The "Soil erosion in agricultural landscapes in the context of dynamic changes of land use patterns and structures" (DYLAMUST) Project, funded by the German Research Foundation (DFG), represents a collaboration between the University of Augsburg and the University of Tübingen. This project embarks on an ambitious journey to develop an hybrid sediment transport model that integrates conventional methodologies with machine learning and artificial intelligence techniques.

Description:

Today's landscapes are largely shaped by human activities. Since the Holocene human-land interactions are driven by environmental attributes that are relevant for the location of settlements. In the recent past, it has been especially the type of ownership and inheritance of land as well as the land consolidation and expansion of the size of single fields due to mechanization that control land use patterns and structures largely.

Today agricultural landscapes face tremendous challenges including sustaining biodiversity, soil and water conservation, sustainability in general, and mitigation of and adaption to climate change. Soil erosion in agricultural landscapes is directly related to dynamic changes in land use patterns and structures. Key soil erosion control variables such as slope length, vegetation cover, erosion control, type of cultivation, and sediment transport pathways depend on it. Soil erosion models have been used for decades to better understand the transport processes and the interrelationships between land use and soil erosion. However, the effects of the above-described dynamics of landscape patterns and structures on soil erosion processes are so far poorly understood or only conditionally represented by soil erosion models. Direct measurements of soil erosion for larger catchments considering landscape patterns and structures are lacking. Thus, it remains largely unclear whether, why, and to what extent changing landscape patterns and structures affect the hydrological and sedimentological connectivity of agricultural landscapes.

The objective of the proposed project is to quantify and better understand these effects on erosion and deposition patterns and sediment input to water bodies. To this end, a classical erosion model will be compared with a data driven model from the field of machine learning and merged into a hybrid modeling approach. Therefore, the innovation potential of machine learning could be applied in soil erosion research by merging the capabilities of a conceptual erosion model with the capability of data driven machine learning modeling.

Finally, scenarios accounting for climate change and different pathways of future agricultural development will be defined and used for modeling to estimate impacts of changing land use patterns and structures on soil erosion, sediment discharge and lateral pollutant transport. Such landscape scale modeling also opens the possibility of identifying optimal land use patterns that reduce on-site and off-site damage from soil erosion.

  DYLAMUST at University of Tübingen

back to research overview