Grantee Research Project Results
Final Report: Assessment and Analysis of Ecosystem Stressors Across Scales Using Remotely Sensed Imagery Reducing Uncertainty in Managing the Colorado Plateau Ecosystem
EPA Grant Number: R825152Title: Assessment and Analysis of Ecosystem Stressors Across Scales Using Remotely Sensed Imagery Reducing Uncertainty in Managing the Colorado Plateau Ecosystem
Investigators: Weigel, Stephanie J.
Institution: Colorado State University , University of Wisconsin - Madison
EPA Project Officer: Packard, Benjamin H
Project Period: October 1, 1996 through September 30, 1999
Project Amount: $251,237
RFA: Ecological Assessment (1996) RFA Text | Recipients Lists
Research Category: Ecological Indicators/Assessment/Restoration , Aquatic Ecosystems
Objective:
The main objective of the project was to develop a standardized analytical algorithm for utilizing multiscale remotely sensed data in the systematic characterization of landscapes at the scale of an ecosystem extent. This involved the creation of ecosystem-extent mosaics of remotely sensed imagery that encompassed the study area, the Colorado Plateaus ecosystem region as defined by Omernik (1987). The focus of algorithm development was the evaluation of four different scale effects analysis techniques: local variance analysis, fractal analysis, global variance analysis, and variogram analysis. These techniques have been proposed as indicators of characteristic scales at which causal processes and factors affecting landscape ecosystem patterns are manifested. Different characteristic scales on the landscape were represented by creating a range of cell size resolutions from remotely sensed imagery for the region. Landsat Multispectral Scanner imagery obtained through the North American Landscape Characterization program was utilized. Pixel sizes were rescaled from the original 60-meter square cell size resolution to pixel sizes of 120, 240, 480, 960 and 1920 meters square. These multiscale data sets were evaluated using the different scale analysis techniques with the aim of determining appropriate techniques for multiscale landscape characterization.Summary/Accomplishments (Outputs/Outcomes):
The analytical algorithm for multiscale scale effects analysis was developed and is presented as Figure 1. This protocol outlines the steps that ecosystem managers/researchers may take in performing a multiscale landscape assessment of ecosystem processes using multiscale remotely sensed imagery. Both full-extent and focused subset-extent analyses are recommended, to most fully ascertain characteristic scales of landscape processes and appropriate scales for analysis.
An aim of the project was to provide a reasonable and feasible approach for ecosystem management teams and researchers to perform multiscale analysis across ecosystem extents. The analytical algorithm presents the steps necessary to perform this type of analysis. The methodologies chosen for rescaling and scale effects analysis are discussed below. They are effective methods and reasonably implemented in most image processing systems, however implementation involve some programming. For both of the scale effects methods recommended, basic code is available through the PI, and is included in the final report for the project submitted to the U S EPA.
Prior to scale effects analysis, imagery must be rescaled from the original resolution to create a multiscale data set (Step IV in the algorithm, Figure 1). The rescaling algorithm recommended is the dampened sine wave (DSW) algorithm developed by Ken McGwire (see McGwire, et al., 1993 for a brief description of the theory behind the algorithm). A detailed description and rationale for the algorithm is included as part of the project final report or available from the PI, as an ERDAS Macro Language file containing the code for implementing the program. The simulation of coarse-resolution image data from fine-resolution data sources requires a reasonable model for the spatial response of the simulated sensor. The model used in this project is preferable to a simple low-pass weighting scheme (averaging) that weights equally all of the cells that contributed to an output lower resolution. DSW is a more realistic simulation of the way a sensor actually works, and results in a series of multiscale images that simulate different resolution sensors having collected image data over the same landscape at the same date and time. Creation of such a data set reduces some of the confounding issues which may arise when trying to compare data obtained from different sensors at different resolutions of the same area, but at different dates and times, which affects things such as sun angle, haze, and other factors which may be difficult to normalize. Thus, the DSW rescaling algorithm allows for the differential weighting that occurs due to the way in which a sensor gathers data. However, if it is not possible to utilize the dampened sine wave methodology, a flat-field filter method that averages all input pixels could be utilized, and has been by many researchers doing multiscale research. Researchers should be aware, however, that the choice of rescaling methodology can affect the final results of subsequent scale effects analyses (Weigel 1996).
Four scale effects methodologies were investigated in this research: local variance analysis, global variance (multiscale) analysis, fractal analysis, and variogram analysis. Two of those methods, local variance analysis and fractal analysis, were assessed to be most effective and feasible, and are recommended for implementation for multiscale analyses for future investigations (Steps VI a and b in the algorithm, Figure 1).
