Using ERS-1 Data to Measure and Map Selected Conditions Related to the Production of Methane in a Wetland Environment: The Nebraska Sandhills, USA

Abstract

Wetlands are important as both sources and sinks for methane, a trace gas implicated in greenhouse warming. Information necessary for estimating fluxes of methane includes plant species and primary production. Measuring and monitoring the spatial and temporal variations in these parameters should facilitate modeling fluctuations in global greenhouse gas amounts. Three ERS-1 and two Landsat-TM datasets, acquired over large wetland sites in the Western Sandhills of Nebraska, were obtained during the 1995 growing season. The SAR images were despeckeled, and both ERS and TM were resampled to a 30m spatial resolution, rectified to a UTM coordinate system, and segmented to exclude uplands using a digital National Wetlands Inventory (NWI) dataset. In-situ reference data were obtained at various sites within the study area in conjunction with satellite overpasses. The ERS data were classified in an attempt to identify and map the distribution of wetlands at two levels: 1) specificity roughly equivalent to that of the NWI; and 2) at the species level, with emphasis on Typha, Scirpus, and Phragmites. The TM scenes were also classified and compared to the SAR result. Finally, a combined ERS and TM dataset was classified. Both SAR backscatter and TM reflectance were correlated to field measures of above-ground biomass. Results were considered with regard to C- and L-band (VV and VH) scatterometer data acquired at close-range over experimental plots containing Typha and Phragmites. Landsat-TM seems better than ERS for generalized classification and mapping, and for emulating the NWI product. However, TM is limited with regard to identification of individual wetland species, and sparse stands of emergent, floating, and submergent macrophytes. ERS seems better than TM for detecting individual species, and sparse stands of emergent, floating, and submergent macrophytes. But, the ERS multi-temporal classification is not easily adapted for most practical mapping applications. Wind must be considered when classifying ERS in shallow lake systems. Potential exists for ERS/TM classification, but more work is needed. Biomass estimation, while encouraging, needs refinement. C- and L-band scatterometer data suggest that polarization is more important in distinguishing Typha and Phragmites than frequency.

Keywords: ERS-1, Wetland species identification, Methane production, National wetland inventory.

1. Background

Our research is aimed at the synergistic use of ERS-SAR and Landsat Thematic Mapper (TM) digital data for characterizing, mapping, and monitoring wetland vegetation in the Sandhills of Nebraska, USA. Classifying and mapping wetlands in that region, generally done for evaluation of wildlife habitat, has been a priority in Nebraska for over 40 years (Seevers et al., 1975; Rundquist, 1983). Therefore, it seemed natural to consider ERS as a potential opportunity to improve on-going efforts to monitor wetlands of the study area.

The work is also linked to another project involving estimation of methane fluxes in the large wetland communities of the Sandhills (Verma et al., 1996). These wetlands are significant as both sources and sinks for methane, a trace gas implicated in "greenhouse warming" (Mitsch and Gosselink, 1993). Parameters for estimating fluxes of methane in a marsh include, among other things, plant species and primary production. Speciation is useful because there are differences in both amount of gas released to the atmosphere and the actual mechanism of transfer, from one species to the next (Vanyarkho and Arkebauer, 1995). Productivity is important because it corresponds to total amount of organic material available for decomposition. Measuring and monitoring the spatial and temporal variations in speciation and production may ultimately facilitate modeling fluctuations in global greenhouse gas amounts.

1.01. Objectives

The general goal of the project was to improve on previous classifications of wetlands in the study area by using ERS. Therefore, we sought, first of all, to test the utility of the SAR for identifying and mapping wetlands at a "general" level. The research consideration, then, was whether or not one could use ERS data, either alone or in combination with Landsat-TM, to classify wetlands at the level of detail provided by the U.S. National Wetlands Inventory (NWI) (Cowardin et al., 1979). At a more specific level, the research focused on two parameters, plant species and primary production, both linked to methane flux. With regard to plant species, we aimed to: 1) examine the utility of ERS data for classifying Typha, Scirpus, and Phragmites; 2) compare ERS results to a similar classification based on multi-temporal TM data; and 3) evaluate the potential for enhancing TM classifications of wetlands by combining the Landsat data with ERS. We also assessed the utility of both ERS and TM in estimating the above-ground biomass in Sandhills wetlands. As a final objective, we evaluated C-and L-band (VV and VH) scatterometer returns for distinguishing Typha and Phragmites.

