Forest Biomass Estimation using Polarimetric SAR Interferometry

Forest biomass is one of the most important parameters for global carbon stock modelling yet can only be estimated with great uncertainties. Unfortunately, conventional remote sensing techniques for the estimation of forest biomass are not able to provide estimates on a global scale. An alternative approach is based on forest height estimates from single frequency polarimetric-interferometric SAR data. Here, forest biomass must be converted from forest height through allometric height-biomass relations. Based on the achieved forest height accuracy, this paper shall critically discuss the accuracy of the forest height-biomass relations as derived from standard forestry tables for temperate European forests. The potential of this approach shall be demonstrated by applying the forest height-biomass allometry to convert a forest height-map – acquired from experimental airborne SAR data over the Fichtelgebirge test site (Germany) – into a forest biomass-map. Introduction Global forest inventory and an accurate forest (above ground) biomass estimation are still critical missing parts in the global climate change discussion [9]. A reliable forest biomass determination is essential for understanding and modelling ecosystem dynamics and global Carbon-fluxes. Parallel to the ecological dimension, under the light of the Kyoto protocol, this problem has also a political dimension that increases the responsibility of the scientific community to provide exact answers. The fact that biomass information is especially missed in remote areas (i.e. tropical and boreal forest ecosystems) combined with the fact that these ecosystems contain the biggest amount of forest biomass make remote sensing techniques a challenge (cf. Tab. 1). However, optical remote sensing sensors are in general not capable of measuring forest biomass or monitoring the dynamics of deforestation and biomass regeneration. There are several reasons for this: 1) insufficient sensitivity to forest structure and above-ground biomass and 2) inadequate temporal frequency as a result of the atmospheric conditions in tropical and boreal ecosystems. On the other hand, conventional Synthetic Aperture Radar (SAR) remote sensing methods based on the evaluation of backscattering amplitudes for the quantification of forest biomass are limited to biomass levels bellow 150 tons/ha [10] and thus, do not allow biomass inventory of forests with higher biomass levels (cf. Tab. 1/ Tab. 2). In this paper we address an alternative methodology for estimating forest biomass from remote sensing data based on forest height estimates obtained from single frequency fully polarimetric-interferometric radar data. Polarimetric Synthetic Aperture Radar Interferometry (Pol-InSAR) is a recently developed radar remote sensing technique, based on the coherent combination of radar polarimetry and SAR interferometry which is substantially more sensitive to forest structural parameters than conventional interferometry or polarimetry alone. In the last years, quantitative model based estimation of forest parameters has been demonstrated using spaceand airborne repeat pass fully polarimetric interferometry at C-, L-, and P-band indicating the potential of this new technology to estimate key forest parameters as forest height and extinction [5,6,7,11,12]. In order to use this technique for forest biomass determination, forest height must be converted to forest biomass using allometric relations. In biology, allometry refers to the study of the proportions of different organs. “Typically, for the morphological development of organs, the relative growth increases differ, but are in a constant relation.” [14] For the allometric determination of forest biomass from forest height, it is necessary to look at the variability of the height-biomass allometry in different forest types. The following pages shall therefore discuss how strong forest heightbiomass relations vary with stand conditions, stand age, thinning intensity and species. The results are based on German standard forestry tables [1,15], but the extrapolation to other ecosystems and forestry systems are discussed. Finally, the potential that lies in the combination of the forest biomass determination with Pol-InSAR and forest heightbiomass allometry, shall be demonstrated by generating a forest biomass map of the Fichtelgebirge test site (Germany). Tab. 1: Biomass concentrations and Carbon-stocks in ecosystems [8]. The three major forest ecosystems