Radar Backscattering From Tropical Forest canopies: Sensitivity to Canopy Structure and Biomass Shakil Ahmad ROMSHOO, Masanobu SHIMADA, Tamotsu IGARASHI Earth Observation Research Center (EORC), NASDA Abstract: Employing the SAR for biomass studies, we nd that the relationship between the sensor response and the biomass is non-linear. The SAR is not only sensitive to the woody biomass but also to the geometry of the trees and stands. This study shows the importance of correcting the structural inuences for the quantitative estimation of the biomass. Theoretical modeling of the radar scattering from the vegetation is employed to simulate the eects of the canopy architecture on the radar backscattering for the radar sensor conguration of the and. The eld data measured from 12 sample plots over a tropical forest area in Chiang Mai, towards north of Thailand is used for driving the simulations. The results show that canopy structure and architecture has a paramount inuence on the radar backscattering from tropical forests for both L-band () and C-band () sensors. For,the infuence from the canopy architecture described by the primary branch diameter, crown depth and the dimeter at breast height (DBH) of the main stems is very appreciable while as for the, the canopy architecture described by the leaf dimensions and the crown depth has a tremendous inuence on the radar backscattering from the vegetation. Both the sensors are sensitive to the woody biomass with, because of better interactions with the woody constituents of the canopy, showing a better sensitivity. In case of,the sensitivity is quite high for the regenerating forests (The observed data set is from the regenerating forest with less than 5 tons/ha) and gets reduced with the increase of biomass. In case of ERS- 2, though we nd that it is sensitive to the biomass upto 5 tons/ha but the sensitivity is appreciably reduced after that. The saturation in case of is reached well before that of the. Key Words : Backscattering coecient,incidence angle, polarization, dielectric constant, tropical forests, canopy architecture, biomass, regeneration, biophysical parameters 1. Introduction The knowledge about the amount and distribution of biomass is important, not only for understanding the carbon budgets, but also for other land surface processes related to water and engergy budget. Carbon sequestration is thought to be promising means of reducing atmospheric carbon dioxide, the focus of Kyoto Protocol, which aims to cut the world`s greenhouse gas emissions such as carbon dioxide and methane. A major constraint to successful forestry based carbon oset programs is the lack of reliable, accurate and cost eective methods for monitoring carbon storage. There is a large uncertainty in the estimates of carbon uxes due to changes in tropical forests. The two estimates of biomass of tropical forests by Brown and Lugu (1991) from two dierent data bases are markedly dierent. In continental Southeast Asia, the forest covers approximately 7 million hectare or 3% of the toal land area. Low altitude forests are estimated to contain 6% of the total above ground carbon in world forest vegetation (Dixon et al.,1994). Nevertheless, for a variety of reasons, there are still considerable uncertainty about carbon stocks in tropical forests at landscape scales (Brown, 1997). A better understanding of the amount and distribution of tropical biomass is therefore of great theoretical and paractical interest. Remote sensing technology, especially the Synthetic Aperture Radar (SAR) systems, have the potential to be useful for estimation and monitoring of forest biomass estimates (Dobson et al., 1992, Le Toan et al., 1992, Kasischke et al., 1994), but the utility of this technology is dependent upon many factors. There are still gaps in the knowledge and before such techniques could be used on an opertional scale, we need to ll up these gaps through rigrous inventories of the unexplored ecosystems and modelling approaches. The radar backscatter is not only sensitive to the total biomass of the trees but is also sensitive to the woody strucuture. This degrades the correlative relationship between the radar backscatter and the biomass. These structural dierences could be used to discriminate structurally dissimilar forests despite having similar forest biomass and would lead to higher accuracy in the biomass estimation. Large scale mapping of forest structure and biomass must rely on remote sensing techniques, more particularly from the SAR systems. Though, Synthetic Aperture radar has been utilized for measuring the above ground woody
