Influence of buffer distance on environmental geological hazard susceptibility assessment

Previous researches seldom studied the selection of buffer distance between geological hazards (positive samples) and non-geological hazards (negative samples), and its reasonable selection plays a very important role in improving the accuracy of susceptibility zoning, protecting the environment and reducing the cost of hazard management. Based on GIS technology and random forest (RF) and frequency-ratio random forest (FR-RF) models, this study innovatively explored the influence of randomly selected non-geological hazard samples outside different buffer distances on the susceptibility evaluation results, with buffer distances of 100 m, 500 m, 1000 m and 2000 m in sequence. The results show that through the confusion matrix and ROC curve test, the accuracy of the model increases first and then decreases with the increase of buffer distance. Both RF and FR-RF models have the highest accuracy when the buffer distance is 1000 m, and the accuracy of the RF model is generally higher than that of the FR-RF model under the same buffer distance. Similar attribute values of positive samples and randomly selected negative samples or “extreme” attribute values of negative samples are the main reasons for the differences in evaluation results of different buffer distances. According to the weight analysis of causative factors, the distance from road, the distance from river and the normalized vegetation index (NDVI) are the main factors affecting the occurrence of hazards. The high and very high susceptibility areas in the study area are mainly distributed on both sides of roads and water systems, which are the key areas for hazard prevention and reduction. The HMC of RF-1000m decreased by 3.55% on average compared with other models. The results of this study improve the accuracy of geological hazard susceptibility assessment, maintain the safety of ecological environment, and provide a scientific basis for the selection of buffer distance index in local and surrounding areas in the future.