In vivo fluorometers use chlorophyll a fluorescence (Fchl) as a proxy to monitor phytoplankton biomass. However, the fluorescence yield of Fchl is affected by photoprotection processes triggered by increased irradiance (nonphotochemical quenching; NPQ), creating diurnal reductions in Fchl that may be mistaken for phytoplankton biomass reductions. Published correction methods are mostly designed for pelagic oceans and are ill suited for inland waters or for high‐frequency data collection. A machine learning‐based method was developed to correct vertical profiler data from an oligotrophic lake. NPQ was estimated as a percent reduction in Fchl by comparing daytime values to mean, unquenched values from the previous night. A random forest regression was trained on sensor data collected coincident with Fchl; including solar radiation, water temperature, depth, and dissolved oxygen saturation. The accuracy of the model was assessed using a grouped 10‐fold cross validation (mean absolute error [MAE]: 7.6%; root mean square error [RMSE]: 10.2%), which was then used to correct Fchl profiles. The model also predicted NPQ and corrected unseen Fchl profiles from a future period with excellent results (MAE: 9.0%; RMSE: 14.4%). Fchl profiles were then correlated to laboratory results, allowing corrected profiles to be compared directly to collected samples. The correction reduced error (RMSE) due to NPQ from 0.67 μg L−1 to 0.33 μg L−1 when compared to uncorrected Fchl data. These results suggest that the use of machine learning models may be an effective way to correct for NPQ and may have universal applicability.

To measure chlorophyll a (Chl a) fluorescence (Fchl), fluorometers use an excitation wavelength that is within the visible spectrum of most zooplankton, and as a result has the potential to cause a phototactic response in zooplankton. The transparent bodies of herbivorous zooplankton may allow viable chlorophyll a within an individual's digestive tract to fluoresce in response to sensor excitation light, resulting in measurement bias. To test for this bias, a fully factorial (± zooplankton and ± light) experiment was conducted in an oligotrophic lake. Excitation light from fluorometers triggered a positive phototactic response during nighttime hours, resulting in swarms of zooplankton congregating beneath the sensor. The maximum hourly mean Fchl from nighttime/open treatments was higher and more variable than nighttime/zooplankton exclusion treatments, with the greatest single hour difference of 7.34 relative fluorescence units (RFU) vs. 0.26 RFU. In open treatments, sustained periods of Fchl exceeded 31x the values of exclusion treatments. A second series of experiments pulsed excitation lights in alternating periods in order to characterize zooplankton response times. Sensor bias was detected in as little as 20 s after initial illumination. Collectively, these results suggest that swarms of phototactic zooplankton can cause substantial bias in Fchl measurements at night. To correct for this bias, post‐processing methods using time series decomposition were demonstrated to remove the majority of Fchl bias.

Ice records at Lake George, an oligotrophic and dimictic freshwater lake in upstate New York, United States reveal that it has failed to freeze over completely 13 times since 1990. This transition from annual to intermittent ice cover is analogous to many other dimictic freshwater lakes globally. Over 60 years of meteorological observations from a nearby airport are analysed and a complicated picture emerges when considering the specific characteristics of each year. For example, Lake George froze over in 1983 and 2007 despite the air temperature having a net warming effect on the lake in the month prior to ice-in. Simple machine learning classifiers are trained using local weather data to predict the presence of complete ice coverage on Lake George and are found to perform adequately compared to observations, with one configuration having an accuracy of 91%. Using downscaled data from a coupled-climate model through to 2,100, projections with the trained classifiers suggest complete ice coverage will be a phenomenon of the past by the mid-to-late 21st Century. Furthermore, by 2080-2,100 the mean air temperature is projected to warm up to +8.0 C under Representative Concentration Pathway 8.5. These projections will be of concern to the communities and policy makers of Lake George responsible for the management of the ecological and socioeconomic systems.