biomass in the temperate and boreal forest ecosystems but very little research, if at all, has been conducted on South East Asian tropical forests. Because of their importance to the terresterial ecosystem, any information about the biomass of these forest biomes would facilitate better understanding of the earth's system. It is important that the tropical forest ecosystem and the potential changes occurring therein,either due to anthropogenic distubances or global climate change, need to be understood. Realizing their importance, NASDA, in continuation of the Global Rain Forest Mapping Project (GRFM) and Boreal Forest Mapping Project (BFM), has initiated the SouthEast Asian Forest Mapping Project. SAR mosiac of the region (Shimada et al., 21) has been recently made public for full exploition by the reseach community. 2. Study Area A few study sites have been chosen in Thailand for the estimation of the forest biomass using spaceborne SAR data. Two sites are in the northern and a few sites are in the western regions of the Thailand. These sites have been studied in the past for improving the silivicultural treatments. But in this article, we would be discussing only the one study site at Huai Rai, towards northeast of Chang Mai in northern Thailand. The forest near Huai Rai covers approximately 4 hectares. A forest area spread over 2 hectares was chosen for forest biophysical parameters measurement. The area is located at an elevation of 485 m a.s.l and receives a annual rainfall of 11 mm. The exact location of this site, the edaphic, climatic, physiographic and vegetation description of this site has been discussed by Timm (2). (1) Field Conditions and Ground Data measurements Dry dipterocarpus forest is a deciduous broad leaved forest community type occurring on relatively drier site conditions and is mainly composed of trees belonging to the family Dipterocarpaceae. It occurs widely in mainlaind southeast Asia. In Thailand, it is found in the north, northeast and central regions where the annual rainfall is 1-15 mm with 5-6 months dry period, which is approximately the period of leaessness. This forest type covers approximately 45% of the total forested area of Thailand (Sanuhulu et al., 1999). The stands show sporadic, discontinuous crown cover with large canopy openings as well as the stratication of whole stand into two layers. At Huai Rai, a total of 12 sample plots were set up to measure a number ofvegetation parameters. The inventory design and the reasons for its choice are discussed in details by Tim, 2. During the period of inventory between 1997-1999, two plots burned in 1998 and all the 12 plots burned in 1999. In a 4 m * 4 m plot, all the adult trees with DBH 5cmwere recorded. while as in the sappling subplots (1 m * 2 m), trees with heights 1.3 m and DBH 5 cm were recorded. In seedling inventory subplot (2 m * 2 m), tree with heights less than 1.3 meters were recorded. All the plots were protected from distubances during the eld work. Investigated tree parameters include specieswise DBH, Density, Height, crown heights, crown radii,horrizontal distribution, crown light availability. In order to get the whole range of parameters used for driving the model simulations, some empirical models (Brown, S., 1997) were used based on the measured data. Some of the parameters were also estimated using the alometric equations from some other literature. 3. Scattering Models Each forest type displays a distinctive backscattering behavior due to the absorption and the scattering properties of its structural elements. There have been quite a few volume scattering models, advocating a rigorous theoretical solution to the electromagnetic scattering problem, developed to model the radar backscatter from the vegetation canopy. They have been based either on radiative transfer theory (Ulaby et al., 199, Karam et al., 1995), distorted born approximation (Lang et al. 1983), Cloud model ( Etema et al., 1978) etc. We used the MIMICS, MIchigan MIcrowave Canopy Scattering model (Ulaby et al. 199) for calculating the backscatter from the vegetation and is based on the theory of radiative transfer. The solution is based on the rst order solution of the radiative transfer equation of a tree canopy comprising a crown layer, a trunk layer and a rough surface ground boundary. This theory is suitable for a medium like vegetation where scatterer have discrete congurations and dielectric constants much larger than that of the air (Ulaby et al., 199). The principal dierence between the other models and the MIMICS backscattering model is that several dierent types of scatterer are considered in the vegetation cover. In addition, considerable eort has been made to establish relationship between the mode parameters and the measurable features of the soil vegetation environment. The mathematical formulation of the wave-vegetation interaction is represented by the following equation: pq = pq:d (o; i)+ pq:dr (o; i)+ pq:rr (o; i)+ pq:gr (o; i) (1) where, i= unit vector of the incident wave, =
