Drone-Based Landfill Survey System

Core

01Removal verification with uncertainty

1. Build a DEM for each survey from the point cloud.
2. Compute the DEM of Difference per grid cell: DoD = DEM(t₂) − DEM(t₁)
3. Estimate each survey’s vertical error from the checkpoints and combine them: σ_DoD = √(σ₁² + σ₂²)
4. Apply a level of detection, LoD95 = 1.96 × σ_DoD. Smaller changes are treated as noise.
5. Use spatially varying uncertainty: steep, rough or poorly covered cells get bigger error bars than flat cleared ground (after Las2DoD and James et al. 2017).
6. Sum cut and fill separately to get removed and added volume, each with a confidence interval.

Volume to tonnage. Payment is per tonne, but drones measure volume. A density range (from weighbridge records or core samples) and the DEM error run together through a Monte Carlo simulation, so the output reads like:

41,200 t removed · 95% interval 38,900–43,600 t
Example of the output format

Why this is the contribution. Drone volume surveys of landfills are already common. Erosion and river researchers routinely report volume change with propagated uncertainty; bringing that discipline to payment verification on an active waste site is the new part.

Fire risk

03Thermal hotspot tracking

• Temperatures from a radiometric thermal camera are projected onto the 3D model, giving a “4D” model (X, Y, Z, temperature). A study at the Meruelo landfill in Spain did this with about 1.6 °C accuracy and found hot areas later confirmed on site.
• Readings are corrected for emissivity and time of day: plastic, soil and wet surfaces read differently at the same temperature.
• A surface hotspot shows heat near the surface, not depth. The output is a list of places to check with probes, not a diagnosis.
• Each hotspot is tracked across flights (area, peak temperature, growth). Growth is the real warning sign.
• The Jetson runs an anomaly detector during the flight and sends grid references to the ground crew immediately.