Total population reports are ill-suited for global biomass estimation of wild animals

Greenspoon et al. (1) used of global population estimates of 392 mammal species to predict the global biomass of mammals. We caution against important limitations in their approach, which likely results in gross underestimations of biomass and its uncertainty. The authors derive >97% of their estimates from the IUCN Red List (RL) database, which is particularly ill-suited for this scope since it is compiled using different standards than scientific ecological investigation and based on inconsistent approaches that are influenced by precautionary to evidentiary attitudes of different RL assessors (2). Because reliable population estimates are available only for relatively small areas (3), RL figures are generally obtained by summing up local estimates based on relatively scarce and biased data, including expert-based guesses with little supporting evidence (Table 1). Depending on the original purpose, RL figures might be over- or under-inflated (3, 4). Some RL’s reported population sizes are derived by applying an average density across an area, which may not align with the area used in Greenspoon et al. to estimate density, hence resulting in biased and unrealistic estimates when compared with field density estimates (Table 1). The RL dataset includes an overrepresentation of threatened species (60 vs. 27% of all mammals). The authors control for this bias by including RL categories as models’ predictors. However, no RL criteria relate to density, with ~80% of mammal species threatened due to small or decreasing ranges. Finally, model extrapolations are based on data heavily biased toward ungulates (40 vs. 4.7% of all mammals) while not controlling for phylogenetic relatedness but reassigning species to other taxonomic order for model predictions. Consequently, predicted density estimates by Greenspoon et al. (1) deviate substantially from published estimates obtained in hundreds of field studies (6) (median absolute orders of magnitude deviation = 0.64, 95% = 0.06 to 2.5, N = 159, Fig. 1A) and are on average underestimated by threefolds (median orders of magnitude deviation = −0.47). This may have profound effects on global estimates (Fig. 1B), leading to a potential underestimation of biomass of ~5.5-fold difference (9.4 vs. 52 Mt for 159 species). Furthermore, we could not replicate the results of 7 of the top 10 marine mammals’ biomass due to inconsistent use of RL data (Table 1). The uncertainty computed by Greenspoon et al. (1) is based on several expert-based intervals and applied to all species, including marine mammals for which estimates of variability are available from literature and the RL but were not used (8). This results in a gross underestimation of the uncertainty around the global biomass prediction (Table 1). While estimating biomass globally provides important insights, we call for a more careful consideration of data quality and consistency and a more robust reporting of uncertainty. We recommend fitting models on empirically derived field density estimates while controlling for phylogeny, environmental variables, and inconsistent sampling methods (e.g. refs. 6 and 9). Uncertainty should be derived directly from the statistical predictive error and, when possible, also from the underlying data. Alternatively, mechanistic eco-physiological models can be used to estimate global biomass following trait-based theory and validated with independent data (9).