Logistic model outperforms allometric regression to estimate biomass of xerophytic shrubs

Allometric model has been applied worldwide to estimate vegetation biomass for decades. However, this model fails to restrict the accelerating increase of biomass as body size grows. That contradicts to the size-related resource limits and intra-specific competitions. Thus, we tested logistic model with limiting factor of the carrying capacity at the 30-year dominant shrub species in Loess Plateau of China, including Caragana korshinskii (51 branches), Salix psammophila (44 branches) and Vitex negundo (28 branches). Our results indicated that logistic model was statistically effective as allometric model indicated by the adjusted code of determination, pvalue, Akaike's Information Criterion and Root Mean Square Error. It was also of more ecological significances by providing the equilibrium growth rate, equilibrium biomass (the asymptotic biomass as branch grew), and pointof-inflections (thresholds for different trends of biomass increase). That had been double-checked with our measured biomass and the published data in previous studies. The unrepresentative samples with the tendency favoring the small- and middle-sized branches, and the consequently biased tendency and overfitting in models from random quirks of samples, might partly explain the repeated validation of allometric model for estimating biomass in previous studies. In general, logistic model outperformed allometric model to estimate shrub biomass of allometric scaling with body size. An appropriate model for biomass estimation benefited to precisely compute carbon sequencing and to assess climate change impact on ecosystems functioning.

Automated summary

The study validated the effectiveness of using logistic model with limiting factor of carrying capacity for estimating biomass of dominant shrub species in Loess Plateau of China. This model not only showed statistical significance, but also provided more ecological meanings and insights into ecosystem functioning. Previous use of allometric model for biomass estimation might have biases and overfitting issues due to unrepresentative samples and random quirks of samples.