INTELLIGENT POST-FIRE HYDROLOGIC AND GEOMORPHIC LANDSCAPE MODELING

Few studies attempt to model the range of possible hydrologic and geomorphic responses following rainfall on a burned basin. Some reasons are the sparseness of data and the complex-coupled, nonlinear, spatial, and temporal relationships among response, initiation, and process variables. In this paper, an unsupervised artificial neural network (ANN) model is developed and used to project multidimensional data from 606 burned basins in the western United States onto a two-dimensional grid called a self-organized map (SOM). The sparsely populated data set included independent numerical landscape categories (weather, land surface form, geologic texture, and post-fire condition), independent landscape classes (bedrock geology and state name), and dependent categorical classes (flooding, sediment flows, landslides, and debris flows). Based on pattern analysis visualized in SOM-based component planes and U-matrix, the relationships among explanatory and response variables were extracted and interpreted. Stochastic cross-validation of the ANN model demonstrated its ability to provide globally unbiased predictions of likely initiation processes (runoff, landslide, and runoff-and-landslide combination) and responses (debris flows, floods, and no events) following rainfall on burned basins, and to quantify the degree of uncertainty in which false positives and negatives occurred. In addition, k-means clustering of the SOM topography identified eight conceptual post-fire regional models. These models represent hypotheses of coupled and nonlinear post-fire hydrologic and geomorphic landscape interaction. These conceptual models provide a basis for future development of regional predictive models using empirical, numerical, and other intelligent discovery techniques.