Determining the best practices for utilizing Unoccupied Aerial Systems (UASs) for monitoring water quality

Abstract

Accurate water quality monitoring techniques are vital for the conservation and consumption of clean water. Traditional in situ sampling practices have limitations that can be augmented with remote sensing techniques. Satellite remote sensing data has been used to support and provide further insight into traditional in situ sampling methods, however, some limitations remain. By incorporating Unoccupied Aerial Systems (UASs) these limitations can be diminished, especially in small freshwater systems. UAS data are not affected by cloud cover, or atmospheric distortions, and have a higher image resolution when compared to satellite data. Many studies have highlighted the significant potential role UASs can contribute to monitoring water quality issues, especially optical properties like turbidity and algal blooms. Several different methods have been used and proven to produce results that are representative of different water quality parameters. However, standardization has not yet been evaluated. Specifically, it is unknown how far into the water column UASs can detect spectral signatures of the water, so it is unknown which type of in situ sample should be collected to validate UAS data (e.g., surface, throughout the water column, or at a certain depth). The objectives of this research were to test previously used methods for ground truthing and explore different sensor capabilities. To achieve this, the first objective was to determine how far into the water column different sensors can detect a Secchi disk, and the best methods for collecting in situ water samples to accurately represent UAS data. UAS imagery was collected from two sites: the Aquaculture ponds at the E.W. Shell Fisheries Center in Auburn, Alabama on 6/11/2022 and at Lake Martin in Dadeville, Alabama on 6/24/2022. A line of Secchi disks was placed into the waterbody prior to the UAS flights. The Secchi depth of the water was taken prior to placing the Secchi line to ensure the disks were above and below the Secchi depth. Water samples were collected using two different methods: surface samples from each waterbody and an integrated depth sample at the Secchi depth of the waterbody. In situ samples were tested for different spectral properties of the water which included chlorophyll-a (chl-a), phycocyanin, and total suspended solids (TSS). ArcGIS Pro, Pix4D, and a Python script were utilized to process and analyze the data. A t-test between the in situ water quality parameters revealed (p = 0.93 for chl-a and 0.51 for phycocyanin) the waterbody was evenly mixed. Therefore, it was not possible to differentiate the relationship between spectral signatures and respective water quality parameters at different depths. However, the UAS imagery revealed that the Secchi disks past the Secchi depth in Lake Martin, an oligotrophic waterbody, were 100% visible in the imagery. The disks in the Aquaculture ponds, which are eutrophic waterbodies, were not 100% visible, even when above the Secchi depth in some instances. The results suggest the trophic state of the water greatly impacts the depth at which physical objects can be detected in aquatic environments from UAS imagery. Thus, different methods for UAS based monitoring should be considered depending on the trophic state of the target waterbody. The second objective was to determine the number of pixels that should be used to calculate an average spectral value from UAS imagery for ground truthing data. The data from the Aquaculture ponds and Lake Martin from the first objective were utilized in addition to data collected from 16 different ponds at the Genetics research lab in Auburn, Alabama on 8/12/2022. Surface samples were collected at each pond and a tube sample up to the Secchi depth at ponds which a Secchi depth greater than 10 cm was collected. The results revealed that Chlorophyll Index Green (CiGreen) was the only index affected by the number of pixels used to represent in situ data as the Pearson correlation coefficient value for TSS increased from 0.87 (p = < 0.01, n = 24) to 0.93 (p = < 0.01, n = 24) when a buffer containing 15 or more pixels was used. The strongest correlation value for Phycocyanin was observed with the Surface Agal Bloom Index (SABI index having a Pearson correlation coefficient value of 0.81 (p = < 0.01, n = 24) for all buffers which ranged from containing 5 pixels to 60 pixels. Chl-a had the lowest correlation values for all buffer sizes and the strongest correlation was seen with the Two-Band Algorithm (2BDA) algorithm and a Pearson correlation coefficient value of 0.59 (p < 0.01, n = 24). Chl-a also saw the most statistically insignificant values with 13 of the 16 correlations having a p-value over 0.05. The strongest Pearson correlation coefficient value was seen with TSS and the CiGreen index of 0.93 and was only present in the buffers ranging from 30 pixels to 60 pixels. Thus, when ground truthing non geolocated in situ values to UAS spectral data, buffers containing at least 30 pixels should be utilized when using the CiGreen index. In addition, an exploratory analysis should be conducted to compare multiple buffer sizes to in situ samples to further ensure accurate data representation for all water quality parameters. As the need for real time water quality data persists, the use of remote sensing techniques and UASs can become a valuable tool for water resource managers. Research is needed to further establish best practices and explore the capabilities of different sensors for different water quality parameters. This research addresses those needs and highlights the importance of future research for UAS water quality monitoring needs.