Abstract - SIR-C/XSAR science activities for this project have included 1) determining forest attributes from AIRSAR and SIR-C/XSAR data, 2) developing, testing and using ecosystem and radar backscatter models to improve understanding of northern forests; 3) studying the spatial structure of forests as revealed by SIR-C/XSAR data; and 4) examining the use of topographic models, developed, from SAR interferometry for future ecosystem studies. This report describes results in each of the areas with an emphasis on most recent results.
Procedures and results of using SIR-C/XSAR data to map forest cover type and above ground forest biomass were developed and applied to the Canadian boreal forest (Prince Albert and Nelson House, BOREAS study areas) and Howland, Maine. As part of the Boreal Ecosystem Atmosphere Study (BOREAS) an investigation was made of the use of Shuttle Imaging Radar-C (SIR-C), X-band synthetic aperture radar (XSAR) and Landsat TM data for estimating total and component above ground woody biomass in boreal forest study sites in Canada. Relationships of backscatter to total biomass and total biomass to foliage, branch, and bole biomass were developed and used to estimate biomass across the landscape. The component biomass procedure required forest type classification with SAR and Landsat data using well-known mapping techniques with combinations of SAR channels. Field measurements included plot level mensuration (species, stem diameter, height, density, and basal area) and tree geometry measurements (leaf, branch, bole size and angle distributions).
The results for the BOREAS Southern Study Area show that above ground biomass can be estimated to within about 1.6 kg/m2 and up to about 15 kg/m2 across the SIR-C images evaluated. A general method produced equivalent results to those obtained by treating forest type (Pine, Spruce and Aspen) separately. The biomass mapping was extended to bole, branch and foliage components from relationships with total above ground biomass developed from detailed tree measurements. Average biomass within the imaged area was estimated to be about 7.3 kg/m2 with biomass components of bole, branch and foliage comprising 83%, 12%, and 5% of the total. Examination of the scaling of biomass estimates from remote sensing images of varying resolution shows that information at scales useful for ecosystem models can be obtained. In addition, the biomass estimation technique provides similar information at different image resolutions. (see Ranson et al., 1997) Similar biomass mapping techniques were applied to the BOREAS Northern Study site with similar results.
The capabilities of AIRSAR and SIR-C/X-SAR images for characterizing a northern hardwood-boreal transitional forest located near Howland, Maine were evaluated and compared. The use of multiple frequency, polarimetric information to produce forest cover classification, biomass estimates and forest spatial pattern analysis were investigated. The results from SIR-C/X-SAR compared to those from AIRSAR were generally similar despite different available frequencies (C-band, L-band, X-band vs. C-band, L-band, P-band) and resolution (25.0 m vs. 8.3 m).
AIRSAR data was able to map stands of hardwood and mixed forests better than SIR-C/X-SAR data. However, SIR-C/X-SAR produced better classification results for conifer forest stands. There was no great benefit from using higher resolution for classification except for forest stands where mixtures of species (i.e., hardwood and softwood) occur. The comparison of the image data also showed that both instruments could provide reasonable estimates of biomass density up to about 15 kg/m2. At higher biomass levels both AIRSAR and SIR-C showed reduced sensitivities Figure 1 shows the SIR-C and AIRSAR biomass maps for a coincident area within the Howland, Maine test site. Finally, spatial character of the image data was examined using perimeter and area relationships and lacunarity analysis. The results were consistent between the two instruments and showed that the forest opening patterns were self-similar for openings greater than about 3 ha. (see Ranson and Sun, 1997)
Table 1 compares the biomass equations used for analysis of the three Backup Supersites. LHV and CHV channels were selected for each equation. The LHH band was selected for the BOREAS SSA site only. The differences in band coefficients indicate that the equations are valid only for the sites where they were developed. From Table 2 it can be seen that the Maine site has the largest average biomass over the test site, the BOREAS SSA the next highest and the BOREAS NSA, the least. These relative differences can be visualized in Figure 2 which shows the biomass for the three areas. Note that the coefficient of variation (Std. Dev/Mean*100) are smaller for the larger biomass sites. The reason for this is not clear at this time, but may be related to the reduced sensitivity of the radar to larger levels of biomass.
|
|
|
| ||
Table 2 . Average dry above-ground biomass averaged over all land (no-water) pixels.
|
|
|
| |
Above ground woody biomass is an important parameter for describing
the function and productivity of forested ecosystems. Recent
studies have demonstrated that synthetic aperture radar (SAR)
can be used to estimate above ground standing biomass. To date,
developing algorithms to estimate woody biomass using radar data
has required extensive ground truth
Figure 1. Howland, Maine AIRSAR and SIR-C biomass maps.
