Estimation of shrub biomass: development and evaluation of allometric models leading to innovative teaching methods

Estimation of shrub biomass: development and evaluation of allometric models leading to innovative teaching methods

Accurate estimation of biomass is becoming vital for selling carbon into national and international markets. Being a dry continent, Australia’s natural forest has several shrub species. However, because of limited availability of methodology and difficulty in estimation they are unaccounted for in many cases. This paper has three objectives: (a) to address the major problem in multiple regressions, (b) to develop the best allometric equation for the biomass estimation of a popular shrub species, wild raspberry (Rubus probus) and (c) to prepare a teaching tool, by following systematic and logical steps, for biomass estimation using ForecastXTM software. We identified the possible explanatory variables, by discussing with experts and citing literature, for shrub biomass and then measured them by destructive sampling at Taabinga, near Kingaroy, Queensland. Our research suggests that careful analysis of correlation matrices gives very important clues to which variables we should select and which we should not for the models. High multicollinearity among the independent variables is a major problem in multiple regressions. This study shows that this problem could easily be solved by using basic scientific formula and applying a single variable instead of applying many highly correlated variables in the model. Unlike most statistical books, our analysis does not suggest to reject that variable from the model whose coefficient is not statistically significantly different from zero as it could be highly influential in another set of combination. Similarly, we recommend using the 'intercept' even if its value is not significantly different with zero as it does not cost extra money to be included but it does help the predictive power of the model. Although we developed a range of biomass prediction models (for wild raspberry) that can be used in different circumstances, our first recommendation is for the model which is based on girth and crown volume. Where cost is the major issues, we prefer the model which employs girth and crown area, as it gives a good result and needs only three variables to be measured. These findings can be helpful in teaching the practical applications of multiple regression in courses such as Data Analysis and Business Forecasting.

File reproduced in accordance with the copyright policy of the publisher/author.