| Under the background of the "Kyoto protocol", evaluate the forest carbon has become a hot research spot. A large number of studies have shown that forest carbon storage in tree layer definitely plays a leading role in aboveground carbon storage, and forest biomass is the prerequisite for further analysis of forest carbon storage. In the current researches, studies of regional (mesoscale) biomass are relatively few, but, regional biomass estimation is the bridge between large scale and small scale biomass estimation, therefore, carry out the quantitative and qualitative research on regional forest biomass in tree layer is of great significance for forest production and management based on "carbon sink" and "carbon sink" trade.However, using the current biomass estimation methods to estimate the regional forest biomass in tree layer has serverl problems as follows:(1) the regional area always has varies species or forest types, using traditional methods to estimate regional biomass needs to establish varies of biomass models based on species or forest types, it has large workload. As forest resources survey data growing increasingly mature and rich, taking full advantange of data of the study ares, it will have better guidance on the macro level to ignore the micro level accuracy appropriately in the context of the big data era. So, it needs to research a method for estimating stand biomass rapidly.(2) When establishing stand biomass model, the current researches do not sufficiently study the biomass-related impact factors, and the established models lack factor information, models built by different scholars have large difference.(3) the current research mainly adopts the method of multivariate regression analysis in modeling, but the method has some flaws when independent variables exitst the multicollinearity problem. There are few studies using neural network to establish model, but these models have some limitations on the generalization ability.(4) there does not have the research work to evaluate forest biomass level, so researching a method to quickly evaluate regional forest biomass could provide decision support for guiding forest production and management with the goal of carbon sink macroscopically.Aiming at the above problems, this paper taking the Mengjiagang Forest as study area, taking Forest Resource Inventory data of Mengjiagang at 2012 as the data source, starts the research of estimating and evaluating regional forest biomass in tree layer which is supported by the "five" rural areas state science and technology plan project "The Sustainable Production and Management Decision Support System of Forest"(2012AA102003-2). The sub-compartment in the study area has been divided into six categories and six kinds of biomass estimation models have been established. On this basis, there establishes forest biomass grade evaluation decision tree model through evaluating the level of sub-compartment biomass at the study area. This study mainly focuses on the extraction of biomass characteristic factor, the cluster analysis of sub-compartment in the study area, the modeling of biomass estimation model and grade evaluation decision tree. The main conclusions are as follows:(1) Appling principal component analysis method to extract eight characteristic factor of forest sub-compartment biomass, the results show that eight main ingredients could express more than 80% information of the original survey data, each factor is independent, the initial factor loading matrix of eight principal components has been rotated, each principal component portrays the obvious meaning and has strong explanatory indicates, which proves the feasibility of extracting biomass characteristic factors. Compared with the parameters of traditional biomass model, the characteristic factor obtained in this paper is independent, and adequately represents the effects of forest biomass-related factor information.(2) Proposing the method building biomass estimation model by distinguishing sub-compartment cluster, using the improved K-means algorithm to analyze forest sub-compartment in the study area, the results showing that the forest sub-compartment in the study area can be divided into six categories, the internal characteristics of each category is obviously different with others and described qualitatively, the qualitative and quantitative analysis of the various types of sub-compartment is finished by combining with BWP cluster validity index. This method is different from conventional means, building biomass estimation model by distinguishing species or forest types has obvious advantages at regional biomass estimate.(3) Building forest biomass estimation model based on sub-compartment cluster, the experiment results show that:by the determination coefficient and the absolute value of the average relative error, six categories of sub-compartment biomass models’ accuracy take support vector regression model as the best. Various models’ determination coefficient range from 0.7 to 0.92 and the absolute value of the average relative error ranges from 11.173% to 23.583%. The next is neural network model, and multiple linear regression models are worst. According to above results, it indicates that there exists a nonlinear relationship between biomass and relative factors, and support vector regression model can fit the relationship well. In addition, the accuracy of support vector regression model based on all sub-compartment data is lower than six categories model. At the same time, the result illustrates that it is feasible to build biomass estimation model based on sub-compartment cluster.(4) Proposing a forest biomass grade evaluation system and evaluation model, using quartile method to divide regional forest biomass into four grades, they are respectively the higher level, high level, medium level and low level, using decision tree method to build two kinds of biomass grade evaluation decision trees based on the total and partial characteristics. The results show that:the accuracy of biomass grade evaluation decision tree based on total characteristics is higher, and the classification accuracy is near 91.03%. It also gets 28 classification rules whose confidence is more than 75%, and provides a new approach for the rapid evaluation of regional biomass.Using biomass grade evaluation tree to estimate every sub-compartment biomass of the study area indicates that the sub-compartment in moderate biomass level has the largest number, and the next is at a higher biomass level. The number of the sub-compartment in high biomass level equals with the low biomass level. Consequently, it provides scientific guidance for forest management with the goal of carbon sink. |
PDF Full Text Request