Landscape Structure, Climate Variability, and Soil Quality Shape Crop Biomass Patterns in Agricultural Ecosystems of Bavaria

Maninder Singh Dhillon^1*Thomas Koellner²Sarah Asam³Jakob Bogenreuther²Stefan Dech^1,3Ursula Gessner³Daniel Gruschwitz¹Sylvia Helena Annuth²Tanja Kraus³Thomas Rummler⁴Christian Schäfer¹Sarah Schönbrodt-Stitt¹Ingolf Steffan-Dewenter⁵Martina Wilde^1,6Tobias Ullmann¹

• ¹Department of Remote Sensing, Institute of Geography and Geology, University of Würzburg, Würzburg, Germany
• ²Department of Ecological Services, Faculty of Biology, Chemistry and Earth Sciences, BayCEER, University of Bayreuth, 95447, Bayreuth, Germany, Bayreuth, Germany
• ³German Remote Sensing Data Center (DFD), German Aerospace Center (DLR), 82234 Wessling, Germany, Wessling, Germany
• ⁴Department of Applied Computer Science, Institute of Geography, University of Augsburg, 86159 Augsburg, Germany, Augsburg, Germany
• ⁵Department of Animal Ecology and Tropical Biology, University of Würzburg, 97074 Würzburg, Germany, Würzburg, Germany
• ⁶Department of Physical Geography and Soil Science, Institute of Geography and Geology, University of Würzburg, 97074 Würzburg, Germany, Würzburg, Germany

Understanding how environmental variability shapes crop biomass is essential for improving yield stability and guiding climate-resilient agriculture. To shed more light on this relationship, biomass estimates derived from a semi-empirical light use efficiency model were compared against those predicted using a machine learning-remote sensing approach that incorporates environmental variables. Specifically, we combined a light use efficiency model with a random forest approach to predict the mean crop biomass across the years 2001-2019 for winter wheat and oilseed rape across the entire region of Bavaria, Germany. Using a 5 km² hexagon-based spatial framework, we integrated light use efficiency-derived biomass with spatial predictors-including land cover diversity, small woody features, elevation, slope angle, aspect, and soil potential-and spatio-temporal climate variables, specifically the seasonal (growing-season) mean and standard deviation of temperature, precipitation, and solar radiation from 2001 to 2019. The machine learning and remote sensing model demonstrated improved predictive accuracy compared to the light use efficiency model alone, particularly for winter wheat. Interpretation of the model revealed that biomass patterns were shaped by both landscape configuration and climatic conditions. Winter wheat was primarily influenced by topographic and landscape features, while oilseed rape was more strongly affected by solar radiation and soil potential. Biomass gains were observed in moderately diverse landscapes, whereas high landscape variability and fragmentation were associated with reduced biomass.Temperature-related thresholds, such as reductions in biomass beyond 21°C for winter wheat and 12°C for oilseed rape, reflect modeled responses under the specific environmental and varietal conditions in Bavaria. These values highlight regional crop sensitivities, rather than universal physiological limits. By combining the strengths of process-based and data-driven approaches, this framework offers a transferable methodology for predicting crop biomass and examining its spatial patterns. The results offer region-specific insights that can inform sustainable agricultural planning in the face of ongoing climate change.