Classification and Monitoring of National Forests of Pakistan using Satellite Imagery and Machine Learning

Abstract

For the ecology and species, deforestation has a wide range of serious effects. Climate change, biodiversity loss, soil erosion, floods, landslides, and other issues may have direct or indirect consequences. The sustainability of agricultural production systems is threatened by forest area loss, endangering the nation’s economy. Large swaths of forested and arable land deteriorate annually and eventually become wasteland owing to natural processes or human interference. It is essential to have up-to-date information that accurately describes the kind and scope of land resources and their evolution. With respect to population, Pakistan is the fifth-largest nation, and it is also the region that is most prone to climate change due to loss of forests. As a result of increased deforestation, it was deemed important to conduct land use studies that concentrated on employing satellite remote sensing (SRS) and Geographic Information System (GIS) technology for timely and efficient solution. This proposed pipeline led to generation of maps showing alterations and changes of woodland in the past and present along with changing pattern of forests and rangeland areas. In this research, imagery of Landsat missions is used for data acquisition based on Enhanced Vegetation Index (EVI) and classification of major forests of Pakistan using Google Earth Engine (GEE) is carried out. Historical data of past 20 years is developed and analyzed using calibrated Top of Atmosphere (TOA) reflectance dataset from Landsat satellite 7, 8 and 9. Analyzing the results shows that 9000 hectares of forested area in Kala Chitta National Park for past two decades was decreased whereas no large difference was observed for Ayub National Park. For Margalla Hills National Park, around 250 sq. Km and 5 sq.km in Pir Lasura National Park of greenland was converted from greenland to build-up area. However, opposite trends were observed for Lal Sohanra National Park where improvement of

v

250 sq. km in forest and shrub land was observed in the region for past twenty years. The findings of the study indicate accuracy of more than 90% for all observed national parks, which is significantly greater than the accuracy levels found in previous studies in similar domain. The highest accuracy was observed for Ayub National Park i.e.; 90% because it spans a smaller region. On the other hand, lowest average accuracy of 92% was observed for Pir Lasura National Park due to the hilly nature of this forest. Moreover, 95%, 96% and 94% accuracies were observed for Margalla Hills, Kala Chitta and Lal Sohanra national parks respectively. The research demonstrates that, for analysis in every field of study, having access to globally accessible dataset for any region is fundamental. The developed dataset is beneficial for understanding the dynamics of forests which can yield knowledge to create forest policy. Moreover, identification of deforestation hotspots at the provincial level offers crucial information into patterns of forests degradation, which aid in the advancement of effective national forest conservation and management initiatives.