This study predicts the reason of unknown-caused fires that occurred in grassland in the east of Inner Mongolia, China.
GIS and logistic regression are used to build the predicting models. The causes of grassland fires were classified as
vehicle, production, living and lighting. The areas were divided into fired and unfired grid cells (500m*500m) with
spatial analysis, in order to determine the spatial factors and weather factors, such as the nearest distance to villages,
roads, fields etc. Logistic regression was used to build predictive models of the probability for each reason of grassland
fires. Four probabilities of each unknown-caused grassland fire were calculated and the maximum value expresses the
fire reason. The results show that natural fires are less than human-caused grassland fires and they can be used in fire risk
models and to support fire management decision-making. These methods would take advantage to the other grassland
fire studies, such as fire ecology, fire weather, fire cycle, etc.