account for almost 80 % of the terrestrial aboveground biomass. Their average biomass concentration exceeds 200 t/ha. Biome Area Abovegr. Biomass Carbon stocks [Gt] [10 km] [Gt] [t/ha] Vegetation Soils Total Tropical Forest 17.6 424 400 212 216 428 Temperate Forest 10.4 118 300 59 100 159 Boreal Forest 13.7 176 200 88 471 55 Tropical Savannas 22.5 132 40 66 264 330 Temperate Grasslands 12.5 18 16 9 295 304 Deserts, Semideserts 45.5 16 16 8 191 199 Tundra 9.5 12 12 6 121 127 Wetlands 3.5 30 30 15 225 240 Croplands 16.0 6 6 3 128 131 Total 151.2 932 70 466 2011 2477 Tab. 2: Biomass saturation limits for C-, Land P-band [9]. The biomass saturation limits for Cand L-band are too low to reach the biomass levels of closed forests, even P-band can only cover biomass-poor closed forests. Radar wavelength Biomass saturation limit % global VegetationForest types that can be mapped [t/ha] -area -biomass C-band 20 25 4 L-band 40-60 35 8 forest tundra, savannahs, woodlands, afforestations, no closed forests P-band 100-150 60-65 20-25 northern boreal forests, thinner southern boreal and temperate forests Test site description The test site “Fichtelgebirge” is a mountain range of 1000 km in NE Bavaria. The highest elevation is the Schneeberg with 1051 m a.s.l.. During 19/20 century – following severe deforestation – the Fichtelgebirge was reforested with Norway spruce (Picea abies), that still today accounts for > 95 % of the forest area. The natural vegetation would probably be dominated by beech forests (Fagus sylvatica), that only make up 2 % of the present forest area. During the late 70’s the Fichtelgebirge became known through the forest dieback in its higher (exposed) regions. The current reforestation tries to promote the autochthonous beech. The terrain varies in topography, edaphic situation, and forest management practices from 650m a.s.l. to 1054m a.s.l.. The variability of terrain, and soil conditions, in combination with two monospecific forest types with a resulting large gradient in tree age, stand height, stand density, and biomass (from 100 tons/ha up to 400 tons/ha) renders it as an excellent test site for the demonstration/validation of the proposed methodology. In May 2001, DLR’s experimental airborne SAR system (E-SAR) was flown over the Fichtelgebirge area, collecting repeat-pass interferometric fully polarimetric data at Lband. Forest height –biomass allometry Forest height and forest biomass are two closely related parameters. In forestry, it is standard to calculate the stem biomass [t/ ha] of a stand from average tree height [m] and basal area (cross-section of all trees in the stand at 1.3m = breast height [m / ha]). The form-factor Fzstand is a factor that depends on stand age, species, thinning a.o.: Biomassstand = Fzstand * (Basal areastand * average tree heightstand) [1] Because radar is not able to supply information for Fz and basal area with sufficient accuracy, a direct relation between stand biomass and stand height is needed. Assuming, forest height-biomass allometry follows the most general law of proportional growth [2], biomass is related to height as follows: Biomassstand = b * (tree heightsstand) a [2] In the following, height-biomass relations will be derived based on German standard forestry tables. Biomass shall be defined as the total standing stem biomass (bark inclusive). It was calculated from the forestry tables by multiplying stem volume with the raw density of the corresponding species (where only usable stem volume was listed, total volume was assumed to make up an additional 0.005m per stem). Stand height refers to the h100-tree height, the height of the 100 highest trees per hectare. The h100 can be regarded as an approximation of the upper canopy height. Since forestry tables are forest type-specific, it needs to be discussed how strong the height-biomass allometry depends on the forest type. These forest types are defined through four parameters: Forest type parameters: (a) forested species (b) forest management (c) stand conditions (d) age (1) An age and stand condition independent height-biomass relation for Norway spruce: The left graph in Fig. 1 shows the stem biomass as a function of stand age. The differences between the yield classes express different stand conditions, namely climate, topography (mainly exposition and slope) and soil. It can be seen that different stand conditions lead to different growth rates. Hence, when estimating stem biomass from stand age, information about the stand conditions is very important. On the