1. INTRODUCTION

Perhaps one of the most important environmental data sets to be collected over the past half-century is the continuous record (since 1955) of the of atmospheric concentration of carbon dioxide at the Mauna Loa conservatory. Because of its role as a key greenhouse gas in the atmosphere, there is a strong consensus that the rising atmospheric carbon dioxide level over the past century has already resulted in an increase in average global air temperature (Hansen et al. 1996), and continuing increases in air temperature should be expected during the next century if atmospheric carbon dioxide levels are not reduced (Houghton et al. 1996).

Development of strategies to adapt to and mitigate the effects of global climate change not only requires an understanding of how changes in the concentrations of atmospheric greenhouse gases influence climate patterns, but also knowledge of how these greenhouse gases are cycled between the atmosphere and the earth's biosphere and geosphere (Watson et al.1996). The scientific community's understanding of this cycling is insufficient to determine how a significant portion of the carbon dioxide being added through the burning of fossil fuel is being absorbed by the biosphere/ geosphere. At the present time, a 2 to 3 Gt yr^-1 carbon sink needs to be identified in order to balance the global carbon cycle. There are compelling arguments that this sink must be terrestrial in origin (Solomon and Cramer 1993).

2. STUDY GOALS AND RATIONALE

The overall goal of this SIR-C/X-SAR investigation was to develop new approaches to estimate long-term carbon source-sink relationships in the forests of central North Carolina using spaceborne remote sensing imagery. The rationale for this goal is as follows:

1. Much of the uncertainty in the global carbon cycle lies in the rates of deforestation and reforestation in the world's forests (Houghton 1996). Over the past several decades, the old-field pine forests found in the southeastern U.S. have represented the single largest source of commercial timber in the world. Thus, the rates of deforestation/reforestation in this region are extremely important in understanding the terrestrial carbon budget.
2. Landsat imagery represents a unique and valuable resource for monitoring long-term, land-cover change. It is a particularly powerful tool for monitoring patterns of deforestation/reforestation (Johnson et al. 1997).
3. Imaging radar data provide a valuable tool for monitoring changes in aboveground biomass in pine forests, particularly in the early stages of forest succession (Dobson et al. 1992; Kasischke 1992; Kasischke et al. 1994a; 1995a; 1997a; 1997b), or in forests where low levels of biomass are found (Harrell et al. 1995).

3. OVERALL APPROACH

Figure 1 outlines the overall approach used in this study, which includes the exploitation of ground-based data sets (stand biomass and ground carbon data) as well as two sources of remote sensing data: (1) time-series Landsat MSS and TM imagery collected between 1973 and 1993; and (2) SIR-C/X-SAR data collected in 1994. The utilization of Landsat data is based on the fact that it is the best source of information on the overall pattern of land and forest cover and provides a historical perspective on the patterns of deforestation and reforestation in this region. Since imaging radar's sensitivity to variations in aboveground biomass is limited to forests with biomass levels of about 150 to 200 tonnes/hectare (Harrell et al. 1997; Kasischke et al.1997a; 1997b), the use of time-series Landsat data provides a unique means to identify those areas with conditions where the use of the SIR-C/X-SAR data results in the lowest errors.

The exploitation of Landsat data for the present study began with the generation of a land-cover map for a time period (fall of 1993) similar to when SIR-C/X-SAR data were collected (spring and fall of 1994). This land-cover map was then modified through a sequential procedure to create a land-cover map for the beginning of the study period in 1973. Eight basic land-cover categories were used for this map: young pine forest, mature pine forests, mature deciduous forest, mixed deciduous/pine forest; agricultural land; water, urban/road, and suburban. Changes were made to the baseline maps using land-cover changes detected in pair-wise analyses of Landsat TM and MSS data collected between 1973 and 1993. These land-cover change maps were created through an automated change vector analysis procedure based on detection of changes in the radiometric brightness in different TM or MSS channels between two dates (Johnson et al. 1997), not on changes in categories created from maps generated from the Landsat data. Change-detection maps were created for the following image pairs: 1973-1979; 1979-1984; 1984-1989; and 1989-1993.

