Investigation of forest area using support vector machine and provide a model for predicting the level of changes

Background and Objective:

 Due to the increasing degradation at the level of the natural ecosystem, the amount and location of land use changes and predicting its future growth trend, I can provide the information I need to planners and managers. In this study, in order to change the current changes and predict the future in the Siahkal range, forecasting and changing the nose were done with Landsat images. There are various methods for predicting land use change. Processes for predicting and modelling land use change, such as urban growth and development, deforestation, etc., are considered powerful tools in managing natural resources and changing the state of the environment. This change reflects how humans interact with their environment, and its modelling has had an impact on settlement and macro-planning. In this research, due to the high capabilities of remote sensing and modelling tools and predicting changes in change using automatic-Markov cells in forests in northern Iran.

Materials and Methods:

 In this research, Landsat 5 images, 2000 TM sensor, Landsat 7 ETM+ sensor 2010 and Landsat 8 OLI sensor 2018 are used. In the preprocessing stage, errors in raw data such as radiometric, atmospheric, geometric, etc. errors are corrected. Was significant but had a radiometric error. 84 points are used for forest use, 76 points for thin forest water, 31 points for consumption and 2 required sensitivities to indicate a specific level of land cover. Land cover is defined into five classes: dense forest, semi-dense forest, sparse forest, urban area and agricultural area. The ENVI Remote sensing Software defines four types of kernels for the support vector machine in the SVM classification section: Polynomial, Sigmoid torsion, and FBCTION (RBF). According to the best kernel studies for land use classification, the radial kernel (RBF) has been proposed. In the present study, this kernel was used for classification. The classification of the appropriate band composition that you want to separate these classes for visual interpretation was selected by the spectral mean plot. This is done by the complex OIF index. After the extraction of land uses by the method, the results were evaluated accurately. Maps are prepared by land use, then with the GPS position of the earth, the map of the situation in the visible area and using the formed error matrix of kappa weakness and its overall accuracy obtained for this work, 200 points are randomly created on the images. The use of these points was determined by field visits and topographic maps of the surveying organization. Land use classification models are prepared, for modelling and land use changes are entered into office software to design land use changes in the required years. Degree of land use change modelling The LCM model was used in the Idrisi software environment. The Markov-CA model is a combination of automated cells, Markov chains, and multi-purpose land allocation. The Markov model also shows each user by generating a set of status probability images from the transfer probability matrix. In the last step of the structural model, using the transfer area matrix in the CA Markov model, a simulated simulation of future land use can be obtained. In this research, the land use map of 2010 and 2018 was used to predict the 2028 map. And in order to accurately review the forecast by CA Markov using the user map for 2000 and 2010, the map for 2018 has been predicted and increased by the map obtained from the classified level for this year.

Results and Discussion:

 The classification accuracy test was obtained using the Kappa coefficient index and overall accuracy. Kappa coefficient and overall accuracy were 0.88 and 0.89 for the image of 2000, 0.91 and 0.92 for the image of 2010, and 0.93 and 0.95 for the image of 2018, respectively. The images are categorized as entered into the software and processed by changing the LCM. Changes in the LCM model showed that during the years 2000 to 2018, more changes were related to the conversion of semi-dense forest land with an area of 42104.27 hectares. Urban land use change has also increased in the years of many studies and amounted to 148.14 hectares. The table of the probability of land use changes in the Markov production model and with the production map at this stage, for the years of Markov forecast studies for 2018 and 2028 showed that in 2028 the urban class area increased to 21293.1 hectares and the valuable land use area of dense forest to 2189.97 hectares will be reduced.

Conclusion:

 In order to prevent the uncontrolled expansion of cities, residential areas and the destruction of forest areas and vegetation, management measures should be taken and management decisions should be made. The level of dense and semi-dense forests in areas with high slopes will decrease further by 2028. Urban land use changes have also increased in the study years and amounted to 148.14 hectares. The results of surveying the area of forecasting classes showed that in 2028, the area of urban classrooms will increase to 21293.1 hectares and the valuable land use area of dense forests will decrease to 2189.97. The ability of the vector machine model in determining land cover/land use, vegetation and forest cover in different regions of Iran has been proven by other researchers. Remote sensing tools can be an important arm in information production in natural resource management.