Enabling an Operational Coupled Modelling and Observing System to Assess Water Quality in the Lake George, New York Watershed

Lake George is located in the Adirondack State Park region of upstate New York, about 350 km north of New York City. It is a glacial, oligotrophic water body and is unique among fresh water lakes because of its ecology, geographic orientation, historical importance, and tourism-driven economic impact. Further, this narrow, dimictic lake (51.5km long with an average width of 2.1km, surface area of 117 km², volume of 2.28 km³, maximum depth of 57m and average depth of 21m) of unusually high clarity has been experiencing ecological changes in recent years with the influx of invasive species and increasing levels of salt (NaCl) from road de-icing agents and the accompanying deterioration of water quality. These changes have been observed through longitudinal studies in the region over the last few decades.

We are studying these changes under the auspices of The Jefferson Project at Lake George, a collaborative effort between IBM Research, Rensselaer Polytechnic Institute and The FUND for Lake George. A core component of this project is the coupled system we developed to both observe and physically model the atmospheric, hydrological, hydrodynamic and ecological aspects of the lake and the surrounding region, which is now running operationally. This system includes a real-time, multi-modal observing system composed of in-situ sensors for atmospheric, stream and lake measurement that supports adaptive sampling driven by forecasted conditions from the models. There are over 50 platforms supporting over 500 sensors in the watershed. These data, as appropriate, are used for model verification, and to improve model initial conditions via data assimilation. Another goal for the coupled system is to drive ecological models that explain the interaction of flora and fauna in the lake as a food web. In addition, the coupled physical models drive geometric modelling to enable fixed and interactive visualizations of the watershed. To support both the modelling and the observing systems, there is an underlying cyberinfrastructure that includes an high-performance computing cluster and storage system, and data acquisition hardware and software. To enable data sharing and software reuse, community data models have been adopted for the data generated by this coupled system. As such, this capability is a testbed for addressing related issues in other lake watersheds, estuaries and similar ecosystems.

Atmospheric forcing is considered for both lake circulation and runoff models, the latter of which hydrologically forces the lake. To address the former, we build upon the on-going work with IBM Deep Thunder^TM. It is based, in part, on the Advanced Research Weather (ARW) dynamical core of the Weather Research and Forecasting (WRF) community model. It is run operationally daily (initialized at 00 UTC) nested to 333m horizontal resolution for 36 hours for the watershed with high vertical resolution in the lower boundary layer. It operates in the very large eddy (VLES) regime with no boundary-layer parameterization.

For the runoff, a two-dimensional hydrological model has been implemented. For its daily update, it is driven operationally by the 333m-resolution weather model and has been scaled to utilize 10m-resolution topographic data for the watershed. It supports fully dynamic routing with flow driven by both precipitation and snow melt, including over 400 stream networks and 78 outlets with a total length of over 1000 km. The model has been extended for the transport of dissolved salt, originating from road surfaces. One of the applications is to consider potential nutrient loading in the lake driven from storm water runoff.

For the lake circulation, we build upon the on-going work with IBM DeepCurrent, a three-dimensional, hydrodynamic model with a vertical hydrostatic approximation operationally implemented at high-resolution for Lake George (approximately 50 m horizontal, and 2.8 m vertical). It is based, in part, on the Environmental Fluid Dynamics Code (EFDC) community model. The model utilizes data derived from an high-resolution bathymetric survey of the lake, and has been extended to address chlorine ion transport as an indicator of water quality given the tendency of sodium to bind with soil en route to the lake. Operationally, it is driven by the aforementioned meteorological and hydrological models to produce a 36-hour forecast, once per day.

There is an inherent scale gap between each of these models and how a food web model would be driven, which needs to be addressed. The first aspect was the operational implementation of Deep Thunder at the aforementioned VLES scale. To consider issues with the sub-surface flow to properly inform the hydrological model, an off-line land-surface model (NOAH-MP) has been deployed to establish the base flow. After an 18-month spinup, it produces a 168-hour forecast each day on the Deep Thunder 333m nest.

To enable better coupling between the meteorology and land surface, and eventually groundwater, a three-dimensional, hydrological model is being implemented by adapting the WRF-Hydro community model. It is running operationally on-line at 33m resolution for the watershed forced by the Deep Thunder output. Eventually, it will replace both the off-line use of NOAH-MP and the aforementioned 2d runoff model.

In order to better understand the movement of biota in the lake, particle tracking and ecological models have been implemented by adapting the Stanford Unstructured Nonhydrostatic Terrain-following Adaptive Navier-Stokes Simulator (SUNTANS) community model. Given the triangular computational mesh derived from the high-resolution survey, the horizontal resolution varies from ~10m (near coastlines) to ~50m (in open water). Operationally, it is similarly driven as DeepCurrent to produce a 36-hour forecast, once per day. The particle tracking is done as an interactive post-process leveraging the simulated velocity fields.

A simple ecological model has also been developed and deployed operationally that considers nutrients, phytoplankton, zooplankton, small and large detritus, and oxygen. It uses the aforementioned hydrodynamic and meteorological forcings to consider relative growth of phyto- and zoo-plankton. Pytoplankton is represented via a proxy, chlorophyll-a.

We will present an overview of each of these models and the observing system along with the results to date, including the model coupling and computing infrastructure required for operations. We will also discuss the automation for the coupled execution, including monitoring, visualization and validation. In addition, we will outline recommendations for future work.