Annual Evapotranspiration of a Forested Wetland Watershed, SC

Abstract

In this study, hydro-meteorological data collected from 1964 to 1976 on an approximately 5,000 ha predominantly forested coastal watershed (Turkey Creek) at the Francis Marion National Forest near Charleston, SC were analyzed to estimate annual evapotranspiration (ET) using four different empirical methods. The first one, reported by Zhang et al. (2001), that takes into account annual precipitation, potential ET (PET), and a vegetation water-use factor. The second method by Lu et al. (2003) uses annual rainfall, elevation, latitude and forest cover. The third method by Turner (1991) uses annual rainfall, coverage of the watershed by forest and non-forest vegetation. The fourth method by Calder and Newson (1979) uses annual rainfall and Penman PET for the grass vegetation, actual forest canopy cover, interception fraction, and fraction of the wet days. Results from each of these methods were compared with the measured water balance in which annual ET is a difference of measured annual rainfall and stream flow. The study period included years with annual rainfall varying from 1853 mm (wet) to 1020 mm (dry), typical to the Southeastern coastal plain. The 13-year measured mean annual ET was 983 mm and the annual ET remained to be near PET (>90% of average Thornthwaite PET of 1079 mm) for the years exceeding the long-term average rainfall and/or the years with just below the average but with the wet antecedent year. Years with consistently below average annual rainfall yielded annual ET equivalent to 80% or less of the annual estimated PET. Based on the statistical evaluation, Turner method yielded the best estimates with a mean annual ET of 974 mm (± 116 mm) and a Nash-Sutcliffe “E” coefficient of 0.64, followed by the Lu et al and Zhang et al methods. The mean annual (MAE) and mean absolute annual (MAAE) errors for both the Turner and Lu et al methods were less than 0.5% and 6.6%, respectively. The Calder-Newson method performed poorest (E = -0.29) among those evaluated. The highest overprediction error (<16%) in all methods, except for the CALDER, was observed in the second of the two consecutive dry (lower than near average rainfall) years, as expected, because none of these methods takes the antecedent soil water storage conditions into account. However, when comparing the mean annual ET over the 13-year period, there was no difference in estimates among the methods although all of them underestimated the measured. Results of assessing the impacts of reduction in forest cover on mean annual runoff using Turner and Lu et al methods indicated an increase of as much as 62% runoff as a result of removal of 90% forest cover on the study watershed. Although these empirical methods can be linked with GIS databases for effectively conducting “what-if” scenario analyses of land use changes, further research is warranted to assess their applications with data from other sites in the region, and to compare their utility relative to process-based ET measurements and water balance models.