Drone-based imaging to assess the microbial water quality in an irrigation pond: A pilot study

Microbial water quality datasets are essential in irrigated agricultural practices to detect and inform measures to prevent the contamination of produce. Escherichia colt (E. coli) concentrations are commonly used to evaluate microbial water quality. Remote sensing imagery has been successfully used to retrieve several water quality parameters that can be determinants of E. coli habitats in waterbodies. This pilot study was conducted to test the possibility of using imagery from a small unmanned aerial vehide (sUAV or drone) to improve the estimation of microbial water quality in small irrigation ponds. In situ measurements of pH, turbidity, specific conductance, and concentrations of dissolved oxygen, chlorophyll-alpha, phycocyanin, and fluorescent dissolved organic matter were taken at depths of 0-15 cm in 23 locations across a pond in Central Maryland, USA. The pond surface was concurrently imaged using a drone with three modified GoPro cameras, and a multispectral MicaSense RedEdge camera with five spectral bands. The GoPro imagery was decomposed into red, blue, and green components. Mean digital numbers for 1-m radius areas in the images were combined with the water quality data to provide input for a regression tree-based analysis. The accuracy of the regression-tree data description with only imagery inputs was the same or better than that of trees constructed with only water-quality parameters as inputs. From multiple cross-validation runs with only imagery inputs for the regression trees, the average (+/- SD) determination coefficient and root-mean-squared error of the decimal logarithm of E. coli concentrations were 0.793 +/- 0.035 and 0.131 +/- 0.011, respectively. The results of this study demonstrate the opportunities for using sUAV imagery for obtaining a more accurate delineation of the spatial variation of E. coli concentrations in irrigation ponds. Published by Elsevier B.V.