*** most updated version found in Google Document linked above ***
Global Aboveground biomass Potential (GAP)
Global aboveground biomass potential (GAP) uptake, and rate (coming late 2021), calculated using the SPEC Lab’s GAP workflow (v45 as of 09/26/21, as described at: http://www.speclab.org/gap and available for download at: http://www.speclab.org/datasets-and-codes are created through a blended approach that integrates custom procedures in the IDL, R, and Python coding languages, with Google Earth Engine, ArcMap, and ENVI workflows, with output products being 100x100m spatial resolution GeoTiff images. First, the variable predicted in this workflow is the aboveground biomass (AGB) as calculated by the European Space Agency (ESA) Climate Change Initiative (CCI) v2.0 (see Santoro and Cartus 2021, and https://climate.esa.int/media/documents/Biomass_D4.3_CCI_PUG_V2.0.pdf). This data provides a global map with a resolution of 100 x 100 m and with units of Mg/ha of above-ground biomass (AGB); which they define as the density, i.e., the amount of living biomass per unit area. as expressed in units of mass of dry matter per unit ground area, i.e., Mg/ha-1 (Megagrams per hectare). In addition to an estimation of the amount of AGB per pixel, the Standard Deviation (SD) uncertainty of that estimation is also provided as a separate map, also with per-pixel units of SD of the AGB estimates in Mg/ha.
To create a wall-to-wall prediction of the Maximum Possible AGB; First, a collection of potential biophysical and bioclimatic predictors of AGB was compiled. Specifically: long-term climatic water deficit (CWD) (https://chave.ups-tlse.fr/pantropical_allometry.htm), Soil Cation Exchange Capacity (CEC) from SoilGrids v2 (see https://soilgrids.org and de Sousa et al. 2020), Global Inundation Extent from Multi-Satellites (GIEMS) (see http://www.estellus.fr/index.php?static13/giems-d15 and Fluet-Chouinard et al. 2015 and Prigent et al. 2007), Elevation (NASA Shuttle Radar Topography Mission (SRTM) Version 3.0 Global 1 arc second), Global SRTM Landforms (e.g., peak/ridge, upper slope, or valley; see Theobald et al. 2015), Global SRTM mTPI (Multi-Scale Topographic Position Index; a continuous variable ranging from bottom of valley to peak of ridge, see Theobald et al. 2015), hydrosheds (e.g., the upstream area (in number of cells) draining into each cell, see Lehner et al. 2008), biomes and ecoregions (see https://ecoregions.appspot.com), and the first 16 Chealsea bioclimatic predictors (described at https://chelsa-climate.org/bioclim and in Karger et al. 2017, Dryad).
In addition to the AGB maps, three levels of binary (1=inside, 0=outside) study area mask area are also calculated, named “Low”, “Medium” and “High”, and which refer to the level of confidence expressed in the “AGB Maximum Possible” values. The “Low” mask includes a latitude filter of 30 and -30 degrees and only terrestrial land pixels as defined by the Copernicus 100m land cover year 2019 discrete classification, and specifically excluding land cover classes 0 (“no input data available”), 80 (“permanent water body”) and 200 (“open sea”). The “Medium” mask further restricts the “Low” mask to exclude GIEMS classes 1, 2, or 3 which are permanent water bodies (Aires et al. 2017). The “High” mask then further restricts the “Medium” mask to include only forested areas, using only classes 111-126 of the same Copernicus discrete land cover data described above.
Locations for 2194 established primary forest plots located between -30 to 30 degrees latitude, and in a forest biome (RESOLVE 2017) were identified in a literature review, specifically from the Forest Observation System (Schepaschenko et al. 2019), Larjavaara and Muller‐Landau 2012, and Mitchard et al. 2014. We then added 56 additional plots across key bioclimatic gradients where plot density was visually determined to be insufficient, for a total of 2050 plots, which were then randomly divided the plots into 60% for calibration and 40% for validation purposes. The calibration plots were then buffered by 5 km and approximately 20 pseudo-plots were randomly located within each buffer region which we referred to as primary forests landscapes, for a total of 29,000 pseudo-plots.
