Journal: Remote sensing of Environment, 2003, Vol. 86, 150-161
Authors: Emma Underwood, Susan Ustin, and Deanne DiPietro
Reviewed by: Vandana Vandanapu
Table of Contents
Research Objectives ----------------------------------------------------- 6
Study site -------------------------------------------------------------------7
Methods -------------------------------------------------------------------- 9
Nonnative species are a current focus of interest of ecologists, biological conservationists and natural resources managers due to their rapid spread, threat to biodiversity and damage to ecosystems. Controlling and managing invasives requires new methods to map and monitor their spread. Remote sensing can aid land manager’s efforts to control and monitor invasive plants by providing detailed information on their location and extent of spread. While digital multiband remote sensing and aerial photography have been available for many years, newer detector technologies have made it possible to accurately acquire a detailed laboratory-like spectrum of each pixel in an image from space. This study used NASA’s Advanced Visible Infrared Imaging Spectrometer (AVIRIS), a 224-band instrument with nominally 10 nm contiguous bands over the 400-2500 nm range for mapping invasive species iceplant (Carpobrotus edulis) and jubata grass (Cortaderia jubata) in Vandenberg Air Force Base, California. This research evaluated several data processing (minimum noise fraction (MNF), continuum removal, and band ratio indices) and classification methods and compared their success in determining the spatial extent of these invasive plants. The MNF and band ratios techniques were able to accurately identify new infestations into native habitats and Continuum Removal method is an efficient way of characterizing presence / absence of iceplant.
Keywords: AVIRIS, mapping, nonnative species, iceplant, jubata grass.
The increasingly rapid redistribution of the world’s species by humans and consequent detrimental invasion of ecosystems by alien species is one of the most serious and challenging threats to the world’s biodiversity, human health, and economy ever faced. An estimated 50,000 non-native species exist in the United States, and the invaders among them cause major environmental damages and losses adding up to more than $138 billion per year (Pimentel et al., 1999). Exotic plants are invading approximately 700,000 hectares of wildlife habitat each year. After habitat loss due to land-use change, biological invasion is one of the major contributors to local and global loss of biodiversity, causing extinction through competition, predation, hybridization, and habitat alteration (D’Antonio, 1997). The phenomenon is so wide-spread and has such far-reaching effects that it may be considered a significant component of global change (Vitousek et al., 1996).
It is now widely recognized that invasions by non-native plants present a serious ecological threat to the already fragmented and greatly reduced native ecosystems of California (Bossard et al., 2000; Barbour et al., 1993). The number of alien plant species in California is estimated at 1,025, representing 17.5% of the flora (Rejmánek et al., 1994). In response to the expanding ranges and increasing damage done by invasive and noxious non-natives, control of invasive species has become a priority for environmental management and an integral component of many habitat conservation efforts.
Reducing the impacts to local ecosystems and biodiversity caused by alien species and employing restoration and other remedy actions has become a trend in conservation (Stein & Flack, 1996). To better understand the status and to support researchers and decision makers to develop strategies and remedies for this threatening problem, it is necessary to obtain accurate spatial information and the progression about the invasions of alien species into native eco community. A key requirement for the effective management of invasive plants is the ability to identify, map, and monitor invasions. Hand-mapping in the field or from aerial photos are techniques commonly used in eradication efforts, but these methods are labor intensive and limited. Hand mapping from field observation requires access to the site from the ground, a prospect that is not always practical, safe, or timely, especially on an active military base.
Unlike field- based investigations, remote sensing provides an timely and economical approach for discriminating invasive plant species from local botanic community, especially in a large - scale investigation. In contrast to field-based surveys, imagery can be acquired for all habitats, over a much larger spatial area, and in a short period of time. Consequently, researchers have sought to exploit unique phenological, spectral, or structural characteristics of the nonnative species in the image to distinguish them from the mosaic of species around them.
Until relatively recently, aerial photographs and multispectral satellite images are the primary sources of remote sensing applications to vegetation mapping and have attained mixed success (for example, Lins et al., 1996; McCormick, 1999). In Aerial photography techniques the problem species is usually mapped by taking advantage of a characteristic such as flower or bract color that distinguishes it spectrally from surrounding plants in a scene. Everitt and Deloach (1990) mapped Chinese tamarisk using aerial photography when the plant was in its yellow phase just prior to leaf-drop in the fall. However, the major disadvantage of this approach is that it relies on the nonnative plant possessing visually detectable unique characteristics as well as requiring extensive manual labor for processing.
