Optimised UAV Flight Parameters for High-Precision Coastal Erosion Monitoring
Comparing different flight modes to determine optimal parameters for coastal monitoring
Abstract
This research evaluates the impact of different UAV flight modes on the accuracy and quality of photogrammetric outputs for coastal monitoring applications. By comparing RTK, Baseline, Automatic Exposure Bracketing (AEB), and Terrain Follow flight modes, this study provides evidence-based recommendations for optimizing data acquisition parameters to improve model accuracy and efficiency.
Key Findings
- RTK mode demonstrated superior performance with highest spatial resolution (2.15 cm/pixel) and point cloud density (271.35 pts/m³)
- Baseline and AEB modes performed similarly with impressive mean vertical displacement values (0.007m and -0.012m respectively)
- Terrain Follow mode showed reduced performance with lowest resolution (2.56 cm/pixel) and longest processing time
- RTK and Baseline modes are recommended for high-precision coastal monitoring applications
Methodology
Study Area and Experimental Design
The study was conducted at the Queensland Government Hydraulics Laboratory (QGHL) site in Deagan, Brisbane, Queensland. All flight missions were conducted on the same day, within a one-hour window under clear and calm conditions, ensuring that external atmospheric factors had minimal influence on data quality.
All datasets for this experiment were collected using the DJI Mavic 3M, thereby standardizing the acquisition process and isolating the effect of flight parameter variations. Four separate flight missions were conducted over the QGHL site, each varying one independent flight parameter to assess its impact on the accuracy and precision of the photogrammetric outputs.
Flight Modes Tested
RTK Mode
Served as the reference flight, using Real-Time Kinematic positioning to enhance geolocation accuracy.
Baseline Mode
Utilized standard, typical field data acquisition parameters with grid-based flight path.
AEB Mode
Employed Automatic Exposure Bracketing parameters to handle varying lighting conditions.
Terrain Follow Mode
Adjusted flight altitude dynamically to follow the topography of the site.
Data Processing and Analysis
All datasets were processed using Pix4Dmapper to generate Digital Surface Models (DSMs), orthomosaics, point clouds, and Digital Terrain Models (DTMs). Consistent processing parameters were maintained across all datasets to ensure fair comparison.
The analysis included quantitative metrics such as horizontal and vertical displacement, point cloud density, processing time, and geolocation accuracy. A post-hoc analysis phase was also conducted to further interrogate the impacts of the varied flight parameters.
Key Results
Performance Metrics Across Flight Modes
Results from the flight mode comparison revealed distinct performance characteristics across RTK, Baseline, AEB, and Terrain Follow modes. The following table summarizes the key metrics for each flight mode:
| Metric | RTK (Reference) | Baseline | AEB | Terrain Follow |
|---|---|---|---|---|
| Resolution (cm/pixel) | 2.15 | 2.26 | 2.25 | 2.56 |
| Mean Vertical Displacement (m) | Reference | 0.007 | -0.012 | 0.054 |
| Total Processing Time | 12m:00s | 13m:07s | 12m:45s | 16m:01s |
| Point Cloud Density (pts/m³) | 271.35 | 225.56 | 229.06 | 152.93 |
| GCP RMS Error (m) | 0.010 | 0.008 | 0.011 | 0.010 |
| Geolocation X/Y Bias (m) | 0.02/-0.03 | -0.28/-1.48 | -0.11/-1.65 | -0.10/-0.12 |
Table 1: Summary of key metrics across flight modes
Point Cloud Comparison
Point cloud comparisons revealed complex patterns in displacement measurements across flight modes. While RMSE values showed notable differences between modes (Baseline: 0.381m, AEB: 0.483m, Terrain Follow: 1.139m), the median displacement values were remarkably consistent across all modes (approximately 0.034m horizontal), suggesting comparable performance in measuring stable ground features.
