Below water (bathymetric) survey data is a critical component of any hydraulic model as well as being critical for carrying out volumetric change analysis, however due to the difficulties involved in collecting this data at high resolutions, many practitioners rely on widely spaced cross section surveys to represent the below water surface.
With rapid advances taking place in drone technology as well as significant advances in the field of computer science relating to ‘machine learning’, we have been actively exploring techniques to enable the rapid and cost effective capture of bathymetric data. The final output of our technique is an accurate, seamless DEM of the above and below water-bed elevation.
Led by Matthew Gardner, who has been working professionally in the UAV sector since 2015, our team have developed a range of workflows and techniques for capturing bathymetric data using a standard drone equipped with only a high resolution digital camera. The results are proving to be very promising with our techniques now having been used commercially throughout New Zealand.
We have several techniques which are suitable for different types of waterways. Whilst the workflow for each job differs slightly, generally we are using two overall techniques, depending on the water characteristics.
Technique 1 - Depth / Colour Relationship
This technique works by finding a mathematical correlation between colour and depth. Each pixel in a digital photograph has a Red (R), Green (G) and Blue (B) value. In essence we find a correlation between R,G,B and depth and then apply this correlation to each pixel of the orthophoto creating a digital elevation model (DEM) of depth. In addition to using RGB, we also try and find additional parameters which can help the model find an even stronger relationship. These features always vary for each waterway, however can be very useful in obtaining a strong correlation.
Our techniques are based on international research and literature however we have further developed these techniques by incorporating automatic image recognition tools and machine learning algorithms into our workflow in order to improve the overall quality and reliability of our results.
The results will depend on the specific waterway, however to date we have successfully achieved string correlations using this technique down to a depth of 50m in a clear mountain lake.
Technique 2: Adjustment for surface refraction
For shallow water bodies with relatively clear water, such as braided river systems, we have found that we are able to utilise techniques which incorporate an allowance for refraction. These techniques utilise modern photogrammetric techniques which require the waterway to be flown at a lower altitude and with more overlap than is the case for the depth / colour technique. In the right water conditions, we are able to achieve a very accurate bathymetric DEM using this technique. To date, we’ve found these techniques are able to work well in depths up to 3 to 4 m.
In some situations, we have found that using a combination of the two techniques will provide the best result, especially when there is significant noise in the dataset due to high winds or rapids etc.
In order to carry out this technique we need two main datasets:
High resolution orthorectified imagery, or overlapping drone images which can be processed in a photogrammetric software package including referenced ground control points (GCPs)
Bed elevation data for calibration purposes - this can be captured by a boat equipped with a depth sounder, or using manual surveying techniques using a total station or RTK GPS system (ie cross section survey). Often we simply survey several cross sections down the length of the waterway – it is important that a wide range of depth data is captured in the calibration dataset. Another way to collect this data is to simply zigzag a boat down the waterway collecting a random sample of a wide range of depths.
Water level information at the time of the imagery capture – as our technique works by finding a correlation between colour and depth, we need to know the water level so that we can convert the depth information back into bed elevation. If this information is not available, it can be estimated from the point cloud, however is most accurate if measured at the time of survey.