Discrimination of wetland and non-wetland community types with multi-spectral multi-angle, polarized data

Discrimination of Wetland and Non-wetland Community Types with Multi-spectral Multi-angle, Polarized Data

V C. Vanderbilt, G.L. Perry and J. A. Steam Ames Research Center, Moffett Field, California, USA S.L. Ustin, M.C. Diaz Barrios, S. Zedler and J. Syder University of California, Davis, California, USA L.A. Morrissey and G.P. Livingston University of Vermont, Burlington, Vermont, USA F.-M. Breon and S. Bouffies CEA/DSM/LMCE, Gif sur Yvette, France M. Leroy UMR CNES-CNRS-UPS, Toulouse, France M. Herman and J.-Y. Balois L.O.A., U.S.T. de Lille, France

Preprint from: Seventh International Symposium Physical Measurements & Signatures in Remote Sensing April 7 -11, 1997 Courchevel - FRANCE

Abstract

The areal representation of boreal wetlands, the source areas for methane, an important greenhouse gas, is poorly known -- estimates differ nearly seven fold. Here our objective was to investigate the feasibility of 1) using POLDER data to discriminate wetlands as distinct from other ground covers and 2) classifying wetlands according to whether they are open-water areas or fens. The results show that the visually blinding glitter of sunlight off ruffled water surfaces provides a strong, unique, angular signature reflection which is characteristic of wetlands and uncharacteristic of other common cover types.

1 INTRODUCTION

The areal extent of boreal wetlands, important natural source areas for methane, a greenhouse gas which affects climate, is poorly known on regional to global scales (Aselmann and Crutzen, 1989; Matthews and Fung, 1987; Morrissey and Livingston, 1992; Roulet et al., 1992; Crill et al., 1988; Klinger et al., 1994; Barlett and Harriss, 1993; Fechner and Hemond, 1992; Roulet et al., 1992; Moore et al., 1990; Crill et al., 1988; Whalen and Reeburgh, 1988). Uncertainties in current wetland areal estimates, which vary seven fold, lead to large discrepancies in estimates of overall exchange of CH4 between the atmosphere and northern sources and sinks (Aselmann and Crutzen, 1989) and thus to uncertainties in projecting climate processes. A satellite-based approach capable of accurately monitoring the global areal extent of northern wetlands is needed to address this issue.

Methane emissions from wetlands are controlled by factors that affect its product ion and consumption by complimentary processes related to soil microbial communities and the mechanism by which it is transported from the soil to the atmosphere. Methane is produced by anerobic bacterial decomposition of dead organic matter in waterlogged and highly reduced soils of wetlands and is consumed by aerobic microbial communities in overlying oxic soils or waters. The presence or absence and characteristics of the wetland vegetation are particularly important in determining exchange rates. The wetland vegetation provides the carbon source used in microbial decomposition through root exudation and dead plant material and, in the case of emergent herbaceous communities, an effective mechanism involving the plant vascular system for transporting methane from the soil to the atmosphere. The latter, termed plant mediated transport, is a secondary consequence of the mechanism vegetation communities use to move oxygen from the atmosphere to their roots in the underlying waterlogged and anaerobic environment. The result of these various controls is that methane exchange rates between the surface and the atmosphere can vary over two to three orders of magnitude but is closely associated with wetland type. For example, methane exchange rates are typically large for inundated herbaceous wetlands (defined here as fens) and relatively small for seasonally non-inundated wooded wetlands and open water areas within which methane exchange is controlled largely by soil- or water-atmosphere diffusion processes (Matson & Harriss, 1995. Morrissey and Livingston, 1992; Klinger et al., 1994; Bartlett and Harriss, 1993, Moore et al., 1990; Moore and Knowles, 1990; Crill et al., 1988, Harriss et al., 1985).

For purposes of estimating methane exchange between wetlands and the atmosphere, it is clearly important to know the areal extent of the various wetland types, particularly that of herbaceous inundated wetlands. The discrimination of these communities using classical optical remote sensing approaches, however, has proven challenging as their spectral signature was often confused with nonor low-methane emitting areas. The potential of optical remote sensing techniques to detect accurately standing surface water depths as shallow as 1 cm remains to be realized. While remote sensing data of clear, deep, open water areas are usually correctly classified, those open water areas which are clear and shallow are typically misclassified as bare soil or vegetation, depending upon the visibility of bottom soil or bottom growing underwater vegetation. No optical techniques to date has used sun glint from the specularly reflecting water surface to identify the presence of surface water and therefore wetlands.

