Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
THE USE OF REMOTE SENSING FOR THE STUDY OF AIR POLLUTION EFFECTS IN FORESTS Barrett N. Rock James E. Vogelmann Nancy J. Defeo Institute for the Study of Earth, Oceans and Space Science and Engineering Research Building University of New Hampshire Durham, NH 03824 ABSTRACT Remote sensing techniques employing satellite and airborne multispectral data sets provide an accurate means of detecting, quantifying, mapping and monitoring damage in high elevation spruce/fir forests in the northeastern United States (US) and the Federal Republic of Germany (FRG). A gradient of montane conifer damage has been detected using Landsat Thematic Mapper data acquired for the Adirondack Mountains of New York (most severe), Green Mountains of Vermont (moderate damage), and the White Mountains of New Hampshire (limited damage). Landsat Multispectral Scanner ciata have been used to detect a drop in near infrared reflectance between 1973 and 1984 in transition zone forests (dominated by red spruce and balsam fir) and portions of hardwood forests in the Green Mountains, while reflectance has not changed in upper elevation fir zones. Based on in site spectrometry, such a drop in reflectance has characterized increasing levels of forest decline damage in red spruce. Advanced high-spectral resolution airborne imaging spectrometers have detected highly-diagnostic spectral features associated with in situ spectral measurements of spruce damage in both the US and FRG. Current and future work on diagnostic spectral signatures may allow such advanced sensors to identify specific kinds of damage (i.e., determine causes of damage). INTRODUCTION The northeastern United States have been experiencing a decline in red spruce and balsam fir since approximately 1960 (1~. At present, no specific causes have been identified as responsible for the damage. The Forest Response Program of the USDA Forest Service has asked a number of specific questions in order to clarify cause-and- effect issues better. Some of the specific questions for which remote sensing studies may provide input are as follows: Are changes in growth and mortality in spruce-fir forests in the eastern United States greater than can be attributed to typical trends and levels of natural variability? What spatial patterns, if any, exist in growth and mortality changes in spruce-fir forests in the eastern United States and how do these patterns relate to spatial patterns of potential pollutant exposure? 183
184 What are the effects of sulfur and/or nitrogen derived pollutants alone or in combination with oxidants on spruce and fir morphology? Previous forest damage and decline studies of Camels Hump in the Green Mountains of Vermont have identified three components of a spectral signature associated with decline in red spruce (Picea rubens): a blue shift of the chlorophyll well/red edge; a drop in reflectance of the near infrared (NIR) plateau; and a relative increase in the short wave infrared (SWIR) reflectance values (2,3~. A damage mapping technique has been developed which utilizes a ratio of SWIR/NIR reflectance values as measured by aircraft and satellite spectral bands (3,4~. This technique has been shown to be an extremely accurate means of detecting, quantifying and monitoring forest damage in conifer stands in both the northeastern and southeastern United States (5,6~. Presently, researchers at the University of New Hampshire are involved in several remote sensing studies that build on this work. One study involves the use of satellite data to determine the change in NIR reflectance associated with amount of damage in the Green Mountains between 1973 and 1984. A second study examines spatial patterns of damage which exist across the Adirondacks, Green Mountains and White Mountains. A third study uses a high-spectral resolution airborne sensor to look at spectral signatures characterizing various types of damage. These studies are summarized below. CHANGE DETECTION STUDIES Remote sensing investigations employing NS-001 Thematic Mapper Simulator (TMS) and Landsat Thematic Mapper (TM) data (3,4,5,6) have shown excellent correlations between ground-based estimates of conifer forest damage and 1.65/0.83 micrometer band (TM 5/4) ratios. Figure 1 shows a damage assessment image made using the TM band 5/band 4 ratio along with a near infrared band (band 5) and a visible band (band 2), color coded red, green, blue, respectively. Red areas in the image represent damaged conifer areas, while dark blue represents healthy conifers and turquoise represents deciduous zones. Numbered and lettered areas are sites for which ground assessments have been made (2,3,4~. Although images produced using this ratio are extremely accurate in mapping and quantifying forest damage levels, it is often difficult to ascertain what proportion of the damage detected is a result of a general forest decline phenomenon ("unnatural" damage) and what is attributed to "natural" conditions, such as those related to poor growing conditions, ice and wind storms, and other natural stresses. One can begin to address the question of what proportion of damage is