Assessing Water Quality
Using drones to estimate water constituents, based on how sunlight is reflected by water.
The goal is to assess water quality, by using spectral reflectance cameras carried by drones to measure sunlight reflected by water
- Adapt remote sensing methods to use drones in assessing concentration of phytoplankton and mineral/organic material (ocean colour)
- Contribute to improvements in accuracy of remote sensing methods by using observation collected by drones
Partners involved:
Data Collection
There are remote sensing methods for assessing water quality. This application is testing application of these methods using data collected using hyper and multi spectral cameras on drones.
The method is known as ‘Ocean Colour’, and it uses remote sensing to extract water quality parameters from the colour of the water. There are some limitations to this remote sensing method, such as scattering in atmosphere and cloud coverage.
We will test using SeaBee to collect higher spatial and temporal resolution data in coastal areas, which we cannot get currently from satellite sources (current satellite for ocean colour gives 300 m resolution).
Ocean colour algorithms work well in ocean settings right now, but not as well in detailed, coastal areas (like along the coast of Norway). In these coastal settings, even if spatial resolution is good enough, the ‘edges’ (from shadows, snow cover) complicate the results. Further, there are not many satellites measuring hyperspectral reflectance, and those that do pass around once a week. We expect that using SeaBee drones will give more flexibility to measure ocean colour, even with cloud
SeaBee Technology
For this application, Seabee’s multispectral (Micasense RedEdge-dual and Altum-PT) and hyperspectral (Specim AFX10) cameras were used. These cameras capture the light signal leaving the water. The light signal contains information about its optical properties of the reflected light.Using the optical property information it is then possible to extract water constituents and their concentration such as phytoplankton and mineral or organic material.
Oslofjord is the sample area, as it provides various natural events, human impacts and weekly observation data collection by FerryBox (LINK). The data are collected by flying drones on a sunny day. In 2024, the missions will collect data that will be uploaded to SeaBee Geonode. From there, the data with be processed automatically using Ocean Colour alogrithms.
Main results so far:
Water Quality
A first evaluation of the methods was conducted in a controlled environment, at NIVA’s field research station.Both the multispectral and hyperspectral cameras were mounted on a XXXX drone and flown over an outdoor tank (50 m3 volume/~5.5 m diameter) filled with seawater. The images were analysedby examining the correlation between reflectance measurements at relevant wavelengths with the corresponding discrete measurements of chl-a and turbidity.
The tank was filled with Oslo fjord seawater and manipulated to modify its optical properties with addition of phytoplankton cultures, lignin and kaolin. The initial concentration of natural-occurring phytoplankton in the Oslo fjord seawater was ~15 µg chl-a l-1, and green alga Tetraselmis sp. was added to maintain chl-a at ~ 8-13 µg chl-a l-1. Lignin, which is a natural polymer contained in the plants which constitute part of the coloured dissolved organic matter (CDOM) that can be found in some marine and freshwater, was added to simulate higher concentrations of CDOM with a 440 (1/m) up to ~4. And kaolin, a clay mineral powder, was used to increase the turbidity up to around 20 FNU and therefore scattering of seawater in tank. Water samples were taken and analysed on a benchtop spectrophotometer as reference.
In addition to the over-tank experiments, both the multispectral and hyperspectral cameras were also flown on test transects using DJI Matrice 300 RTK and DJI Matrice 600 Pro UAVs over Oslo fjord, and water samples for chl-a and turbidity were also taken simultaneously with the drone observations and measured with laboratory instruments.
The camera data processing includes corrections for geometric aberrations, the dark signal from the raw image (mean value of the dark pixels) was removed and the signal was normalised with the exposure time, and calibration coefficients were applied to convert the raw values into absolute radiance.
Figure 3: NIR-RE-R-G-B Altum PT multispectral camera map from 80 m (left) and RGB image from AFX10 hyperspectral camera (right) over NIVA’s research station.
Ocean colour algorithms can then be applied using reflectance data to retrieve the concentration of seawater constituents of interest. In this first study, empirical relationships have been examined between reflectances from the cameras and reference chl-a and turbidity measurements.
For Chl-a matchups, both the multispectral and hyperspectral cameras performed well across the experimental range of 0.78-17.44 µg chl-a L-1. The hyperspectral camera data also exhibit chl-a absorption at 670 nm and sun-induced chl-a fluorescence at 760 nm.
For turbidity matchups, both cameras also performed well across the experimental range of 1.06-18.49 FNU. Despite the turbidimeter measuring in the NIR (860nm), the best correlation with the multispectral camera was in the red edge band (717 nm), and therefore 717 nm was used for both cameras. Turbidity is strongly correlated with total suspended material (TSM) concentration.
Plastic Litter
A test scenario of simulated plastic litter was also conducted where plastic pieces of different sizes were placed on a nearby beach and has confirmed that the UAV-camera packages were able to capture images with image spatial resolution better than 1 cm to detect small object and spectral patterns from different polymer types was discernible. The images are partitioned using the Deep Endmember Hierarchy technique into constituent spectra.
Contact Us
Financed by:
