You are here:
A SUB-PIXEL ACCURACY ASSESSMENT FRAMEWORK FOR DETERMINING LANDSAT TM DERIVED IMPERVIOUS SURFACE ESTIMATES.
Citation:
Jennings, D B., D J. Williams, AND S T. Jarnagin. A SUB-PIXEL ACCURACY ASSESSMENT FRAMEWORK FOR DETERMINING LANDSAT TM DERIVED IMPERVIOUS SURFACE ESTIMATES. Presented at A Remote Sensing and GIS Accuracy Assessment Symposium, Las Vegas, NV, December 11, 2001.
Impact/Purpose:
Overarching Objectives and Links to Multi-year Planning
This research directly supports long-term goals established in ORD's multi-year research plans related to GPRA Goal 2 (Water Quality) and Long Term Goal WQ-2 Assessment of aquatic systems impairment. Relative to the GRPA Goal 2 Water Quality multi-year plan, this research will "provide tools to assess and diagnose impairment in aquatic systems and the sources of associated stressors" and "provide the tools to restore and protect aquatic ecosystems and to forecast the ecological, economic, and human health outcomes of alternative solutions" (Water Quality Long Term Research Goals 2 and 3).
Subtask 1 - Impervious Surface Evaluation
This subtask addresses the development of impervious surfaces estimators for local to regional scale assessments of watersheds and their landscape relationship to stream ecology. The amount of impervious surface area in a watershed is a key indicator of landscape change. As a single variable, it serves to integrate a number of concurrent interactions that directly influence a watershed's hydrology, stream chemical quality, and in-stream habitat. It is our working hypothesis that impervious surface area within a watershed, as an independently mapped predictor variable, can be used to generally track a range of watershed ecological parameters (e.g., NPS pollution, biological integrity, TMDLs) that are of concern to local, state and federal environmental managers. The specific objectives of this research are: 1) to quantitatively evaluate the varying remote sensing methods used in mapping impervious surfaces at multiple scales (local to regional), and 2) to relate the varying levels of impervious surface area in watersheds to the environmental condition of multiple water resource endpoints such as streamflow, temperature, and biota.
Subtask 2 -- Landscape Assessments and Evaluations of Best Management Practices: Watershed Demonstrations
Best Management Practices (BMP) encompass a range of strategies to reduce water pollution related to urban and agricultural activities. EPA, through Section 319(h) of the Clean Water Act [PL 92-500], provides grants to states to implement BMPs in areas with suspected or known water-quality problems. Grants for implementation of BMPs have not been tracked or monitored to document their effectiveness. Although effectiveness can be measured in many different ways, one straightforward but important measure is existence. Implementation of BMPs is a voluntary process and actual implementation is not always executed (Nowak 1992). The primary objective of this project is to assess the feasibility of using high-resolution aerial photography and other remotely sensed data to identify the existence of BMPs that were planned under the 319 program. An additional objective is to evaluate the effectives of BMPs implemented by examining monitoring data from about 5 sites in the OW National NPS monitoring system.
There are several potential benefits to determining the feasibility of using the aerial photography for identifying BMPs: 1) since BMP implementation is voluntary and some may not be implemented due to a variety of social and economic factors (Nowak 1992), remote detection of BMPs can provide data to estimate the ratio of BMPs implemented to BMPs planned; 2) remote detection of BMPs provides validation data that can be input into EPA's Grants Reporting and Tracking System (GRTS), and 3) remote monitoring of BMPs over time could be used to develop data on BMP lifespans, providing important data related to social- and cost-effectiveness.
Subtask 3 -- TMDL Non-point Source Assessment Tool
This subtask involves the development of a software tool to assess the potential risks of water bodies to exceed TMDL threshold values established by States. When completed, the tool will allow the user to evaluate watersheds over entire regions. The too
Description:
The amount of impervious surface in a watershed is a landscape indicator integrating a number of concurrent interactions that influence a watershed's hydrology. Remote sensing data and techniques are viable tools to assess anthropogenic impervious surfaces. However a fundamental problem exists with respect to impervious surface determination from coarse resolution data such as LANDSAT TM. The heterogeneous spatial nature of impervious
surfaces on the landscape, in concert with the sensor resolution, serves to produce "mixed pixels". This inherent processing problem forces an aggregation of multiple landscape features into general classes such as "high density residenfial", "low density residential", etc. The goal of this research is to ascertain the per pixel impervious surface percent associated with differing methodologies and categories of LANDSAT derived classification maps. The assessment strategy is based on the per pixel correlation of high-resolution data and geo-registered LANDSAT classification data. Metho& involving spatial aggregate and spatial disaggregate processing of raster data are utilized to assess the accuracy of evolving remote sensing classification techniques such as sub-pixel analysis and GIS fusion methods. In addition, these methods can be utilized within an accuracy assessment sampling framework to derive percent imperviousness related to general LANDSAT classification categories (e.g., National Land Cover Data) via post-processing analysis. Spatial aggregation techniques utilize impervious surface data derived from high-resolution . geo-spatial data sets that are resampled into lower resolution cells (e.g., I m to 30m). The high- resolution data is utilized to derive an impervious percent per given 30m cell coincident to the classified LANDSAT data. The classification of the aggregated high-resolution data may then be compared to the classification of the LANDSAT classified data. Spatial disaggregation techniques utilize LANDSAT classified impervious surface data sets that are re-sampled into higher resolution constituent cells (e.g., 30m to I m) while maintaining the same class attribute. These re-sampled high-resolution cells are then compared to a truth dataset derived from higher resolution data on a pixel-to-pixel level or by utilizing a sampling window if co-registration errors are a concern. For this study, we obtained several independent impervious surface and general land use/land cover classification datasets derived from LANDSAT TM data. Truth data was derived from large-scale aerial photography and rasterized county-level GIS parcel layers. Accuracy for either the aggregate or disaggregate approach may be derived from 1) a standard contingency matrix, 2) a correlation of coincident pixels or 3) a total impervious percent per geographic unit (e.g., watershed).