other hand (Fig. 1 right graph), when estimating stem biomass from tree height, the influence of stand conditions is almost negligible: a specific height corresponds to a specific biomass, no matter wether it is at a young age by fast growing forest (under favourable stand conditions) or at an old age by a slow growing forest (under adverse stand conditions). This phenomenon is known as the extended law of Eichhorn [4]. Hence, an allometric height-biomass relation for Norway spruce (Picea abies, Assmann, moderate thinning, cf. Fig. 1) can be established and is independent of stand conditions and age. Biomassstand = 0.68 * (tree heightstand) [3] (2) The variability of the height-biomass relation for Norway spruce: So far, forest tables were only established for managed forest systems, where the growth rate is maximized through a moderate to strong thinning of weaker trees. Therefore allometric relations will always possess a certain variability due to the missing information about the forest management. Also, an absolute independence of the height-biomass relation from stand conditions and age is not given [1,4,14]. In order to quantify the total variability of the height-biomass regressions for Norway spruce (eq. 3), Fig. 2 shows regressions for 7 different forestry tables, of which the Assmann regression is generally regarded as the most reliable one. The overall variability of the height-biomass regressions varies approx. +/15 % from the Assmann regression for moderate thinning. This estimate corresponds to the findings of Assmann, too [1]. (3) The effect of species: Fig. 3 shows height-biomass regressions for 4 important (temperate European) forestry species: Norway spruce (Picea abies), Scots Pine (Pinus sylvestris), European beech (Fagus silvatica) and oak (Quercus robur). The height-biomass relations between species also only vary within a limited range of +/15 %. Results from Fichtelgebirge Fig. 4 and Fig. 5 show results for the Fichtelgebirge test site. The data were acquired with DLR’s experimental airborne SAR system (E-SAR) in May 2001. The height image was obtained through the inversion of single baseline interferometric polarimetric L-band data [6]. At three different sites elaborate ground measurements were carried out for height validation. It was found that the extracted heights matched the upper canopy height (h100) very well [13]. Therefore the biomass image was calculated with the height-biomass relation of eq. 3 for Norway spruce. Since the Fichtelgebirge consists to 98 % of spruce forests and has mainly been managed in a conservative (table-conform), the bias of the height-biomass allometry can be assumed to be less than 15 %. Fig. 1: Stem biomass as a function of forest age (right) and h100tree height (left) for Norway spruce (Picea abies, Assmann, moderate thinning [1]) under different stand conditions (yield classes oh40 (black): fast growing to yield class oh20 (light gray): slow growing). Fig. 2: Stem biomass as a function h100tree height for different forestry tables of Norway spruce (Picea abies) [1,15] Fig. 3: .Stem biomass as a function of h100tree height for Norway spruce (Picea abies), Scots Pine (Pinus sylvestris), European beech (Fagus silvatica) and oak (Quercus robur).[1,15] Fig. 4: forest height image of a part of the Fichtelgebirge where forest height was extracted from the Pol-InSAR data using the random volume over ground scattering model. It is overlayed with a vector layer that contains the stand borders. Where two stands of different height meet, the stand border becomes clearly visible. Fig. 5: forest stem biomass image calculated from forest heights. Assuming that the extracted height of each pixel equals the h100 of that pixel, the height-biomass allometry for Norway spruce (eq. 3) was applied to calculate stem biomass. Since the test site is dominated by managed Norway spruce stands the accuracy should not fail by more than +/15 %. Conclusions It was shown that forest height is a key parameter for estimating forest stem biomass. As shown from forestry tables, the relation between height and biomass follows the law of proportional growth [2]. This is not self-speaking since managed forests are not left to grow naturally but are altered (i.e. thinned out) in order to maximize the growth rate. A height-biomass relation that was derived for Norway spruce (Picea abies, Assmann, moderate thinning) was shown to vary only 15 % for a number of different spruce forestry tables, despite a wide range of stand conditions and thinning methods. Also