BIOMASS Campaigns

The coregistration of heterogeneous geospatial raster maps is useful in various remote sensing applications.
Since the number of available data increases and the resolution improves, it is interesting to have an approach as automated, fast, robust and accurate as possible. In this paper, we propose to apply a solution based on optical-flow computation, originally used in Computer Vision. This algorithm allows the registration of images in a non-parametric way. After an analysis of the physical specificity of remote sensing images, we show that it is necessary to apply a preprocessing step that produces images similar enough for the flow algorithm. It results in a generic algorithm that handles the registration of large data in a very efficient way. Next, an analysis of the relevant parameters is investigated for several scenarios.

We demonstrate the efficiency and accuracy of our coregistration chain on radar, lidar and optical images of forest from the BIOSAR 2010 Campaign over Remningstorp (Sweden). Data considered are airborne forest images with resolutions of about one meter for radar, 50 cm for optical image, 10 m or 50 cm for LIDAR data. The algorithm was applied and compared to coregistration made using only georeferencing information. It was shown to be more accurate on all our examples. In the cases presented, the registration precision of our approach opens new opportunities for studies showing the gain of data fusion for forestry, for example:

- in applications involving the counting of trees;

- in vegetation specie classification;

- in biomass estimation.

This registration algorithm is intended to be used for other ESA campaigns whenever heterogeneous data (LIDAR, optical images) can be used as ground truth for results obtained from POLINSAR techniques. Moreover, it is adapted to the new context of Big Data in Earth Observation.

Tropical forests contain the major part of the world forest biomass and their changes by deforestation and/or by forest degradation strongly affect the terrestrial carbon budget. To measure with accuracy tropical forest biomass and its temporal change is one of the objectives of the ESA Biomass mission.

The retrieval algorithms currently developed for Biomass are based on advanced radar data processing: PolSAR, PolInSAR and TomoSAR.

As Biomass will be the first Synthetic Aperture Radar (SAR) using P-band (435 MHz), one issue is the lack of spaceborne P-band SAR data for developing and validating the retrieval algorithms. As a result, So many scientific campaigns activities are currently on-going to improve our knowledge of radar backscattering of forests.

AfriScat experiment is the second experiment based on long-term radar backscattering measurement in P-band. It will provide a feedback on two key issues linked to mission operations and Level-2 product quality:

The coherence of P-band radar as a function of time of day, time intervals and environmental conditions which directly impacts the quality of the Level-2 products;

The scattering characteristics of P-band radar within the tropical forest to support the development of robust widely-applicable retrieval algorithms and the development of algorithm theoretical baseline documents.

AfriScat campaign is built on the expertise gained from the previous TropiScat campaign which took place in French Guiana from 2011 to 2013, extending the results to the African tropical rain forest, whose structure and environmental conditions are different. The Ankasa Conservation Area Tower in Ghana, operating as part of the CarboAfrica eddy covariance network and EU ERC Africa GHG grant, has been selected for hosting the AfriScat experiment. Ankasa site characteristics differ from those of the TropiScat experiment. The Ankasa tower is located on a plateau, the ground topography is flat, on the contrary to French Guiana, with a terrain slopes opposite to the radar line of sight. Because of less topographic effects an higher double bounce contribution is expected and is likely to impact the results related to temporal decorrelation. In addition, forest structure, AGB and floristic composition also differ between Africa and South America forests.

The AfriScat experiment is based on two types of mesures:

Radar data: the scatterometer is located on a 60 meters high tower. This instrument is acquiring polarimetric and tomographic radar data. The measurement configuration is identical to the TropiScat one.

In-situ data: By using the instruments on the flux tower, the system is collecting a large set of environmental data like rain level, air and ground temperatures, soil moisture, visible/infrared radiation fluxes …

By analyzing these long-term data and the correlation between them, the AfriScat campaign will provide feedback on:

The long-term statistical characterization of temporal coherence at P-band over african tropical forests (diurnal, weekly, monthly, yearly);

The identification and characterization of the main sources of temporal decorrelation (environmental, instrument, geometrical, structural);

The 3-dimensional distribution of radar scatterers and their backscattering characteristics;

The impact of double bounce effects in the radar backscattering of tropical forests;

The validity of simplified physical and statistical scattering models (intensity, polarimetry, PolInSAR) used as basis for geophysical information retrieval.

The experiment has been started in July 2015 and will run until July 2016.

