Methodology Flowchart - Website Architecture

Short Description

We take 3 overlapping aerial survey images into ERDAS IMAGINE. Using this tool, we generate Tie Points, perform Aerial Triangulation and then generate 2 DSMs from the two overlapping regions. These 2 DSMs are then merged together in CloudCompare to produce a final DSM as a LAS file which is then rasterized. A similar process is performed with the 2 DSM LAS files acquired from the AHN website to produce a single LAS file as well as a rasterized file. From both these LAS files, we also generate DTMs as raster files using the Cloth Simulation Filter. Next, we add checkpoints to compare the DSM and the DTM generated by our team against those from the AHN website. Based on the checkpoints, we then calculate the overall RMSE as well as the Uncertainty per checkpoint for both the DSM and DTM. The output rasters are of 1 meter resolution.

Our website relies on Firebase as the backend service for storing most of our geospatial data through a storage bucket (hosted by Google Cloud Platform). Further, we rely on Netlify for hosting the website. The website is rendered using HTML, CSS, and JavaScript. Firstly, we add a basemap from OSM using Leaflet. For rendering the rasters we use GeoRaster, while we render the vector files using Leaflet. Additionally, the map controls are also created using Leafelt. Additionally, we use Potree to provide a 3D visualization of the DSM generated by our team. The rendered 3D representation of the DSM (using potree) is stored on the ITC server due to space constraints.

Key Sources of Uncertainty in DSM and DTM

1. Variability in Tie Point Extraction: The automatic extraction of tie points can produce inconsistent results across different processing runs, particularly in areas with low texture or repetitive features. These variations can lead to misalignments and reduced DSM accuracy.

2. Imprecision in GPS/IMU-Based Georeferencing: The accuracy of direct georeferencing depends on the quality of GPS and IMU data, which can be affected by signal interference, calibration issues, or environmental conditions—resulting in positional errors in the final model.

3. Residual Lens Distortions: Despite calibration, minor distortions from the camera lens may persist, impacting the geometric accuracy of images and the fidelity of the derived surface model.

4. Temporal Differences Between Datasets: Natural or human-induced changes, such as vegetation growth or new construction, can cause mismatches when comparing current imagery to older DSMs, leading to discrepancies that reflect real-world changes rather than processing errors.

5. Artefacts Introduced During Processing: Software used in photogrammetry may introduce subtle distortions during steps like image matching, point cloud generation, or interpolation, which can compromise the overall quality of the DSM and DTM.