ESA’s GlobBiomass project and datasets Maurizio Santoro · 2020. 8. 25. · CCI Biomass 1stUser...
Transcript of ESA’s GlobBiomass project and datasets Maurizio Santoro · 2020. 8. 25. · CCI Biomass 1stUser...
CCI Biomass 1st User Workshop,Paris, 25 Sept. 2018
ESA’s GlobBiomass project and datasets
Maurizio Santoro
Gamma Remote Sensing
On behalf of GlobBiomass project team
What is GlobBiomass?
● GlobBiomass (2015-2017) was an ESA-funded project, part of the Data User Element (DUE).
The DUE has the aim of favoring the establishment of a long-term relationship between the
User communities and Earth Observation.
● The main purpose of GlobBiomass was to better characterise and to reduce uncertainties of
AGB estimates by developing innovative mapping approaches using EO and in-situ data
○ in five regional sites for the epochs 2005, 2010 and 2015 and
○ for one global map for the year 2010
Why a global map of forest biomass?
● Datasets available based on remote sensing○ Global AGB: Kindermann et al., 2014; GEO-CARBON, 2014; Liu et al., 2015; Hu et al., 2016
○ Biome AGB: Saatchi et al., 2011; Baccini et al., 2012; Thurner et. al., 2014; Avitabile et al.,
2016
● Most datasets use data from around year 2000 or represent AGB at coarse resolution
● Cross-comparisons reveal divergent estimates at local scale
● Errors and uncertainties often not (fully) described
● Weaknesses:○ Handful of remote sensing datasets used, often sub-optimal to derive biomass
○ Strong requirement on reference data for training retrieval models
Data and methods: issues and proposed solutions
● Issue 1: EO does not quantify biomass → The signals of EO data available for 2010 are only
weakly affected by biomass-related forest attributes
● Issue 2: wealth of models relating EO signals to “biomass” → classical approach to retrieve
biomass: train a model with in situ data or surrogate data → unrealistic approach at global
scale to capture spatial variability of the EO signal correctly
● Solution 1: use EO data to exploit as much as possible the information content on “biomass”
● Solution 2: (i) select a well-known modelling framework, (ii) that allows tuning of the model
parameters in space and time, and (iii) does not require in situ data for training (self-calibration
of model)
The GlobBiomass global retrieval method (EO2GSV)
The GlobBiomass global retrieval method (GSV2AGB)
Examples of Water Cloud Model
Boreal: GSV 300 m3/ha @ AGB: 150 Mg/ha (BCEF @ 0.5)
Wet tropics: GSV 300 m3/ha @ AGB: 250 Mg/ha (BCEF @ 0.85)
Envisat ASAR, HH or VV-pol(largest dynamic range)
ALOS PALSAR, HV-pol
Forest aboveground biomass, AGB (Mg/ha) @ 100m
Color bar constrained to 0 – 350 Mg/ha to enhance contrast
Examples of AGB estimates (Mg/ha)
North Poland Riau, Sumatra
Known caveats of AGB estimates Data processing issues → uncompensated topography in ALOS mosaic
West Sumatra DRC
Known caveats of AGB estimates Signal-related issues → Biomass of dense mangroves often underestimated
Matang, Malaysia
Known caveats of AGB estimates Signal-related issues → Biomass of flooded vegetation overestimated
Along Congo River, DRC
AGB standard error (%) @100m
Color bar constrained to 0 – 100% to enhance contrast
Contribution to standard errorTAr = Tropical rainforestTAwa = Tropical moist dec. forestTAwb = Tropical dry forestTBSh = Tropical shrublandTBWh = Tropical desertTM = Tropical mountain
SCf = Subtropical humidSCs = Subtropical drySBSh = Subtropical steppeSBWh = Subtropical desertSM = Subtropical mountain
TeDo = Temperate oceanicTeDc = Temperate continentalTeBSk = Temperate steppeTeBWk = Temperate desertTeM = Temperate mountain
Ba = Boreal coniferousBb = Boeal tundra woodlandBM= Boreal mountainP = Polar
Validation protocolInventory plots
Plot vs. pixel 0.1 deg averages of plots and pixels
Regional statistics
Total volume and above-ground biomass for 2010 Total volume in forest Average GSV in forestGlobBiomass: 694.6 109 m3 GlobBiomass: 142.7 m3/haFAO FRA 2010: 495.6 109 m3 (*) FAO FRA 2010: 121.8 m3/ha
Total above-ground biomass in forest Average AGB in forestGlobBiomass: 522.6 Pg GlobBiomass: 107.3 Mg/haFAO FRA 2010: 469.4 Pg (**) FAO FRA 2010: 115.4 Mg/ha
Forest areaGlobBiomass (based on CCI Land Cover): 4.87 109 haFAO FRA 2010: 4.06 109 ha
No data in FAO FRA 2010 for major countries:(*) Australia, Dominica, Ecuador, El Salvador, Paraguay, Togo, Venezuela (**) Dominica, Ecuador, El Salvador, Paraguay, Togo, Uruguay, Venezuela
Comparison with FAO FRA 2010 AGB statisticsAfrica: Countries adopting BCEF > 2
Asia:Right: Countries with topographyLeft: SE Asian countries
Europe: Forest fragmentation
Central America: countries adopting BCEF > 1.5
South America: Guyana, Fr. Guyana and Suriname, different FRA values
OceaniaPNG based on lowland dataNZ based on commercial forestNote: Size of dot proportional to forest area
Pakistan
Argentina
Ivory Coast
Cuba
PNG
New Zealand
Comparison with EO-based AGB estimates
• GlobBiomass generated the first global dataset of forest biomass at moderate resolution
• RS does not „see“ biomass → combination of available data streams mandatory to limit
estimation errors
• Strong confidence on the spatial distribution of biomass and its levels globally
• New set of estimates that may impact the global carbon budget so far assumed
• The estimates have local systematic errors BUT we understand these errors ○ EO data sub-optimal to estimate biomass
○ Ready-to-use EO data products often only choice, not the best one though
○ One global model, strongly adaptive, achieved a fairly decent result but we could not avoid
local over/underestimation due to the simplicity of the inversion model
Conclusions
• We need multiple sources of EO data that senses structure, species and moisture are
envisaged à currently, these are not available
• We need EO data as clean as possible from errors
• We need to explore the EO signals to understand how to “best” set up retrieval models
• Biomass retrieval models need in situ data for development but not necessarily for
operations
• We need to explore the impact of scales (remote sensing vs. in situ) in what we see
• We need a solid statistical framework for accounting for errors and uncertainties
• We need to move from a single epoch to a sequence of maps
A perspective from a “data producer”
• GlobBiomass global data products of AGB and GSV @ 100 m (version of 2018-05-31)
available at http://globbiomass.org/products/global-mapping/
• Cite as: Santoro, M. (2018): GlobBiomass – global datasets of forest biomass. PANGAEA,
https://doi.pangaea.de/10.1594/PANGAEA.894711
• For questions, comments and issues, please refer to
Maurizio Santoro
Data release