Bluebell Index completes independent validation of Dynex QCHP canopy-height prediction model

Bluebell Index has completed an independent assessment of the Dynex QCHP (Quantum Canopy Height Prediction) model, evaluating its performance across five geographic validation sites and comparing its outputs with historical canopy-height information derived from the ETH Global Canopy Height 2020 dataset through Google Earth Engine.

The assessment examined predicted mean canopy height, spatial prediction coverage, historical reference support and consistency with recent land-cover conditions.

Unlike conventional canopy-height products that represent vegetation conditions at a fixed historical point in time, Dynex QCHP is designed as a forward-looking prediction system, incorporating temporal information to estimate future changes in forest canopy structure.

Strong operational performance

Across the five-site assessment, Dynex QCHP demonstrated significant operational advantages compared with conventional canopy-height assessment workflows.

According to the validation study, the solution delivered the assessment:

• approximately 5× faster than the equivalent classical workflow;
• at approximately 90% lower delivery cost per unit area;
• completing the five-site assessment in approximately three months;
• with a mean prediction coverage of 85.62% across the validation programme.

Prediction coverage exceeded 99% at three of the five validation sites, demonstrating the ability of the system to generate spatially extensive canopy-height assessments while explicitly reporting where sufficient prediction confidence is available.

Comparing canopy-height estimates

The validation compared Dynex QCHP predictions with historical canopy-height statistics derived from the ETH Global Canopy Height 2020 reference dataset.

Two sites showed particularly close site-level mean agreement:

Site 3

Historical reference: 10.69 m
Dynex QCHP:           12.92 m
Difference:           +2.23 m
Coverage:              99.97%

Site 5

Historical reference: 9.05 m
Dynex QCHP:            6.74 m
Difference:            -2.31 m
Coverage:              99.69%

These sites combined high spatial prediction coverage with comparatively small differences in mean canopy height.

Detecting change rather than reproducing a historical map

One important aspect of the study was the ability to distinguish between differences caused by model behaviour and differences that may reflect actual changes in land cover.

At Site 4, the historical reference indicated a mean canopy height of 4.17 m, while Dynex QCHP predicted 0.66 m. Recent imagery reviewed during the assessment was consistent with cleared or agricultural conditions.

This illustrates an important characteristic of forward-looking environmental monitoring: a model should respond to real changes in vegetation rather than simply reproduce a historical reference layer.

Toward scalable forest monitoring

The operational implications extend beyond canopy-height estimation itself.

Lower processing costs and shorter assessment cycles could allow canopy monitoring to move from occasional surveys toward a repeatable monitoring capability, supporting applications such as:

• forest asset valuation;
• biomass and carbon assessment;
• biodiversity monitoring;
• deforestation and degradation detection;
• ESG and environmental reporting;
• portfolio-scale environmental risk monitoring.

The validation concluded that the combination of speed, cost efficiency, explicit coverage reporting and scalability makes the approach particularly relevant for large-area and recurring forest-monitoring programmes.

An important methodological consideration

The validation also identified an important limitation of the available reference data.

The ETH Global Canopy Height dataset represents vegetation conditions primarily from 2019–2020, while the Dynex QCHP prediction target was approximately August 2026.

For this reason, differences between the two datasets were treated as descriptive comparisons rather than direct prediction errors.

Formal metrics such as RMSE, MAE, bias and R² were intentionally deferred until a future validation phase using temporally aligned independent measurements, such as airborne LiDAR, recent GEDI or ICESat-2 observations, or field measurements.

Next steps

The validation recommends expanding future assessments using contemporaneous independent reference data and instrumented benchmarking of processing time and costs.

Such work would complement the operational evidence already observed and enable formal predictive-accuracy statistics to be established alongside the demonstrated scalability and efficiency characteristics of the platform.

Download the full validation report

The complete independent validation report provides additional information on the methodology, validation sites, canopy-height comparisons, prediction coverage, operational performance and recommendations for future validation.

Download the full Bluebell Index – Dynex QCHP Validation Report (PDF)