In contrast, the use of digital multispectral imagery offers the opportunity for automated image processing, access to recent historical data for time series analyses, and large spatial coverage. The consideration of spatial resolution is also very important in detection and mapping of individual species. It has been found that LANDSAT Thematic Mapper and SPOT data, with ground resolution of 30 meters and 20 meters, respectively, are not generally considered useful for mapping at the species level (Carson et al., 1995) unless stands of the weed are both large enough to fill a pixel and strikingly different from surrounding vegetation, as in the case of false broom weed mapped by Anderson et al. (1993) in rangelands in south Texas. Because these types of data can provide only limited spectral information, they may not be able to produce results with high quality and confidence, especially when dealing with detection and mapping tasks down to the species level (Chen et al., 2003).
Fortunately, the availability of hyper spectral imagery provides researchers an opportunity to pursue more complex analysis. Hyper spectral imagery consists of tens to hundreds of contiguous spectral bands therefore can provide more complete coverage of spectral information about targets. Previous studies have demonstrated the possibility to perform species - level vegetation classifications using hyper spectral data (Cochrance, 2000; Laba et al., 2003; Schmidt & Skidmore, 2003). AVIRIS (airborne visible/infrared imaging spectrometer) is the principal hyper spectral instrument now in use. It is an across-track scanner that collects radiance measurements in 224 contiguous bands approximately 10nm wide (JPL, 2002). The range of the electromagnetic spectrum sampled is approximately 400 – 2500nm, corresponding to the visible and near to short-wave infrared regions. With increased spatial and, more critically, fine spectral resolution, offers an enhanced potential for mapping invasive species. Because of the large number of wavebands (224), image processing is able to capitalize on both the biochemical and the structural properties of the target invader.
The objectives of this research were:
to map the spatial location, distribution and abundance of 2 nonnative plants i.e., ice plant (Carpobrotus edulis) and jubata grass (Cortaderia jubata) in VAFB, California.
to test the efficacy of AVIRIS imagery for nonnative plant mapping.
to compare three techniques for processing the imagery: minimum noise fraction (MNF), continuum removal, and band ratio indices. These three processes were compared for their ability to delineate the spatial extent and the density of ice plant and jubata grass and also to critically evaluate the relative ease of processing and repeatability of each method.
The information collected through this effort will:
increase the ability of resource managers to analyze and prioritize invasive plant management needs, enhancing the time and cost-effectiveness of management actions;
serve as a baseline for long-term monitoring, assist with the evaluation of changes in alien plant populations over time and detecting new infestations; and
serve as a critical tool for increasing public and political awareness and education on invasive plant issues.
In addition, the data collected through this study will provide the basis for an invasive plant management plan for Vandenberg Air Force Base, California.
The focal area for this study is Vandenberg Air force Base (VAFB) located along the central coast of California (Fig 1). It is the third largest Air Force Base in the nation, encompassing 39,800 ha, and for nearly half a century it has served as a launch and test site for medium- to long-range ballistic missiles, as well as government and commercial satellites. The vegetation is diverse and characteristic of central California coastal habitats (Holland, 1986).
Fig. 1. Vandenberg Air Force Base.
Of the 836 vascular plants documented at VAFB, almost a quarter is invasive species. In particular, C. edulis and C. jubata have successfully invaded two native community types: coastal dune scrub community and maritime chaparral (Keil & Holland, 1998). Hence the focus of this research is the encroachment of ice plant and jubata grass into these native communities and specifically on the ability of AVIRIS to identify pixels of different densities of these species.
