Automatic change detection of forest cover change in Sri Lanka from 2008 to 2018 using satellite images

Abstract

According the statistics by United Nations FAO(Food and Agriculture Organization) in 2010, Sri Lanka there are 167,000 hectares as primary forest and 185,000 hectares forest as planted forest. But, unfortunately the amount of forest cover is decreasing and it stated that total forest cover reduction is 17.7% per year. Calculating the forest lost is still in the manual process in Sri Lanka and it was identified as time consuming and expensive task. Among the forest areas in Sri Lanka, Willpattu has largest coverage and it contributes to the rain and water and air purification. This study was conducted to investigate change detection of ―Wilpattu‖ forest cover by using the satellite remote sensing data from 2000 to 2019. Here only geo-referenced RGB images of corresponding satellite images were taken which are Landsat 7 and 8. RGB values and HUE values of those satellite images were used to identify the forest cover by using statistical model and neural network. In statistical model, samples of forest, water and land area were obtained using the human perception and then the calculated minimum and maximum values were used for the classification of images. In Neural network models, Green, Red and Hue pixel values of images were used as the input to neural network. K-means clustering was used to divide the clusters as forest, water and Land. The results show 55% accuracy when compared with the data of forest department data. HUE values are used same particular images using mean and variance and results shows 65% accuracy. But it shows 352.59 km2 forest cover reduction in statistical model and 589.61 km2 in neural network model (RGB) and 555.45 km2 in HUE from 2000 to 2019. Although it shows the decrease pattern of forest cover, both classification methods misclassified other vegetation types as forest. This can conclude by using only Red, Green and Hue it is not enough to identify the accurate forest cover and further development of model is still required.