On the Retrieval of Forest Biomass from SAR Data

The question of retrieving forest biomass from SAR data has been subject of numerous research works since the early 1990s. The first results using AIRSAR data showed that it was possible to invert P band SAR data into biomass maps. Using the currently available spaceborne SAR data, namely ERS, JERS and RADARSAT, the biomass retrieval appears limited in terms of biomass range and, in certain cases, site specific. To provide an overview on issues related to forest biomass retrieval, this paper will present a survey of the possibilities and limitations of SAR data at different frequencies, polarisations and incidence angles. First the definition of biomass and biomass parameters (age, height, stem volume, dbh) will be addressed, and the information requirements in forest management and in global environment studies for biomass information will be discussed. The information content of SAR data in terms of the retrieval of biomass parameters will be assessed, based on an understanding of the underlying scattering mechanisms, which in turn are derived from observations and modelling results. For this purpose, an analysis of data acquired by multiple frequency, incidence and polarisation systems and by systems with interferometric capabilities has been carried out. The results, which have shown that the sensitivity to biomass parameters differs strongly at different frequencies, polarisations and incidences, have been interpreted using theoretical models. The aim was to determine the scattering mechanisms involving different categories of tree elements such as trunks, branches and leaves. This understanding is necessary to explain the relationships between the radar measurements with the dielectric and geometrical properties of the constituent scatterers, which in turn are related to the required biomass parameters (e.g. above ground biomass or stem volume) via allometric equations. The theoretical models appeared useful to assess the validity domains of the observed relationships, when the effects of ground characteristics, tree species, mixture of species, topography ... need to be accounted for. The information content relevant to biomass retrieval of ERS, JERS, RADARSAT data, including interferometric measurements will be summarised. I BIOMASS PARAMETERS Measurements of biomass and related forest parameters (age class, height, basal area ...) are of prime importance in forestry applications and in environment and climate studies. However, the type of measurements and their required scale and accuracy differ in these applications. Forest management operations such as timber exploitation, forest thinning, afforestation monitoring require quantitative, accurate information at a local level, on the timber volume and growth. At the regional level (or country level), requirements are mainly related to the mapping of forest resources and the identification of forest changes relevant to a forest resource inventory. In global environment and climate studies, biomass is a key variable in annual and long-term changes in the terrestrial carbon cycle, and needed in modelling carbon uptake and redistribution within the ecosystems. The amount and distribution of biomass over the earth’s surface is one of the major uncertainties in our ability to understand the global carbon cycle. In forest resource inventory and management the following main biomass and structural forest parameters are addressed: Timber volume: above ground volume of standing trees, for all diameters down to a certain limit, typically 12 cm in Europe. The information is used for wood, fibre and fuel supply. Timber annual increment and timber annual cut and storage are the derived parameters. Stem volume: the volume over bark of the stem, down to a given diameter, does not include the volume of branches. Diameter at breast heigh (dbh): diameter of the stem at 1.3 m height from the ground level, recorded down to a certain limit, e.g. 12 cm. Mean height: ground to top of tree measured on single trees. In forest management practices, biomass parameters are estimated in the following steps: sample plots or transects are selected in a forest region, for example plots of 0.5 ha or transects of 10x100 m; inside the plot, the species composition is established, diameter and height of all living trees down to a certain limit (e.g. 12 cm) are measured. Then the timber volume or stem volume, expressed in cubic metres per hectare, are estimated for each tree using existing allometric equations established for each species and age. For global change studies, the parameter of interest is the above ground dry weight (biomass, expressed in tonnes per hectare), estimated also using allometric equations, and a wood density value, varying typically from 0.5 g/cm to 0.7 g/cm It can be understood that uncertainties in in-situ biomass estimates can be very important. Also, differences between measurements and estimation methods may cause large discrepancies between estimates. For example, the limit of recorded diameters of stem and branches can vary from 5 cm to 12 cm depending on the standard used. Also, for a given forest, discrepancies between values using different allometric equations can be very large . In an example of secondary forest, the differences between biomass estimates have been found as large as 50% for a 18 year stand, up to 400% for younger stands (Alves, 1997). In monospecific forests, the uncertainties are reduced. 