Figure 2. Comparison of biomass maps for three SIR-C/XSAR Backup
Supersites.
measurements to construct relationships between biomass and SAR backscatter. Models were used to help develop a relationship between biomass and radar backscatter. A gap-type forest succession model was used to simulate growth and development of a northern hardwood-boreal transitional forest typical of central Maine, USA. Model results of species, and bole diameter at breast height (dbh) of individual trees in a 900 m2 stand were used to run discontinuous canopy backscatter models to determine radar backscatter coefficients for a wide range of simulated forest stands. (see Ranson et al., 1997).
Using model results, relationships of co-polarized backscatter
to forest biomass were developed and applied to an airborne SAR
(AIRSAR) image of the forest. Figure 3 shows simulated PHH and
CHH backscatter for the 3450 modeled forest stands. The different
colors represent soil drainage classes. The algorithm selected
by stepwise regression using all AIRSAR co-polarized channels
selected PHH and CHH. Figure 4 (upper) shows the scatter plot
for biomass vs. a linear combination of PHH and CHH. Note that
the effects of soil drainage class (viz. Figure 3) are
reduced. A relationship derived totally from model results was
found to underestimate biomass. Calibrating the modeled backscatter
with limited AIRSAR backscatter measurements improved the biomass
estimation when compared to field measurements (Figure 4 (lower).
The approach of using a combination of forest succession and
remote sensing models to develop algorithms for inferring forest
attributes produced comparable results with techniques using only
measurements. Applying the model derived algorithm to SAR imagery
produced reasonable results when mapped biomass was limited to
15 kg/m2 or less. These results
can be improved through improvements in the forest growth model
and radar backscatter model treatment of cross-polarization backscattering.
2.2 Three-dimensional radar backscatter modeling of forest cutting patterns
The understanding of ecosystem processes and the impacts of natural and anthropogenic induced change upon these processes required techniques to monitor changes in the structure, composition and function of ecosystems. Radar backscattering is a useful monitoring structure since it is sensitive to vegetation water content, its distribution within the canopy as leaves, branches and boles and the underlying soil surface. In retrieval of forest parameters from radar images, one of the major concerns is the uncertainty or error induced by the variation of tree size and position. The Goddard 3-D radar backscatter model (see Sun and Ranson, 1995) was used to model the effects of forest structure on radar backscatter. Lacunarity spatial analysis was performed on simulated and actual SAR images of forest areas with different cutting patterns. Spatial pattern analysis discriminated between forest spatial structures better than analysis of average backscatter. The results showed that the model worked well for our predominately coniferous forest test site by giving good prediction of total backscattering averaged over the site. Significant spatial correlations between AIRSAR and simulated images were also produced for all frequencies (P, L and C-band) and polarization combinations (HH, VV, HV). Forest cutting patterns typical to those found in the Howland Forest were simulated with the 3-D backscatter model. Mean backscatter was similar among the cutting patterns, whereas spatial analysis revealed differences (See Sun and Ranson, 1997).
Results of mapping forest cover and biomass in our study area
using SIR-C/XSAR data are encouraging and should be useful for
the study of above ground carbon in boreal and other ecosystems.
Comparison of AIRSAR and SIR-C
image data over northern forest in Maine, USA showed similar results
for forest type classification and woody biomass estimates. Sensitivity
to biomass was sufficient to map most of the types of stands in
our area.
Figure 3. Simulated PHH and CHH backscatter vs simulated biomass
for 3450 modeled Howland, Maine forest stands. Colors indicate
ten different soil drainage classes that were used in the simulations.
Figure 4. Top. Model results showing a scatter plot of biomass
vs. backscatter using a linear combination of radar channels.
Bottom. Comparison of SAR estimated biomass using model derived
allgortithm and field measured biomass.
A comparison of the April and October data sets was also conducted to understand the effects of seasons on the analysis. Frozen trees and wetter background contributed to increased backscattering observed in the April data . SIR-C/XSAR data acquired during the growing season would be useful to determine the effects of deciduous leaves on biomass estimation. The usefulness of X-band for forest type mapping may be greater during the summer when deciduous trees have leaves.
Currently, radar models do not simulate cross polarized backscatter very well because of the lack of treatment of multiple scattering. This in an obvious area of future work. Our radar backscatter models will be further improved by incorporating higher-order scatterers. Radar backscatter modeling will be continue to be used to investigate the radar responses to successional forest structures by using the simulated forests from forest growth models as inputs. These modeling results will aid in improvement of forest parameter retrieval algorithms.
