2. BPS and AGB Processing Overview

2.1 BPS Overview

The BIOMASS Processing Suite (BPS) is in charge of processing BIOMASS Level-0 data up to Level-3, generating a wide set of products. Product generation is depicted in separate figures for the TOM phase and the INT phase. The processing foresees the following steps:

• L1 processing: the BIOMASS image formation step, taking as input one L0 slice and generating as output N ≥ 1 L1 frames;

• Stack processing: coregistration and stacking that defines the required 7-image tomographic stacks or the 3-image interferometric stacks, including stack phase calibration;

• L2a processing: per-stack processing that provides inputs to tile generation performed by the L2b processors. For one global cycle the number of stacks varies mainly with latitude, approximately from a minimum of two at the Equator (ascending/descending) to about six at higher latitudes (three ascending/descending pairs);

• L2b processing: aggregates all stack-based L2a products over the same ground location (tile) in the same time interval and generates a single BIOMASS L2b product. The number of stacks available per DGG tile varies based on tile dimensions and swath coverage [RD8];

• L3 processing: consolidation processing removing discontinuities in L2b products.

Fig. 1 — Biomass high-level processing scheme from L0 data to L3 products, TOM phase.

Fig. 2 — Biomass high-level processing scheme from L0 data to L3 products, INT phase.

2.2 AGB Processing Overview

The AGB processing has as main goal the generation of maps of above ground biomass density at a spatial resolution roughly equivalent to ~200 m × 200 m (i.e. 4 ha) at the Equator [AD2]. It receives as input a stack of BIOMASS Level-1c images (SCS co-registered and calibrated stack), including its relevant annotated data [AD3]. The AGB processing is divided into three levels:

• L2a, generates a ground cancelled GN product, corresponding to L1 acquisitions filtered in order to attenuate ground scattering and enhance canopy scattering; interferometric diversity is used to perform filtering, so that only polarimetric diversity remains after filtering;

• L2b, generates above ground biomass density map from all the L2a products on a connected group of DGG tiles, as well as a quality map, training on external reference AGB [RD10];

• L3, consolidates the result removing discontinuities in L2b products.

Note

Provided that FD algorithm proper validation will only be possible with BIOMASS in orbit, the strategy proposed to deal with FH/AGB dependency on FD is to initially input to FH/AGB a forest/non-forest mask generated outside of BPS [AD1] [AD2]. After testing FD performance with BIOMASS in orbit, FD dependency can be switched on again, provided it has a satisfactory performance. The diagram of Fig.2 reports nonetheless FD input for completeness.

Note

L2 processor receives from Stack processor L1c products with two co-polarization (CP) channels and one cross-polarization (XP) channel. Details on the aggregation of XP channels into one XP are reported in [RD1].

Note

Regarding stack images cardinality, their number can vary (not always 3 or 7) due to contingency reasons and it can go up to 8 in TOM phase.

2.3 AGB Processing Workflow and Algorithmic Overview

The main steps of the AGB processing workflow are depicted in the following figure and described in detail in the following sub-sections. This algorithm is a tailoring of the one described in [RD11].

Fig. 3 — AGB processing detailed workflow. 5°x5° neighbourhood and 3°x3° processing block are shown in Fig.12.

2.3.1 Ground cancellation (L2A_P)

The main steps of the stack-based L2a block are:

• forest coverage check: predetermine which input data (as a whole image) should need to be processed;

• calibration: application of phase calibration and ground steering (terrain at zero elevation) screens to the stack;

• ground cancellation: removes the contribution of the ground scattering from the stack;

• sigma naught calibration: removes the first-order dependence on incidence angle and normalizes to resolution cell area;

• geocoding: projects the processed data on a geographic map, accounting for the different forest point locations with respect to the reference DEM.

2.3.2 Above-ground biomass density estimation (L2B_AGB_P)

The main steps of the L2b block are:

• selection of training data: checking for availability of training AGB data within the L2a stacks and selecting the appropriate source (either external reference AGB data or previous version of the BIOMASS AGB map);

• training data extraction: extracting data from the extent of a tile and its neighborhood, matching pixels with reference AGB and forest type information with pixels from L2a stacks;

• parameters estimation: estimating AGB model parameters;

• AGB mean and standard error estimation: extracting L2a stack data within the current tile and creating maps of AGB mean and standard error.

2.3.3 Note on interpretation of inputs and outputs tables

Inputs and outputs for each processing sub-step are summarised in specific subsections (two subsections for each sub-step). A specific tabular format is defined to improve access to the document. The meaning of each column is exemplified in the following table.

Column	Meaning
Symbol	Mathematical symbol of the quantity appearing in formulas
Name	Name of the quantity as related to other BPS documentation (e.g. AUX_FMT, L1 ATBDs)
Origin / Destination	BPS document name describing the origin of the quantity, processing step, or destination product
Size	Dimensions of the quantity
Variability	Range of values or update rate of an internal resource
Default	Default value for configuration parameters
Description	Verbal description of the quantity and its characteristics (e.g. measurement units)