP, Dependence Plots).
Work 2025 and 2026
The database is to be further completed and documented so that it can also be used for follow-up projects. The data input of the various models will be standardised wherever possible and the adaptation to local site conditions will be further improved. Suitable measures will be defined to assess the accuracy of the model forecasts and to compare the models.
As a first application, the models for analysing the variance of volume yields are used. The suitability of the different models for this problem will be tested and compared in terms of accuracy, detail, data requirements and availability as well as interpretability. This comparison of methods is to be published in a scientific journal.
With the developed analysis concept, various relationships between agricultural land use and environmental effects can subsequently be analysed. A first analysis could deal with the effects of regionally increased drought on yields, growth periods and soil properties. The possibility of using remote sensing to close data gaps and improve the accuracy of the models will also be investigated. In addition, synergies with other BAB projects at the interface between agriculture and the environment (‘Systemic considerations in the agricultural, environmental and food sector’ (BAB 056/22), ‘Economic analyses and consulting in the agricultural, environmental and food sector’ (BAB 056/22), ‘Economic analyses and consulting in the agricultural, environmental and food sector’ (BAB 056/22)) are to be developed.
Work package yield model for financial soil estimation
The influence of site conditions on the yield level plays a significant role in the financial soil estimate. Soil climate curves determine a discount or addition to the expected yield depending on soil type, precipitation and heat conditions. Starting in 2025, the soil climate curves are to be recalculated in this work package in cooperation with the BMF. Initially, methodological approaches and data requirements will be discussed and a decision will be made on whether to involve other partners. Implementation will then begin. The final result should be available by December 2025.
Timetable
Project start: 01/2020
Project end: 12/2026
This text has been automatically translated with www.DeepL.com/Translator.