Measurement of tree structure is important for various reasons: estimating above ground biomass (AGB) i.e. the standing stock of carbon (C) stored per unit area; assessing tree function, particularly light, water and nutrient transport; metabolic scaling; competition; characterisation of age, type and history; response to climate and change/disturbance. In addition, tree structure is an important determinant of the Earth Observation (EO) signal, and is thus the focus of many parameter retrieval, calibration and validation efforts. Accurate measurements of tree shape and size are extremely hard to make, particularly non-destructively and for trees of any size. As a result many properties, particularly AGB, are inferred from empirical allometric relationships between easily measurable parameters such as diameter-at-breast height (dbh), and/or tree height, and tree size [1]. Allometric relationships also underpin estimates of forest properties from current and future EO missions; the forthcoming ESA BIOMASS and NASA GEDI missions will both rely on height allometries to infer AGB. Tropical forest allometries are based on relatively few destructive harvest measurements (few 10³ trees in total), heavily biased towards smaller trees, and often from only partial tree measurements. This results in allometries with poorly-quantified uncertainty, which increases with tree size [2]. As a result, AGB, and forest structure more generally, are highly uncertain across the tropics. This results in large uncertainties in estimates of tropical forest C stocks and fluxes, as well as major differences between estimates made using slightly different allometries and other assumptions [3].

Here, we show how terrestrial laser scanning (TLS) methods have developed to the point where they can be routinely deployed to measure tree structure of tropical forest trees accurately and non-destructively [4]; this is not currently possible any other way. We present results from TLS field campaigns across the tropics, including Africa, S. America and Australia. We use the TLS measurements to derive estimates of tree volume and, combined with wood density estimates, AGB. We show that these estimates agree to within a few % of destructively measured values [5]. We also show that TLS measurements can capture the size and structure of all trees within measured plots, resulting in size estimates that are significantly different from existing allometries. We discuss the potential for (and implications of) including TLS-derived measurements into tropical forest allometries, for improving estimates of C stocks, particularly for large trees, and particularly from forthcoming EO missions that will depend on tropical forest allometry. 

The ESA BIOMASS mission was selected in 2013 as the 7th Earth Explorer mission. BIOMASS will provide estimates of forest biomass and height with full coverage over the tropical areas exploiting the penetration capabilities of P-band. In order to further support the BIOMASS mission development, especially concerning the mission concept verification and the development of geophysical algorithms, ESA funded the P-band AfriSAR campaign following the previous INDREX-II (Indonesia, 2004) and TropiSAR (French Guiana, 2009). AfriSAR is being carried out in Gabon by ONERA (SETHI system, July 2015, dry season) and DLR (F-SAR system, February 2016, wet season). Four different test sites (Lopè, Mondah, Rabi and Mabounie) have been chosen including different physical forest structure types, biomass levels, growth stages, and different levels and kinds of disturbance.

Key AfriSAR objectives are: (i) to assess penetration conditions, (ii) to evaluate the performance in forest height and biomass estimation in the 100-300 t/ha range, (iii) to quantify temporal decorrelation, (iv) to evaluate the temporal variability of polarimetric signatures and vertical reflectivity estimates in two seasons (wet and dry). To reach these objectives, in all of the four test sites, fully polarimetric multibaseline data sets are being acquired in order to allow the application of polarimetric (Pol-SAR), polarimetric interferometric (PolInSAR) and tomographic (TomoSAR) techniques. The Pol-InSAR baselines (2 to 3 depending on the test site) have been optimized in order to provide the necessary sensitivity for forest height inversion. Besides, the TomoSAR baselines (7 to 9) have been designed to provide a vertical Rayleigh resolution between 10 and 15m. The selected area in the Lope national park will be definitely the most covered in space and time due to the large biomass range (from 60 to 600 t/ha), structural diversity (mosaic of savannah and dense primary forest) and terrain slope variation. Beyond the TomoSAR stack, two additional data sets with a reduced number of tracks (3-4) will be collected with 4 days revisit time to quantify temporal decorrelation effects over PolInSAR parameter inversion. Furthermore, the same baselines will also be acquired with different headings to evaluate the effect of topography.

In this contribution, the ONERA and DLR AfriSAR campaigns will be presented. The selected test sites and the related acquisitions will be presented. Furthermore, the results of the evaluation of the data quality will be discussed. Temporal coherence within both seasons (dry and wet) will be discussed and first PolSAR, Pol-InSAR and TomoSAR results regarding biomass estimation, forest height inversion and vertical structure parameters extraction will be shown. A preliminary validation of these quantities against ground measurements (where available) will be presented as well.

Not available