natural vs. unnatural by using multitemporal remote sensing data sets to monitor forest conditions through time. The following is a summary of a study to evaluate the potential of using Landsat Multispectral Scanner (MSS) data to detect long-term reflectance changes indicative of high-elevation coniferous forest health (7~. Data from August 29, 1973 (Landsat 1) and August 21, 1984 (Landsat 5) from the Green Mountains of Vermont were used in this study. Sun elevation was 48° for both data sets, and solar azimuth was similar for both scenes (134° and 136° for the 1973 and 1984 data sets, respectively). Multispectral Scanner data were computer-processed at the Jet Propulsion Laboratory (Pasadena, CA, U.S.A.) using the VICAR processing system installed on a VAX 11/780 computer. Bands used in the study were centered at 0.65 (0.60-0.70; MSS Band 5) and 0.95 (0.80- 1 .10; MSS Band 7) micrometers. Following co-registration of portions of the data sets including coverage of the Green Mountains, data sets were standardized by use of 20 forested targets. These sites represented relatively mature stands, most of which were located at low elevations, and were presumed to have undergone minimal spectral change between 1973 and 1984. Sites were field-checked in August of 1987 to verify that these areas had not been logged or selectively thinned between 1973 and 1984. Standardization targets included six coniferous sites and 14 deciduous sites. Mean digital numbers were extracted from each site for the 0.95
~ ~~ - ~~ ~ - ~ - - ~ ~ at A- ~ - - - - -I figure 1. Damage assessment image made uslDg NS-OO1 Ibemadc Flapper Slmulalor data. Numbered sad lettered area are study sites. Red grew indicate beavy forest damage. hiodlfled from Rock et aL all. Repclated with permission from Rock e1 AL 1986. Copyclght1986 by ~ merlcan Insthute of BlologlcalSclences.
186 micrometer band from both 1973 and 1984 data sets. The 1973 vs. 1984 values regressed against each other yielded an r2 value of 0.971 for the 0.95 ,um band. This indicates that an essentially linear relationship exists between 1973 and 1984 data sets for this band, and implies that the MSS band 7 spectral properties for these standardization sites had not changed significantly during the time period. The linear regression equation derived from the relationship between the 1973 vs. 1984 vegetation standardization targets was then used to convert digital number values from 1984 data into units comparable in value to 1973 data for the 0.95 Am band. Following standardization of the 1984 data set for the 0.95 Am band, a difference image was produced by subtracting 1984 from 1973 data sets, and adding an offset of 100 to eliminate negative numbers. Pixels for which values were greater than 100 showed a decrease in NIR reflectance from 1973 to 1984 relative to the 20 vegetation standardization targets. Pixels for which values were less than 100 showed an increase in NIR reflectance between these dates, relative to the standardization targets. A three-color composite using the 1973 0.65 Am and 0.95 Am bands in conjunction with the 0.95 Am difference data set (Figure 2) in the order of blue, green and red was produced. The 0.65 and 0.95 Am bands were linearly stretcher! using standard methodology to enhance contrast, and the difference data set was linearly stretched to enhance decreases in reflectance between 1973 and 1984. This image not only indicates where major reflectance decreases have occurred but also depicts topographic relief. Areas of red or dark orange generally indicate where coniferous areas decreased in reflectance, whereas yellow to light orange areas indicate where deciduous vegetation decreased in reflectance, in relation to the standardization targets. Areas that are blue to green showed either no major near-infrare~i reflectance changes, or increases in reflectance. Field and laboratory spectral ciata acquired for red spruce at the Camels Hump study area and for Norway spruce in West Germany suggest that a decrease in the near infrared reflectance accompanies an increase in needle damage associated with forest decline (S,9,10~. Within the montane coniferous areas, decreases in reflectance were most apparent in the transition zone forests on the western lower slopes, where balsam fir and red spruce dominate. Near-infrareci reflectance at the upper elevations, where balsam fir dominates, was relatively unchanged. A general trend of decreasing basal area and inferred biomass loss through time has been documented for the montane forest on the west facing slopes of Camels Hump (the northernmost mountain seen in Fig. 2) (11,i2~. It is presumed that this decrease in basal area and inferred loss of green leaf biomass anti the concomitant increase in amounts of dead branches/trees results in the observed decreases in reflectance in the coniferous portions of the difference image. It should be noted that it has been found that tower reflectance in the near-infrared implies lower amounts of biomass as estimated by leaf area index (LAI) measurements for some species (13~. However, it has not been documented that lower levels of biomass (or LAI) correlate well with near-infrared