Z= Ground Term Direct Term Crown Layer Z=-d Direct Reflected Term Reflected Term Trunk Layer Z=-(d+ht) (a) 1-2 -3-4 -5-6 1 15 2 25 3 35 4 45 5 55 6 Incidence Angle Total Cr_Gnd Dir_cr Trnk_Gnd Dir_Gnd (c) Gnd_Trunk Term Transmissivity -2-3 -4-5 -6.9.8.7.6.5.4.3.2.1 ERS JERS 1 15 2 25 3 35 4 45 5 55 6 Incidence Angle (b) 1 15 2 25 3 35 4 45 5 55 6 Incidence Angle Total Dir_Gnd Dir_Cr Trnk_Gnd Cr_Gnd (d) Fig. 1 This composite gure shows : (a) The dierent volume scattering terms incorporated in the MIMICS model; (b) the transmissivity of and SAR system as a fucntion of incidence angle; (c-d) the contribution of dierent scattering mechanisms for and SAR as a function of incidence angle unit vector in the direction of the receiving antenna (o=-i for backscatter), p= polarization of the scattered wave, q=polarization of the incident wave, d = direct; dr; direct-reected; rr=reected; and gr=ground reected, attenuated by overlying canopy. As shown in the equation 1 and illustrated by Figure 1(a), the radar backscattering from canopy is made up of four terms: a direct backscatter term pq:d (o; i) corresponding to volume scattering, which is caused by direct reection of the incident wave from individual scatterers. The direct-reected term pq:dr (o; i) results from two similar mechanisms. In one case, the wave gets scattered from the scatterer to the ground and back to the radar, whereas in the second case, the wave is rst reected from the ground surface and then scattered towards the observer. The third term pq:rr (o; i) represents the reected term and involves twice the reection from the ground and once from the canopy. However, if the canopy is mature and uniform, the reected term is usually negligible (Lang, 1981; chauhan et al., 1994). The fourth term pq:gr (o; i) is the direct backscattering of the ground attenuated by the canopy. We used the dicontinuous version of the MIM- ICS model wherein eects of the gaps within the forest stands are taken into consideration. Mc Donald et al. (1993) have shown that the gaps occuring in the canopy layer may signicantly aect the signal transmissivity and backscatter. The forest canopy in our study area is quite transparent and the proportion of gaps in such a type of young regenerating dry Dipterocarpus forests could be as high as 4% (Sahunalu et al., 1993). Due to this structure of the forests, only single layer has been modelled with the assumption that the understory would have small eects for SAR (Silva et al., 1997). 4. Results and Discussion The study has been carried out to quantify the eects of the vegetation structure and biomass on radar backscattering from and SAR systems. The carries a L-band SAR with HH polarization and operates at an incidence angle of 35 where as the is C-band SAR system with VV polarization and operating at an incidence angleof23.for analyzing the eects of a particular vegetation parameter on the radar backscattering, all input parameters except the one under study are kept constant. This is important to know the eects of structural elements while keeping the total biomass constant. Though, we conducted in- uences of a large range of vegetation characteristics related to dierent branch orders and leaves
-2-4 -6-11 -13-15 -16 4.5 8.5 12.5 16.5 2.5 24.5 28.5 32.5 (a) 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8-6 DBH (cm) Crown Depth (m) (c) 3-9 -11-13 -15-5 -7-9 -11-13 -15-17 -19-21 -23-25 -6.5.7.9 1.1 1.3 1.5 1.7 1.9 Primary Branch Radius (cm) (b) 2.1 2.3 1 1.5 2 2.5 3 3.5 4 Leaf Diameter (cm) (d) 2.5-16 -18-16 -2.3.35.4.45.5.55.6.65.7-18 7 14 21 28 35 Leaf Moisture (e) Wood Volume (m 3 /ha) (f) Fig. 2 This is the composite plot of the gures that shows the sensitivity of and SAR to dierent attributes: (a) Diameter at Breast Height (DBH); (b) primary branch radius;(c) crown depth; (d) Leaf diameter; (e) Leaf moisture; and (f) wood volume. but due to the space limitations, we would be discussing only a few and important results. Figure 1(b) shows the transmissivity of the L- and C-band radars under the given eld conditions. The C-band transmissivity is also appreciable due to the presence of gaps in the canopy. The canopy density of the dry dipeterocarpus forests is low compared to the other moist tropical forests due to poor site conditions and relatively long dry spells. This makes the forest oor quite visible to the radar signals and under such conditions, the surface conditions would have a signicant impact on the total radar backscattering (Dobson et al., 1992). The Figure 1 (c & d) shows the contribution of the dierent scattering mechanisms as a function of incidence angle for nand ERS- 2 congurations respectively. For SAR, the main contribution to the total scattering comes from the trunk-ground term followed by the direct ground term. The contribution from the direct ground term is signicantly reduced at higher angles of incidence while at lower angles of incidence, the main contribution is from the ground. The longer wavelengths are usually scattered in forests by large tree components such as tree boles and main branches. This indicates that the SAR sensor conguration is suitable for biomass studies in forests as the trunks constitutes a significant proportion of the biomass in mature forests. For ERS SAR conguration, the main contribution
