Smartphone-derived optical proxies for estimating toxicity risk of Microcystis aeruginosa complex in inland waters

Toxic blooms dominated by cyanobacterial colonies of Microcystis aeruginosa complex (MAC) accumulate in the water surface, so they can be tracked by remote sensing. The abundance of toxic MAC cells is related to colony size, a parameter affecting water-leaving radiance (L_w). Here, we design a strategy to estimate the in situ abundance of toxic MAC based on optical remote sensing. Bio-geo-optical models were constructed for a large subtropical reservoir (Salto Grande in the Uruguay River) to estimate the abundance of toxic MAC using remote sensing reflectance (R_rs or L_w normalized by downwelling irradiance) indices. Spectrally weighted R_rs ratios were derived from smartphone measurements and related to three aggregated proxies of MAC (chlorophyll a, biovolume, and the abundance of mcyE gene copies, C^mcyE) using regression models and machine learning techniques. A classification tree (CART) model identified a threshold in the red/green reflectance ratio of 1.216 to predict the presence of high (> 1000 cells mL^-1) or low (< 1000 cells mL^-1) C^mcyE (average accuracy = 0.66, sd = 0.14). Our results suggest that MAC toxic and non-toxic cells can be discriminated in Salto Grande reservoir by using first-order optical proxies derived by a smartphone. The resulting R_rs ratio thresholds can be exploited to develop a cell phone-based tool able to detect toxic blooms, allowing citizen monitoring of aquatic ecosystems and provide a fruitful avenue to advance in the remote prediction of harmful blooms. The generalization of the method to other water bodies requires validation with local measurements.