Clark, R.N., King, T.V.V., Ager, C., and Swayze, G.A., 1995, Initial vegetation species and senescence/stress mapping in the San Luis Calley, Colorado using imaging spectrometer data. Proceedings: Summitville Forum '95, H.H. Posey, J.A. Pendelton, and D. Van Zyl Eds., Colorado Geological Survey Special Publication 38, p. 64-69.

Initial Vegetation Species and Senescence/Stress Indicator Mapping in the San Luis Valley, Colorado Using Imaging Spectrometer Data


Roger N. Clark, Trude V.V. King, Cathy Ager and Gregg A. Swayze U. S. Geological Survey

Mail Stop 964
Box 25046 Federal Center
Denver, CO 80225


Spectroscopic analysis of Airborne Visual and Infra-Red Imaging Spectrometer (AVIRIS) data is presented to show vegetation species and senescence/stress mapping. AVIRIS data were acquired over the San Luis Valley up to the Summitville abandoned mine in Colorado on September 3, 1993. The data were first corrected to surface reflectance using a combined radiative transfer model plus empirical ground calibration. Reference spectra were extracted from the AVIRIS data set over known crops and in adjacent pasture land. Continuum-removed spectral features from the reference spectra were compared to each spectrum in the AVIRIS data set using a modified least-squares spectral feature matching algorithm. Ground truth data were compared to the resulting identifications and the method proved to be 96% accurate.

A method of analyzing the position of the blue shift of the chlorophyll absorption red edge is presented. Shifts smaller than 1% of the bandwidth of the spectral data can be detected. In the case of AVIRIS, that corresponds to 0.1 nm. The shift was mapped for the data set and numerous fields were found to have varying degrees of shift. Some were in senescence, some had been sprayed with a defoliant, and in others the shift may be caused by lack of water or other nutrients. This analysis shows the AVIRIS instrument wavelengths are stable to better than 0.1 nm over at least one 512 line scene.


Vegetation covers a large portion of the Earth's land surface. Obtaining quantitative information about vegetation with remote sensing has proven difficult. To first order, all vegetation is chemically similar, and most healthy plants are green showing absorption bands that are almost identical. Plant species are generally characterized by the leaf and flower or fruit morphology, not by remote sensing spectral signatures. However, to the human eye, many plants show varying shades of green, so there is direct evidence for spectral differences between plant types. Quantifying these subtle differences in a predictable manner has not been easy, and has been made more difficult by lack of adequate instrumentation.

Imaging spectroscopy is a new mapping tool and the next generation in remote sensing technology. The narrow spectral channels of an imaging spectrometer form a continuous reflectance spectrum of the Earth's surface. That contrasts with the 4 to 7 channels of the previous generation of imaging instruments, like the Landsat Thematic Mapper (TM) and Multispectral Scanner (MSS). While systems like Landsat can distinguish general brightness and slope differences in the reflectance spectrum of the surface, imaging spectroscopy not only does that, but also resolves absorption bands in the spectrum which can be used to identify specific species. Spectroscopic analysis of imaging spectroscopy data allows any material to be mapped, whether mineral, vegetation, manmade, water, snow, or other if it displays unique absorption features in the measured spectral region (e.g. see Clark et al., 1993a, Clark et al., 1992) .

NASA is now flying the "Airborne Visual and Infra-Red Imaging Spectrometer" (AVIRIS) instrument. AVIRIS acquires data in the spectral range from 0.4 to 2.45 microns in 224 spectral channels. The instrument is flown in an ER-2 aircraft (a modified U-2) at 19,800 meters (65,000 feet). The pixel spacing on the ground is ~17 meters, with a 20-meter square pixel, the swath width is about 10.5 kilometers (614 pixels) and the swath length can be up to about 1000 kilometers with the current tape recorder capacity on the aircraft.

Clark et al., (1990a, 1991) developed a new analysis algorithm that uses a digital spectral library of known materials and a fast, modified-least-squares method of determining if a single spectral feature for a given material is present. This algorithm, now called " Tricorder," compares continuum-removed spectral features (Clark and Roush, 1984) from the imaging spectrometer data set to corresponding continuum-removed spectral features from a reference spectral library. Multiple features from multiple materials are compared and the closest match is selected. The algorithm does not force a detection like many other algorithms. For example, many algorithms take a set of curves and best fit them to the observed data, often requiring a set of parameters (like mineral fraction) to sum to one. The tricorder algorithm has no such constraint. If the materials do not exist in a given pixel, such that there are no spectral features from those or similar materials, the algorithm produces zeros, indicating they are not detected.