Given the emphasis of topography for SRTM the use of interferometric SAR for our forest ecosystem research is being explored. The Western Sayani Mountains test area we are working on with Dr. Slava Kharuk of the Sukachev Forestry Institute in Siberia poses topographic challenges to our ecosystem analyses. Repeat pass SRL-2 data was obtained and analyzed, but there appears to a problem with the baseline. The topographic information interferometrically derived from these data as well as DTM data (if available) will be used to improve and extend our current biomass mapping methods to mountainous areas through data correction for local incidence angle, layover and shadows. The use of combined Shuttle Laser Altimeter and interferometry is also being investigated.
The SIR-C/XSAR missions are demonstrating the unique capabilities
of multiple frequency and multipolarization SAR data for studies
of the Earth. Similar measurements over a significant portion
of the Earth's biomes will produce a unique and valuable source
of data for terrestrial ecologists and climatologists. Global
maps of important terrestrial vegetation parameters derived from
these data sets can be used directly by scientists to gain further
understanding of the current "state of the Earth" and
develop insight into the global consequences of climate change.
Thus we look forward to a third SIR-C mission that would map
significant areas of the worlds forests with L-band co- and cross
polarized channels in addition to the topographic data.
Refereed Journal Articles
Chauhan,N.S., R.H. Lang and K.J. Ranson, (1991), Radar modeling of a northern forest, IEEE Trans. Geosci. Remote Sens, 29:627-638.
Daughtry, C.S.T, K.J. Ranson and L.L. Biehl, (1991), C-band backscattering from corn canopies. Int. J. Remote Sensing, 12:1097-1109.
Kimes, D.S., K. J. Ranson, and G. Sun . Inversion of a forest backscatter model using neural networks, (1996), Accepted by Int. J. Remote Sensing
Lang, R.H., N.S. Chauhan and K.J. Ranson, (1994). Modeling of P Band SAR returns from a red pine stand. Remote Sens. Environ. 47(2):132-141.
Ranson, K. J., R. Lang, G. Sun, N, Chauhan, R. Cacciola and O. Kilic, (1997), Mapping of boreal forest biomass from spaceborne synthetic aperture radar, Accepted JGR
Ranson, K. J., G. Sun , J.F. Weishampel , and R.G. Knox (1997) Forest biomass from combined ecosystem and radar backscatter modeling. Remote Sensing of Environment 59:118-133.
Ranson, K. J., Saatchi, S. and G. Sun. (1995), Boreal forest ecosystem characterization with SIR-C/XSAR. IEEE Trans. Geosci. Remote Sens. 33:867-876.
Ranson, K.J. and G. Sun, "Mapping biomass for a northern forest using multifrequency SAR data," IEEE Transactions on Geoscience. Remote Sensing, 32(3):388-396, (1994).
Ranson, K.J. and G. Sun, (1994). "Northern forest classification using temporal multifrequency and multipolarimetric SAR images," Remote Sensing of Environment, 47(2):142-153
Ranson, K.J. and S.S. Saatchi, (1992), C-band microwave scattering from small balsam fir, IEEE Trans. Geosci. Remote Sens. 30:924-932.
Salas, W.A., K.J. Ranson, B.N. Rock and K.T. Smith. (1994). Temporal and spatial variations in dielectric constant and water status of dominant conifer species from New England, Remote Sensing Environ. 47(2):109-119.
Sun, G. and K. J. Ranson. (1995), A three-dimensional radar backscatter model of forest canopies, IEEE Trans. Geosci. Remote Sens. 33:372-382.
Sun, G. and K. J. Ranson, (1997), Radar modeling of forest spatial structure, Submitted Int. J. Remote Sensing
Waring, R., J. Way, R. Hunt, L. Morrisey, K.J. Ranson, J. Weishampel and R. Oren, (1995) Imaging radar for ecosystem studies, BioScience, 45(10) 715-723.
Weishampel, J.F., G. Sun, K.J. Ranson, H.H. Shugart and K.D. Lejeune. (1994). Inference of spatial resolution on forest texture derived from simulated SAR images. Remote Sensing Environ. 47(2):120-131.
Weishampel,J.F., K.J. Ranson and D. Harding. (1995). Remote sensing of forest canopies. Selbyana. 17:6-14.
Book Chapters
Leckie, D, (1997), Manual of Remote Sensing: Radar Volume, Forestry Applications Chapter (J. Ranson contributing author) In press.
Sun, G. and K.J. Ranson, Forest ecological studies and radar backscatter modeling, in Guo Huadong and Zheng Lizhong (eds.), (1995), Microwave remote sensing for earth observation, Science Press, Beijing, pp. 213-223.