of the backscattered signals comes from the ground followed by direct crown term. Towards lower angles of incidence, the dominant contribution comes from the ground because of the appreciable gaps in the canopy layer. The signals at C-band wawelenth are scattered mainly from the small tree components. As discussed earlier, radar backscattering besides being sensitive to the biomass is also sensitive to the structural attributes of the vegetation. This could have adverse eects on the biomass estimation using SAR data. Figure 2 shows the sensitivities of the L- and C-band radar to the structural and biomass attributes of the forest vegetation. The L-band radar ( conguration) is sensitive to the branch dimensions while being almost insensitive to leaf structure and leaf moisture. While as the C-band radar ( conguration) is insensitive to the branch characteritics and very sensitive to the leaf structure and moisture. Both and ERS shows appreciable sensitivity to the change in the crown depth. The Figure 2 (a) shows the sensitivity of the JERS and ERS radars to DBH (Diameter at Breast Height). For conguraion, the main contribution comes from the crown and the direct ground interactions Fig. 1(d), and thus is insensitive to the structural change in the tree boles while on the other hand, is sensitive to the bole structural changes because the backscattering is main contributed from the trunk-ground interactions Fig. 1(c). Similarly, for same explantions, the shows signicant sensitivity to the changes in the primary branch structure while as the ERS shows feeble sensitivity. This indicates that in forests, where most of the biomas is concentrated in the tree boles, shows a promise for biomass estimation. On the other hand, ERS is very sensitive to the changes in the leaf structure and composition while the shows no sensitivity to these attributes. The Figure 2 (d & e) shows the sensitivity ofthetwo SAR systems to the leaf dimensions and leaf moisture respectively. Both the sensors show appreciable sensitivity to crown depth as indicated by Fig. 2(c). This is because the changes in the canopy depth bring about concurrent changes both in the bole length as well in the number of the leaves and twigs. The being sensitive to the former while ERS is sensitive to the later and this manifests in high sensitivity to the changes in the crown depth for both sensors. Fig. 2(f) shows the and SAR backscattering relationship with the wood volume. It is again reiterated here that the forests under study are regenerating forests with biomass less than 5 ton/ha. The higher levels of biomass have beensimulated by incrementing the various growth parameters taken from the literature. It could be seen from the gure that shows higher sensitivity to the biomas as compared to the. The sensitivity is quite good even upto higher levels of biomass while as the backscattering shows saturation at very low biomass levels. Further, the dynamic range of the backscattering is quite higher than that of the. The sensitivity of the C-band SAR to biomass could be due to the smaller DBH of the regenerating forest crop (less than 15 cm). Kasischke et al. (1991) also found C-band VV polarization data from AIRSAR ights over Duke Forest to be sensitive to changes in biomass, but only when the diameters were less than 1 cm. The sensitivity of the ERS to the lower biomass levels could be also due to the ground contribution at lower levels of the biomass as there are appreciable gaps in the canopy. A few other researchers have also found that C-band is less reponsive to tree biomass with a smaller dynamic range than the lower frequencies (Le Toan et al. 1992 and Kasischke et al., 1994) 5. Conclusions The scattering model simulations tell us that there is a structural control of the backscattering behavior for both and ERS radars. Because of the longer wavelength, the L-band SAR is senstive to the branch elements while as C-band SAR is sensitive to the leaf structural and biomass attributes. We would have to derive relationships between the radar backscatter and biomas which are independent of the structural