| Subcompartment is the basic unit in the forest resources investigation which participates in the calculation on many biological parameters like quantity, quality, distribution and dynamic growth, is an essential indicator for forest researcher. A workflow for Chinese continuous forest inventory (second class survey) has been well-developing for over 60 years. In second class survey, we focus on elaborate investigation on subcompartments and expect to extract surveying data with high quality. Therefore the research focuses on this unit are usually four folded.1st is on the field survey. Its key task is to measure biometric data with rapid, cost-effective and accurate approach; 2nd is on data mining from inventory data in order to improve allometric equation together with present data; 3rd is on experience extension into third class survey and expect empirical models in research can be helpful for peer researches.In this research, we presents a fusion approach with data mining. knowledge discovery, non-destructive measurement using electronic theodolite and 3D angle gauge to process subcompartments database and develop corresponding allometric model. Following are key contributions in this research:(1) Innovated 3D angle gauge technology in field survey ① We present a new gauge system to meet different measurement situation. This system is combined by 3D gauge, high precision gauge and traditional gauge for various measurement subjects. combining with the forest management table, we designed 3D gauges with Fg=1.0,3.0,5.0 and 6.0. Those 3D gauges were examined by in-stu data. It has an average relative error in 1.65%when investigated average tree height. It has an average relative error in 1.99% when investigated tree number in a hectare. It has an average relative error in 0.34% when investigated growing volume stock in a hectare. Comparing to traditional gauge, this new system significantly decreased errors and numbers of tree measurement to about 10.② We present a fusion approaches collaborating with 3D gauges which includes DTM data, high resolution remote sensing images and unmanned aerial vehicle (UAV). This fusion provides 3 perspectives from space- borne sensor, air- borne sensor to directly ground survey which has a comprehensive understanding to subjects in second class survey. At meanwhile it can be easily utilized in a larger scale beyond second class survey with full flexibility and low cost.(2)Data mining and knowledge discovery from forest inventory for developing allometric equation.①Two step clustering are used in the data mining procedure to process inventory data for Huangnihe district local forest administration which collected by the second class survey over years. Multiple factors were taken part into preprocessing to locate which are dominant factors. Then the filtered data were processed by two step clustering. After a comparing with traditional methods, we classified all subcompartments into 8 bins. This classification has advantage in statistic and practice than traditional one and can contribute to forest management.②For each subcompartment type in inventory data, complete factor regression model, the screening factor model and LM-BP neural network training model were used to process data and examined by independent samples. Results shows that a. complete regression model result of all types of subcompartment has an good fitting to in-stu data(R2>0.83) with a good test precision. All MSE absolute values are less than 0.1%, mostly MPE values are less than 5% (except the V class)b.Sub-Compartment type screening factor regression model determination coefficient also has an good fitting to in-stu data(R->0.79) with a good test precision. All MSE absolute values are less than 0.16%, mostly MPE values are less than 5% (except the V class).c. For the Sub-Compartment type fully factor LM - BP neural network model, all its MSE value are less than 1.7%, training precisions are more than 0.86 simulation precisions are more than 0.90.d. For the Sub-Compartment type screening factor LM - BP neural network model, all its MSE value are less than 2.4%. training precisions are more than 0.83 simulation precisions are more thanTo sum up, results shows that multiple factor precision is higher than the screening factor.Thus we can find out number of factors plays an important role to the regression precision for GSV at scale of subcompartment. Form the viewpoint from methology, LM - BP neural network is better than traditional allometric methods. However its. algorithm is under a black-box and not suitable for many cases.(3) long term growing prediction for subcompartmentFor a certainly forest, the period for forest continuous inventory is 10 years. How to supervise forest between this gaps is another focus in this research. We use Huangnihe’s inventory data from 1995 to setup basic models on DBH, height and other biological parameters. Richard equation is considered as a best model in this situation. yD-T=17.07(1-e-0.10A)3.67 yH-T=16.85(1-e-0.11A)4.23Its coefficient of determinations are R2=0.88 and R2=0.84 which represents a good fitting. Its coefficient of applicability analysis of decision are R2=0.76 and R2=0.84. Its SEE,MSE,MPE are all less than ±3%. Those indicates that this model is qualified for growing prediction.LM-BP neural network model can also be used in this part of research. Its H-T model has a correlation coefficient at 0.95. Its H-T model has a correlation coefficient at 0.94. Both of them shows good fitting with in-stu data. But its black-block character is still a drawback for correlational research.(4) Forest sub-compartment Data application:research of Felling timber species of forest sub-compartment rate calculation model. According to the requirement of forest felling operational inventory, combined with forest resources operational inventory, Established a nondestructive measurement of tree height diameter model by using the technology of arbitrary tree:Conifers (for example in Larch) model::d=0.7775d1.3 1.1415h-0.3437 (R=0.88)、d=0.5728d1.3 0.8874 H0.3930h-0.3468 (R=0.90); Broadleaf trees (Populus tomentosa as an example) model:d=1.04878d1.3 1.0315 h-0.2803 (R=0.91)、d=0.6985d1.3 0.9331 H0.2330 h-0.2821(R=0.91). Various species of trees model fitting is better. Calculation of sectional volume and total volume formula for forest felling operations, wood material ratio calculation has been done. The method for the nondestructive accurate determination of small timber produced rate provides methods and technical support.To sum up, subcompartment data acquisition, mining and prediction are used as the main line, from the survey equipment, method and technology and the analysis of the mining method and technology two aspects to improve the idea is feasible, and can obtain more wider forest resource status data, digging more conform to the objective reality and subjective demand information knowledge, and for the management of forest resources, forest resources information development provides new methods and new way of thinking, to improve the quality of forest resources survey number especially, improve the investigation system has practical significance. |
PDF Full Text Request