Metabolism, or carbon reduction through photosynthesis (gross primary production, GPP) and oxidation through respiration (R), is an important feature of ecosystems. In lakes, metabolic balance can be quantified by comparing autochthonous GPP and ecosystem R of autochthonous C, as well as allochthonous sources derived principally from the surrounding landscape. This balance influences critical ecosystem characteristics, such as lake color and mixing depth, pH and dissolved gas concentrations, and aquatic food webs. At a landscape scale, lake metabolism influences how lakes store, export or vent carbon to the atmosphere.
Quantifying lake metabolism has challenged ecologists for decades, because observable variables important to metabolism are controlled by a complex suite of processes. For example, dissolved oxygen (DO) is a variable commonly measured in lake sensor networks, and as a substrate of R and a product of GPP has been used to estimate metabolism (Odum 1956). However, DO time series show complex patterns at multiple scales, suggesting multiple processes control its dynamics, confounding our ability to infer metabolism (graph at right)
How complicated does a metabolism model need to be to explain DO variability?
In this study, we fit five different metabolism models to sensor network data from two lakes, Crystal Bog Lake and Trout Lake, both of which are study systems of the NTL LTER. Alternative models reflect a range of complexities (Table 1), and include as the simplest case F, GPP as a linear function of light and constant R based on night-time measurements (Model 1, two parameters). In the most complex case, GPP and R were modeled as non-linear processes (Model 5, five parameters). Model parameters were estimated using PEST software (Parameter Estimation, Inc., Sandy, UT). To improve the probability of locating the global minimum, we performed minimization nested within a range of the possible starting conditions for parameters. The goal was to determine whether complex metabolism models better predicted observed DO.
Results: When models were fit to observed DO data from sensors, the simplest model predicted DO as well as the most complex one (graph at left), even though lakes differed substantially in size and trophic state. Trophic differences were reflected in metabolic balance. For Crystal Bog, a dystrophic lake, R and GPP were 4.6 and 2.7 mg DO L-1 d-1, respectively. In Trout Lake, which is oligotrophic, mean GPP (0.19) was slightly greater than R (0.14). Estimates among models for a lake were similar, and tended to differ from the mean by less than 5%.
Although metabolism explains much of the diel pattern, some variance in DO remained unexplained. Internal waves, diurnal mixing, and weather-induced changes in thermal structure appear to account for some residual variance. The implications for biological process we were unable to identify in fitting models to observed DO are that their affects on DO fluxes may be subtle and integrated into the parameter estimates of the simple models. Their contributions to dynamics in DO are probably less pronounced than the contributions of physical processes not modeled.
Hanson, P.C., S.R. Carpenter, N. Kimura, C. Wu, S.P. Cornelius, and T.K. Kratz. 2008. Evaluation of metabolism models for free-water dissolved oxygen methods in lakes. Limnol. Oceanagr. Methods. 6:454-465.
Odum, H. T. (1956), Primary production in flowing waters, Limnol. Oceanogr., 1, 103-117.