reflectance for conifer species (14,15~. Therefore, at present, it cannot be stated that decreases in the near-infrared reflectance noted for much of the conifer zone at Camels Hump are directly related to decreases in green leaf biomass, or with the increases in dead branches and trees which accompany loss of biomass, or both. The data sets being compared were acquired during approximately the same time of year (late August), and thus potential problems due to different solar angles and azimuths have been negated. However, it should be noted that annual phonological differences due to rainfall and temperature variations represent potential problems in multitemporal studies. Phenological differences may be minimized, but not totally eliminated as factors contributing to reflectance differences between data sets, by selection of data from the same date from year to year. Rainfall anti temperature data for the areas and dates in question should be used to determine if annual phonological
L - - . lo., _ A. it' ~ ANT I' Figure 2. False color composite image from Multispectal Scanner (MSS) data of a portion of the Green Mountains of Vermont using 1973 0.65 and 0.95 Am bands, and the 0.95 ,um difference data set. Areas of red, orange or yellow indicate where near-infrared reflectance has decreased from 1973 to 1984 in relation to 20 deciduous and coniferous targets. Modified from Vogelmann (7~.
differences are likely to be being compared. 188 f ~ . , major factors influencing the remotely sensed data sets It is likely that the lower reflectance of the deciduous vegetation in the higher elevations in 1984 vs. 1973 was due at least in part to rainfall differences affecting green leaf biomass production. More rainfall occurred during the growing season of 1973 as compared with that of 1984. However, it is not felt that rainfall differences alone can explain the reflectance differences noted in the high elevation coniferous regions. Some of these areas showed reflectance changes (e.g., the west facing transition zone), whereas adjacent regions (e.g., high elevation areas dominated by balsam fir) that presumably were under similar climatic conditions did not. The differences in rainfall might be expected to have different effects on conifer leaf flush in 1984 vs. 1973. However, since conifers retain their needles for several years, these first-year needles make up only a portion of the total conifer foliage influencing reflectance. Such rainfall differences would not be expected to affect greatly the phonologic state of the older needles. Thus, coniferous vegetation would not be as susceptible to annual variation in rainfall patterns as would deciduous vegetation. At present, it is felt that the decrease in near-infrared reflectance noted in the 1984 data set as compared to the 1973 data set for the high-elevation coniferous regions is attributed to the general forest decline process, being related to the increased levels of mortality and decreased levels of green biomass that have been documented in this region. SPATIAL STUDIES The detection and quantification of spatial patterns of conifer forest damage in the eastern United States may be done accurately and objectively using remote sensing techniques. Remote sensing data can be used to detect large, regional variations in forest condition that can then be correlated with patterns of pollutant exposure, soil types, geology and other factors that may affect the condition of forest communities. Previous studies have found that the ratio of TM band ~ to band 4 is strongly correlated with ground-based measurements of forest damage in the northeastern United States. The higher the level of forest damage, the higher the ratio value (3,4,5,6~. A Thematic Mapper scene (that included coverage of the Green Mountains and the Adirondack Mountains) acquired August 4, 1984, and a second scene (that included the Green Mountains and the White Mountains) acquired June 10, 1984 were used. From these data sets, values of the damage assessment ratio (TM band 5 / band 4) were compared among the three mountain ranges. The two TM scenes were standardized by calibrating pixel values based on homogeneous ground targets and by using the same parameters to stretch the band 5 / band 4 ratio over the full O - 255 dynamic range. Coniferous portions of the image were isolated from non-coniferous forest regions using a method in which a mask was placed over all regions of the image which did not correspond to coniferous forest. A complete description of the method can be found in Vogelmann and Rock (6~. A damage rating scale was developed using the TM band 5 / band 4 ratio to assess relative damage levels of montane conifer areas among selected mountains in the Green Mountains of Vermont and the White Mountains of New Hampshire (6~. The same procedure was used to assess relative damage levels of conifer areas in the Adirondack Mountains. Low, medium and high damage study sites located on Camels Hump in the Green Mountains were used as standards of reference. Damage levels for each of these reference sites were determined by visually assessing percentage foliar loss at each study site (4,6~.
