Improving the precision of lake ecosystem metabolism estimates by identifying predictors of model uncertainty

Abstract

Diel changes in dissolved oxygen are often used to estimate gross primary production (GPP) and ecosystem respiration (ER) in aquatic ecosystems. Despite the widespread use of this approach to understand ecosystem metabolism, we are only beginning to understand the degree and underlying causes of uncertainty for metabolism model parameter estimates. Here, we present a novel approach to improve the precision and accuracy of ecosystem metabolism estimates by identifying physical metrics that indicate when metabolism estimates are highly uncertain. Using datasets from seventeen instrumented GLEON (Global Lake Ecological Observatory Network) lakes, we discovered that many physical characteristics correlated with uncertainty, including PAR (photosynthetically active radiation, 400-700 nm), daily variance in Schmidt stability, and wind speed. Low PAR was a consistent predictor of high variance in GPP model parameters, but also corresponded with low ER model parameter variance. We identified a threshold (30% of clear sky PAR) below which GPP parameter variance increased rapidly and was significantly greater in nearly all lakes compared with variance on days with PAR levels above this threshold. The relationship between daily variance in Schmidt stability and GPP model parameter variance depended on trophic status, whereas daily variance in Schmidt stability was consistently positively related to ER model parameter variance. Wind speeds in the range of ~0.8-3 m s -1 were consistent predictors of high variance for both GPP and ER model parameters, with greater uncertainty in eutrophic lakes. Our findings can be used to reduce ecosystem metabolism model parameter uncertainty and identify potential sources of that uncertainty.

LakeMetabolizer: an R package for estimating lake metabolism from free-water oxygen using diverse statistical models

“…First, the model can be run unconstrained and the impossible estimates can be classified as nonsensical and removed from subsequent analysis. These impossible results are often from days when physical processes (e.g., wind mixing) dominate the DO signal and therefore are days when the biological signal is overwhelmed by other sources of DO variability (Rose et al 2014). Second, the model can be written to constrain the parameters and force the estimation of positive GPP and negative R using a priori information about the possible values of GPP and R.…”

“…In these situations, estimates of uncertainty can help guide the investigator, but there is not a consensus on an optimal strategy. Strategies include, but are not limited to, keeping all metabolism estimates except for extremely uncertain values (Solomon et al 2013), setting more rigorous thresholds of uncertainty (Cremona et al 2014), down-weighting poorly fit metabolism days based on uncertainty estimates (Rose et al 2014), or fitting metabolism parameters over multiple days (Van de Bogert et al …”

Carbon quality regulates the temperature dependence of aquatic ecosystem respiration

“…If the increase in temperature between the environment and the experiment harmed the autochthonous treatment microbial community, we would have expected a greater impact and thus lower ER in the heated autochthonous treatment. Rather, the variability in daily estimates of ecosystem metabolism from free water dissolved oxygen sensors likely contributes a great deal to our uncertainty (Rose et al., ). While in situ measurements of dissolved oxygen are ecologically realistic compared to bottle incubations, diel curves in dissolved oxygen often are not perfectly sinusoidal and thus estimates of metabolic parameters often contain some uncertainty (Winslow et al., ).…”

Consequences of gas flux model choice on the interpretation of metabolic balance across 15 lakes

“…All data were collected between 2007 and 2010, and data for each lake include between 133 and 338 days of observation. The dataset was collected through GLEON and has been used previously in large-scale limnological analyses (Solomon et al 2013, Rose et al 2014. Additional information about the lakes can be found in Solomon et al (2013).…”