1.02. Study Areas

The study area was in the Western Sandhills of Nebraska (Figure 1). The region consists of a virtual "sea" of sand dunes that are generally stabilized by a veneer of grass. Deflation hollows are present where the grass cover has been interrupted. Low-lying, flat-floored valleys, some wet and some dry, are prevalent throughout the region. The wet meadows contain lush vegetation, which exists in sharp contrast to the sparsely vegetated uplands and dry valleys. Rainfall results in virtually no runoff because of the sandy soils, so recharge is rapid and extensive. As a result, there is a tremendous reserve of ground water beneath the Sand Hills, and the hundreds of shallow lakes and wetlands in the western portion of the region are, for the most part, surface expressions of the water table. Dominant species of wetland emergents in the study area are broadleaf cattail (Typha latifolia), hardstem bulrush (Scirpus acutus), and common reed (Phragmites communis).

The specific study site for this paper was at the Crescent Lake National Wildlife Refuge (CLNWR). Our brief discussion will focus primarily on the expanse of wetlands surrounding Island Lake (refer to Figure 1).

A secondary study site was at the University of Nebraska Agricultural Research and Development Center (ARDC), located near Lincoln, Nebraska. Our research group operates a facility at ARDC for collecting spectral data over homogenous stands of wetland species under tightly controlled conditions. The cultivation of wetland species at ARDC began in early 1992, when four experimental plots, each approximately 5x5m were constructed. In more recent years, plot size has increased beyond 30x30m. Rhizomes of several species including Typha, Scirpus, Phragmites, and others were purchased from commercial nurseries and planted in separate plots. Optical data (Spectron SE-590) have been acquired numerous times in each growing season since original planting.

2. Methodology

2.01. Satellite Data

Three ERS-1 images, acquired over the Western Sandhills on June 3, July 8, and September 16, 1995, comprise the principal datasets for the study. Corrections to data prior to our receipt included adjustments for slant angle, antenna pattern, and radiometric fall-off. We were unable to calibrate backscatter because we lacked proper ground-calibration targets. The ERS images were despeckeled using a Lee-sigma filter, and a 3x3 moving average. Each image was normalized so that its minimum value was always zero. This was accomplished by simply subtracting the lowest data value within an image from all data comprising that image.

Two Landsat TM scenes, acquired over the study area on June 14 and August 17, 1995 were obtained to supplement the ERS data. Digital numbers were converted to reflectance using standard procedures (Markham and Barker, 1986). The ERS and TM datasets were both resampled to a 30m spatial resolution using a nearest-neighbor algorithm, and rectified to a UTM coordinate system.

2.02. Field Radar-Scatterometer Data

Radar data at the ARDC experimental site were acquired at approximately weekly intervals during the 1996 growing season over pure stands of Typha and Phragmites. The Scirpus plot was deemed too sparse for meaningful data collection in 1996. The van-mounted radar scatterometer acquired data at both C-band (5.3 Ghz) and L-band (1.275 Ghz) at incidence angles of 0, 12, 24, 36, and 48 degrees. Polarizations included HH, HV, VH, and VV. All measurements were made at 2.5m above the top of the canopy. Ancillary data included spectroradiometer data (Spectron SE-590) in visible and near-infrared spectral regions, canopy-light extinction (Li-Cor LAI-2000), and digital photography of the canopies (Kodak DC-40).

2.03. In-Situ Vegetation Data

In-situ above-ground biomass data were obtained at 41 sites along 7 transects around the Island Lake site concurrent, or nearly so, with satellite overpasses. The location of each sampling site was determined to approximately 2m using differential GPS. All vegetation in an area of 0.25m squared was clipped and bagged. Other measurements, such as height, numbers of green plants, etc. were made. Upon returning to Lincoln, material was sorted and to separate green from senesced vegetation and then weighed.. Samples were oven-dried and re-weighed. Identical procedures were employed at the ARDC-controlled site, except that the areas sampled were 0.10m squared, due to relatively small plot sizes. Surface cover and species identification were also recorded at 46 sites within and around Island Lake during the 1995 growing season. The location of each site was again determined using differential GPS. These sites were ultimately used as control points to check the accuracy of the classified images. Unfortunately, Phragmites sites were not identified during the 1995 field survey.

