%0 Journal Article
%9 ACL : Articles dans des revues avec comité de lecture répertoriées par l'AERES
%A Razakamanarivo, R. H.
%A Grinand, Clovis
%A Razafindrakoto, M. A.
%A Bernoux, Martial
%A Albrecht, Alain
%T Mapping organic carbon stocks in eucalyptus plantations of the central highlands of Madagascar : a multiple regression approach
%D 2011
%L fdi:010053589
%G ENG
%J Geoderma
%@ 0016-7061
%K Above-and below-ground biomass ; Soil organic carbon ; Spatial ; variability ; Boosted regression tree ; Short rotation forestry ; Eucalyptus robusta
%M ISI:000291139900014
%N 3-4
%P 335-346
%R 10.1016/j.geoderma.2011.03.006
%U https://www.documentation.ird.fr/hor/fdi:010053589
%> https://www.documentation.ird.fr/intranet/publi/2011/06/010053589.pdf
%V 162
%W Horizon (IRD)
%X Recent concerns about global warming have resulted in more concerted studies on quantification and modeling of carbon (C) storage in different ecosystems. The aim of this study was to assess and map the carbon stocks in above (ABG), below-ground (BLG) biomass and soil organic carbon contained in the 30 centimeter top-layer (SOC) in coppices of eucalyptus plantations in the central highlands of Madagascar in an area of 1590 ha. Relationships between C stock and various biophysical (stool or shoot stockings and ages, circumferences) and spatial (elevation, slope, and soil type) factors that may affect C storage within each pool were investigated. Three different modeling techniques were tested and compared for various factor sets: (i) simple linear regression (SLM), (ii) multiple linear (MLM) models and, (iii) boosted regression tree (BRT) models. Weights of the factors in the respective model were analyzed for the three pool-specific models that produced the highest accuracy measurement. A regional spatial prediction of carbon stocks was performed using spatial layers derived from a digital elevation model, remote sensing imagery and expert knowledge. Results showed that BRT had the best predictive capacity for C stocks compared with the linear regression models. Elevation and slope were found to be the most relevant predictors for modeling C stock in each pool, and mainly for the SOC. A factor representing circumferences of stools and their stocking (stools.ha(-1)) largely influenced BLG. Shoot circumference at breast height and shoot age were the best factors for ABG fitting. Accuracy assessment carried out using coefficient of determination (R-2) and ratio of standard deviation to prediction error (RPD) showed satisfactory results, with 0.74 and 1.95 for AGB, 0.85 and 2.59 for BLG, and 0.61 and 1.6 for SOC respectively. Application of the best fitted models with spatial explanatory factors allowed to map and estimate C contained within each pool : 32 +/- 13 Gg C for ABG, 67 +/- 15 Gg C for BLG and, 139 +/- 36 Gg C for SOC (1 Gg=10(9) g). A total of 238 +/- 40 Gg C was obtained for the entire study area by combining the three C maps. Despite their relatively low predictive quality, models and C maps produced herein provided relevant reference values of C storage under plantation ecosystems in Madagascar. This study contributed to the reducing of uncertainty related to C monitoring and baseline definition in managed terrestrial ecosystem.
%$ 068