189 Ranges of ratio values corresponding to low, medium and high damage categories were defined, and numbers of conifer pixels falling within each damage category were totaled. The level of damage for each mountain was then summarized using the following equation: Damage\ Rating = (100 - % Low Damage Pixels + % High Damage Pixels) /2 Low% %Medium %High Elevation Damage Damage (meters) Pixels Damage Damage Site Pixels Pixels Rating ADIRONDACKS Whiteface Mt. 1484 S.6 12.7 78.7 85.1 High Peaks Area 1268 2.3 6.S 90.4 94.3 GREEN MOUNTAINS Camels Hump 1244 26.5 20.7 52.S 63.2 Mt. Abraham 1260 25.6 25.6 48.9 61.7 Breadloaf Mt. 1165 37.9 24.6 37.5 49.S WHITE MOUNTAINS Mt. Moosilauke 1464 72.3 16.S 10.9 19.3 Lafayette Mt. 1585 63.9 19.0 17.0 26.6 Table 1. Conifer damage in the Adirondack Mountains, Green Mountains and White Mountains. Table 1 is a summary of conifer damage for several high elevation areas in the Adirondack Mountains, Green Mountains and White Mountains. It should be made clear that these damage ratings are relative measures of forest health based on the field work done at Camels Hump. Thus, a damage rating does not correspond to percent mortality, but is merely a relative measure that can be used to compare damage levels among individual mountains. It is apparent that there is a trend of decreasing damage from the westernmost range (Adirondacks) to the easternmost range (the White Mountains). It is also evident from the table and from field studies that elevation, slope and aspect alone are not factors which account for the relative levels of damage in coniferous forests in New York and New England. The National Acid Precipitation Assessment Program sponsored studies to determine the spatial patterns of wet deposition pH values in North America. Results are shown in Figure 3. The pattern of pH values found in the mountains of the northeastern United States correlates with our damage assessment: the lowest pH values in North America are centered over the Adirondacks and pH values increase (indicating less acidic conditions) in all directions from this area. The lowest pH values (in the Adirondack Mountains) correspond to the highest damage ratings, while higher pH values (in the White Mountains of New Hampshire) correspond to the lower damage ratings of the three ranges studied. The Green Mountains in Vermont, which have intermediate damage levels, are inferred to have intermediate pH values.
190 ~ I' eat, ~ ~ ~ '~'~ ~ ,, EXPLANATION · S.4 pH at sample site 5.0 Line of approximately equal pH value - ` 5.4 ' ~ s.6. /r ~ ~ ~ ~~~ ~ r - - 0'0~t 7'5s';~ , `'r 1 1 ' r _ ,' J ~;3: 0 to Figure 3. pH measurements for North America, measured in 1982 (16). FLUORESCENCE LINE IMAGER STUDIES To date, high-spectral resolution in situ and airborne sensor data sets have been acquired for forest decline sites in the northeastern United States (3,9), and the Federal Republic of Germany (S,10~. Although these high-resolution data sets provide a great deal of fine-spectral feature information relating to specific symptoms of forest decline (chlorosis, canopy dryness, and foliar loss), as yet such symptoms have not been related to exposure to specific pollutants such as sulfur and/or nitrogen compounds, either alone or in combination with oxidants. To develop spectral signatures characteristic of spruce response to specific pollutant exposure, high-resolution in situ spectral assessment studies must be conducted in association with
