Comparison of data mining and allometric model in estimation of tree biomass.

BACKGROUND: The traditional method used to estimate tree biomass is allometry. In this method, models are tested and equations fitted by regression usually applying ordinary least squares, though other analogous methods are also used for this purpose. Due to the nature of tree biomass data, the assumptions of regression are not always accomplished, bringing uncertainties to the inferences. This article demonstrates that the Data Mining (DM) technique can be used as an alternative to traditional regression approach to estimate tree biomass in the Atlantic Forest, providing better results than allometry, and demonstrating simplicity, versatility and flexibility to apply to a wide range of conditions. RESULTS: Various DM approaches were examined regarding distance, number of neighbors and weighting, by using 180 trees coming from environmental restoration plantations in the Atlantic Forest biome. The best results were attained using the Chebishev distance, 1/d weighting and 5 neighbors. Increasing number of neighbors did not improve estimates. We also analyze the effect of the size of data set and number of variables in the results. The complete data set and the maximum number of predicting variables provided the best fitting. We compare DM to Schumacher-Hall model and the results showed a gain of up to 16.5% in reduction of the standard error of estimate. CONCLUSION: It was concluded that Data Mining can provide accurate estimates of tree biomass and can be successfully used for this purpose in environmental restoration plantations in the Atlantic Forest. This technique provides lower standard error of estimate than the Schumacher-Hall model and has the advantage of not requiring some statistical assumptions as do the regression models. Flexibility, versatility and simplicity are attributes of DM that corroborates its great potential for similar applications.

On the use of data mining for estimating carbon storage in the trees.

Forests contribute to climate change mitigation by storing carbon in tree biomass. The amount of carbon stored in this carbon pool is estimated by using either allometric equations or biomass expansion factors. Both of the methods provide estimate of the carbon stock based on the biometric parameters of a model tree. This study calls attention to the potential advantages of the data mining technique known as instance-based classification, which is not used currently for this purpose. The analysis of the data on the carbon storage in 30 trees of Brazilian pine (Araucaria angustifolia) shows that the instance-based classification provides as relevant estimates as the conventional methods do. The coefficient of correlation between the estimated and measured values of carbon storage in tree biomass does not vary significantly with the choice of the method. The use of some other measures of method performance leads to the same result. In contrast to the convention methods the instance-based classification does not presume any specific form of the function relating carbon storage to the biometric parameters of the tree. Since the best form of such function is difficult to find, the instance-based classification could outperform the conventional methods in some cases, or simply get rid of the questions about the choice of the allometric equations.