Events Calendar

PREDICTING DEBRIS YIELD USING ARTIFICIAL INTELLIGENCE MODELS

Zhiqing Kou, Ph.D. Candidate
Astani Dept. of Civil and Environmental Engineering Abstract:
Artificial Neural Network is a very powerful computational tool for modeling very complicated and highly nonlinear problems in various fields. In this study, it is first applied to estimate accumulated debris yield in 14 debris basins within Los Angeles County, California from 1984 to 2003 as a result of a series of storm events from watersheds partially or totally burned by wildfires. ANN models achieve very satisfactory modeling results when compared to a statistical model. The ANN technique is then applied to forecast unit debris yield resulting from a significant storm event collected from 36 small debris basins from 1938 to 1983 within the county. The same unit debris yield data is simulated by another two artificial intelligence models, Adaptive-Network-Based Fuzzy Inference System (ANFIS) and Generalized Dynamic Fuzzy Neural Network (GD-FNN) model. In addition to four basic input parameters: drainage area, watershed relief ratio, maximum one-hour rainfall intensity, and fire factor, six watershed morphological parameters including elongation ratio, drainage density, hypsometric index, total stream length, mean bifurcation ratio, and transport efficiency factor are included as input parameters and their relative importance are determined through sensitivity analysis. ANN models are also developed for modeling unit debris yield at 80 small debris basins which are classified into five groups based on the slopes of their upstream collection watersheds: mild slope, steep slope, steeper slope, extreme steep slope, and the steepest slope. In addition to four aforementioned basic input parameters, three soil properties such as soil erodibility factor, permeability rate, and liquid limit are included as input parameter one by one to study their impact on the simulation. Unit debris yield collected from large watersheds with area between 10 and 25 mi2, between 25 and 50 mi2, and between 50 and 200 mi2 are also simulated by neural networks. The modeling results show ANN models are able to reproduce most unit debris yield very close to their measured values and the accuracy of unit debris yield estimated by ANN models is significantly higher than those obtained from ANFIS, GD-FNN model, and empirical equations developed by US Army Corps of Engineers.

Location: Kaprielian Hall (KAP) - 209

Audiences: Everyone Is Invited

Contact: Evangeline Reyes