The following projects demonstrate the use of GLEON CDI technologies and ideas in lake science. Each project is highly collaborative with many GLEON co-authors, uses data from GLEON sites, depends on GLEON CDI infrastructure to obtain data and/or execute models, and implements sophisticated numerical simulation or other modeling approaches innovated for the specific project. These projects have not only benefited from advanced cyber-infrastructure, but have also informed the evolution of GLEON cyber-infrastructure.
Predicting Phytoplankton Blooms
Emily Kara (Univ. Wisconsin), and many other GLEON co-authors
Phytoplankton dynamics in lakes are notoriously difficult to predict at the sub-seasonal scale, because they are influenced by a complex set of physical, chemical, and community interactions, each of which responds differently to a suite of exogenous drivers. At what space and time scales can we predict phytoplankton dynamics well? What are the essential ecosystem processes necessary for making good predictions, and what processes are we missing? To what extent are our traditional sampling techniques, as well as new measurements from sensor networks, informative to the modeling effort? In this study, we use a coupled hydrodynamic-water quality numerical model (DYRESM-CAEDYM) to simulate phytoplankton dynamics in Lake Mendota, Wisconsin. We test multiple nutrient load and weather scenarios, as well as initial phytoplankton community composition to generate a broad range of ecosystem predictions. Model predictions were compared with traditional limnological observations and sensor measurements to determine the time scales at which the model predicted well. Although the simulations recreated well the observed variability in lake physics and chemistry at scales ranging from days to months, the model did not reproduce well the short-term dynamics in phytoplankton. Changes in external loads and weather did not change phytoplankton dynamics, rather they changed the timing and magnitude of maximum phytoplankton biomass. The model reproduced well the seasonal changes in phytoplankton biomass, however, controversy remains over explaining the variability in shorter time periods for which chlorophyll fluorescence from sensors was substituted for biomass. To what extent do the model’s shortcomings at high resolution reflect model uncertainty versus unknown relationships in the observation data between chlorophyll fluorescence and phytoplankton biomass?
Kevin Rose (Miami Univ.), Jordan Read (Univ. Wisconsin, CEE), Luke Winslow (Univ. Wisconsin, CFL), Emily Kara (Univ. Wisconsin, CEE).
Diel fluctuations of dissolved oxygen (DO) in the surface mixed layer of lakes are often used to quantify the metabolic balance of these ecosystems. The general metabolism model is based on the assumption that daytime increases in DO are driven by gross primary production (GPP) and nighttime drops in DO are driven by respiration (R). The balance of these two processes is called net ecosystem production (NEP). Critical to interpreting estimates of lake metabolism is an understanding of the conditions when the traditional metabolism model succeeds and the conditions in which the model fails to accurately predict diel DO dynamics, as well as the drivers that facilitate uncertainty in parameter estimates within the model. Here we use data from 22 GLEON sites to test how characteristics of thermal stability predict uncertainty in these parameter estimates. We use outputs from Lake Analyzer, another GLEON product, to estimate stability parameters. Additionally, we highlight the gradients of characteristics of lakes across which we observe differences in mean and median uncertainty parameters. This collaborative manuscript, an output from discussions at the GLEON 10 meeting in Torres Brazil in May 2010, is currently in preparation with plans to submit to Limnology and Oceanography Methods by December 2010.
Regional Carbon Cycling
Paul Hanson (Univ. Wisconsin, CFL), David Hamilton (Univ. Waikato, NZ), Emily Stanley (Univ. Wisconsin, CFL), Nick Preston (Univ. Wisconsin, SAGE), Owen Langman, Emily Kara (Univ. Wisconsin, CEE)
What is more important in lake carbon cycling – the nature of the load or the nature of the lake? The fate of organic carbon (OC) loads to lakes depends on both the recalcitrant nature of the load and the internal lake characteristics, such as residence time, exposure to light, temperature-mediated respiration, and the productivity of the lake itself. Unfortunately, few of these characteristics have been quantified at the ecosystem scale, leading to high uncertainty about the fate of OC in lakes. Is it stored in the lake, mineralized and sent to the atmosphere as CO2, or flushed downstream to the next ecosystem? To answer these questions, we simulated 16 lakes across orthogonal gradients of two key lake attributes, lake size (which influences residence time, and mixing, for example) and lake color (which influences water temperature and light penetration, for example). The gradients approximately covered those found in a survey of the 7,500 lakes in the Northern Highland Lake District of northern Wisconsin. For each lake, we simulated three different load recalcitrance levels, resulting in 48 simulations. Four of the simulations were compared with study lakes from the NTL LTER and used as calibration points in the gradients. We found that the nature of the lake and the nature of the load were about equally important in determining the fate of the OC load. An assumed recalcitrance level of the OC was just as important as the lake’s water temperature, color and mixing regime combined. For lakes with extremely short or extremely long residence times, uncertainty in other load or lake characteristics was not important. However, in lakes with moderate residence times of about 1-3 years, uncertainty in our quantification of lake characteristics or uncertainty in load recalcitrance lead to high uncertainty about the fate of the carbon load.
Loons and Water Quality
Paul Hanson (Univ. Wisconsin, CFL), John Walker (USGS), Randy Hunt (USGS), Mike Meyer (WI DNR)
Will northern Wisconsin lakes still support loon populations 50-100 years from now? In collaboration with scientists from the USGS and the WI Department of Natural Resources (DNR), who are studying the future of the common loon, we are modeling lake water quality under future climate scenarios. Loons are visual feeders, relying on high water clarity to spot and capture their prey. By using climate predictions and the resulting changes in surface- and ground-water hydrology, we are modeling lake water quality far into the future. A fundamental challenge is the absence of predictions from watershed models for organic carbon, nitrogen, and phosphorus exports from watersheds to lakes. As a result, viable scenarios for future hydrology and weather do not have commensurate predictions for two factors that affect water clarity the most in this region – lake dissolved organic carbon concentration and phytoplankton biomass. We can, however, use simple steady state models to generate all the possible combinations of nutrient loads and their steady-state values in lakes, in effect creating a future ‘nutrient surface’ over which lakes might exist. For each of the 30 study lakes, we will use climate scenario and hydrology data to drive simulations, and run the simulations over the full surface of nutrient load and steady state conditions. This will yield a large variety of predictions for water clarity, anoxia, phytoplankton blooms, and water temperatures that could influence invasibility by less desirable fish prey, as well as water quality conditions that affect how well loons can spot their prey.
Automatic Model Discovery
Ken Chiu group (SUNY Binghamton)
There has been significant advancements in harnessing computational power to perform simulations at unprecedented spatial and temporal scales. Less well-developed, however, is leveraging computational thinking at higher stages of the scientific process. In particular, often the scientist is not sure of the mathematical form of the equations themselves. In this project, we seek to use evolutionary computing algorithms to automatically explore different model equations, testing each model against the observed data. Differential equations that fit the data well are allowed to continue, while ones that are inaccurate die out.
General Purpose computation on Graphical Processing Units (GPGPU) for Individual- and Particle-Based Modeling
Ken Chiu group (SUNY Binghamton)
In recent years, hardware companies such a Nvidia have realized that graphical processing units (GPUs) are ideal platforms for massively parallel computation, and have created languages and APIs to allow users to run their computations on the GPU. These GPUs approach 1000 processors in a single board, potentially greatly speeding up simulations. Particle simulations have show been amenable to GPGPU, but significant work remains to be done to make this technology available to the domain scientist.