Ice plant (Carpobrotus edulis)
The California Exotic Pest Plant Council's (CalEPPC) List of Exotic Pest Plants of Greatest Ecological Concern in California (1999) rated C.edulis as an A-1 species (The Most Invasive Wild land Pest Plant: Widespread). Carpobrotus edulis (Fig 2) has been invading native and non-native plant assemblages in California since its introduction in the early 1900s.This succulent perennial was introduced from South Africa to United States in the early 1900s for erosion control. The fleshy indehiscent fruits are widely dispersed by several native animals .As a result, Carpobrotus edulis has moved away from areas where it was planted and is invading a variety of coastal plant communities through out the pacific United States. D’Antonio & Mahall (1991) demonstrated that the invasive perennial Carpobrotus edulis can directly compete with native coastal California shrub species for soil resources. The species’ success is due to its tolerance of a range of soil moisture and nutrient conditions, and utilizing a number of mammals for seed dispersal (D’Antonio, 1993). Ecological impacts of ice plant include aggressive competition with native species, such as Tidestrom’s lupine (Lupinus tidestromii), destabilizing native dune communities and modifying soil pH (Moss, 1990). Economic impacts stem from time and financial costs associated with both manual and mechanical controls.
Fig 2: Ice plant (Carpobrotus edulis)
Jubata grass (Cortaderia jubata).
Another invasive plant with the potential to
significantly alter mediterranean-type ecosystems
in California is Cortaderia jubata (jubata grass).
C. jubata is a large perennial tussock grass native
to the Andean regions of Ecuador, Peru and Bolivia.
In the later half of the 19th century it was introduced
into California for use as a landscaping ornamental
(Lambrinos, 2000). It has subsequently escaped from
cultivation and expanded in coastal habitats along
the Pacific coast. Jubata grass poses a significant
threat to Mediterranean ecosystems because of its
prolific Wind dispersed seeds, tolerance of a broad
range of habitats, and its competitiveness for light,
moisture, and nutrients (Cowan, 1976).
Fig. 3. Jubata grass (Cortaderia jubata)
1. Description of fieldwork and GPS data collection
The GIS database included topographic, vegetation, land use history and road layers. Sampling was undertaken in five community types identified a priori: intact coastal dune scrub, intact maritime chaparral, iceplant invaded scrub, iceplant invaded chaparral, and chaparral invaded by jubata grass. Ecological data was collected at three scales: plot, site and community, including % cover by species, canopy height, disturbance type and size, and soil characteristics. GPS points (Trimble Pro-XRS, Trimble Navigation, Inc.) were made in plot centers and polygons for iceplant, jubata grass, and intact community types. Field measured reflectance spectra of the dominant native species, invasives, and soils were acquired coincident with AVIRIS using a GER 2500 (Geophysical Research, Corp.) over 400-2500 nm. Owing to almost continual coastal fog during the period around the over flight, only 80 individual spectra of target species at VAFB were acquired. AVIRIS data were acquired by the NOAATwin Otter aircraft on September 9, 2000 at 3810 m, providing a nominal pixel resolution of 4.5 m.
Fieldwork at VAFB
2. Data processing techniques
AVIRIS data was calibrated to surface reflectance using the atmospheric correction program, ACORN (Analytical Imaging & Geophysics). Image analyses were performed using ENVI (Environment for Visualizing Images; Research Systems, Inc.). The AVIRIS scenes were spectrally and spatially subset for georeferencing. Noisy bands were removed and masks were created for vegetated areas (NDVI>0.2).
A supervised maximum likelihood classification was performed on the results of each of the three processing techniques (MNF, continuum removal, and band ratio indices).
(a) Minimum Noise fraction classification
A “Minimum Noise Fraction” (MNF) Transform is used to reduce the number of spectral dimensions to be analyzed. The MNF transformation is a linear transformation related to principal components that orders the data according to signal-to-noise-ratio (Green et al., 1988). It can be used to determine the inherent dimensionality of the data, to segregate noise in the data, and to reduce the computational requirements for subsequent processing (Green et al., 1988; Boardman and Kruse, 1994). As a standard image processing technique, the authors were interested in assessing how well it performs for identifying the focal species.