The human eye sees different plant leaves as shades of green, as characterized by the "green peak" in reflectance spectra. The eye/brain color system differentiates shades of green under different lighting conditions. For example, under indoor incandescent light, there is a strong red slope to the spectral signal received by the eye due to the red spectral shape of the light source (as compared to the same plant in direct sunlight). We are still able to distinguish the plant as green and distinguish different plants as various shades of green.

The Tricorder algorithm is very sensitive to the shape of spectral features and has the potential to distinguish more subtle differences in the visible spectrum of plants than the human eye. We have applied this method, along with reference spectra of specific plants, to map vegetation species in AVIRIS scenes.

The continuum removal to isolate diagnostic spectral features is an important step, particularly when a pixel contains spectral information from green plants, dry vegetation, and soil. In such a case, the combination of materials within a pixel changes the color perceived by the human eye. However, by isolating absorption features with continuum removal, the position and shape of the continuum removed spectral feature remains constant, although its depth changes with absorber fractional areal coverage in the pixel. The Tricorder algorithm normalizes the absorptions so that overall shape is compared.


We analyzed AVIRIS data obtained over agricultural areas in the San Luis Valley of Colorado. The data were acquired on September 3, 1993. A combined method of radiative transfer modeling and ground calibration site reflectance were used to correct the flight data to surface reflectance (Clark et al., 1994a). This method, called Radiative Transfer Ground Calibration, or RTGC, corrects for variable water vapor in the atmosphere and produces spectra free of artifacts with spectral channel to channel noise approaching the signal to noise of the raw data. The calibration site soil samples were obtained on the day of the overflight and measured on our laboratory spectrometer (Clark et al, 1990b). The site was near the center of the AVIRIS scene and the spectra of the soil is spectrally bland, especially in the region of the chlorophyll absorption in the visible portion of the spectrum.

The center of the scene is located at approximately 106 degrees 03' longitude, 37 degrees 23' latitude, and the scene covers about 92 square kilometers. This scene is one of 28 in the area for a general project to study the Summitville abandoned mine site in the mountains above the San Luis Valley and its effects on the surrounding environment.


The study area includes farmland producing potatoes, alfalfa, barley, oat hay, canola, and open fields containing chico, and other unidentified weeds. Ideally, one would have a digital spectral library of reference spectra of the plant species to be mapped. Such a library does not exist for vegetation, as it does for minerals (e.g. Clark et al., 1993b). It is not known how many spectra as a function of growing season would be required to represent the changing spectral signatures.

Some plant species were measured with a portable field spectrometer, but due to windy conditions and limited availability of the instrument, sufficient data to form a library of all crops could not be obtained. The main objective of the field operations was to provide spectral information on sites with detailed geochemical analysis. Because of the limitation of obtaining field spectra, reference spectra were obtained for sites of known species directly from the AVIRIS data. AVIRIS, having been well calibrated to surface reflectance, is an excellent field spectrometer, providing data for large areas. The reference spectra obtained from the AVIRIS data set are shown in Figure 1. The alfalfa, canola, oat hay, and nugget potato spectra (Figure 1a) show the plants to be green and healthy. The barley had lost all its chlorophyll signature (Figure 1b). The Norkotah potatoes were not being irrigated as they were about to be harvested, and consequently they showed a weak chlorophyll and cellulose absorptions with soil (clay) absorptions from exposed soil. These potatoes were also being sprayed with a defoliant. Thus, they should show decreased chlorophyll absorption along with a shift of the red edge of the absorption to shorter wavelengths. The chico and pasture spectra show combinations of chlorophyll and cellulose (dry vegetation) absorptions. There was rain in the valley in the few days before the flight so the chico/pasture may not show much water deprivation stress (being native plants they are hardy and can also withstand some reduced precipitation compared to the crops).

Figure 1a: 12K GIF

Figure 1b: 12K GIF

Figure 1. Reference spectra used in the mapping of vegetation species. The field calibration spectrum is from a sample measured on a laboratory spectrometer, all others are averages of several spectra extracted from the AVIRIS data. Note that the noise is extremely low, comparable to the lab spectrum of the field calibration site. In Figure 1a (top), each curve has been offset from the one below it by 0.05. In Figure 1b (bottom), each spectrum has been offset by 0.04 from the one below it, except the top spectrum is offset 0.06. The offsets are cumulative, so the field calibration spectrum is offset a total of 0.18 for clarity.


The continuum-removed chlorophyll-containing spectra are shown in Figure 2. The shape differences between each species enable the Tricorder algorithm to discriminate between them. The tricorder algorithm produced maps of the distribution of each material. If there are other plants present in the image with similar spectral features, they could be misidentified as one of those in Figure 1. Thus, it is important to have a complete set of reference spectra. The extensive work by many workers on the Summitville project provided knowledge of the crops in the area, thus our spectral library for this case should be complete for crops. As the time of data acquisition was early fall for this mountain valley (elevation ~7650 feet, 2330 meters), some crops and/or pasture are senescing. The crops are changing spectrally, so this is a difficult test, and it was not known how representative a single reference spectrum for each crop would be. Each crop exists in several fields, each of which may have been in a different stage of senescence.

