This the multi-page printable view of this section. Click here to print.

Return to the regular view of this page.

Science and Products

The science behind the generation of BIOMASS products.

This section briefly illustrates the algorithms for the generation of BIOMASS products and implemented in BioPAL. For further information refer to the original source For specific topics see Publications.

The following table reports current target figures for the products.

Level 2 Product Resolution Accuracy
AGB 200 m <20% (or 10 t/ha for AGB<50 t/ha)
FH 200 m <30% for trees higher than 10 m
FD 50 m detection at a specified level of significance

The products are intended to be global, covering forested land areas between 75°N and 56°S but subject to United States Department of Defense Space Object Tracking Radar (SOTR) restrictions, as illustrated.

BIOMASS coverage Map of coverage of ESA and NASA satellite measurements of forest structure and biomass. The background shows the global coverage area of the NISAR mission and the sensitivity of NISAR to aboveground biomass values < 100 Mg/ha (green and yellow). The BIOMASS mission coverage includes the tropical belt and a portion of the northeast Siberia and the GEDI LiDAR sampling coverage from the International Space Station (ISS) between ±50 degrees latitude. Original source Forest Observation System.

1 - Forest Disturbance

Forest disturbance algorithm description

Forest Disturbance (FD) is defined as an area where an intact patch of forest has been cleared, expressed as a binary classification of intact versus deforested or logged areas.

The disturbance product generation is based on Level 1 products: the restriction to severe disturbance allows the spatial resolution to be much finer than the other products since the associated changes in backscatter are expected to be several dBs in each intensity channel. The detection of changes in the polarimetric time series is based on hypothesis testing, where the null hypothesis is that in a time series of polarimetric data no change has occurred (at a given position and up to a given time). If this hypothesis fails at a given level of significance, then we assume a change has occurred. Note that it is closely related to the constant false alarm rate approach to target detection, in which the probability that an undisturbed pixel is incorrectly classified as disturbed is fixed. The estimates are updated each time a new acquisition is added to the stack, and significance levels can be attached to the test statistics.

An open issue in generating the BIOMASS disturbance product in this way is that changes in the signal caused by disturbance occur against a general background of non-forest and environmental changes. Further work is required to quantify how much these nuisance changes will increase the false detection rate. An important requirement is also to have an initial forest mask, derived from the BIOMASS data themselves or from some other source, in order to mask out detections in the non-forest areas.

2 - Forest Height

Forest Height estimation

Forest Height (FH) is defined as upper canopy height according to the H100 standard used in forestry expressed in meters. H100 is defined as the average height of the 100 tallest trees/hectares.

The baseline methodology for BIOMASS interferometric phase is implemented by means of PolInSAR. Polarimetric-interferometric correlations estimated from data are linked to forest structural parameters such as forest height, ground-to-volume ratio, canopy extinction and temporal decorrelation through the Random Volume over Ground (RVoG) model. This model assumes that forest scattering comes from an extended layer of height equal to the canopy height above an opaque ground layer. The vertical distribution of scatterers is weighted by an extinction function, accounting for electromagnetic attenuation through the vegetation. The propagation through the volume is assumed to be independent of polarization.

The main challenges for BIOMASS are the presence of ground scattering in all polarizations due to the limited extinction at P-band, the limited resolution available at 6 MHz and temporal decorrelation due to the three-day repeat cycle. During the tomographic phase, FH will be estimated from the upper envelope of the tomographic voxel intensity, as well as the Digital Terrain Model (DTM) from the lower envelope to be ingested to the RVoG model as a-priori.

3 - AGB

Above Ground Biomass estimation

AGB stands for Above Ground Biomass and is defined as dry weight of woody matter per unit area, expressed in t/ha = Mg/ha. AGB includes the mass of live organic matter above the soil including stem, stump, branches, bark, seeds and foliage, it does not include dead mass, litter and below-ground biomass.

The relationship between P-band SAR backscatter intensity of forests and forest AGB is governed by a large number of factors related to both forest characteristics (3D structure, species, growth stage, leaf and wood water content, etc.) and the environment (topography, soil moisture and surface roughness). AGB retrieval accounting for all these factors is ill-posed. This often led in the past to the formulation of very complex models and/or using a lot of reference data, which are usually not available on a global scale.

Level 2 Studies brought a new conceptual design of the retrieval, making full use of the BIOMASS interferometric observation capabilities to circumvent these limitations. This revision is based on three research results of the mission preparatory activities:

  • the observational and modeling evidence that 30–40% of the total AGB in a dense tropical forest (BIOMASS main target area) is contained in the canopy region 25–35 m above the ground, and that the biomass in this region is highly correlated with the total AGB;
  • the development of signal processing techniques (i.e., ground cancellation) to cancel out the backscatter signal from the ground layer using interferometric stacks of P-band data and Digital Terrain Model (DTM) to isolate the volume scattering element of the forest canopy;
  • the development of an optimization approach to solve the model that minimizes the need for reference data.

All these results translate in the CASINO algorithm (CAnopy backscatter estimation, Subsampling, and Inhibited Nonlinear Optimisation). The algorithm inverts a power-law function relating AGB to ground cancelled backscatter data, through non-linear iterative minimization. Dimensionality of the problem is reduced by constraining the parameters according to phyisical considerations. The need for reference calibration data is minimized, though reference data still play a crucial role for absolute retrieval.