Machine Learning: How Can It Save Mangrove Forests?

Mangrove forests are an essential component of the coastal zones in tropical and subtropical areas, providing a wide range of goods and ecosystem services that play a vital role in ecology.

They are also threatened, disappearing, and degraded across the globe. One way to stimulate effective mangrove conservation and encourage policies for their protection is to carefully assess mangrove habitats and how they change, and identify fragmented areas.

But obtaining this kind of information is not always an easy task. In this essay, I will explore how deep learning, a subset of machine learning that employs artificial neural networks with multiple layers to model complex patterns in data1, can help us save mangrove forests by improving their mapping and monitoring.

Mapping mangrove ecosystems with deep learning

(Photo : BAGUS SARAGIH/AFP via Getty Images)

Since mangrove forests are located in tidal zones and marshy areas, they are hardly accessible23. Therefore, remote sensing techniques that use satellite images to capture information about the earth's surface are often used to map these fragile ecosystems, as per Phys.org.

However, different types of satellite images have different spatial resolutions, spectral bands, and temporal frequencies, which can affect the accuracy and reliability of the land use/land cover classification.

Moreover, different classification techniques have different strengths and limitations in dealing with complex and heterogeneous landscapes such as mangroves.

Deep learning can offer a solution to these challenges by using the object-oriented classification of fused Sentinel images.

Sentinel images are freely available multispectral remote sensing data that have a high spatial resolution (10 m) and temporal frequency (5 days).

Fusing these images means combining them with other types of images, such as panchromatic or radar images, to enhance their quality and information content.

Object-oriented classification means grouping pixels into meaningful objects based on their spectral, spatial, and contextual features, rather than classifying each pixel individually.

A recent study by Dr. Neda Bihamta Toosi, a postdoc at Isfahan University of Technology in Iran in the journal Nature Conservation2 compared the performance of different combinations of satellite images and classification techniques for mapping mangrove ecosystems on the northern coast of Qeshm island, Iran.

They found that object-oriented classification of fused Sentinel images can significantly improve the accuracy of mangrove land use/land cover classification compared to pixel-based classification or using Sentinel images alone.

They also found that among different machine learning algorithms, such as maximum likelihood classification, support vector machine, random forest, and decision tree, random forest was the best one for classifying mangrove ecosystems.

Also Read: Mangrove Loss Degradation Can Be Attributed To Socioeconomic and Biophysical Factors

Monitoring mangrove ecosystems with deep learning

Mapping mangrove ecosystems is not enough to ensure their conservation. It is also important to monitor their condition and how they change over time due to natural or human-induced factors.

This can help identify areas that are degraded or fragmented, and evaluate the effectiveness of management and restoration actions, as per Newswise.

Deep learning can also help with this task by using model-based landscape metrics and principal component analysis techniques.

Landscape metrics are quantitative measures that describe the spatial structure and composition of a landscape.

A principal component analysis is a statistical technique that reduces the dimensionality of a data set by extracting the most important features that explain its variance.

Bihamta Toosi used these techniques to assess the spatial disturbance of mangrove ecosystems in Qeshm island based on the land use/land cover maps derived from the object-oriented classification of fused Sentinel images.

They calculated several landscape metrics for each land cover class, such as patch density, edge density, mean patch size, shape index, aggregation index, and fragmentation index.

They then applied principal component analysis to reduce these metrics into two components that explained 95% of the variance in the data set.

They used these components to generate disturbance maps that showed the degree of disturbance for each pixel in the study area.

The results showed that most of the mangrove forests in Qeshm Island were moderately disturbed (40%), followed by low disturbance (30%), high disturbance (20%), and very high disturbance (10%).

The disturbance maps also revealed the spatial patterns of disturbance across the island, showing that the eastern part was more disturbed than the western part due to human activities such as urbanization, agriculture, aquaculture, and tourism.