We used a multi-step process to calibrate and validate the GAP workflow. First, we ran a single random forest model (see http://haifengl.github.io/api/java/smile/regression/RandomForest.html for detailed information on the random forests model) with all potential predictors to generate a map of plot AGB, henceforth referred to as Maximum Possible AGB. We then manually reviewed the outlier pseudo-plot locations to check if recent high-resolution satellite imagery indicated they were degraded or deforested between when the plot was included in the literature and 2017 when the AGB Present was derived, resulting in the removal of 56 pseudo-plots (coincidentally the same number as we had added). Lastly, we filtered out any points having a proportion of Maximum Possible AGB less than 80%, which indicated points that had either had degradation to AGB occur, or were subject to any of a variety of remote sensing error sources in the AGB map development process.
These were then input into an ‘iterative random forest regression with bias correction’. Specifically, consecutive random forest regressions (each with 1000 trees) were used to first predict the Maximum Possible AGB and then improve this by subsequently modeling the remaining prediction error. In five iterations, this process reduced the out Of Bag Error Estimate from 33 to 10. The final most significant explanatory variables were (ordered highest to lowest for top 5 only): Bio_3, Topo_pos, Bio_12, Bio_4, and Elevation. The bias correction step adjusted for differences among broad geographic regions, specifically the Americas, Africa, and Asia, by creating a separate linear regression between calibration pseudo-plot AGB and the final predicted Maximum Possible AGB, and these three regressions were then used to create the final map of Maximum Possible AGB for the entire global study area.
Validation was performed at the global study area scale using the 40% validation plots (n = 870; note not pseudo-plots) and consisted of a linear regression equation between plot measured AGB and that predicted by our approach which was highly significant (N=870, p-value < 0.001, Adj-R2 = 0.74).
AGB potential uncertainty was calculated as separate min and max AGB potential uptake values that bounded the predicted AGB potential, with calculations for the four key output maps performed using the following equations:
“AGB Proportion” = (“AGB 2017” / “AGB Predicted Max Possible”) * 1000 (eq. 1)
“AGB Potential” = “AGB Predicted Max Possible” – “AGB 2017” (eq. 2)
“AGB 2017 Minimum” = “AGB 2017” – “AGB 2017 SD” (eq. 3)
“AGB 2017 Maximum” = “AGB 2017” + “AGB 2017 SD” (eq. 4)
“AGB Potential Min” = “AGB Predicted Max Possible” – “AGB 2017 Minimum” (eq. 5)
“AGB Potential Max” = “AGB Predicted Max Possible” – “AGB 2017 Maximum” (eq. 6)
For example, for a pixel located in an old-growth Amazon forest. We see values of “AGB Predicted Max Possible” = 362, “AGB 2017” = 316, “AGB Potential” = 46, “AGB Potential Min” = 279, “AGB Potential Max” = 0, “AGB Proportion” = 872 (remember, this is in 1000 = 100% scale, so the value of 872 indicates that “AGB Potential” value of 46 is equal to 1000-872 = 128, or 12.8%). An additional old-growth Amazon forest pixel provides similar results of Max Possible AGB: 363, Potential AGB: 165 (62 to 268), and Proportion Max AGB: 545. And an adjacent pasture pixel provides results of Max Possible AGB: 364, Potential AGB: 361 (356 to 362), and Proportion Max AGB: 8. Lastly, we assess the relationship between “AGB Proportion” and AGB uncertainty (calculated as the (“AGB Pot” / “AGB Potential Max”)/”AGB_Potential”), using 5000 randomly distributed global locations, and limiting points to those in the “Medium Mask”, and find a highly significant linear positive linear regression (N=689, p-value < 0.0001, Adj-R2 = 0.52, ‘“AGB Uncertainty” = 0.1257882 + 0.001046 * “AGB_Proportion”’). This result highlights that while uncertainty is higher in more intact stands, in areas of restoration potential, such as young secondary forests or pastures, the uncertainty becomes very low.
