This study evaluates the vertical accuracy of several widely used global Digital Surface Models (DSMs) within the urban context of Campo Grande, Brazil. By comparing free and commercial satellite-derived elevation data against high-precision ground truth, the researchers provide a roadmap for engineers and urban planners to determine which datasets are reliable enough for hydrological modeling and cartographic applications.
TL;DR
The study finds that the freely available ALOS AW3D30 and the commercial Tandem-X are the most accurate global elevation models for urban Brazilian environments, achieving vertical errors (RMSE) under 3 meters. While ALOS AW3D30 is suitable for detailed 1:25,000 scale mapping, older standards like SRTM and Aster GDEM are significantly less precise, limited to broader 1:100,000 scale applications.
What Problem Does This Paper Address?
Digital Surface Models (DSMs)—which represent the elevation of the Earth's surface including buildings and vegetation—are the backbone of urban flood modeling, landslide susceptibility analysis, and infrastructure planning. While global DSMs like SRTM are freely available, their accuracy is often compromised in urban areas due to "signal noise" from buildings and complex topography.
In Brazil, cartographic products must adhere to the PEC-PCD (Padrão de Exatidão Cartográfica), a national standard that dictates what scale a map can reliably represent based on its error margins. Before this paper, there was a lack of rigorous testing to see how newer models like Tandem-X and ALOS AW3D30 stack up against these Brazilian standards in a specific urban setting like Campo Grande.
Methods
The researchers employed a "ground truth" comparison method. They collected two high-precision reference datasets using GNSS (Global Navigation Satellite System) technology:
- Kinematic Dataset: A vehicle-mounted dual-frequency receiver captured over 14,000 points along city streets.
- RTK Dataset: Static Real-Time Kinematic surveying captured ~3,000 points across diverse environments (vegetation, buildings, and water bodies).
The core challenge was ensuring a "like-for-like" comparison. Because different models use different vertical datums (some use the GRS80 ellipsoid, others use the EGM96 geoid), the team performed rigorous pre-processing to convert all measurements into a compatible format before calculating the Root Mean Square Error (RMSE) and classifying the models according to the PEC-PCD scales.
Datasets
The study evaluated six major DSMs:
- Tandem-X (12m resolution): A high-resolution radar mission.
- ALOS AW3D30 (30m resolution): A photogrammetry-based model from the Japanese Space Agency.
- SRTMc & SRTM v.3 (30m resolution): The industry-standard radar topography data.
- TOPODATA (30m resolution): A refined version of SRTM specifically processed for Brazil.
- Aster GDEM v.2 (30m resolution): An optical-based global model.
Key Insights and Ideas
- Resolution isn't everything: Despite having a lower spatial resolution (30m) than Tandem-X (12m), the ALOS AW3D30 model actually showed the highest vertical accuracy. This suggests that the photogrammetric processing used for ALOS may handle urban surface textures better than radar-based interferometry in certain contexts.
- The "Stadium Outlier": The researchers identified a specific outlier in the Tandem-X data—a -25m discrepancy located at a sports stadium. This highlights a critical limitation of satellite DSMs: large, smooth, or uniquely shaped urban structures can cause significant local errors that "average" statistics might miss.
- SRTM is showing its age: While SRTM revolutionized the field 20 years ago, this study confirms it is no longer the "gold standard" for urban work. Its derivatives (SRTMc and TOPODATA) performed better than the raw version but still lagged far behind ALOS and Tandem-X.
Key Results
The models were ranked based on their RMSE (lower is better):
- ALOS AW3D30: RMSE of 1.57m to 1.77m. It was the only model compatible with the 1:25,000 scale (Class A) in the second dataset.
- Tandem-X: RMSE of 2.68m to 2.80m. It consistently met the requirements for 1:50,000 scale mapping.
- SRTMc / TOPODATA / SRTM v.3: RMSEs ranged from 3.4m to 5.0m. These are only recommended for 1:100,000 scale projects.
- Aster GDEM v.2: The worst performer with an RMSE of ~8.0m, showing high variability and unreliability for urban engineering.
Limitations and Future Work
The authors acknowledge that while their GNSS datasets were extensive (over 18,000 points), they were primarily collected at ground level (streets and open areas). Urban "canyons" created by high-rise buildings can still interfere with satellite signals, potentially leading to higher errors than those reported here.
Future research is suggested to specifically target areas with high-density vertical construction to see if the accuracy of ALOS and Tandem-X holds up when the "surface" becomes significantly more complex. Additionally, comparing these results against airborne LiDAR—the gold standard for elevation data—would provide an even more granular look at these satellite models' performance.
Summary updated