High-frequency data within a modeling framework: On the benefit of assessing uncertainties of lake metabolism

We used a Bayesian metabolic model for assessing the gross primary production (GPP), ecosystem respiration (ER) and their uncertainties in lake Võrtsjärv, a large eutrophic lake in Estonia (North-eastern Europe). Diel cycle modeling was based on high-frequency (10-min) measurements of irradiance, water temperature and dissolved oxygen during most of the growing season (from May to August 2011). Posterior distribution of production and respiration was successfully simulated with the model and displayed with highly credible intervals (2.5 and 97.5 percentiles). Considering the mean GPP and ER values, the lake was autotrophic from May to June, at equilibrium in July, and heterotrophic in August. However, adding the uncertainty to metabolism estimates revealed that an ambiguous metabolic state (no clear monthly predominance of auto- or hetero-trophy) represented between 12 and 32% of the period. It is thus incautious to conclude about lake metabolic state in these conditions. A comparison with the existing classical model based on dissolved oxygen measurements showed that metabolic dynamics differed between the two approaches. Though the classical model recorded highest ecosystem productivity in midsummer, the Bayesian model predicted that productivity peaked earlier in the season and gradually declined as the irradiance dropped and the water temperature rose. Coupling between GPP and ER during the whole study period was very variable, resulting that, depending on the month, 50-100% of primary production was consumed in the lake. This coupling variability was caused by extensive diel fluctuation of irradiance-dependent production compared to relatively stable water temperature and respiration. The background respiration was high in spring and declined progressively in summer, reflecting lower inputs of allochthonous organic matter to the lake. With a wider use of high-frequency techniques for measuring lake ecological parameters, this kind of performant models that are able to assess lake productivity within small time steps and take into account the uncertainty, will be increasingly needed in the future.