Merremia Vanuatu Dashboard

1 Satellite Image Acquisition

High-resolution multispectral imagery is sourced from the European Space Agency's Sentinel-2 constellation (10 m spatial resolution, 5-day revisit cycle). Additional very-high-resolution validation imagery (sub-metre) is obtained from Planet Labs SkySat tasking when available. Cloud-free composites are generated for each quarterly survey window using median pixel compositing across available scenes.

• Sentinel-2 Level-2A (surface reflectance) bands: B2, B3, B4, B5, B6, B7, B8, B8A, B11, B12
• Cloud masking via ESA Scene Classification Layer (SCL) and Fmask algorithm
• Temporal composites generated per quarter (Jan–Mar, Apr–Jun, Jul–Sep, Oct–Dec)

2 Spectral Index Computation

Multiple vegetation and canopy indices are computed from the multispectral composites to enhance Merremia detection. The vine's rapid canopy-smothering growth habit produces a distinct spectral signature — high NDVI combined with low canopy texture and reduced red-edge variability compared to native forest.

• NDVI (Normalized Difference Vegetation Index) — overall greenness
• EVI (Enhanced Vegetation Index) — canopy density in high-biomass areas
• NDMI (Normalized Difference Moisture Index) — leaf water content differences
• Red-Edge Position (REP) — distinguishes vine canopy from native broadleaf forest
• GLCM Texture Metrics — homogeneity and contrast computed from NIR band to detect the smooth, uniform canopy blanket characteristic of Merremia overgrowth

3 Machine Learning Classification

A Random Forest classifier (500 trees, max depth 20) is trained on labelled ground-truth polygons to distinguish Merremia-affected canopy from native vegetation, agricultural land, bare soil, and water. The model uses all spectral bands plus derived indices and texture features as input variables.

• Training/validation split: 70/30 stratified by island and species
• Feature importance ranking used to prune redundant inputs
• Cross-validation (5-fold) overall accuracy: 89.3% (M. peltata: 93.1%)
• Post-classification morphological filtering to remove isolated pixels (<0.5 ha)

4 Field Validation & Ground-Truthing

Quarterly field surveys are conducted by DEPC rangers and trained community observers across 34 monitoring sites on 6 islands. GPS-tagged photographs and species identification are recorded using the KoBoToolbox mobile platform, then uploaded for comparison against classification outputs.

• Stratified random sampling: 15 points per threat zone (High, Moderate, Low)
• Species-level identification confirmed by University of the South Pacific (USP) botanists
• Producer's accuracy (commission error): 87.6%
• User's accuracy (omission error): 91.2%

5 Change Detection & Trend Analysis

Quarterly classification maps are compared using post-classification change detection. Net area change, expansion direction, and growth rate are computed per island. A linear regression model is fitted to the time-series data to project future spread and estimate time-to-threshold for management triggers.

• Change maps produced for each consecutive quarter pair
• Expansion vectors computed using centroid shift analysis
• Growth rate modelled via ordinary least squares regression on cumulative area
• Threshold alerts set at 3,000 ha national total

6 Dashboard Reporting

Classification outputs, field validation data, and trend analyses are aggregated and pushed to this web dashboard. A background Web Worker runs automated risk scoring, projection modelling, and priority ranking each time the page loads. PDF reports can be generated on-demand for offline use by DEPC officers and NBSAP reporting.

Satellite Imagery

Sentinel-2 and Landsat are open-access, peer-reviewed, and operationally validated datasets maintained by government space agencies. Planet SkySat is a commercial source used only for targeted validation.

Training & Ground-Truth Data

Ancillary & Reference Data

Credibility Assessment Notes

• High — Peer-reviewed, operationally validated, or produced by authoritative government / academic institutions with established QA/QC procedures.
• Medium — Citizen science or crowd-sourced data with some quality controls (e.g., iNaturalist research grade, KoBoToolbox with photo verification), but may contain spatial inaccuracies or identification errors. Used to supplement, not replace, primary training data.
• All training data undergoes manual review by Vanuatu Spatial Solutions analysts and USP botanists before inclusion in the classification model.
• Model accuracy metrics are recalculated each quarter as new ground-truth data is added.

Key Objectives

Intended Users

• DEPC officers and rangers — for operational planning, resource allocation, and field survey coordination
• NBSAP focal point and reporting team — for national and CBD reporting on invasive species management progress
• Provincial government environment units — for island-level situational awareness
• Conservation NGOs and donors — for project monitoring and impact assessment
• Research institutions (USP, CSIRO, SPC) — for scientific analysis and collaborative research
• Community leaders and landowners — for awareness of Merremia impacts on their land and forests

Prototype Disclaimer

This is a prototype dashboard intended to demonstrate the feasibility of satellite-based Merremia monitoring for Vanuatu. Data presented is based on the best available sources as of the last survey cycle, but should be validated with on-ground surveys before making management decisions. The classification model is under active development and accuracy may vary by island, season, and cloud cover conditions. Vanuatu Spatial Solutions welcomes feedback from all users to improve the system.

Live Dashboard

View all field-collected records on an interactive map with auto-refreshing data. The live dashboard updates every 5 seconds and shows records as they are synced from the field.

Field Collector App

The Merremia Field Collector is an offline-first Progressive Web App (PWA) for rangers to record Merremia sightings in the field. It captures GPS location, species, threat level, photos, and more — then syncs to GitHub when connected.

How It Works

• Collect — Rangers use the Field Collector PWA to record Merremia observations with GPS, photos, and threat assessments
• Sync — Data syncs automatically to the merremia-field-data repository on GitHub
• Visualize — The Live Dashboard polls GitHub every 5 seconds and displays all records on a map with stats
• Manage — Records can be deleted from the Field Collector with password protection