Weishampel,J.F., R.G. Knox, K.J. Ranson, D.L. Williams and J.A. Smith, (1996). Integrating remotely sensed spatial heterogeneity with a 3-D forest succession model, in Gholz, H.L., Nakane, K. and Shimoda, H. (eds), The use of remote sensing in the modeling of forest productivity at scales from the stand to the globe. Kluwer Acad. Publishers, Dordrecht
Published Proceedings
Chauhan, N.S., R. Lang, J,. Ranson and O. Kilic, (1994), Multistand radar modeling from a boreal forest: Results from the BOREAS Intensive Field Campaign- 1993. Proc. IGARSS'94, Pasadena, CA, August 8-12.
Lang, R.H., K.J. Ranson, N.S. Chauhan and O. Kilic, (1994), Radar analysis and modeling of forest stands for biomass estimation, Progress in Electromagnetic Radiation Symposium, The Netherlands, July.
Lee, J.S., D.L. Schuler, R.H. Lang and K.J. Ranson, (1994), K-distribution for multilook processed polarimetric SAR imagery, Proc. IGARSS'94, Pasadena, CA, August 8-12.
Ranson, K. J. , J.A. Smith, G. Sun, J.F. Weishampel and R.G. Knox, (1995), Forest structure from combined optical and microwave modeling and measurements, by submitted to Combined Optical-Microwave Earth and Atmosphere Sensing (CO-Meas'95), Atlanta, GA April 1995.
Ranson, K. J., Guoqing Sun, John F. Weishampel and Robert G. Knox, (1995), Interfacing forest succession and remote sensing models for forest ecosystem studies, SPIE Conf. Multispectral and microwave sensing of forestry, hydrology, and natural resources, Rome, Italy, Sept 26-30, 1994. Vol. 2314, pp. 526-537.
Ranson, K. J., R. H. Lang, G. Sun N.S. Chauhan and R.J. Cacciola, (1995), Mapping of boreal forest biomass using synthetic aperture radar measurements and modeling, Retrieval of Bio- and Geophysical Parameters from SAR for Land Applications, Toulouse France, 10-13 October.
Ranson, K.J. and G. Sun (1993), Retrieving spatial patterns from SAR image data. Proc. IGARRS'93, Tokyo Japan. pp. 1213-1215.
Ranson, K.J. and G. Sun, (1995), Dependence of radar backscattering on northern forest structure observed from AirSAR and SIR-C/XSAR. IGARSS'95
Ranson, K.J., B.N. Rock, W.A. Salas, K. and D.L. Williams,,(1992), Analysis of the dielectric properties of trunk wood in dominant conifer species from New England and Siberia. Proc. IGARRS'92, Houston, TX, pp. 1283-1285.
Ranson,K.J. and G. Sun, (1992), Mapping biomass data for a northern forest ecosystem using multi-frequency SAR data. Proc. IGARRS'92, Houston, TX, pp. 1220-1222.
Salas, W.A., K.J. Ranson, B.N. Rock and D.M. Moss, (1991), Diurnal changes in the dielectric properties and water status of hemlock and red spruce from Howland, ME. IGARSS'91, Espoo, Finland.
Sun, G. and K. J. Ranson (1993), Polarimetric radar data decomposition and interpretation, 4th Annual JPL Airborne Geosciences Workshop, Vol. 3 AIRSAR Workshop, JPL Pub. 93-26, Pasadena, CA, pp.65-67.
Sun, G. and K. J. Ranson, (1994), Spatially explicit modeling of radar backscatter from forest canopies, Proceedings of EUROPTO, Vol. 2314, pp. 559-570. The European Symposium on Satellite Remote Sensing, 26-30 September, NRC of Italy Headquarters, Rome, Italy.
Sun, G. and K.J. Ranson, (1993), Modeling of radar responses to forest physical parameters. Proc. IGARRS'93, Tokyo Japan. pp. 384-386.
Sun. G. and K. J. Ranson, (1992), Relating multifrequency radar backscattering to forest biomass: Modeling and AIRSAR measurement. 3rd JPL Airborne Science Workshop, Vol 3. AIRSAR, Pasadena, CA, pp. 27-29.
Williams, D.L., V.I. Kharuk, V.M. Jhirin, B.N. Rock, K.J. Ranson, (1992), Sayani'91: A joint United States/Commonwealth of Independent States field campaign to investigate forest decline damage in the Krasnoyarsk region of southcentral Siberia. Proc. IGARRS'92, Houston, TX, pp. 1286-1288.