Here our goal was to develop an improved approach for estimating the areal extent of the various types of wetlands, viz. m order of priority, inundated herbaceous wetlands (e.g., fens), open water (without emergent vegetation), and woody, seasonally inundated wetlands. As a first step to this end, we identified inundated areas directly, taking advantage of the specular reflecting properties of the surface waters (Cox and Monk, 1954; Breon and Deschamps, 1992). We tested three hypotheses. First, the visually blinding glitter of sunlight off ruffled water surfaces provides a strong, unique, angular signature reflection characteristic of wetlands and uncharacteristic of other cover types. Second, fens and open water areas display different, wind dependent glitter signatures as a function of view angle. Finally, analysis of remotely sensed data, collected in and near the specular direction, will allow discrimination of fens, open water areas and other cover types.

2 METHODS

2.1 Data collection

POLDER data (Breon and Deschamps, 1992) were collected, Fig. 1, from a C-130 airplane flying approximately five km above ground level on July 21, 1994 during the Boreas Experiment Intensive Field Campaign 2 (IFC-2). The Southern Study Area (SSA) of the Boreas Experiment, located near (54°N,105°W) in central Canada not far from Hudson's Bay, was sampled along numerous parallel flight lines, Fig. 1, each flown either toward or away from the solar azimuth direction

The POLDER data were collected in five spectral bands in the blue (443 nm), green (550 nm), red (665 nm) and near infrared (865 nm and 910 nm) spectral regions. The field of view of each image was approximately +/-50° along track (in the principal plane) and +/-43° across track (in the perpendicular plane). Approximately 95% of each image overlapped adjacent images.

2.2 Data analysis

Data were calibrated with reference to May 1994 sensor calibration To improve image band-to-band registration accuracy, the spatial resolution of each image was degraded using 3x3 pixel averaging, providing a ground pixel footprint sue of approximately 150m. Data were displayed, Fig. 2, using a Lambert conformal projection, removing the effects of changing sensor altitude and au-craft roll, pitch and yaw. The pixel data of all imagery collected within a flight line were redistributed to create fifteen new images, each representing a constant view angle in the principal plane along the flight line, Fig. 2. Only data within a few pixels of the principal plane were classified, using a parallelepiped or "levels.' classifier, in order to obtain angular signatures of the three classes, open water, fen and vegetation (non-inundated). The Normalized Difference Vegetation index (NDVI) was computed (865 am baud - 665 nm band)/(865 nm band + 665 nm band) for the various information classes as a function of view angle.

3 RESULTS

Fig. 2 shows one flight line viewed at zenith angles of +50°, 46°, 42° (specular direction), 36°, 30°, 7° (near the nadir direction), -38° and -43° (hot spot direction). The principal plane forms a horizontal line along the vertical centerline of each image. Glitter from the open water area at the left end of each of the images appears in each of the five zenith view angles between 30° and 50°. However, glitter from fens appears along the principal plane only for zenith view angles 42° and 46°, a much narrower range of angles.

These results show the difference in the angular spread of the glitter from fens and open water areas, suggesting this difference may be used to distinguish between them in remotely sensed data. The results, Fig. 2, show that all tens specularly reflecting sunlight are located within a few degrees of the principal plane. While one presumes there are fens located in the imagery at angles away from the principal plane, no fens are apparent, illustrating first the importance of the specular direction for fen detection and second the fact that fens appear to be green vegetation at angles away from the specular direction.

Fig. 3 illustrates our hypothesis that during moderate wind conditions the angular spread of the glitter from open water areas will be much greater than the angular spread of the glitter from vegetation covered fens. The results, Fig. 2, suggest that for intermediate wind speeds, the wind protecting effects of the vegetation growing above the water surface of fens will allow separation of fens and open water areas in remotely sensed data. Conversely, w hen wind speed is zero and the surfaces of both fens and open water areas are flat, then the angular spread of the glitter would be identical, causing them to be inseparable in remotely sensed data. And presumably at large wind speeds, the water surfaces of both fens and open water areas will likely be equally ruffled; the angular spread of the glitter will be equally large and fens and open water areas again would be inseparable in remotely sensed data.

Fig. 4 shows that for most view angles, values of the NDVI of fens are similar to the NDVI of vegetation, all being above approximately 0.75, except in the specular direction where it is less than 0.30, NDVI values are low in the specular direction because the glitter from the water surface of the fen is large in both wavelength bands, red and near infrared, which provides, when NDVI is calculated, a near zero number divided by a large number. Away from the specular direction, the fens appear to be green vegetation and display au appropriately large value of NDVI. This suggests that NDVI, when measured both in the specular direction and away from the specular direction will provide a metric sensitive to detection of fens.