Aires, F., Miolane, L., Prigent, C., Pham, B., Fluet-Chouinard, E., Lehner, B. and Papa, F., 2017. A global dynamic long-term inundation extent dataset at high spatial resolution derived through downscaling of satellite observations. Journal of Hydrometeorology, 18(5), pp.1305-1325.
Chave J, Rejou-Mechain M, Burquez A, Chidumayo E, Colgan MS, Delitti WBC, Duque A, Eid T, Fearnside PM, Goodman RC, Henry M, Martinez-Yrizar A, Mugasha WA, Muller-Landau HC, Mencuccini M, Nelson BW, Ngomanda A, Nogueira EM, Ortiz-Malavassi E, Pelissier R, Ploton P, Ryan CM, Saldarriaga JG, Vieilledent G. Improved allometric models to estimate the above ground biomass of tropical trees. 2014. Global Change Biology. DOI: 10.1111/gcb.12629
de Sousa, L.M., Poggio, L., Batjes, N.H., Heuvelink, G.B., Kempen, B., Riberio, E. and Rossiter, D., 2020. SoilGrids 2.0: producing quality-assessed soil information for the globe. Soil Discuss., 1.
Dmitry Schepaschenko, Jérôme Chave, Oliver L. Phillips, Simon L. Lewis, Stuart J. Davies, Maxime Réjou-Méchain, Plinio Sist, Klaus Scipal, Christoph Perger, Bruno Herault, Nicolas Labrière, Florian Hofhansl, Kofi Affum-Baffoe, Alexei Aleinikov, Alfonso Alonso, Christian Amani, Alejandro Araujo-Murakami, John Armston,, Luzmila Arroyo, Nataly Ascarrunz, Celso Azevedo, Timothy Baker, Radomir Bałazy, Caroline Bedeau, Nicholas Berry, Andrii M. Bilous, Svitlana Yu. Bilous, Pulchérie Bissiengou, Lilian Blanc, Kapitolina S. Bobkova, Tatyana Braslavskaya, Roel Brienen, David F.R.P. Burslem, Richard Condit, Aida Cuni-Sanchez, Dilshad Danilina, Dennis del Castillo Torres, Géraldine Derroire, Laurent Descroix, Eleneide Doff Sotta, Marcus V.N. d'Oliveira, Christopher Dresel, Terry Erwin, Mikhail D. Evdokimenko, Jan Falck, Ted R. Feldpausch, Ernest G. Foli, Robin Foster, Steffen Fritz, Antonio Damian Garcia-Abril, Aleksey Gornov, Maria Gornova, Ernest Gothard-Bassébé, Sylvie Gourlet-Fleury,, Marcelino Guedes, Keith Hamer, Farida Herry Susanty, Niro Higuchi, Eurídice N. Honorio Coronado, Wannes Hubau,, Stephen Hubbell, Ulrik Ilstedt, Viktor V. Ivanov, Milton Kanashiro, Anders Karlsson, Viktor N. Karminov, Timothy Killeen, Jean-Claude Koffi Konan, Maria Konovalova, Florian Kraxner, Jan Krejza, Haruni Krisnawati, Leonid V. Krivobokov, Mikhail A. Kuznetsov, Ivan Lakyda, Petro I. Lakyda, Juan Carlos Licona, Richard M. Lucas, Natalia Lukina, Daniel Lussetti, Yadvinder Malhi, José Antonio Manzanera, Beatriz Marimon, Ben Hur Marimon Junior, Rodolfo Vasquez Martinez, Olga V. Martynenko,, Maksym S. Matsala, Raisa K. Matyashuk, Lucas Mazzei, Hervé Memiaghe, Casimiro Mendoza, Abel Monteagudo-Mendoza, Olga V. Morozyuk, Liudmila Mukhortova, Samsudin Musa, Dina I. Nazimova, Toshinori Okuda, Luis Claudio Oliveira, Petr V. Ontikov, Andrey F. Osipov, Stephan Pietsch, Maureen Playfair, John Poulsen, Vladimir G. Radchenko, Kenneth Rodney, Andes H. Rozak, Ademir Ruschel, Ervan Rutishauser, Linda See, Maria Shchepashchenko, Nikolay