Bayesian Networks and GIS-based analysis for flood risk assessment in agriculture during heavy rainfall events

Abstract

The increasing frequency and intensity of heavy rainfall events due to climate change pose a growing risk to agriculture, and its prediction remains a substantial challenge for the scientific community. Leveraging the availability of high-quality open geospatial data provided by governmental institutions and existing probabilistic risk models, this study proposes a systematic approach for analyzing the risks associated with heavy rainfall events in agriculture. The objective is to develop and test a methodology that integrates open geospatial data, considers uncertainties, and conducts scenario analyses based on the historical heavy rainfall event of June 2, 2024, in the Rems-Murr district, Germany. Initially, variables such as Rainfall Intensity, Temperature, Land Use (from land use maps), Soil Type, Soil Moisture, Slope, Elevation, and River Discharge—obtained from official institutions—along with variables like Proximity to River, Road Density, and Proximity to Forest (derived from GIS analysis), were conceptually integrated into a Bayesian Network (BN). This integration was based on theoretical foundations and quantified using conditional probability distributions (CPD). The results demonstrate that the methodology combining BN and GIS analyses, along with scenario analysis, sensitivity analysis, optimization, and model validation, was successfully applied to the wider area of the Rems-Murr district. When tested on the heavy rainfall event of June 2, 2024, in the study area of Rudersberg, it provided qualitatively convincing results for flood risk assessment in agriculture. Validation in Rudersberg yielded a Root Mean Square Error (RMSE) of 23%. The methodology was also successfully applied across regions in Miedelsbach, where validation with official data collected by the county resulted in an RMSE of 30%. These findings indicate that the methodology is applicable not only within the study area but also across different regions. It is recommended to improve the model by incorporating additional variables such as surface parameters, roughness values, and drainage systems to improve accuracy. Furthermore, integrating meteorological forecasts could provide a basis for forward-looking risk predictions.