(b) Continuum removal for water bands
The continuum removal technique isolates spectral features and standardizes reflectance across the liquid water absorption features so that they may be intercompared (Clark & Roush, 1984). In the continuum removal technique, a spectral absorption feature is selected from pixel profiles and normalized by fitting a curve (or continuum) across the profile as if to remove the "valley" created by the feature. The area of the curve is then calculated and the depth of the absorption features compared (Research Systems, 1999). The values were then fed into an unsupervised classifier for grouping into classes. In this study, a water absorption feature 15 around 985 nm (between the wavelengths 924–1061 nm) and 18 around 1184 nm (1108– 1254 nm) was used based on the hypothesis that plant water content would be a distinguishing feature for Ice plant. Fig 4 shows a comparison of the reflectance spectra of pixels having two different densities of ice plant compared to a pixel of intact coastal scrub. The graph shows two distinctive absorptions for dense ice plant centered on 985 nm and 1184 nm wavelengths. The use of the water absorption feature at 985 nm to map the ice plant resulted in the identification relatively well than the 1184 nm feature due to the background atmospheric water vapor absorption at 1240 nm.
Fig. 4. Reflectance spectra showing absorption in the water band.
(c) Band ratio indices
Remote sensing analysts have found that a wealth of information can be extracted from the spectral bands by multiplying or dividing the reflectance values of one band by those of another (i.e., creating a band ratio). The resulting band is called an index. Numerous vegetation indices have been developed and are routinely used by remote sensing scientists. Perhaps the most common vegetation index is the Normalized Difference Vegetation Index (NDVI) developed by Rouse et al. (1974). The NDVI is a band-ratio index that utilizes the characteristics of the visible red band and the near infrared band to provide a relative measure of green biomass and chlorophyll.
This study investigated the use of selected vegetation indices (NDVI, water absorption, greenness, and pigment properties) to use in the classification, which emphasize the biochemical and the biophysical properties of the vegetation contained in physiologically important bands.
A visual comparison of the three classification results (Fig.5) shows that the iceplant is clearly distinguished in all images with a similar spatial configuration: densest adjacent to the coastline and tapering off eastwards. An essential component of any classification routine is the assessment of the accuracy of the results obtained. One of the most common ways of determining accuracy is to use a confusion matrix (Fuentes et al., 2001). This approach uses pixels from the classified image and checks their labels against a reference data source or “ground truth.” In this method, accuracy can be expressed in three ways: (1) producer’s accuracy or omission errors; this refers to the probability (%) that the correct class has been identified on the basis of the ground truth reference; (2) user’s accuracy or commission errors; this is the probability that a pixel of a class has been classified correctly based on the total number of pixels classified as that class; and (3) overall accuracy or the total number of image pixels classified correctly. Another measurement of map accuracy is the kappa coefficient, which is derived using the elements of the error matrix. The kappa coefficient gives an estimate of overall accuracy based on both omission and commission errors (Richards and Jia, 1999).
Fig 5: Results from mapping ice plant and jubata grass at VAFB.
The overall accuracy results showed that the supervised classification of the MNF results performed best. [MNF=76.2% (kappa = 0.70), continuum removal classification = 54.9% (kappa = 0.44) and band ratio technique =58.8% (kappa = 0.49)]. In contrast, the users’ accuracy of the MNF was the lowest, 83%, compared to 94% and 89% for the band ratios and continuum removal, respectively. This implies that while the MNF is able to produce a more accurate map of multiple vegetation classes, the continuum removal and the band ratio techniques are better suited for detecting species with distinct characteristics, such as iceplant, with its distinctive succulent characteristics.
A key requirement for the effective management of invasives is to be able to delineate both the spatial extent and the severity of infestation, which are captured to different degrees by the processing methods (MNF, Continuum removal and Band ratio indices). The author evaluated these three approaches in terms of accuracy, logistics of processing, and ease of interpretation, which are all necessary considerations for management.
A confusion matrix was performed using only ROIs collected from the intact scrub and also areas in which iceplant has invaded the scrub and exists in different densities and NOT of the entire image and other classes. Accuracy assessments were performed using ROIs from the scrub class and also the different densities of iceplant. The high values for MNF, Band ratios and Continuum removal not significantly different for scrub and each density class. However, when comparing the accuracy of scrub and three classes of iceplant density the MNF process worked best. This is expected as the MNF draws from all available bands for the processing, whilst the band ratios and the continuum removal are relying on a limited amount of information. However, in terms of classifying one of the target species, ice plant, the continuum removal and the band ratio methods actually performed better.