inuences. This would be especially important when regional or global biomass mapping is attempted. We would have to stratify the forests types on the basis of structure and then attempt biomass estimation on specic forest types which would eliminate or minimize the eects of the structure on biomass estimation. The sensitivity of both these radars to biomass is also appreciable at lower levels of biomass, with being more suitable for biomass estimation and monitoring due to higher sensitivity, especially in the tropical dry dipterocarpus regenerating forests. The sensitivity of the C-band radar to the biomass at lower levels of biomass is related to the prevalence of gaps in these young tropical forests and also due to the small dimensions of the tree boles to which C-band has been reported to be sensitive. In presence of gaps, the radar signals can penetrate the canopy and interact with the standing stems. As expected, the sensitivity to the higher biomass levels is signicantly dampened while as the JERS shows some sensitiv-
ity, even though week, towards higher values of the biomass (around 2 tons/ha). The data from a number of other sites in the Southeast Asian tropical forest belt, with dierent species composition, physiographic and climatic conditions, is being analysed and correlated with the space-borne SAR measurements to understand the inuence of the structure on the radar estimation of biomass. The complete understanding from both theoretical, satellite observations and eld experimentation would facilitate the development of an index which could account for the structural in- uences of the canopy on the radar backscattering. The development of that index or indices would facilitate the radar estimation of the biomass at regional and global scales with acceptable degree of certainity. REFERENCES 1) Attema, E. P. W, Ulaby, F. T (1978). Vegetation Modelled as a Water Cloud, Radio Science, Vol. 13, No. 2, Pages 357-364, March-April 1978. 2) Brown, S. (1997). Estimating biomass and biomass change of tropical forests: a Primer, Forestry paper No. 134, FAO, Rome. 3) Brown, S. (1991). Biomass of tropical forests of south and southeast Asia, Can. J. For. Res. vol.21, pp. 111-117. 4) Dobson, M. C., Ulaby, F. T., LeToan, T., Beaudoin, A, Kasischke, E. S., and Christensen, N (1992). Dependence of radar backscatter on conferous forest biomass, IEEE Trans. Geosci. Remote sens., GE-3, pp. 412-415. 5) Dixon, R. K., Brown, S., Houghton, R. A., Solomon, A. M., Trexler, M. C., Wisniewski, J., (1994). Carbon pools and ux of global forest ecosystems, Science, Vol. 263, pp. 185-19. 6) Karam, A. M., Amar, F., Fung. A. K. et al.(1995). A microwave polarimetric scattering model for forest canopies based on vector radiative transfer theory, Rem. Sens. Environ., vol. 53, pp 16-3. 7) Kasischke, E. S., Bourgeau-Chevez, L. L and Christensen, N. L., and Dobson, M. C.(1991). The relationship between aboveground biomass and radar backscatter as observed on airborne SAR imagery, in proceedings of the Third AIRSAR Workshop, JPL publication 91-31, Jet Propulsion Laboratory, Pasadena, CA, pp. 11-21. 8) Kasischke, E. S., and Christensen, N. L., Bourgeau- Chevez, L. L, and Haney, E (1994). Observations of the sensitivity of ERS-1 SAR image intensity to changes in above ground loblolly pine forests, Int. J. Remote Sens., vol. 15, pp. 3-15. 9) Lang, R. H., and Sidhu, J., (1983). Electromagnetic backscattering from a layer of vegetation : A discrete approach, IEEE Trans. Geosci. Rem. Sens., vol. 21, No. 1, pp. 62-71. 1) Le Toan, T., Beaudoin, A., Riom, J., and Guyon, D. (1992). Relating forest biomass to SAR data, IEEE Trans. Geosci. Rem. Sens., vol. 3, pp. 43-411 11) Mc Donald, K. C. and Ulaby, F. T (1993). Radiative transfer modelling of discontinuous tree canopies at microwave frequencies, Int. J. of Remote Sensing, vol. 14, No. 11, pp.297-212. 12) Sahunalu, P., and Dhanmanonda, P. (1993). Structure and dynamics of dry dipterocarpus forest, Sakaerat, northeastern Thailand, In Tuxen, R and Leith, H. (ed.) Handbook of Vegetation science, pp. 465-494, Kluwer Academic Publishers, Boston. 13) Shimada, M. and Isoguchi, O (21). SAR mosiacs of Southeast Asia using calibrated path images, Int. J. Remote Sensing (In press). 14) Silva, T. A. M. and Dias, J. M. B (1997). The eects of forest understory on synthetic aperture radar, Procedings Int. Geo. Symp. on Rem. Sens. (IGARSS'97), Singapore, vol. II, pp. 773-777. 15) Tennigkeit, T (2). Transformation of degraded forests into semi-natural production forests in northern Thailand, Ph.d thesis, university of Freiburg, Germany. 16) Ulaby, F. T.,Sarabandi, McDonald, K., Whitt, M., and Dobson, M. C. (199). Michigan Microwave Canopy Scattering Model, Int. J. Remote Sensing, Vol. 11(7), pp 122353.