To assign landscape carbon levels to the 1973 baseline map, a look-up-table approach has been adopted. The aboveground and soil carbon levels in this look-up table were derived from a variety of sources, including: (1) state forestry records; (2) soil maps and surveys; (3) records from ongoing research, including that collected during this experiment; and (4) scientific literature. Gross changes in the carbon budget for this region are then derived through monitoring changes in land-cover using the time series Landsat data.

A refined estimate of carbon fluxes will be derived using the SIR-C/X-SAR data. First, algorithms have been developed to use SIR-C/X-SAR data to monitor aboveground biomass. The initial biomass algorithms were derived from JPL AIRSAR data (Kasischke et al. 1995). Initially, a wide range of biomass estimation approaches were evaluated using two different dates of SIR-C data (Harrell et al. 1997), while a second analysis considered SIR-C and X-SAR data collected over a number of different dates (Kasischke et al. 1997a). The second step in this analysis involved identification of all sites within the test area which were determined to be deforested and reforested between 1973 and 1993. The biomass estimation algorithms were then exercised for these specific sites to estimate the aboveground biomass levels at these sites in 1994. Based on the biomass and ground carbon data, two sets of curves were generated; (1) one set of curves to estimate patterns of aboveground biomass as a function of stand age (Haney 1995; Kasischke et al. 1994b); and (2) a second set of curves which describe patterns of ground-carbon accumulation as a function of stand age and aboveground biomass for the different soil types in the test region. The information on aboveground biomass derived from SIR-C/X-SAR and date of cutting derived from Landsat is then combined with these curves to determine how much carbon has been sequestered per year for the different deforested/reforested test sites.

4. RESULTS

This project is scheduled for completion at the end of 1997; thus, only interim results are ready for presentation at this time.

The analysis of the SIR-C/X-SAR data sets collected over the Duke Forest Supersite are nearing completion. The initial studies of these data focused on evaluating various biomass estimation approaches, as well as the differences between using data collected in the fall and spring. Data collected at an incidence angle of 32 were used in this initial study. During the second phase of the biomass studies, data from a variety of incidence angles were used, and X-SAR imagery were also included in the analyses.

The results of these initial studies showed that multi-stage approach developed for the Duke Forest by Kasischke et al. (1995) performed better than approaches developed at other test sites (Harrell et al. 1997). In addition, using data collected in the fall resulted in lower errors than using data collected in the spring. It was concluded that the high soil moistures present during the spring data collections resulted in more variation in scattering from the test stands, which in turn, resulted in lower accuracies for biomass estimation. The RMS errors for the biomass estimation using the fall data were 4.14 kg m^-2 for stands with biomass levels less than 10 kg m^-2, 5.91 kg m^-2 for stands with biomass less than 20 kg m^-2, and 8.66 kg m^-2 for all stands.

Figure 2 presents an initial attempt to create a biomass map for the Duke Forest region using SIR-C/X-SAR imagery. This image was derived using data collected on 3 October 1994 at an incidence angle of 32. This map is for those areas covered by pine trees as determined through a land-cover map generated from a 1993 Landsat Thematic Mapper image.

The next step of the analysis involves using deforestation maps generated from analysis of time-series Landsat data to identify specific stands to estimate biomass using the SIR-C data. Figure 3 presents an image depicting patterns of deforestation for the western portion of the study area. This image covers an area of 21 by 15 km in size, or 31,500 hectares. Of this area, in 1973 a total 4590 hectares were covered by pine forest. During the period of 1973 to 1993, an estimated 1877 hectares of pine forests were cleared, or 41% if the total. The most land was cleared in the 1973 to 1979 time period, when a total of 611 hectares of pine forest was cut.