The author mentioned that the three processing techniques succeeded in capturing some of the pertinent characteristics of iceplant and successfully classified it. For example MNF worked extremely well identifying different densities of iceplant. This is because it actively creates new bands using the most spectral information in the imagery. However, it is difficult to interpret. Band ratio indices are intuitive as ratios emphasize important ecophysiological information on land cover. Continuum removal works well for areas of dense iceplant, but poorly for complex mosaics with other scrub species and is excellent and reliable for classifying presence / absence of iceplant. Its greatest advantage is the ease and efficiency of processing using a standard ENVI process. One disadvantage with continuum removal is that the results were speckled and even when results of band ratios was sieved and clumped it only improved the overall accuracy of the confusion matrix by 3%.
This research illustrated that AVIRIS imagery offers improved opportunities for mapping invasive plants in a matrix of other vegetation types. The improved spectral resolution of the AVIRIS imagery permits identification of vegetation characteristics that are not possible using multispectral wavebands traditionally used in remotely sensed imagery.
This study showed that the invasive plants iceplant and jubata grass in California’s Mediterranean-type ecosystems can successfully be mapped using hyper spectral imagery (AVIRIS data). The immediate benefit of this research has been to contribute to the knowledge base of land managers at VAFB by providing improved information on the spatial extent and the density of the iceplant and jubata grass, which will lead to better protection of the native biodiversity. This research also described some encouraging findings for applying hyper spectral imagery to mapping iceplant at Vandenberg Air force base that can be repeated over time to detect change and that they can use to monitor control efforts.
The authors acknowledged:
-John Brooks, Amparo Rifa, and George Scheer for helping to conduct fieldwork.
-George Scheer, Pablo Zarco-Tejada, and Karen Olmstead from CSTARS at U.C. Davis for assistance in processing of the imagery.
-Teresa Magee at Dynamac for providing input advice on the project.
-Pablo Zarco-Tejada from CSTARS for helpful comments in reviewing the paper
-And Finally to the Strategic Environmental Research and Development Program and NSF’s IGERT program for providing funding for their research.
Anderson, G. L., Everitt, J. H., et al. (1993). Using satellite data to map false broomweed (Ericameria austrotexana) infestations on south Texas rangelands. Weed Technology 7: 865-871.
Barbour, M., Pavlik, B., Drysdale, F., Lindstrom, S. (1993). California's Changing Landscapes. Sacramento, California Native Plant Society.
Bossard, C., Randall, J. M., Hoshovsky, M. C. (2000). Invasive Plants of California's Wildlands. Berkeley, University of California Press.
Boardman J. W., and Kruse, F. A. (1994). Automated spectral analysis: A geologic example using AVIRIS data, north Grapevine Mountains, Nevada: in Proceedings, Tenth Thematic Conference on Geologic Remote Sensing, Environmental Research Institute of Michigan, Ann Arbor, MI, p. I-407 - I-418.
CalEPPC. (1999) Exotic pest plants of the greatest ecological concern in California. California Exotic Pest Plant Council, San Juan Capistrano, California.
Carson, H. W., Lass, Lawrence W., Callihan, Robert H. (1995). Detection of yellow hawkweed (Hieracium pratense) with high resolution multispectral imagery. Weed Technology 9(3): 477-483.
Chen, C. F., P. F. Lee, F. Tsai, L.-H. Liang, S.-M. Hsu, S.- L. Cheng, 2003, "Applications of Airborne Multispectral Scanner Images" (in Chinese), COA Technology Research Report No. 922029, Council of Agriculture, Taipei, Taiwan.
Clark, R. N., & Roush, T. L. (1984). Reflectance spectroscopy: Quantitative analysis techniques for remote sensing applications. Journal of Geophysical Research, 89, 6329– 6340.
Cochrane, M. A., 2000, "Using vegetation reflectance variability for species level classification of hyper spectral data", Int'l J. Remote Sensing, 21(10), pp. 2075- 2087.
Cowan, B. (1976). The menace of pampas grass. Fremontia, 4(2), 14–16.
D’Antonio, C.M. & Mahall, B.E. (1991) Root profiles and competition between an invasive, exotic, perennial, Carpobrotus edulis, and two native shrub species in California coastal scrub. American Journal of Botany 78, 885–894.
D’Antonio, C. (1993). Mechanisms controlling invasion of coastal plant communities by the alien succulent Carpobrotus edulis. Ecology, 74(1), 83–95.