Figure 1: Box and Whisker plots showing horizontal, vertical and total displacement between RTK reference and other flight modes (AEB, Baseline, and Terrain Follow)
Post-hoc Analysis
Post-hoc pairwise comparisons revealed minimal differences between AEB and Baseline modes across all displacement types, with small absolute differences (≤0.019m) and similar variance ratios (~0.60). However, comparisons involving Terrain Follow mode showed more substantial disparities, with larger effect sizes (-0.1328 to -0.1573) and notably higher variance ratios (3.36 to 11.66).
| Displacement Type | Measurement | AEB vs Baseline | AEB vs TF | Baseline vs TF |
|---|---|---|---|---|
| Horizontal | Effect Size | 0.0188 | -0.1429 | -0.1547 |
| Absolute Diff (m) | 0.0060 | 0.1030 | 0.1090 | |
| Variance Ratio | 0.63 | 7.33 | 11.66 | |
| Mean Difference (%) | 9.0% | 84.8% | 92.0% | |
| Vertical | Effect Size | -0.0658 | -0.1384 | -0.1034 |
| Absolute Diff (m) | 0.0190 | 0.0660 | 0.0470 | |
| Variance Ratio | 0.60 | 3.36 | 5.61 | |
| Mean Difference (m) | -0.019 | -0.066 | -0.047 |
Table 2: Post-hoc pairwise comparison analysis for flight mode displacement accuracy
Orthomosaic Comparison
Comparative analysis of orthomosaics generated from the RTK reference and Terrain Follow flights demonstrated remarkable geometric consistency despite representing the most divergent flight configurations tested. Precise measurements of fixed structures revealed only a 1mm difference between the two outputs, suggesting that orthomosaic generation remains robust regardless of the flight configuration employed.
Discussion
The comparative assessment of RTK, Baseline, AEB, and Terrain Follow flight modes revealed distinct performance characteristics and operational implications for coastal monitoring applications. RTK mode served as the reference dataset due to its superior accuracy metrics and optimal post-processing results, allowing for robust comparison across flight configurations.
Performance Patterns
While Terrain Follow showed higher mean displacement (0.173m) and standard deviation (σ=0.956m) compared to other modes, its performance metrics for non-outlier measurements were similar to Baseline and AEB modes. This pattern suggests that Terrain Follow's apparently poorer performance in aggregate metrics like RMSE is primarily driven by its different handling of vegetation and structural edges rather than a systematic deterioration in accuracy.
Processing Efficiency
Processing metrics revealed consistent performance in key photogrammetric parameters across modes, with successful image calibration (159-160 images) and comparable feature matching (70,965-72,034 key points). However, significant variations emerged in processing efficiency and point cloud density. RTK mode achieved optimal results with the highest point cloud density (271.35 points/m³) and fastest processing time (12:00), while Terrain Follow showed reduced performance (152.93 points/m³, 16:01 processing time).
Limitations
Several limitations should be noted. The study was conducted at QGHL rather than the intended coastal environment due to weather constraints, potentially affecting the direct applicability to coastal monitoring scenarios. This limitation is particularly significant for the evaluation of the AEB mode, which was specifically included to assess its capability in handling high-reflectance surfaces like beach sand. Additionally, the analysis is based on a single test flight for each mode; multiple flights under varying conditions would provide more robust statistical validation.
Conclusion
The results indicate that RTK and Baseline modes provide the most reliable data for coastal monitoring applications, with comparable accuracy levels to established benchmarks in coastal photogrammetry literature. While AEB mode demonstrated acceptable accuracy for most applications, the significantly higher variance ratios and localized deviations seen in Terrain Follow mode suggest its use should be carefully evaluated and potentially removed from standard operating procedures.
The flight mode optimization investigation revealed that RTK and Baseline modes provide optimal performance for coastal surveys, with a mean vertical displacement value of 0.7cm for Baseline mode. AEB mode showed overall impressive accuracy with a mean vertical displacement of -1.2cm. Terrain Follow mode demonstrated degraded performance, notably in its effect size and variance. Consequently, Terrain Follow mode should be avoided for high-precision applications.
These findings provide clear operational guidance for future coastal monitoring protocols, helping to standardize data collection methodologies and improve the consistency and reliability of photogrammetric outputs for environmental monitoring applications.
Related Research
Contact Us
Interested in discussing this research or exploring similar projects? Our team of experts is ready to help.
Get in Touch