5. CONCLUSIONS

The Duke Forest SIR-C Investigation is the culmination of a series of studies carried out over the past decade to develop approaches to use radar imagery to monitor forested ecosystems. These studies have included both theoretical and empirical analyses and have involved research on data collected by airborne and spaceborne SAR systems. They have clearly demonstrated that imaging radar systems can be used to monitor patterns of forest regrowth in the pine forests found throughout the southeastern United States. Figure 4 contains a series of L-band (HH) radar images of a several forest stands which were cleared and replanted in the early 1980's. The 1984 image was collected by the CCRS Convair-580 Airborne SAR; the 1989 image was collected by the U.S. Navy's P-3 Airborne SAR and the 1994 image was collected by SIR-C. The fact that SAR is sensitive to forest regrowth is clearly evident on this imagery.

The ongoing studies at the Duke Forest have covered a wide range of topics, including:

1. Investigation of the basic relationships between variations in aboveground, woody plant biomass and radar backscatter (Dobson et al. 1992; Harrell 1990; Kasischke 1992; Kasischke et al. 1994a; 1995), including development of approaches to characterize forest ecosystems in ways to which the principal forest stand characteristics resulting in the microwave backscatter can be meaningfully compared to the radar signatures (Haney 1994; Kasischke et al., 1994b).
2. Comparison of results from theoretical models to observed variations in radar backscatter to develop a better understanding of the primary sources of variation in SAR images of forested areas (Sheen and Johnson 1992; Wang et al. 1994, 1995).
3. Development of algorithms to estimate aboveground biomass from multi-channel radar imagery, including the accuracies associated with these algorithms (Kasischke 1992; Kasischke et al. 1995; 1997a; Harrell et al. 1997). The Duke Forest Supersite was unique in terms of biomass estimation algorithms for several reasons. First, the range of biomass in the different test stands spanned a range from 0 to 44 kg m^-2. Second, there were a large number of stands used in this study (50). This large number allowed for the creation of both training and testing test stands. And third, the approaches used to sample the test stands involved measurement of trees from a statistically significant number of trees within each stand.
4. Development of approaches to use radar imagery to study a range of forest ecosystem processes (see, e.g., Kasischke and Christensen 1990; Kasischke et al. 1997b; Ustin et al. 1991). The overall modeling approach described in Figure 1 represents one manner in which SIR-C/X-SAR data will be used to develop a better understanding of carbon fluxes from forest ecosystems.

REFERENCES

Dobson, M.C., F.T., Ulaby, T. Le Toan, A. Beaudoin, E. S. Kasischke, and N.C. Christensen, Dependence of radar backscatter on conifer forest biomass, IEEE Trans. Geosci. Remote Sens., 30, pp. 412-415, 1992.

Hansen, J., R. Ruedy, M. Sato, and R. Reynolds, Global surface air temperature in 1995: Return to pre-Pinatubo level, Geophysical Research Letters, 23, 1665-1668, 1996.

Haney, E., The effects of competition on the morphology and patterns of biomass accumulation in loblolly pines (Pinus taeda L.), Duke University M.S. Thesis, Durham, North Carolina, 1994.

Harrell, P.A., Monitoring aboveground biomass in deciduous forests using multichannel airborne SAR imagery, Duke University M.S. Thesis, Durham, North Carolina, 1990.

Harrell, P.A., L.L. Bourgeau-Chavez, E.S. Kasischke, N.H.F. French and N.L. Christensen, Sensitivity of ERS-1 and JERS-1 radar data to biomass and stand structure in Alaskan boreal forest, Remote Sens. Environ., 54, pp. 247-260, 1995.

Harrell, P.A, E.S. Kasischke, L.L. Bourgeau-Chavez, E. Haney, and N.L. Christensen, Comparison of approaches to estimate of aboveground biomass in southern pine forests using SIR-C data, Remote Sensing Env. 59, pp. 223-233, 1997.

Houghton, R.A., Land-use change and the temporal record. In Forest Management and the Global Carbon Cycle. M.J. Apps and D.T. Price, editors, NATO ASI Series, subseries 1, Vol. 40 "Global Environmental Change", Springer-Verlag, Berlin, pp. 117-134, 1996.

Houghton, J.T., L.G. Meira Fihlo, B.A. Callander, N. Harris, A. Kattenberg, and K. Maskell. 1996. Climate Change 1995 - The Science of Climate Change. Cambridge University Press, Cambridge, England, 572 pp.