D'Antonio, C. (1997). Introduction. Assessment and Management of Plant Invasions. 324 In: J. O. Luken, Thieret, J. W. New York, Springer.
Everitt, J. H., DeLoach, C. J. (1990). Remote sensing of Chinese tamarisk (Tamarix chinensis) and associated vegetation. Weed Science 38(3): 273-278.
Fuentes, D.A.; Gamon, J.A.; Qiu, H.; Sims, D.A.; Roberts, D.A. (2001). Mapping Canadian Boreal Forest Vegetation Using Pigment and Water Absorption Features Derived from the AVIRIS Sensor. Journal of Geophysical Research. . 106, 33,565-33,577.
Green, A. A., Berman, M., Switzer, B., and Craig, M. D. (1988). A transformation for ordering multispectral data in terms of image quality with implications for noise removal: IEEE Transactions on Geoscience and Remote Sensing, v. 26, no. 1, p. 65 - 74.
Holland, R.F. (1986). Preliminary classification of the terrestrial plant communities of California. California department of Fish and Game.Sacremento, California, USA.
JPL (2002). The AVIRIS Homepage. http://makalu.jpl.nasa.gov/aviris.html
Keil, D. J., & Holland, V. L. (1998). The vegetation of Vandenberg Air Force Base. Report prepared for The Nature Conservancy and Vandenberg Air Force Base.
Kruse, F. A., (1996a). Geologic Mapping Using Combined Analysis of Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) and SIR-C/X-SAR Data: In Proceedings, International Symposium on Optical Science, Engineering, and Instrumentation, SPIE Proceedings, August 1996, Denver, v. 2819, p. 24 - 35.
Laba, M., F. Tsai, D. Ogurcak, S. Smith, M. E. Richmond, 2003, "Use of Derivative Analysis for the Spectral Discrimination of Invasive Wetland Plant Species", submitted to PE & RS.
Lambrinos, J.G. (2000) The expansion and impact of alien pampas grass in California. PhD Dissertation, University of California, Los Angeles, California, USA.
Lambrinos, J.G. (2000). The impact of the invasive alien grass Cortaderia jubata (Lemoine) Stapf on an endangered mediterranean-type shrubland in California. Diversity and Distributions 6,217-231.
Lins, K. F. and R. Kleckner, 1996, "Land use and land cover mapping in the in the United States: An overview and history of the convept" in GAP Analysis: A Landscape approach to Biodiversity Planning, eds. J. M. Scott, T. Tear and F. Davis, pp.57- 65.
McCormick, C. M., 1999, "Mapping Exotic Vegetation in the Everglades using large- scale aerial photographs", PE&RS, 65(2), pp. 179- 184.
Moss, T. K. (1990). Ice Plant eradication and native landscape restoration. Asilomar State Park Newsletter, 155– 156.
Pimentel, D., Lach, L., Zuniga, R., Morrison, D. (1999). Environmental and economic costs associated with non-indigenous species in the United States. Cornell University News Service.
Rejmánek, M. and J. M. Randall (1994). Invasive alien plants in California: 1993 summary and comparison with other areas in North America. Madroño 41(3): 161-177.
Research Systems Inc. (1999). ENVI user's guide. Research Systems Inc., Boulder, CO.
Richards, J. A., and X. Jia, Remote Sensing Digital Image Analysis, 3rd ed., 400 pp., Springer-Verlag, New York, 1999.
Rouse, J. W., R. H. Haas, J. A. Schell and D. W. Deering. (1974). Monitoring vegetation Systems in the Great Plains with ERTS, Proceedings, Third Earth Resources Technology Satellite-1 Symposium, Greenbelt: NASA SP-351, 3010-317.
Schmidt, K. S., A. K. Skidmore, 2003, "Spectral discrimination of vegetation types in a coastal wetland", Remote Sensing of Environ. , Vol. 85, pp. 92- 108.
Stein, B. A., S. R. Flack, (eds.), (1996). America's Least Wanted: Alien Species Invasions of U.S. Ecosystems, The Nature Conservancy, Arlington, Virginia USA.
Vitousek, P. M., C. M. D'Antonio, et al. (1996). Biological invasions as global environmental change. American Scientist. 84: 468 (11 pages).