height-biomass relations for four common forestry species did not vary more than 15 %. Over the Fichtelgebirge that consists of 98 % spruce forests and has mainly been managed in a conservative (tableconform) way, the stem biomass was calculated using a height-biomass regression for Norway spruce. The estimated error for the height-biomass allometry can therefore reach 15 % for some stands. A 10 % error in the height determination [13] results in an error of 17-18 %, leading to a maximum possible error of 25-30 %. It is likely that the ground measurements of conventional forest inventories do not achieve such an accuracy. The tree height determination of forest inventories is expected to yield an error 10 %. Also ground measurements need to extrapolate from few samples (typical grid: 0.5-1 ground measurement/ ha). Deviations will arise for non-table-conform forests, i.e. natural forests, forests cultivated using a different forestry concept or disturbed forests (calamities, ...). Here, the height-biomass relation will vary with the complexity of the forest structure: different layers, tree densities, etc.. In such cases, additional information about species (obtained for example from optical data) and/or structural information like forest density or layer structure (extracted from radar data) are needed to raise the estimation accuracy. Naturally growing forests e.g. can be expected to exhibit a stronger exponent as can be seen in Fig. 6. Fig. 6: Height-biomass allometry for a montane tropical rain forest (Peru [3]) and a temperate spruce forest (Germany [1]) The proposed approach to estimate above ground biomass from forest height estimates obtained from polarimetricinterferometry shows some significant advantages compared to the conventional backscatter saturation basedmethodology: (1) significantly higher biomass levels can be measured. No saturation limit has been found for temperateforests up to 400 t/ha. (2) Instead of a direct relation of backscatter amplitude to biomass, important forest parameters asforest height, stand and/or canopy density and underlying topography are directly estimated. These parameters areuseful for a wide spectrum of forest applications apart of biomass estimation. References1. Bayer. Ministerium f. Landwirtschaften u. Forsten, BMELF (1979). Hilfstafeln fuer die Forsteinrichtung.Eching 2. Bertalanffy v. L. (1951). Theoretische Biologie, II. Band, Stoffwechsel, Wachstum. A. Francke AG Verlag. 3. Boerner, A. (2000). Classification of premontane tropical Forests at the Eastern Slope of the Andes in the RioAvisado Watershed, Alto Mayo Region, Northern Peru. Univeristy of Bayreuth, Nov. 2000 4. Burschel, P., Huss, J. (1997). Grundriss des Waldbaus. Berlin 5. Cloude, S.R., Papathanassiou, K.P. (1998). Polarimetric SAR Interferometry. IEEE Transactions onGeoscience and Remote Sensing, vol. 36, no. 5, pp. 1551-1565, September 1998 6. Cloude, S.R., Papathanassiou, K.P., (2002). A 3-stage Inversion Process for Polarimetric SAR Interferometry.Proceedings of European Conference on Synthetic Aperture Radar, EUSAR’02, pp. 297-282, Cologne,Germany, 4-6 June 2002 7. Cloude, S.R., Papathanassiou, K.P., Woodhouse, I., Hope, J., Suarez Minguez, J.C., Osborne, P., Wright, G.(2001). The Glen-Affric Project: Forest Mapping using Polarimetric Radar Interferometry. Proceedings ofIGARSS’01, Sydney, Australia, July 9-13, 2001 8. Dixon, R.K. et al (1994). Carbon Pools and Flux of Global Ecosystems. Science, vol. 263, Jan. 14, 1994 9. FAO (2001). State of World’s Forests. NY10. Imhoff, M.C. (1995). Radar Backscatter and Biomass Saturation: Ramification for Global Biomass Inventory.IEEE Transactions on Geoscience and Remote Sensing, vol. 33, no.2. pp. 511-518, 1995 11. Papathanassiou, K.P., Cloude, S.R. (2001). Single Baseline Polarimetric SAR Interferometry. IEEETransactions on Geoscience and Remote Sensing, vol. 39, no. 11, pp. 2352-2363, November 2001 12. Papathanassiou, K.P., Reigber, A., Cloude, S.R. (1999). Vegetation and Ground Parameter using PolarimetricInterferometry Part I/II: The Role of Polarization/ Parameter Inversion and Optimal Polarizations. ProceedingsCEOS SAR Workshop, Toulouse, 26-29 October 1999 13. Papathanassiou. K.P,. Mette, T., Hajnsek, I. (2003). Model based Forest height Eestimation from SingleVaseline Pol-InSAR Data: the Fichtelgebirge test case. Proceedings of POLinSAR Conference, Frascati, Italy,Jan. 2003 14. Pretzsch, H. (2001). Modellierung des Waldwachstums. Parey, Berlin. 15. Schober, R. (1995). Ertragstafeln wichtiger Baumarten. Frankfurt