Monitoring Urban Land Cover Changes By UsingMachine Learning Methods on Medium ResolutionSatellite Imagery

Abstract

The variation in land use and land cover allows us to track the growth and change indensely populated areas of Pakistan specifically Islamabad. The results in this studyexamine land cover changes and show how multi-temporal Landsat data can be usedto analyze land cover changes over time. This data can be utilized as a source ofinformation for land-use planning and policy-making decisions including urbanization,water quality management, forest management, and other environmental institutions.In this research, we perform temporal analysis for the region of Islamabad, to studythe change in the last 20 years by using medium-resolution satellite imagery.To achieve our objective, we use supervised machine learning approaches and the algo-rithms we incorporate in this study are Random Forest(RF)and Maximum LikelihoodEstimation(MLE). The accuracy assessment is achieved through stratified randomsampling and an error matrix is used to evaluate classification accuracy. The accuracyof the classifier lies between 79% - 86% and the kappa coefficient is between 0.78-0.85.The overall accuracy of the supervised machine learning algorithm i.e. Random Forestis much higher than the Maximum Likelihood Estimation, and the pipeline of GoogleEarth Engine will be very effective and will reduce the total time of data collection andpre-processing by approximately half as compared to other software-based approacheslike ArcGIS.This research indicates that the net loss in the green areas is 236.74 km2, meanwhile,the buildup area is increased by approximately 314 km2, barren land reduces by 83km2and the water bodies are reduced by 0.71 km2. This study gives us insight intohow land use and land cover have changed over our study period (2000-2020).