Carbon stock in the Atlantic Forest: uncertainties behind allometric equations

Background/Question/Methods

Reliable data in forest carbon stock is crucial for disentangling the processes behind land use impacts on climate change. The Brazilian Atlantic Forest (BAF) covers 15 million hectares, ranging 29° of latitude, but only two allometric equations are available in the literature. These models were done with restricted sample sizes due to the difficulties in obtaining this type of data, and can be highly biased. Here, we want to know how accurate are local and pan-tropical models in estimating BAF aboveground biomass (AGB) and how large are biases in different forest successional stages. We established 38 plots (0.1 ha each) in an area of ~25km². Forest age ranged from 0 to 30 years, but 3 plots were estimated to be older than 80 years. All stems (height>1.5m) were measured. In two plots (~30 years), 66 trees were cut down. We then oven-dried leaves, branches and wood to determine total dry weight. First, we calculated the mean absolute error (ME_trees, %) relative to differences between modeled (5 published equations) and measured AGB (cut down trees). Finally, we compared estimation biases over successional stages using predictions generated from the best performing model in the previous step with the other equations.

Results/Conclusions

The local and pan-tropical models showed different accuracy in theirs predictions of our study-site AGB. Relative errors for individual tree biomass estimations ranged from 0 to 413%. The equation Scatena_1993 (Biotropica,v.25) most accurately predicted AGB (ME_trees=25%). The two local models, Burger_2008 (Biota Neotrop,v.8) and Tiepolo_2002 (Taiwan Forestry Research Institute, v.153), showed weak performances estimating AGB (ME_trees=35% and ME_trees=74%, respectively). Two pan-tropical equations, Brown_1989 (Forest Science, v.35) and Chave_2005 (Oecologia,v.145), showed a lower error in their predictions than the local ones. Because Scatena_1993 better estimates AGB for large and small trees, we selected this equation as reference to evaluate AGB biases in different successional stages. The plot AGB ranged between 24 and 282 Mgha^-1. The two local equations showed contrasting performances in AGB estimation at plot level. For example, Burger_2008 increased and Tiepolo_2002 decreased prediction errors as plot AGB increased. Brown_1989 showed constant errors among successional stages and Chaves_2005 showed increased estimation error as the plot AGB increased. The choice of models can change AGB at individual tree or plot level because the predictions differ strongly. Our results show the challenges in making accurate biomass and carbon predictions for BAF and indicate the need for future large-scale validation of these equations.