Abstract: Karst geological hazards pose a significant challenge for the urban construction and underground space development and utilization in the Guangdong-Hong Kong-Macao Greater Bay Area, especially in Guangzhou and Shenzhen. Karst exploration generally involves identifying and assessing caves through a combination of drilling and geophysical information. In recent years, the cross-hole computed tomography (CT) geophysical method has been widely used in karst exploration in the Greater Bay Area due to its ease of operation and strong ability to obtain geological information. However, the accuracy of this method in identifying caves still needs to be quantitatively evaluated. This paper collected a large amount of data on karst drilling and exploration, and the accuracy of cross-hole CT karst identification was statistically analyzed using the model factor method. The results showed that this method could accurately detect the buried depth of the cave roof, floor and height, with an average error less than 5%. The predictive accuracy of the buried depths of the cave roof and floor has very low variability, only 5%, while the predictive accuracy of the cave height has medium variability, exceeding 35%. The accuracy stability of the cross-hole CT karst identification method is satisfactory and not affected by the factors such as CT method type, cave filling condition, emission and reception point distance, drilling type, cave roof thickness, drilling distance, and verification hole distance. This paper also conducted a simple correction of the current cross-hole CT method, which increased the average accuracy of the model by 4% and reduced the variability by 3% without increasing the computational complexity. Finally, the analysis confirmed that the model factors for predicting cave height follow a Weibull distribution. The research results can provide theoretical support for karst cave exploration and risk assessment in karst areas.

Key words: cross-hole CT method, karst identification, accuracy evaluation, statistical analysis, risk management