Figure 2: 16K GIF

Figure 2. The continuum-removed chlorophyll absorption spectra from Figure 1 are compared. Note the subtle changes in the shapes of the absorption between species.

The Tricorder algorithm examines the spectrum of each pixel in the image and chooses the best match to the set of reference spectra. The images from each crop/soil/pasture match were then color coded and a color coded map produced (Figure 3). Results of field checking of crops in the area were used to produce a color coded map of what Figure 3 image might look like if completely correct (Figure 4, "the answer map"). Field verification data were supplied by Maya ter Kuile of Argo Engineering (1993, personal communication), and our own work. Every crop and field could not be checked due to limited resources for the large area.

Figure 3: 151K GIF

Figure 3. The Tricorder analysis of vegetation species. See text.

Figure 4: 117K GIF

Figure 4. The field verification data. Crops were identified by traveling by car to each site and visually identifying the crops. Crops not color coded in this image had not been checked. Only one pasture is indicated, although many others are obvious and are indicated correctly in Figure 3, except for the unexplained canola in the fields. While it is possible that there is some canola growing wild, it is more likely that the pasture reference spectra shown in Figure 1 are incomplete. The fields marked with a "P" are partially correct in the AVIRIS analysis. The Field marked "C" is the location of the ground calibration site. Additional spectra will be used in future mapping studies.

Examination of Figures 3 and 4 shows the tricorder-derived map to be highly accurate. Of 43 verification fields (not including chico/pasture), 7 included the sites of our reference spectra, 33 were identified correctly, another 3 were identified as mixed by the tricorder analysis (but were indicated as one crop type in the field data), and no fields were incorrect. The fields identified as mixed were of two species, one of which was correct in each case. The field check was done by driving the roads and identifying the crops visually. Because of the time required to develop this analysis (AVIRIS data receipt followed by calibration to surface reflectance in January, 1994, initial mapping in February, 1994), and the fact that it was winter, we were unable to investigate the cause of the discrepancies in the three fields mapped as mixed. If we give a score of half to the three fields identified as mixed, the success score of 96% shows the method is accurate.

The accuracy is made even more impressive when one considers several fields were mapped correctly even though they had already been harvested. There was apparently enough plant material still left on the fields to still make a correct identification. This is again possible because the algorithm normalizes the absorptions, so even though the vegetation cover may be only a few percent, a correct identification can be made. The harvested areas can be identified in Figure 5 in circular plots where the colored pixels are sparse and/or low in intensity. Clark et al., (1992) were also able to differentiate vegetation communities on the arid Colorado Plateau, another sparsely vegetated case. The harvested areas should also be under stress, and are indicated as such in the senescence/stress map (below).

Canola was mapped in many of the areas known to be pasture. While it is possible that canola seeds have been blown into surrounding fields and are growing, it is more likely that the canola spectral signature is close to other (native?) plants in those open fields. Complete discrimination of crops and surrounding wild vegetation requires a more complete set of reference spectra, beyond the scope of this initial study.

Figure 5: 179K GIF

Figure 5. The vegetation senescence/stress anomaly map


The long-wavelength side of the chlorophyll absorption (~0.68 to ~0.73 micron) forms one of the most extreme slopes found in spectra of naturally occurring common materials, plant or mineral (see Figure 1). The absorption is usually very intense, ranging from a reflectance low of less than 5% (near 0.68 micron) to a near infrared reflectance maximum of ~50% or more at ~0.73 micron). The properties of reflectance spectra (e.g. Clark and Roush, 1984) indicate such an absorption band is "saturated." In such a case, the absorption band minimum will not change much with increased or decreased absorption, but the wings of the absorption will change. When the chlorophyll absorption in the plant decreases, the overall width of the absorption band decreases. The short wavelength side of the chlorophyll absorption is not observed in reflectance as is the long wavelength side because of other absorptions in the ultraviolet (UV). The result of this combination appears as a shift to shorter wavelengths as the chlorophyll absorption decreases. This has popularly become known as the "red-edge shift" or the "blue shift of the red edge" and can be caused by natural senescence, water deprivation, or toxic materials (e.g. Collins et al., 1983; Rock et al., 1986).

The ratio of two spectra, one shifted in wavelength, the other not, and each with steep slopes as seen at the "red edge," will produce a spurious feature when there is only a small shift between the two. If the blue shifted spectrum is divided by an unshifted spectrum, a peak will be observed in the ratio. For a spectrum of green vegetation (from Figure 1a), a 1 nm shift will produce a residual feature of about 6%. The AVIRIS data have a signal to noise of several hundred in this spectral region, so red-edge shifts of less than 0.1 nm are possible to detect. We used field spectrometer spectra for the San Luis data set, and computed a ratio cube which would show a peak when a shift occurs.