Johnson, R.D., J. Colwell, N. Roller, and E.S. Kasischke, automatic mapping of land cover changes using change vector analysis - a review, Int. J. Remote Sens. (in review) 1997.

Kasischke, E.S., Monitoring changes in aboveground biomass in loblolly pine forests using multichannel synthetic aperture radar imagery, University of PhD Dissertation, Ann Arbor, Michigan, 1992.

Kasischke, E.S. and N. L. Christensen, Jr., Connecting forest ecosystem and microwave backscatter models, Int. J. Remote Sens., 11, pp. 1277-1298, 1990.

Kasischke, E.S., L.L. Bourgeau-Chavez, N.L. Christensen, Jr., and E. Haney, Observations on the sensitivity of ERS-1 SAR image intensity to changes in aboveground biomass in young loblolly pine forests, Intern. J. Remote Sens., 15, pp. 3-16, 1994a.

Kasischke, E.S., Christensen, N.L., Jr., and E. Haney, Modeling of geometric properties of loblolly pine tree and stand characteristics for use in radar backscatter models, IEEE Trans. Geosci. Remote Sensing, 32, pp. 800-822, 1994b.

Kasischke, E.S., Christensen, N.L., Jr., and L.L. Bourgeau-Chavez, Correlating radar backscatter with components of biomass in loblolly pine forests. IEEE Trans. Geosci. Remote Sensing, 33, 643-659, 1995.

Kasischke, E.S., K. Smith, P. Harrell, and L.L. Bourgeau-Chavez, Optimum methods for estimating aboveground biomass in southern U.S. pine forests using SIR-C/X-SAR data, Int. J. Remote Sens. (in review), 1997a.

Kasischke, E.S., J.M. Melack, and M.C. Dobson, The use of imaging radars for ecological applications - a review, Remote Sens. Env. 59, pp. 141-156, 1997b.

Sheen, D.L. and L.P. Johnston, Statistical and spatial properties of forest clutter measured with polarimetric synthetic aperture radar (SAR), IEEE Trans. Geosci. Remote Sensing, 33, 578-588, 1992.

Solomon, A.M. and W. Cramer. 1993. Biospheric implications of global environmental change. In Vegetation Dynamics and Global Change, A.M. Solomon and H.H. Shugart, editors, Chapman and Hall, New York, pp. 25-52.

Ustin, S.L., C.A. Wessman, B. Curtiss, E. Kasischke, J. Way, and V. Vanderbilt, Opportunities for Using the EOS Imaging Spectrometers and Synthetic Aperture Radar in Ecological Models, Ecology, 72, pp. 1934-1945, 1991.

Wang, Y., F.W. Davis, J.M. Melack, E.S. Kasischke and N.L. Christensen, Jr., The Effects of Changes in Forest Biomass on Radar Backscatter from Tree Canopies, Int. J. Remote Sens., 16, pp. 503-513, 1995.

Wang, Y., E.S. Kasischke, F.W. Davis, J.M. Melack, and N.L. Christensen, Jr., The Effects of Changes in Loblolly Pine Biomass and Soil Moisture Variations on ERS-1 SAR Backscatter - a Comparison of Observations with Theory, Remote Sens. Env., 49, pp. 25-31, 1994.

Watson, R.T., M.C. Zinyowera, R.H. Moss and D.J. Dokken. 1996. Climate Change 1995 Impacts,

Adaptations and Mitigation of Climate Change: Scientific-Technical Analyses, Cambridge

University Press, Cambridge, England, 878 pp.LIST OF FIGURES

1. Overall Approach for Estimating Carbon Flux in Southern Pine Forests.

1. SIR-C Derived Estimates of Pine Forest Biomass within The Duke Forest Supersite.

1. Locations of Pine Forest Clearing in the Western Portion of the Duke Forest Supersite as Determined Through Analyses of Landsat MSS and TM Data.

1. Observation of Forest Regrowth over a Ten-Year Time Span using Airborne and Spaceborne L-Band (HH) SAR Imagery.