Airborne Drones for Water Quality Mapping in Inland, Transitional and Coastal Waters—MapEO Water Data Processing and Validation

Abstract

Using airborne drones to monitor water quality in inland, transitional or coastal surface waters is an emerging research field. Airborne drones can fly under clouds at preferred times, capturing data at cm resolution, filling a significant gap between existing in situ, airborne and satellite remote sensing capabilities. Suitable drones and lightweight cameras are readily available on the market, whereas deriving water quality products from the captured image is not straightforward; vignetting effects, georeferencing, the dynamic nature and high light absorption efficiency of water, sun glint and sky glint effects require careful data processing. This paper presents the data processing workflow behind MapEO water, an end-to-end cloud-based solution that deals with the complexities of observing water surfaces and retrieves water-leaving reflectance and water quality products like turbidity and chlorophyll-a (Chl-a) concentration. MapEO water supports common camera types and performs a geometric and radiometric correction and subsequent conversion to reflectance and water quality products. This study shows validation results of water-leaving reflectance, turbidity and Chl-a maps derived using DJI Phantom 4 pro and MicaSense cameras for several lakes across Europe. Coefficients of determination values of 0.71 and 0.93 are obtained for turbidity and Chl-a, respectively. We conclude that airborne drone data has major potential to be embedded in operational monitoring programmes and can form useful links between satellite and in situ observations.

CoastalWQL: An Open-Source Tool for Drone-Based Mapping of Coastal Turbidity Using Push Broom Hyperspectral Imagery

“…Variants of the algorithm involving its applicability on red to NIR wavelengths were reported in various studies, where [4,21] used 660 nm and 668 nm, and [41] chose 710 nm instead. [39] suggested 681 nm, which has the lowest error rate for suspended particle matter (SPM) retrieval.…”

“…Turbidity retrieval from UAV imagery with machine learning (ML) models appears to show promise. In another UAV study by [50], where various ML models were evaluated for the retrieval of suspended solids (SS), SS prediction in ranges from R 2 = 0.91 to 0.99, which performed significantly better than some studies using semi-analytical algorithms, e.g., [4,21]. With the development of CoastalWQL, its application, together with ML models, was used to extensively monitor turbidity plumes associated with land reclamation activities with an R 2 = 0.75 (see [5]).…”

“…For instance, the results obtained in this study using Nechad et al's (2010) [39] semi-analytical algorithm at 715 nm showed close correspondence with the results obtained in a similar study that estimated dredge-induced turbidity plume using the same semi-analytical algorithm at 660 nm, where the study achieved a RMSE of 3.39 with a UAV-borne MSI (Parrot Sequoia) for turbidity values ranging from 3.3 to 72.4 FNU [4]. Another study using the same semi-analytical algorithm for turbidity retrieval at 668 nm obtained an RMSE of 10.13 FNU with the MicaSense Dual Camera System [21].…”

“…The recalibration of the coefficients is required since A τ and C relate to IOPs such as absorption by particle and non-algal particles and particulate backscatter, which depends on the morphology and characteristics of the particles, as well as turbidity concentration. Recalibration of A τ and C coefficients were similarly conducted by [21] where the derived coefficients are 366.14 and 0.1956 for A τ and C at 668 nm.…”

“…Recent studies by [21,22] adopted a direct georeferencing method for high-resolution UAV RGB/multispectral snapshot imagery, in which the MosaicSeadron from [22] achieved an error rate (standard deviation) of 2.51 m at a ground sample distance (GSD) of 0.5 m/px, as well as increased the ground coverage to 33 hectares as compared to 13.45 hectares with the SfM to mosaic snapshot images of water bodies. However, the direct-georeferencing method for snapshot imagery may not always be applicable to push-broom imagery where the image output, metadata, and the temporal resolution of the captures and measurement of flight parameters are vastly different from that of snapshot imagery due to the nature of the imaging principles that differ significantly between push-broom and snapshot imaging (see Appendix A) [23,24].…”