2.04. Classification Procedures

Data for topographic uplands were removed from all ERS and TM scenes using the NWI as a "cutter file." Therefore, only lowland areas were digitally classified. Three separate classifications were executed: 1) three dates of C-band, like-polarized ERS (i.e., three channels of information); 2) two dates of green, red, and near-infrared TM (i.e., six channels); and 3) two dates of green, red, and near-infrared TM combined with one date (July 8) of ERS (i.e., seven channels). The clustering was accomplished with an unsupervised minimum spectral distance algorithm in Erdas IMAGINE. Results were evaluated in two ways: 1) by overlaying the digital NWI classification for the Island Lake area; and 2) by relating pixel locations to GPS reference points and corresponding documentation acquired during in-situ vegetation sampling.

3. Results and Discussion

3.01 Multi-Temporal ERS

The first classification used three separate ERS scenes, June 3, July 8, and September 16, as input. The lake surface was rough in the June and September images and smooth in the July image. Backscatter from areas of emergent vegetation was relatively low to moderate in June, moderate to high in July, and low to moderate in September. Backscatter from areas of near-surface submergents was relatively high in June, very low in July, and moderate in September. Therefore, we believe that it may be possible, under some conditions, to distinguish between emergent and submergent vegetation, especially by using multi-temporal ERS images where one image contains a smooth lake surface and another a rough lake surface. It seems that the necessary conditions for detecting submergents include plants being at or very near the surface where they impact wave action.

From the classification, a total of 50 clusters resulted, and these were reduced to seven, based on both ground-truth and other first-hand knowledge of the site. Results are depicted in Figure 2. Notice the "salt-and-pepper" appearance of the classified multi-temporal ERS data. This is due to the fact that the radar backscatter is impacted by all vegetation, however small, "emerging" from water surface, as well as any floating algal mats or debris. The result is a classified image that seems "noisy," with an appearance of "almost too much information." Our conclusion is that ERS data are not easily adapted for classifying wetlands at the generalized NWI level (compare Figures 2 and 3).

We believe the ERS to hold potential for classifying wetlands at the level of individual species. However, when the ERS classification is compared to ground-reference sites, the result is rather unimpressive. It should be noted, though, that the relatively low number of ground-control points, the difficulty of isolating pure stands of individual species in the field and the resulting necessity of including a "mixed class" of vegetation, the lack of reference sites specifically for Phragmites, and the 30m pixel size may all have contributed to the low classification accuracy.

A positive finding in using multi-temporal ERS-1 data was that it allowed detection of sparse stands of emergent, floating, and submergent (near-surface) macrophytes. It has, in the past, been difficult for us to detect the presence of these vegetative groups.

We also determined that wind is a factor in classification of ERS data in Sandhills wetlands. Areas with emergent vegetation and areas with large amounts of near-surface submergents show a lower radar backscatter, compared to the water itself, when the water surface is rough. Conversely, when the water surface is smooth, areas with emergent vegetation show a higher backscatter than the water. The decrease of backscatter in vegetated areas, when the water surface is rough, is probably due to the dampening of wave action by the plants. Thus, while wind seems, at first to be detrimental to successful classification, it can be beneficial when classifying wetlands in shallow lakes using orbital SAR.

3.02 Multi-Temporal TM

The classification using TM data was certainly "smoother" than that for ERS, and it was easily adapted to emulate the NWI map for the area (compare Figures 3 and 4). On the "negative side," however, Island Lake contains innumerable small, sparse stands of emergent macrophytes (mostly Scirpus), floating, and submergent macrophytes. The preponderance of those stands were invisible to the TM. Despite these omissions, the TM classification was "statistically" slightly better at the "species level" than that for ERS (refer to Table 1). Once again, however, the reader should be reminded of the potential accuracy-assessment shortcomings noted in a previous paragraph (section 3.01).

3.03 Combined ERS and TM

The classification involving six TM (green, red, near-infrared for two dates) and one ERS (July 8) datasets is shown as Figure 5. The result seems, from a practical mapping point of view, to be a rather nice "compromise" between the relatively high level of detail provided by the ERS and the relatively lower level of specificity provided by the TM. The high percentage of incorrectly identified land cover at the control-point sites (Table 1) seems rather incongruous, and further work on accuracy assessment is warranted. The identification of submerged / near surface vegetation was diminished when the TM data were merged with ERS. Most of the error seems to have occurred in three areas: misidentification of bulrush as water or mixed vegetation; inability to distinguish between the mixed and cattail classes; and lack of submergent detection. It seems reasonable that the single date radar image was "overwhelmed" by the TM data. In retrospect a better classification may have been attainable if fewer TM bands had been used.

