A critical review of forest biomass estimation models, common mistakes and corrective measures

Abstract The choice of biomass estimation models (BEMs) is one of the most important sources of uncertainty in quantifying forest biomass and carbon fluxes. This review was motivated by many mistakes and pitfalls I encountered in the recent literature regarding BEMs. The most common mistakes were the arbitrary choice of analytical methods, model dredging and inadequate model diagnosis, ignoring collinearity, uncritical use of model selection criteria and uninformative reporting of results. Sometimes, errors in parameter estimates were not checked and model uncertainty was ignored when interpreting and reporting results. Consequently, biologically implausible and statistically dubious equations such as ln ( M ) =  ln ( a ) +  b ( lnD ) +  c ( lnD ) 2  +  d ( lnD ) 3  +  e ( lnρ ) have been published as allometric models. These are perpetuated in the literature, databases and field manuals and will pose a serious threat to the integrity of future forest biomass estimates. Through worked examples, I also illustrate that (1) allometric coefficients can be biased by the choice of analytical procedures and methodological artefacts; (2) collinearity of predictors can result in coefficients with unacceptable levels of error; (3) the R 2 and Akaike information criterion (AIC) have been misused and have resulted in the selection of implausible BEMs; and (4) differences in the definition of model “bias” has sometimes led to contradictory reports. I propose corrective measures for most of these problems and provide suggestions for prospective authors on how to avoid pitfalls in interpretation and reporting of results.