20% is the figure generally given in biomass estimates for forest plantations. Also, above all, biomass measurements on the ground are very time consuming, labor-intensive and often constrained by lack of access. II RADAR REMOTE SENSING OF FOREST BIOMASS For the above reasons, the perspectives of using remote sensing techniques to estimate forest biomass presented an important interest. Remote sensing data available at different scales, from local to global, and from various sources, from optical to microwave, are expected to provide information that could be related indirectly, and in different manners, to biomass information. Numerous studies have demonstrated that approaches using optical remote sensing data do not work for most terrestrial biomass densities. In areas of low biomass , the NDVI has been used to derive LAI and biomass value. However, the relationship depends on the soil reflectance and NDVI saturates when the canopy reaches full coverage. Similar results have been obtained with mid-infrared channels . With their longer wavelength, microwaves are expected to penetrate into the canopy, to interact with different components of the trees, and to provide information about biomass up to a higher saturation level compared to optical systems. The relationships between microwave backscatter and total above ground biomass have been studied extensively. Experimental results have been obtained since early 1990’s at different forest sites (Le Toan et. al., 1992, Dobson et al., 1992, Imhoff et al., 1995). The data were collected using multifrequency polarimetric SAR systems such as the AIRSAR and the Shuttle Imaging Radar SIR-C/X-SAR. These studies all showed that 1) biomass dependence of radar backscatter varies as a function of radar wavelength, polarisation and incidence angle. Longer wavelength is the most sensitive, cross polarisation HV is more sensitive than like polarisations HH and VV, 2) the sensitivity of radar backscatter intensity to variations of biomass saturates after a certain level of biomass is reached. The saturation level is higher for longer wavelengths. It was found that for C band HH or VV, the saturation is reached at 30 t/ha , whereas at P band HV, biomass as high as 200t/ha can be reached. More recently, ERS interferometric coherence has also been analysed as a function of forest biomass. The coherence decreases as a function of forest biomass and saturates also at low biomass. According to these results, the currently available spaceborne systems ERS, RADARSAT, operating at Cband, HH or VV are not the most adapted to biomass retrieval. JERS 1, operated until October 1998 at L band HH, provided slightly better sensitivity and higher saturation point. Despite their limitations, the C-band SAR systems are and will be providing data on a regular basis with ERS, RADARSAT and ENVISAT ASAR. In terms of biomass retrieval, the possibilities of using the available data to get a low / high biomass class discrimination, or to go further and have more than one class in the low range of biomass, could be an advantage compared to optical systems, but only if the methods prove to be more robust, since both systems are sensitive only to biomass in a limited biomass range. The question to be addressed at present is to know how the radar measurements from ERS, RADARSAT, ENVISAT ASAR, and to a lesser extent JERS-1, are related to biomass parameters in different forests ,and what is the possibility of inversion. An alternative to the use of existing spaceborne optical and radar remote sensing methods is to prospect other sensor types, either airborne systems which will be used as complementary to spaceborne operations, or new concepts which need to be considered for future spaceborne systems. Among the future systems that will be implemented in space in the next few years, polarimetric SAR at L band presents the best system for biomass retrieval, as shown by results obtained with SIR-C/X-SAR. Research and technical investigations aiming at developing a spaceborne Pband SAR are underway. Single pass interferometry, as in X-SAR/SRTM will be an alternative to have the information on the canopy height, in areas where precise DEM is available. From there, biomass information could be derived. High resolution airborne systems can be developed to complement the spaceborne systems. Using a VHF airborne system, recent results demonstrated that the SAR intensity is still sensitive to biomass up to 300 tons/ha. III -FOREST BIOMASS RETRIEVAL USING ERS, RADARSAT AND JERS DATA 3-1 Physical background: The first step was to understand the main interaction mechanisms between the radar wave at C-band, VV polarisation (ERS), C-band HH (RADARSAT), L-band HH (JERS), and the forest canopy. The understanding will be based on experimental observations and on physical modelling. To better understand the effect of frequency, polarisation, incidence, research results using multifrequency polarimetric SAR such as those on airborne systems or on the spaceborne SIR-C/ X-SAR were used. To illustrate the relationships between SAR measurements and forest biomass, the results using SIR-C/X-SAR data obtained over the Landes forest were chosen. The forest is the largest plantation forest in France, totally formed of maritime pine in large homogeneous stands of various ages.The advantage to use this type of forest is that the uncertainties of in-situ parameters are reduced compared to natural forests, and also, the physical understanding of the interaction process is facilitated by a more accurate description of the canopy. Fig 1 shows the backscattering coefficients as a function of forest biomass at HH polarisation, L and C band, and incidence angles of 26° and 54°, close to that of ERS (23°) and JERS (35°). Different behaviours can clearly be observed. The backscattering exhibits no variation with increasing biomass at C band 54° and an increasing trend at L band 54°, whereas a decreasing trend is observed at C-band 26° until 10-15 years (30-50 tons/ha), where the signal becomes stable. At L-band 26°, a small decrease is also observed at young ages. The same results were observed at VV polarisation. At this point, these empirical data can be used to draw conclusion on the use of ERS and JERS data for Landes forest observations. In both case, it would be possible to discriminate young (<10 years) from old forests, and to detect low biomass up to 30 t/ha. However, the results could be site and time specific. The effects of tree varieties, stand composition, stand homogeneity, weather and environmental conditions need to be assessed. Total Biomass (t/ha) 0 33 65 95 130 150 -10 -9 -8 -7 -6 -5 -4 -3 -2 C -B an d h h ( d B m 2 / m 2 )