Advancing the Spatiotemporal Assessment of Mangrove Ecosystem using
Machine Learning Approaches -- Case Study of a Coastal Megacity, Mumbai,
India
Abstract
Mangrove ecosystems are an essential component of tropical and
subtropical urban coastal regions where they provide critical ecosystem
services and ensures climate mitigation while playing a pivotal role in
the livelihoods of coastal communities. However, growing anthropogenic
pressures from rampant urbanization and infrastructural demands are
leading to an unparalleled loss and degradation of mangrove cover
especially in coastal cities of the global south. Addressing the
immediate need for monitoring, protection and restoration of the
ecologically stressed mangroves, this study uses earth observations,
machine learning and cloud computing methods for advancing timely and
accurate spatiotemporal mangrove mapping and change detection. Image
classification through four different models i.e. Classification and
Regression Trees (CART), Random Forest (RF), Gradient Tree Boost (GTB)
and Support Vector Machine (SVM) was performed using Google Earth Engine
to classify mangrove extent along the coastal regions of Mumbai, India.
Spatially explicit temporal trend in mangrove extent was studied and
used to estimate the rate of change of mangrove extent over a period of
30 years. Accuracy assessment was conducted to validate the robustness
of trained classifier models alongside their comparative performance.
Classification accuracies on the order of 95% were achieved through the
machine learning-based classifier models in distinguishing mangrove
areas from other land cover types. The time-series analysis combined
with image classification reveals the pattern and causes of
spatiotemporal changes in mangrove cover and highlights the hotspots of
mangrove loss and gain. This approach can aid stakeholders in the
management and restoration of mangrove ecosystems through periodic and
cost-effective monitoring of mangrove cover particularly in data
deficient coastal cities. The outcomes of the study will contribute
towards efficient decision-making in achieving the localization of
Sustainable Development Goals 6 and 11 of the United Nations.