Authors: Kimberly Blazzard*, East Tennessee State University, Ingrid Luffman, East Tennessee State University, T. Andrew Joyner, East Tennessee State University
Topics: Hazards, Risks, and Disasters, Geographic Information Science and Systems, Geomorphology
Keywords: Climate, sinkholes, GIS, R, spatial statistics, regression, MaxEnt
Session Type: Poster
Start / End Time: 1:20 PM / 3:00 PM
Room: Napoleon Foyer/Common St. Corridor, Sheraton, 3rd Floor
Presentation File: No File Uploaded
Sinkholes often develop shortly after periods of heavy rain and may be connected to larger, macro-climatic patterns. For example, recent research indicated that a 200-m-long collapse zone, in Guanxi, China, was preceded by a year-long drought followed by a heavy, single day rain event (469.8 mm total). This study compared two techniques to examine possible relationships between Tennessee sinkholes and climatological patterns: a logistic regression created with ESRI ArcMap and R, and a Maxent “species distribution” model. Climate normals averaged over the period 1960-2000, available from WorldClim, were used for analysis and included annual minimum temperature, annual maximum temperature, annual average temperature, average temperature range, average wind speed, average solar radiation, average water vapor pressure, and annual precipitation. Tennessee sinkhole records, available through the University of Tennessee, were recorded from Tennessee topographic quadrangles. Results showed highly significant (p < 0.001) correlations between sinkhole events and precipitation, maximum temperature, wind speed, and solar radiation. Regression results indicate that some sinkhole formation variability could be explained by these climatological patterns combined with other long-term variables. The variable with the strongest correlation with sinkhole formation was wind speed (-.217 correlation coefficient); wind speed was determined to be a good surrogate variable for topography. The logistic regression produced an overall better estimate of sinkhole risk locations than the MaxEnt model, although both models produced similar output. The models correctly assigned 96.2% (logistic regression model) and 98.4% (MaxEnt model) of sinkholes in the testing dataset to medium, high, or very high risk zones.