Developing Rainfall Intensity-Duration-Frequency Curves for Alabama under Future Climate Scenarios Using Artificial Neural Networks

Abstract

Hydrologic design of water management infrastructures is on the basis of specific design storms derived from historical rainfall events available in the form of intensity-duration-frequency (IDF) curves. However, it is expected that the frequency and magnitude of future extreme rainfalls will change due to the increase in greenhouse gas concentrations in Earth’s atmosphere. This study evaluated potential changes in current IDF curves for Alabama under projected future climate scenarios. Three-hour precipitation data simulated by five combinations of global and regional climate models were temporally downscaled using artificial neural networks (ANNs). A feed-forward, back-propagation model was developed to estimate maximum 15-, 30-, 45-, 60-, and 120-min precipitation. The results were compared with disaggregated rainfall derived using a stochastic method. Comparison of these two methods indicates that the ANN model provides superior performance in estimating maximum rainfall depths, whereas the stochastic method tends to under-predict maximum rainfall depths. Developed IDF curves indicate that future rainfall intensities for the events with duration <2 h are expected to decrease by 33–74% compared with those of current events when the ANN model is used, whereas large uncertainty exists in the projected rainfall intensities of longer-duration events. This result was independent of the temporal downscaling method used.