191 controlled-exposure experiments. As a means of portraying the power of airborne high-spectral resolution data in assessing types of forest damage, work presented in detail elsewhere (9) is cited below. An airborne imaging spectrometer, known as the Fluorescence Line Imager (FLI), and owned and operated by Moniteq, Ltd., Toronto, Canadai, has been used to detect reflectance features associated with the chlorophyll well/red edge blue shift characteristic of in situ spectral measurements (9~. These spectral fine features are not detected by broad-band sensor systems such as the MSS, TM or TMS, but rather require the high-spectral resolution capabilities of imaging spectrometers such as the FLI and NASA's Airborne Visible Infrared Imaging Spectrometer (AVIRIS). Figure 4 presents in situ and FLI reflectance data acquired for a low and a high damage site on Camels Hump (sites 1 and 7, respectively, Fig. 1~. In both, normalized plots are presented and the blue shift is readily seen. In addition, spectral reflectance in the visible green and red regions of the electromagnetic spectrum (0.50-0.69,um) characteristic of chlorosis is also seen in both data sets. The FLI data acquired from several forest decline (Waldsterben) study sites in the Federal Republic of Germany are currently being assessed. The blue shift is also seen in these European data and appears to be highly diagnostic of levels of damage in Norway spruce and white fir. Using red edge parameters, a false color image of the FLI flight line may be compared with a similar portion of the TMS image for the same area. A comparison of the two images suggests that the broad-band TMS data are detecting generic damage in both red spruce in the transition zone forest (lower elevation) and balsam fir (higher elevation) in the conifer forest zone on Camels Hump (9~. The FLI image appears to be mapping only the damage, based on red edge parameters, occurring in the transition zone red spruce. Winter damage and fir wave damage is known to occur in the upper elevation, fir-dominated conifer forests above the transition zone on Camels Hump. It has also been shown that balsam fir on the mountain has undergone a less severe (although statistically significant) decline in vigor and biomass than has the red spruce ( 12~. This suggests that the use of both sensor systems (TM/TMS and FLI) may provide information which allows separation of different damage types: forest decline damage in red spruce and winter damage/fir-wave damage in balsam fir. High-spectral resolution remote sensing systems currently available on airborne platforms will eventually be available on orbital platforms such as the NASA/ESA Earth Observing System (Eos). Once in situ spectral signatures have been identified which are diagnostic of specific natural or pollutant damage, such airborne and/or spaceborne sensor systems may provide forest assessment capabilities which will relate specific spectral signatures (effects) to specific causal agents so that direct cause-and-effect relationships may be remotely detected and monitored on a regional/global scale. SUMMARY Each remote sensing system has limitations based on spatial/spectral resolution, band placement, software availability, and other complicating factors. Used alone, each sensor can only be used to solve a portion of the forest damage and decline/atmospheric pollutant puzzle. When data are acquired with many sensors over the same region, cause-and-effect issues can be addressed more readily. In the above studies we found that the changes in health in certain portions of the spruce-fir forests in the eastern United States were greater than can be attributed to ~ Reference to specific manufacturers is for clarity and does not constitute endorsement of product by NASA or the University of New Hampshire
0.8 Lo of ~ 0.6 J IL 0.4 0.^ o 100 50 ~ B) at I S53 R ED E DO E MAX SLOPE = 710.91 nm (:HLOROPHYLL WELL MIN = 693.80 nm ~111~ / _ 1/ // //: RED EDGE MAX SLOPE // = 726.04 nm // CHLOROPHYLL WELL MIN _ / = 698.62 nm ~ ~ ~( 1 1 1 1 500 540 580 620 660 700 740 78() 820 860 900 WAVELENGTH, nm NORMALIZED ~ , 1 623 694 -7D WAVEi-£bJGTH, Am Figure 4. Ground spectral data (A) and aircraft spectral data (B) for high and low damage sites on Camels Hump, Vermont. In each, the solid curves are from low damage areas (site #1 in Fig. 1) and the dashed curves are from high damage areas (site # 7 in Fig. 1). Both sets of curves show a diagnostic "Blue Shifts in the position of the chlorophyll red edge located at approximately 710-740 nm.