The red-edge shift was mapped using the tricorder algorithm (Figure 5). The fact that the resulting image shows no horizontal scan line striping attests to the superb wavelength stability of the AVIRIS instrument. As indicated earlier, the senescence/stress crop fields are those which have been harvested, sprayed with a defoliant, deprived of water, or may have other toxic influences. The area covered by this scene is not affected by acid mine drainage from the Summitville mine, thus it provides a control for the region of what might be expected for normal conditions. There is a curious senescence/stress indication along the banks of the stream that enters the image at the top middle of the image to the lower left (stream flow is to the southeast). The other stream, flowing from the bottom middle of the image to the right middle edge (also flowing to the southeast) shows no similar anomaly. The cause is as yet undetermined.


Vegetation species mapping is possible with high precision using spectral feature analysis of data from airborne imaging spectrometers. Once calibrated, and after reference spectra have been selected, species mapping, along with senescence/stress indicator mapping can be achieved in less than 1 second of CPU time per square kilometer per species (or soil/mineral) on a 10 million floating point operation per second (MFLOP) workstation.

Imaging spectroscopy data can be used for environmental application, monitoring vegetation cover and its health, monitoring vegetation species, and providing a rapid overview of large areas. When applied to large areas, the cost to derive these maps is low in cost relative to field checking and monitoring on the ground. The species maps might be used for more accurate crop yield predictions.

Space limitations prevent us from presenting the entire AVIRIS data set of the Summitville and San Luis Valley study area in this paper, and processing is not yet complete. Additional areas, including vegetation species and senescence/stress indicator maps for more of the study area will be shown at the meeting.


Collins, W., S.H. Chang, G. Raines, F. Canney, and R. Ashley, Airborne biogeochemical mapping of hidden mineral deposits: Econ. Geol., 78, p. 737, 1983.

Clark, R.N. and T.L. Roush, Reflectance Spectroscopy: Quantitative Analysis Techniques for Remote Sensing Applications: J. Geophys. Res., 89, p. 6329-6340, 1984.

Clark, R.N., A.J. Gallagher, and G.A. Swayze, Material absorption band depth mapping of imaging spectrometer data using a complete band shape least-squares fit with library reference spectra: Proceedings of the Second Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Workshop, JPL Publication 90-54, p. 176-186, 1990a.

Clark, R.N., T.V.V. King, M. Klejwa, G. Swayze, and N. Vergo, High Spectral Resolution Reflectance Spectroscopy of Minerals: J. Geophys Res. 95, p. 12653-12680, 1990b.

Clark, R.N., G.A. Swayze, A. Gallagher, N. Gorelick, and F. Kruse, Mapping with Imaging Spectrometer Data Using the Complete Band Shape Least-Squares Algorithm Simultaneously Fit to Multiple Spectral Features from Multiple Materials: Proceedings of the Third Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) Workshop, JPL Publication 91-28, p. 2-3, 1991.

Clark, R.N., G.A. Swayze, C. Koch, A. Gallagher, and C. Ager, Mapping Vegetation Types with the Multiple Spectral Feature Mapping Algorithm in both Emission and Absorption: Summaries of the Third Annual JPL Airborne Geosciences Workshop, Volume 1: AVIRIS Workshop. JPL Publication 92-14, p. 60-62, 1992.

Clark, R.N., G.A. Swayze, and A. Gallagher, Mapping Minerals with Imaging Spectroscopy: U.S. Geological Survey, Office of Mineral Resources Bulletin 2039, p. 141-150, 1993a.

Clark, R.N., G.A. Swayze, A. Gallagher, T.V.V. King, and W.M. Calvin, The U. S. Geological Survey, Digital Spectral Library: Version 1: 0.2 to 3.0 m: U.S. Geological Survey, Open File Report 93-592, 1340 pages, 1993b. (Also being published as a USGS Bulletin, 1300+ pages, 1994 in press.)

Clark, R.N., G.A. Swayze, K. Heidebrecht, R.O. G.A.F.H. Goetz: Calibration to Surface Reflectance of Terrestrial Imaging Spectrometry Data: Comparison of Methods: Applied Optics in review, 1994a.

Rock, B.N., J.E. Vogelmann, D.L. Williams, A.F. Voglemann, T. Hoshizaki, Remote detection of forest damage: Bio. Sci. 36, p. 439, 1986.

U.S. Geological Survey, a bureau of the U.S. Department of the Interior
This page URL=
This page is maintained by: Dr. Roger N. Clark
Last modified November 18, 1998.