Predicting impacts of changing land use/cover on streamflow in ungauged watersheds

摘要

Southeastern US has experienced substantial land use/cover (LULC) changes over the past 200 years with the following LULC transition: Forest → Agriculture → Forest → Urban. Currently, Southeastern US has the highest population growth and urbanization rates in the US and this trend is expected to continue. Understanding the potential implications of these LULC changes, specifically urbanization, on streamflow hydrology is of paramount importance. The focus of this study was comparing two conceptually different modeling approaches in predicting (ⅰ) streamflow and (ⅱ) effect of LULC modifications on streamflow in ungauged watersheds. The SWAT model (a quasi-process-based watershed model) and ANN (black box type, lumped model) were applied to 10 watersheds in West Georgia. The LULC in these watersheds varied from heavily forested, mixed to urban. The ANN model employed in this study is actually a hybrid model that combines the strengths of data driven approaches with physical process. Soil data from SURRGO, LULC data from 2006 NLCD, and daily radar precipitation and temperature data for the 2003-2006 period were used as inputs to both models. Separate SWAT models were developed for forested and urban watersheds to better understand the relationship between LULC and streamflow. The calibrated models were then applied to two nearby watersheds with similar characteristics, one urban and one forested, to predict daily streamflow. The ANN model was developed by training the ANN algorithm with streamflow data from all the watersheds except for the two test watersheds. The modeling results indicated that calibrated feed-forward ANN model was capable of predicting streamflow in test watersheds with higher accuracy compared to the SWAT.