3.04 Biomass Estimation

Additional research is clearly required in the instance of biomass estimation with ERS, TM, and/or ERS/TM. Scatterplots show distinct clusters of points by species, but correlations were relatively low. Despite the low correlations, it seems that potential exists for using ERS and TM synergistically to predict biomass over the study area.

3.05 Field Radar-Scatterometer

Scatterometer data from the 1996 growing at the ARDC site provide some insight into potential problems encountered when classifying ERS data for the Sandhills study area. Both C- and L-band VV and VH data are summarized in Figure 6. The graphic suggests that wavelength is not as important as polarization in distinguishing Phragmites from Typha. However, results are very preliminary, and more data collection is required. We anticipate continuing our scatterometer work over controlled wetland sites during the 1997 field season.

4. Summary and Conclusions

Although the results are preliminary, the work suggests to us that Landsat-TM is superior to ERS if one is concerned with mapping wetland communities at a very general level, such as the classes provided in the U.S. National Wetland Inventory. However, TM does not allow easy identification of individual species, and sparse stands of emergent, floating, and submerged macrophytes. Despite our very limited quantitative assessment of accuracy in classification, multi-temporal ERS would seem to hold great potential for detecting individual wetland species. The ERS backscatter should ultimately allow us to better characterize the architecture of vegetation canopies in Sandhills wetlands. Despite the lack of quantitative proof, we believe that merged ERS and TM datasets will ultimately improve our ability to analyze, map, and monitor the spatial extents of conditions pertinent to the natural production of trace gases, including methane.

Other findings can be briefly reiterated. Our cursory evaluation of biomass estimation with both ERS and TM yielded inconclusive results, and more work is clearly necessary. Wind may be considered both detrimental and beneficial in classifying wetlands in shallow lakes using orbital SAR. On the negative side, water surfaces roughened by wind complicates classification; on the positive side, a comparison of backscatter when lake surfaces are rough versus smooth may facilitate the identification of sparce stands of emergent, floating, and submergent species. Preliminary field-scatterometer results suggest that the single-polarization (VV) characteristic of the ERS may inhibit successful classification of certain wetland species.

5. References

Cowardin, L.M., V. Carter, F.C. Golet, and E.T. LaRoe, 1979:

Classification of Wetlands and Deepwater Habitats of the United States. Office of Biological Services, Fish and Wildlife Service, U.S. Department of the Interior, Washington, D.C.

Markham, B.L. and J.L. Barker, 1986:

Landsat MSS and TM post-calibration dynamic ranges, exoatmospheric reflectance and at-satellite temperatures. EOSAT: Landsat Technical Notes, No. 1, August.

Mitsch, W.J. and J.G. Gosselink, 1993:

Wetlands (2nd Ed.). New York: Van Nostrand Reinhold.

Rundquist, D.C., 1983:

Wetlands inventories of Nebraska’s Sandhills. Resource Report #9, Conservation and Survey Division, University of Nebraska-Lincoln, 46 pp.

Seevers, P. R. Peterson, D. Mahoney, D. Maroney, and D. Rundquist, 1975:

An inventory of Nebraska wetlands with the use of imagery from the Earth Resources Technology Satellite. Proceedings, 4th Annual Remote Sensing of Earth Resources Conference, 281-292.

Vanyarkho, O. and T.J. Arkebauer, 1995:

Responses of Phragmites australis and Scirpus acutus leaf and canopy gas exchange to environmental conditions. Ecological Society of America, 80th Annual Meeting, Snowbird, Utah.

Verma, S.B., T.J. Arkebauer, F.G. Ullman, D.P. Billesbach, J. Kim, R.J. Clement, D.W. Valentine, D.S. Schimel, and E.A. Holland, 1996:

An integrated investigation of carbon dioxide and methane fluxes in mid-latitude prairie wetlands: micrometeorological measurements, process-level studies and modeling. Third Annual Report, Great Plains Regional Center for Global Environmental Change, U.S. Department of Energy, 87-92.

Figure 1. Study area.

Table 1.

Figure 2. ERS-1 Classification

Figure 3. National Wetland Inventory classes

Figure 4. TM classification	Figure 5. TM/ERS-1 classification

Figure 6. C-Band scatterometer data

Keywords: ESA European Space Agency - Agence spatiale europeenne, observation de la terre, earth observation, satellite remote sensing, teledetection, geophysique, altimetrie, radar, chimique atmospherique, geophysics, altimetry, radar, atmospheric chemistry