The local variance method has been reported widely in the literature as a method for interpreting the spatial structure of images, and for selecting an appropriate scale of data (see Woodcock and Strahler, 1987, for an early application). Local variance is a texture analysis technique that measures the mean value of the standard deviation within a 3 x 3 moving window across the image pixel cells. The standard deviation of the nine values is computed, and the mean of these values over the entire image is taken as an indication of the local variability in the image. The reasoning behind the measure is that when the spatial resolution (pixel size) is considerably finer than the objects in an image scene, most of the measurements of variance will be low, since there will be high correlation between neighbors. When the objects in the scene become similar to the size of the pixels, likelihood of similarity with neighbors becomes higher, and local variance rises. As the size of pixels increase and many objects are found in a pixel, mixing occurs and the local variance decreases (Woodcock and Strahler, 1987). Thus, graphs of local variance versus resolution can be used to measure spatial structure in images, and to indicate dominant scales of patterns/objects in a scene, and appropriate scales of data.
The local variance algorithm was implemented in the ERDAS Imagine image processing software (ERDAS, Atlanta GA). The local variance model was derived from an existing Imagine model, and is implemented using the Model Maker tool in Imagine. Most image processing software should incorporate a type of texture analysis algorithm, which could be modified to perform local variance analyses.
Fractal analysis has been shown to be a useful tool in the characterization of landscapes using remotely sensed imagery (Bian and Walsh, 1993; Lam, 1990; Weigel, 1996; Qui, et al., 1999). Fractals are recognized as a robust method for understanding landscape complexity, especially across scales, since landscapes become more heterogeneous with changing scale of observation and measurement (Quattrochi, et al., 1997.) As was done in this analysis, the fractal dimension D can be derived from the pixel digital numbers in each band of remotely sensed imagery, and plotted versus spatial resolution (Step VIII, Figure 1). "Breaks" in scale are often observed in these plots, indicating possible scale-dependencies of landscape processes on the ground and/or scales of significant geographic interest.
The specific method employed to calculate fractal dimension in this project was the isarithm method, which has been characterized as compared to other fractal analysis methodologies by its robustness, accuracy and relative lack of sensitivity to input parameters (Emerson et al. 1999). This method is recommended for researchers using the fractal methodology for multiscale analyses of remotely sensed data. Briefly, each band of the remotely sensed image is considered as a surface, and the surface variability is measured using a walking-divider logic of the surface isarithms (this procedure is explained in Lam and De Cola, 1993, and in Qui, et al., 1999). The program used here was adapted from the ICAMS (Image Characterization and Modeling System) software developed as a NASA collaboration to "measure, characterize, and model multiscale remotely sensed data" (Quattrochi, et al., 1997). Other versions of isarithm programs are available in the literature. The ICAMS program was modified somewhat to allow for the large file sizes and for the double precision floating data type that was the result of the ASPR radiometric correction. Thus the program was run independent of the ICAMS structure, as a C program. This program version was developed with the programming assistance of Ionut Aron.
Ecosystem landscape processes occur across a range of scales, both in terms of the extent of the area impacted and the resolution and operational scale at which a process occurs. This research dealt with these different meanings of scale, by establishing a protocol for looking at landscapes using multiscale methodologies. The protocol addresses the full extent of a research study area as well as selected sub-regional study areas of interest. The assumption is that ecosystem managers/researchers will bring their understanding of a region and its diversity to the choice of sub-regional study areas (performed in Step V, Figure 1), to enhance the full extent analyses undertaken, and provide insight and opportunity for unique subareas of an ecosystem.
To look more closely at resolution and operational scale, the research first provides a rescaling methodology, which allows remotely sensed imagery to be resampled to lower cell size resolutions, i.e. to a coarser resolution. This has multiple implications in terms of actual studies: lower resolution data conserves storage space, which can be an issue over large geographic extents, and when using multispectral or hyperspectral imagery with many bands (layers) representing the same area. Lower resolution data sets may also be more representative of products that will be available from newer sensors, designed to assess land use and global change from a more global perspective. When data are resampled across a range of cell resolutions from higher to lower, a multiscale data set is created which establishes a number of different scale filters through which a landscape can be examined.
The operational scale of different processes examines the level at which a process is detected and manifested. For example, the operational scale of erosional processes may vary from that of the erosion from a devegetated construction site at the size of a residential lot, to the erosional processes that create large geologic features. Multiscale data allows both a visual representation of the different filters for detecting processes of interest (the series of multiscale images that are generated) as well as providing the input data sets for scale effects analysis methodologies, which work to identify points where critical scales or scale breaks in a data set may occur. This is exemplified by graphing scale effects indices (e.g. local variance values or fractal dimension D) versus resolution level. The resultant graphs may indicate breaks in scale or peaks where there are changes in the ways in which the landscape responds to the indices, at a particular resolution.
