Estimation Of Forest Volume Based On Remote Sensing Information

Forest is an important material resource of national,the development of forestry economy is animportant part of national economic development.How to timely grasp the dynamic changes of forestresources information is one of the most important tasks of the forest resource monitoring.Forestvolume is the important part of forest resources survey, traditional forest resources survey need a longperiod, the task is heavy and labor-intensive, cost is relatively large. Based on Longquan which is one ofthe city of Zhejiang province as the study area, to explore how to quickly and accurately obtain thevolume of forest resources information. The government and the forestry related department could usethese dynamic changes information to specify and adjust forestry policy, compile forestry plan andevaluation its economic benefit. By this way the forest resources is fully utilized in the nationaleconomy, and constantly improve its potential productivity.The research chooses Longquan2007ETM+Remote Sensing image and forest management inventory as the foundation data,combined with BPneural network to establish model for estimating forest volume.Firstly,based on Longquan ETM+remote sensing images utilized ENVI5.0software to carry beltrepair, coordinate transformation, geometric correction, radiometric correction and image cropping, inorder to extract the interested research area of remote sensing image data.Secondly, using Zonal Statistics of ArcGIS9.3synthesize the pretreatment of remote sensing imageand the forest resources subcompartment data to make overlay analysis. Take the subcompartment as aunit extracted band, band ratio value and age as a variable factor, forest volume as the index. Used SPSSsoftware to analysis the extracted factors. Then, in Matlab R2011b, by polynomial fitting with empiricaldata obtained each variable factor membership.Last, after the above treatment, take variable factors as the input layer vector, the average unitstock volume as the output layer vector and divided the fir sample set into training data and predicteddata. In Matlab R2011b, based on improved BP neural network model to establishLevenberg-Marquardt optimization algorithm.The results show that: the computing community relative error is13.04%, the individual averagerelative error is18.9%, the prediction results can be achieved with high accuracy, close to the overalleffect of simulation model and the actual situation. In addition, owing to the low cost, large sample volume,the data of the present study, the model built in this paper has a high reference value in thevolume of forest resources.