Application of Machine Learning Methodology to Track History of Mangrove Forest Change in Macajalar Bay, Northern Mindanao (1950-2020)

Abstract

Mangrove forest in Macajalar Bay is regarded as an important ecosystem as it provides numerous ecosystem services. Despite their importance, deforestation has been rampant and has reached critical rates. Addressing this problem and further advancing conservation requires accurate mapping of mangroves, and understand the historical land cover changes. However, such information is sparse and insufficient to understand the change dynamics. In this study, mangrove cover change dynamics for Macajalar Bay, Philippines was determined using Landsat data and machine learning techniques. Vegetation maps derived from aerial photographs and satellite images were used to quantify mangroves and to monitor the rates of deforestation over a 70-year period. In 2020, the mangrove forest cover was estimated to be 187.67 ha, equivalent to only 58.00% of the 325.43 ha that was estimated in 1950. Original mangrove forest that existed in 1950 only represents 8.56% of the 2020 extent, suggesting that much of the old-growth mangrove have been cleared before 2000 and that contemporary mangrove extent is mainly composed of secondary forest. Highest deforestation rates occurred between 1950–1990 where it recorded a total of 258.51 ha, averaging a clearing rate of 6.46 ha/year. Clearing has been driven by large-scale aquaculture pond developments. Mangrove gains were evident in 2000 but it plateaued as it approaches 2020, while loss simultaneously increased since 2010. This indicates that mangroves gained since 2000 have experienced low survival rates. Promoting site-species matching, biophysical assessment, and verification of fishpond availability for mangrove rehabilitation programs are necessary undertakings to address such problems.