You are here:
Evaluation of a Moderate Resolution, Satellite-Based Impervious Surface Map Using an Independent, High-Resolution Validation Dataset
Citation:
Jones, J. W. AND S. JARNAGIN. Evaluation of a Moderate Resolution, Satellite-Based Impervious Surface Map Using an Independent, High-Resolution Validation Dataset. JOURNAL OF HYDROLOGIC ENGINEERING. American Society of Civil Engineers (ASCE), Reston, VA, 14(4):369-376, (2009).
Impact/Purpose:
Increases in watershed impervious surface area (ISA) have long been known to affect the quantity and timing of watershed runoff (Hammer 1972; Dunne and Leopold 1978; Jennings and Jarnagin 2002). Studies of aquatic ecology and water quality have shown that increases in watershed ISA adversely impact stream biological integrity (Schueler 1994). Given these linkages, ISA has been proposed (Arnold and Gibbons 1996) and widely adopted (CWP 2003) as an important, integrative environmental indicator. While popular for this purpose, periodically mapping and quantifying ISA over large areas is expensive.
Description:
Given the relatively high cost of mapping impervious surfaces at regional scales, substantial effort is being expended in the development of moderate-resolution, satellite-based methods for estimating impervious surface area (ISA). To rigorously assess the accuracy of these data products high quality, independently derived validation data are needed. High-resolution data were collected across a gradient of development within the Mid-Atlantic region to assess the accuracy of National Land Cover Data (NLCD) Landsat-based ISA estimates. Absolute error (satellite predicted area-- “reference area”) and relative error (satellite (predicted area-- “reference area”)/ reference area”) were calculated for each of 240 sample regions that are each more than 15 Landsat pixels on a side. The ability to compile and examine ancillary data in a geographic information system environment provided for evaluation of both validation and NLCD data and afforded efficient exploration of observed errors. In a minority of cases, errors could be explained by temporal discontinuities between the date of satellite image capture and validation source data in rapidly changing places. In others, errors were created by vegetation cover over impervious surfaces and by other factors that bias the satellite processing algorithms. On average in the Mid-Atlantic region, the NLCD product underestimates ISA by approximately 5%. While the error range varies between 2 and 8%, this underestimation occurs regardless of development intensity. Through such analyses the errors, strengths, and weaknesses of particular satellite products can be explored to suggest appropriate uses for regional, satellite-based data in rapidly developing areas of environmental significance.