193 typical trends and natural variability. Also, a distinct spatial pattern of greater damage in the Adirondack Mountains, decreasing to the east, has been detected and this pattern of damage corresponds to spatial patterns of wet deposition pH values. Finally, current research indicates that spectral signatures characteristic of damage exist, and these can be used to identify various damage symptoms. The work described in this paper was not funded by the U.S. Environmental Protection Agency and therefore the contents do not necessarily reflect the views of the Agency and no official endorsement should be inferred. REFERENCES 1. U.S. Department of Agriculture, Forest Service. 1985. Cooperative survey of red spruce and balsam fir decline and mortality in New York, Vermont and New Hampshire, 1984, Broomall, PA: U.S. Dept. of Agriculture, Forest Services Northeastern area, 53 pp. 2. Rock, B.N., Williams, D.L., and Vogelmann, J.E. 1985. Field and airborne spectral characterization of suspected acid deposition damage in red spruce (Picea rubens) from Vermont. Proceedings of the 11 th International Symposium on Machine Processing of Remotely Sensed Data, Purdue University, West Lafayette, IN, pp. 71-81. 3. Rock, B. N., Vogelmann, J.E., Williams, D.L., Vogelmann, A.F., and Hoshizaki, T. 1986. Remote detection of forest damage. BioScience, 36:439-445. 4. Vogelmann, J. E,. and Rock, B. N. 1986. Assmssing forest decline in coniferous forests of Vermont using NS-OO1 Thematic Mapper Simulator data. Int. I. Remote Sensing, 7:1303- 1321. 5. Rock, B. N., Defeo, N. J., and Vogelmann, J. E. 1987. Vegetation survey pilot study: detection and quantification of forest decline damage using remote sensing techniques. Final Report to the USDA Forest Service, Jet Propulsion Laboratory Document D-4669, Pasadena, California, 30 pp. plus appendices. 6. Vogelmann, I. E., and Rock, B. N. l9X8. Assessing forest damage in high-elevation coniferous forests in Vermont and New Hampshire using Landsat Thematic Mapper data. Remote Sens. Environ., 24:227-246. 7. Vogelmann, J. E. 1988. Detection of forest change in the Green Mountains of Vermont using Multispectral Scanner data. Int. I. Remote Sensing, 9:1187-1200. 8. Rock, B. N., Hoshizaki, T., _ . ~ . . . Lichtenthaler, H., and Schmuck, G. 1986. Comparison of In site spectral measurements of forest decline symptoms in Vermont (USA) and the Schwarzwald (ERG). Proc. of Int. Geosci. and Remote Sensing Symposium (IGARSStS6), IEEE 86CH2268-1, IEEE, New York, Vol 3: 1667-1672. 9. Rock, B. N., Hoshizaki, T., and Miller, J. R. 1988. Comparison of in situ and airborne spectral measurements of the blue shift associated with forest decline. Remote Sens. Environ., 24:109- 127. 10. Herrmann, K., Rock, B. N., Ammer, U., and Paley, H. N. 1988. Preliminary assessment of Airborne Imaging Spectrometer and Airborne Thematic data acquired for forest decline areas in the Federal Republic of Germany. Remote Sens. Environ., 24: 129-149. 11. Vogelmann, H. W., Bliss, M., Badger, G. and Klein, R. M. 1985. Forest decline on Camels Hump, Vermont. Bull. Torrey Bot. Club, 112:274-287.
194 12. Vogelmann, H. W., Perkins, T., Badger, G. and Klein, R. M. 1988. A 21-year record of forest decline on Camels Hump, Vermont. Eur. I. For. Path.: in press. 13. Wiegland, C. L., Richardson, A. I., and Kanemasu, E. T. 1979. Leaf area index estimates for wheat from Landsat anti their implications for evapotranspiration and crop modeling. Agron. I., 71:336-342. 14. Peterson, D. L., Spanner, M. A., Running, S. W., and Teuber, K. B. 1987. Relationship of Thematic Mapper Simulator data to leaf area index of temperate coniferous forests. Remote Sens. Environ., 22:323-341. 15. Franklin, J. 1986. Thematic mapper analysis of coniferous forest structure and composition. Int. J. Remote Sensing, 7:1287-1301. 16. NAPAP (National Acid Precipitation Assessment Program). 1983. Annual report to the President and Congress, Washington, D.C.