Of special interest were the different ways in which different landscape subsets of the Colorado Plateau ecoregion responded when assessed using the scale effects analysis methodologies evaluated. There were also differences detected in response of the different bands of the remotely sensed imagery. Since different wavelengths are often associated with different types of processes, researchers with particular interests in particular types of processes may wish to focus on a specific range of wavelengths to look for characteristic scales of landscape processes. The research protocol developed provides a framework within which to conduct that work. The selection of appropriate scales (data scales and/or research extents) can be elucidated using the procedures presented. The use of scale effects methodologies on multiscale data sets at both full-extent and on selected subsets provides ecosystem managers/researchers with a useful tool for evaluating the scales at which processes occur, stressors may manifest, and remediation may be most appropriate.
Figure 1. Analytical algorithm for multiscale ecosystem analysis using remote sensing and scale effects analysis techniques.
References:
Bian L, Walsh SJ. Scale dependencies of vegetation and topography in a mountainous environment of Montana. Professional Geographer 1993;45:1-11.
Emerson CW, Lam NS-N, Quattrochi DA. Multi-scale fractal analysis of image texture and pattern. Photogrammetric Engineering and Remote Sensing 1999;65:51-61.
Lam NS-N. Description and measurement of Landsat TM images using fractals. Photogrammetric Engineering and Remote Sensing 1990;56:187-195.
Lam NS-N, De Cola L. Fractals in Geography. PTR Prentice Hall, Englewood Cliffs, NJ, 1993.
McGwire K, Friedl M, Estes JE. Spatial structure, sampling design and scale in remotely-sensed imagery of a California savanna woodland. International Journal of Remote Sensing 1993;14(11):2137-2164.
Omernik JM. Ecoregions of the conterminous United States. Annals of the Association of American Geographers 1987;77(1):118-125.
Quattrochi DA, Lam NS-N, Qui H-L, Zhao W. Image characterization and modeling system (ICAMS): a geographic information system for the characterization and modeling of multiscale remote sensing data. In: Quattrochi DA, Goodchild MF, eds. Scale in Remote Sensing and GIS, CRC Press-Lewis Publishers, New York, 1997.
Qui H-L, Lam NS-N, Quattrochi D, Gamon JA. Fractal characterization of hyperspectral imagery. Photogrammetric Engineering and Remote Sensing 1999;65:63-71.
Woodcock CE, Strahler AH. The factor of scale in remote sensing. Remote Sensing of Environment 1987;21:311-332.
Weigel SJ. Scale, resolution and resampling: representation and analysis of remotely sensed landscapes across scale in geographic information systems. Ph.D. Dissertation, Louisiana State University, Baton Rouge, LA, 1996.
Journal Articles:
No journal articles submitted with this report: View all 4 publications for this projectSupplemental Keywords:
ecosystem, scaling, integrated assessment, scale effects analysis, Landsat, NALC., RFA, Scientific Discipline, Geographic Area, Ecosystem Protection/Environmental Exposure & Risk, Ecology, Ecosystem/Assessment/Indicators, Ecosystem Protection, exploratory research environmental biology, Chemical Mixtures - Environmental Exposure & Risk, State, Ecological Effects - Environmental Exposure & Risk, Ecological Effects - Human Health, Environmental Monitoring, Ecological Risk Assessment, Ecology and Ecosystems, Ecological Indicators, ecological exposure, analytical algorithm, multi-scale biophysical models, remote sensing, scaling, ecosystem assessment, variance analysis, Colorado Plateau ecosystem, environmental stressor, multiple stressors, ecological assessment, ecological impacts, assessment methods, environmental stress, landscape characterization, fractal analysis, Colorado (CO)Relevant Websites:
The project is included in the Environmental Health Advanced Systems Laboratory (EHASL) Web site at http://ehasl.cvmbs.colostate.edu . The site currently is under revision, but the link to the project can be found at http://ehasl.cvmbs.colostate.edu/remote . The PI no longer works at EHASL, so future updates to the project page would occur at another site.Progress and Final Reports:
Original AbstractThe perspectives, information and conclusions conveyed in research project abstracts, progress reports, final reports, journal abstracts and journal publications convey the viewpoints of the principal investigator and may not represent the views and policies of ORD and EPA. Conclusions drawn by the principal investigators have not been reviewed by the Agency.