Why Spatial Resolution Matters: Comparing AI Approaches to Mangrove Mapping with UAS Data

How satellite-derived AI, UAS-derived AI, and manual interpretation compare to each other.

Mangrove ecosystems sit at the intersection of biodiversity conservation, disaster resilience, and climate action. They protect coastlines from storm surges, store significant amounts of blue carbon, and provide critical habitat for marine species. Yet despite their importance, accurately mapping mangroves remains a technical challenge.

Many mangrove forests consist of narrow shoreline corridors, fragmented patches, and isolated trees that are difficult to detect using medium-resolution satellite imagery. As artificial intelligence becomes increasingly common in geospatial analysis, an important question emerges:

How much does imagery resolution influence AI detection accuracy?

To explore this, Help.NGO compared three different mapping approaches across two mangrove sites in Catanduanes, Philippines:

• PhilSA's satellite-based AI detection

• Count from Above's drone-based AI detection

• Manual digitization from high-resolution orthomosaics

The results demonstrate that while all three approaches have value, spatial resolution remains one of the most significant factors affecting AI performance.

The Challenge of Mapping Mangroves

Unlike forests with relatively continuous canopies, mangroves form highly irregular coastal environments.

Typical characteristics include:

• narrow vegetation belts

• fragmented stands

• isolated tree clusters

• complex shoreline geometry

• gradual transitions between land and water

These characteristics make boundary detection particularly difficult when imagery resolution is limited. An AI model can only identify details that exist within the imagery itself. Even sophisticated algorithms cannot recover information that is absent from the source data.

Three Approaches, Three Different Objectives

Rather than asking which approach is "best," it is more useful to understand what each method is designed to accomplish.

1. Satellite AI Detection (PhilSA)

PhilSA combines Sentinel-2 optical imagery, ALOS PALSAR-2 SAR data, and AI classification to produce nationwide mangrove maps.

With a spatial resolution of approximately 10 meters per pixel, this approach provides outstanding regional coverage while maintaining a fully automated workflow.

For national inventories and large-scale ecosystem monitoring, this level of detail is often sufficient.

However, the limitations become apparent when mapping smaller coastal features. Because every pixel represents a 10 × 10 meter area, narrow mangrove belts and isolated patches frequently become generalized or disappear entirely.

The resulting polygons often follow the satellite pixel grid rather than the true canopy boundary.

2. Drone-Based AI Detection by Count from Above

The Count from Above workflow uses centimetre-level drone imagery together with AI segmentation models.

Increasing image resolution fundamentally changes what the AI can detect.

Instead of identifying broad vegetation blocks, the model can distinguish:

• individual shoreline structures

• narrow mangrove corridors

• fragmented vegetation

• isolated trees

• complex canopy edges

  
  

The comparison showed that drone-based AI produced boundaries that closely matched observed canopy edges while detecting many small mangrove features absent from the satellite-derived results.

The AI-assisted workflow also significantly reduced the amount of manual work required. In Borongan, analysts were able to review and refine Count from Above's AI-generated mangrove detections at approximately 24 hectares per hour, compared with approximately 4.5 hectares per hour for fully manual digitization in Catanduanes. This represents a more than fivefold increase in mapping speed. As a result, approximately 730 hectares in Borongan were processed and refined in around six days, while processing and manually digitizing 184 hectares in Catanduanes took approximately five days.

3. Manual Digitization

Manual interpretation by expert analysts remains the standard for detailed mapping.

Experienced analysts can identify subtle vegetation boundaries that automated workflows occasionally miss, making manual digitization valuable for quality assurance and AI training datasets.

Its drawback is scalability. Producing high-quality datasets manually requires significant expertise and time, making it impractical for operational mapping across large geographic areas.

A Practical Comparison

The study highlighted a clear relationship between image resolution and mapping performance.

For one representative mangrove study area (in Magnesia, Catanduanes), the mapped extents were:

Method	Mapped Area
PhilSA AI	11.63 ha
Count from Above AI	9.99 ha
Manual Digitization	9.72 ha

The drone AI result differed by only 2.78% from manual digitization, while the satellite-based approach mapped nearly 20% more area, reflecting generalized boundaries associated with lower-resolution imagery.

These differences illustrate how spatial resolution directly influences AI outputs.

Higher-resolution imagery enables more precise boundary delineation and improved detection of fragmented coastal ecosystems.

The Right Tool for the Right Scale

One important takeaway from this comparison is that these approaches are complementary rather than competitive.

Satellite AI remains the most efficient solution for:

• national inventories

• regional monitoring

• long-term environmental assessments

  
  

Drone AI becomes particularly valuable for:

• restoration planning

• conservation projects

• blue carbon assessments

• disaster risk reduction

• local ecosystem management

  
  

Manual interpretation continues to play an essential role in validating automated outputs and improving future AI models.

Beyond Mangrove Mapping

The findings extend beyond mangrove ecosystems.

Many humanitarian, environmental, and disaster management applications involve detecting small or fragmented landscape features, including damaged infrastructure, coastal erosion, landslides, informal settlements, and agricultural impacts.

In each case, automated tagging model performance depends not only on the model itself but also on the quality and resolution of the input imagery.

Selecting the appropriate data source is therefore just as important as selecting the AI algorithm.

Final Thoughts

Artificial intelligence is transforming environmental mapping, but higher model complexity does not automatically translate into higher accuracy.

As this comparison demonstrates, imagery resolution remains one of the strongest determinants of detection quality.

The results also demonstrate the potential of AI-assisted workflows such as Count from Above to significantly reduce the time required to turn high-resolution UAS imagery into operational environmental data.

Help.NGO continues to map mangroves around the Philippines together with partners like Rare.org Philippines, Visayas State University, Philippine Disaster Resilience Foundation (PDRF), and various Provincial, Municipal, and City Disaster Risk Reduction and Management Offices. Leveraging technologies from Amazon Web Services (AWS), DroneDeploy and OpenAerialMap.

Help.NGO publishes all its raw data into OpenAerialMap, for all to freely use.

Authors:

Rosalyn Mateo, UAS and GIS Specialist

Matthew Cua, Director of IT and Innovation

Agata Klat, Director of Communications

David Beatty, Project Coordinator Officer