Fig. 5 shows the radiance of vegetation, open water areas and fens as a function of view angle. Even in the hot spot direction the radiance of vegetation is much, much less than the radiances of open water areas and fens, which display peak values in the specular direction more than eight times those of vegetation Unlike the signature for open water areas, the signatures of vegetation and fens measured away from the specular detection do appear similar with each displaying a local maximum in the hot spot direction. The angular width of the specular peaks of the fens and the open water areas are approximately 6° and 30°, respectively suggesting that when represented in remotely sensed data fens and open water areas may be easily separated according to their differing responses with view angle.

A scatter diagram, Fig. 6, shows the radiance of fens, vegetation and open water areas measured both in the specular direction as well as 6° away from the specular direction. Data of open water areas, which appear bright in both these view directions, scatter in the upper right of Fig. 6. Data of vegetation which appears dark in both of these view directions, is clustered just above zero. Data of fens, which appear bright in the specular direction but comparatively dark 6° away from the specular direction, scatter along a line Just above the horizontal axis. The results show that the three information classes -- vegetation, fens and open water areas -- maybe easily separated using these two directional measurements: [the specular radiance] and [the radiance measured 6° from the specular direction]. The results suggest that analysis with a simple 'levels' or parallelepiped classifier will provide adequate classification accuracy for these data.

Fig. 7 shows results of a 'levels' classification of pixels in the principal plane of the segment. The results show that the open water area is classified as open water and vegetation as vegetation. From photo interpretation analysis, some areas of rivers appear to be misclassified as fens. Misclassification of rivers as fens might be due to the wind sheltering effects of tall woody vegetation located next to the river surface. Only pixels near the center line of the flight line are classified because the results, Fig. 5, show that glitter from fens, the key to the analysis process, appears only within +/-3° of the principal plane m these data.

4 CONCLUSION

Glitter and its observation with POLDER is a powerful tool for classifying and estimating the areal extent of northern, high latitude wetlands (specifically open water areas and fens) and non-wetlands.

REFERENCES

Aselmann, I., and P. J. Crutzen, Global distribution of natural freshwater wetlands and rice paddies, their net primary productivity seasonality and possible methane emissions, Journal of Atmospheric Chemistry, 8, 307-358, 1989.

Bartlett, K. B., and R. C. Harriss, Review and assessment of methane emissions from wetlands Chemosphere, 26, 261-320, 1993.

Breon, F.M. and P.Y. Deschamps, Optical and physical parameter retrieval from POLDER measurements over the ocean using an analytical model, Remote Sens. Environ. 43:1-20, 1992.

Cox C. and W. Munk, Measurements of the roughness of the sea surface from photographs of the sun's glitter, J. Opt. Sot. Am. 44:838850, 1954,

Crill, P. M., K. B. Bartlett, R. C. Harriss, E. Gorham, E. S. Verry, D. 1. Sebacher, L. Madzar, and W. Sanner, Methane flux from Minnesota peatlands, Global Biogeochemical Cycles, 2, 371-384, 1988.

Fechner, E.J., and H F. Hemond, Methane transport and oxidation in the unsaturated zone of a sphagnum peatland, Global Biogeochemical Cycles, 6, 33-44, 1992.

Harriss, R. C., E. Gorham, D. I. Sebacher, K. B. Bartlett, and P. A. Flebbe, Methane flux from northern peatlands, Nature, 315, 652-654, 1985.

Klinger, L.F., P.R. Zimmerman, J P. Greenberg, L.E~ Heidt, and A.B. Guenther, Carbon trace gas fluxes along a successional gradient m the Hudson Bay Lowland, J. Geophys. Res, 99, 1469-1494, 1994.

Matson, P.A. and R.C. Harriss, Biogenic Trace Gases: Measuring Emissions from Soil and Water, Blackwell Science Ltd., Cambridge, MA, 1995.

Matthews, E., and L Fung, Methane emission from natural wetlands: global distribution, area, and environmental characteristics of sources, Global Biogeochem. Cycles, 1, 61-86, 1987.

Moore, T.R. and R. Knowles, Methane emissions from fen, bog, and swamp peatlands in Quebec, Biogeochem, 11, 45-61, 1990.

Moore, T., N. Roulet, and R. Knowles, Spatial and temporal variations of methane flux from subarctic/northern boreal fens, Global Biogeochem. Cycles, 4, 29-46, 1990.

Morrissey, L A., and G.P. Livingston, Methane emissions from Alaska Arctic tundra: an assessment of local spatial variability, J. Geophys. Res., 97, 16661-16670, 1992.

Roulet, N.T., R. Ash, and T.R. Moore, Low boreal wetlands as a source of atmospheric methane, J. Geophys. Res., 97, 3739-3750, 1992.

Whalen, S. C., and W. S. Reeburgh, A methane flux time series for tundra environments, Global Biogeochemical Cycles, 2, 399-410, 1988.

1998, Center for Spatial Technologies and Remote Sensing (CSTARS)
University of California, Davis