Land Use/Land Cover Detection Based On Remote Sensing:A Case Study In Phonxay District,Lao PDR

Understanding the current status and dynamic changes of land use/land cover in a region is necessary for the formulation of regional development planning.Since remote sensing images can quickly obtain information from the Earth surface,it has become the primary technology for the regional land use/land cover detection.Meanwhile,improving accuracy and efficiency of land use/land cover detection from remote sensing images has attracted more and more attention in the field of remote sensing.In the northern part of Lao PDR,most of agricultural productions are still in a traditional style,referred as slash and burn agriculture,which consists of four procedures:deforestation,agricultural farming,land abandonment,and vegetation growth.This traditional agricultural style has been questioning its impaction on the environment and is considered as the primary reason leading to the deforestation in the tropics.In order to balance the agricultural production against the forest protection and provide decision-making basic for rational use of resources,it is necessary to accurately monitor the current situation and dynamic changes of land use in these areas.In this paper,Phonxay District,northern Lao PDR has been selected as the study area.Multi-temporal Landsat satellite remote sensing data were used for land use/land cover change detection and analysis,Main contents and significant contributions include:1)Due to the shortcomings of the existing software in remote sensing image classification,the Python programming language is used to develop a series of tools that can be run with the ArcGIS Desktop to improve the efficiency of land use/land cover remote sensing classification.This tool includes the rapid generation of precision analysis report,batch classification using different combinations of classification features,batch classification using different combinations of training samples,and decision tree generation based on the machine learning framework.2)After comparing and analyzing the land use/land cover results from three classifier(maximum likelihood method,support vector machine and decision tree)and the same classifier but different parameters,it is found that using the same classification features and training samples,the support vector machine and decision tree classifier has a higher accuracy than maximum likelihood method.In the maximum likelihood method,the classification features and training samples have certain influence on the classification result.The range of Kappa coefficients using different classification feature is 0.05,and the classification feature is not positively correlated with the classification accuracy.The training samples are randomly and proportionally divided into several training sample groups.The results show that the accuracy(Kappa coefficient)of classification results is different.In the support vector machine classification,the cost coefficients impact the accuracy of classification results.In the decision tree classification,based on the two different decision tree constructions(artificial way and machine learning method),the classification results are compared.The result shows that using machine learning method has a better accuracy than using artificial way.3)The land use/land cover results derived from remote sensing data in three periods(1992,2000 and 2016)in Phonxay District illustrate that there are few changes in the terms of the land use/land cover area.Each proportion of the forest land in the three years is 89.94%,89.78%and 90.58%respectively.The forest land has the highest proportion of area among all land use/land cover types,follow by the open forest and grassland,which has the area ratio of 6.14%,5.67%and 4.96%in three years.The proportion of agricultural land is 3%,while the residential and water areas occupy about 1.5%and has no significant temporal changes.However,the distribution of each land use/land cover type has an obvious shift.In these three years,the area retained forest land is about 77%of the total area of the study area,account for 85%of the forest area in 2016,which implies that 15%of the forest land is not the original forest.Because of the long time interval between of the two phases(2000 and 2016),the actual proportion of forest land converted from non-forest land should be higher.The proportion of area keeps to be agricultural land is almost 0%in these three years,which means that all agricultural lands are converted from other land use/land cover type.