Shevchenko, Anatoly Shvidenko,, Marcos Silveira, James Singh, Bonaventure Sonké, Cintia Souza, Krzysztof Stereńczak, Leonid Stonozhenko, Martin J P Sullivan, Justyna Szatniewska, Hermann Taedoumg, Hans ter Steege, Elena Tikhonova, Marisol Toledo, Olga V. Trefilova, Ruben Valbuena, Luis Valenzuela Gamarra, Sergey Vasiliev, Estella F. Vedrova, Sergey V. Verhovets,, Edson Vidal, Nadezhda A. Vladimirova, Jason Vleminckx, Vincent A. Vos, Foma K. Vozmitel, Wolfgang Wanek, Thales A. P. West, Hannsjorg Woell, John T. Woods, Verginia Wortel, Toshihiro Yamada, Nur Hajar Zamah Shari, Irié Casimir Zo-Bi (2019). A global reference dataset for remote sensing of forest biomass. The Forest Observation System approach. [Data Collection]. DOI: 10.22022/ESM/03-2019.38.
Fluet-Chouinard E., Lehner B., Rebelo L.M., Papa F., Hamilton S.K. (2015): Development of a global inundation map at high spatial resolution from topographic downscaling of coarse-scale remote sensing data. Remote Sensing of Environment 158: 348-361. dx.doi.org/10.1016/j.rse.2014.10.015
Karger, D.N., Conrad, O., Böhner, J., Kawohl, T., Kreft, H., Soria-Auza, R.W., Zimmermann, N.E., Linder, P., Kessler, M. (2017). Climatologies at high resolution for the Earth land surface areas. Scientific Data. 4 170122. https://doi.org/10.1038/sdata.2017.122d
Karger D.N., Conrad, O., Böhner, J., Kawohl, T., Kreft, H., Soria-Auza, R.W., Zimmermann, N.E, Linder, H.P., Kessler, M.. Data from Climatologies at high resolution for the earth’s land surface areas. Dryad Digital Repository. http://dx.doi.org/doi:10.5061/dryad.kd1d4
Larjavaara, M. and Muller‐Landau, H.C., 2012. Temperature explains global variation in biomass among humid old‐growth forests. Global Ecology and Biogeography, 21(10), pp.998-1006.
Lehner, B., Verdin, K., Jarvis, A. (2008): New global hydrography derived from spaceborne elevation data. Eos, Transactions, AGU, 89(10): 93-94.
Mitchard, E.T., Feldpausch, T.R., Brienen, R.J., Lopez‐Gonzalez, G., Monteagudo, A., Baker, T.R., Lewis, S.L., Lloyd, J., Quesada, C.A., Gloor, M. and Ter Steege, H., 2014. Markedly divergent estimates of Amazon forest carbon density from ground plots and satellites. Global Ecology and Biogeography, 23(8), pp.935-946.
Prigent, C., Papa, F., Aires, F., Rossow, W. B. & Matthews, E. (2007). Global inundation dynamics inferred from multiple satellite observations, 1993-2000. Journal of Geophysical Research, 112(D12107), 1-1
Santoro, M.; Cartus, O. (2021): ESA Biomass Climate Change Initiative (Biomass_cci): Global datasets of forest above-ground biomass for the years 2010, 2017, and 2018, v2. Centre for Environmental Data Analysis, 17 March 2021. doi:10.5285/84403d09cef3485883158f4df2989b0c. http://dx.doi.org/10.5285/84403d09cef3485883158f4df2989b0c
Theobald, D. M., Harrison-Atlas, D., Monahan, W. B., & Albano, C. M. (2015). Ecologically-relevant maps of landforms and physiographic diversity for climate adaptation planning. PloS one, 10(12), e0143619