Project Summary Sheet

Project Title: Hydrology Monitoring Network: Data Mining and Modeling to Separate Human and Natural Hydrologic Dynamics
Project Start Date: 2005 Project End Date: 2007
Web Sites: http://sofia.usgs.gov/ http://sofia.usgs.gov/projects/hydro_monnet/
Location (Subregions, Counties, Park or Refuge): Dade and Monroe Counties, Everglades National Park, Central Everglades, A.R.M. Loxahatchee National Wildlife Refuge
Funding Source: USGS Greater Everglades Priority Ecosystems Science (GE PES) Initiative
Principal Investigator(s): Paul A. Conrads, Edwin A. Roehl
Project Personnel: Mark Lowery, Larry Harrelson
Supporting Organizations: USGS-South Carolina Water Science Center

Associated / Linked Projects: Estimation of Critical Parameters in Conjunction with Monitoring of the Florida Snail Kite Population, The Everglades Depth Estimation Network (EDEN) for Support of Ecological and Biological Assessments, Freshwater Flows into Northeastern Florida Bay, Southern Inland and Coastal Systems (SICS) Model Development, Enhanced Water Quality Monitoring and Modeling Program for the A.R.M. Loxahatchee National Wildlife Refuge

Overview & Objective(s): The emerging field of Data Mining addresses the issue of extracting information from large databases. It is comprised of several technologies that include signal processing, advanced statistics, multi-dimensional visualization, machine learning (including artificial neural networks (ANN)), and Chaos Theory. Data Mining can solve complex problems that may be unsolvable by any other means. The data from the CERP monitoring is a tremendous resource for addressing the critical questions for restoring the South Florida ecosystem. Estuarine systems are difficult systems to analyze due to the complexity of environmental factors occurring simultaneously. To enhance the evaluation of the CERP data base, there is an immediate need to apply new methodologies to systematically analyze the data set to answer critical questions such as relative impacts of controlled freshwater releases, tidal dynamics, and meteorological forcing on streamflow, water level, and salinity. This project will directly address the data analysis issues outlined above.

The first year of the Data Mining Analysis Project addressed these issues by demonstrating how data mining techniques can be applied to the Everglades data bases and ecological studies. Three studies were selected for the demonstration work - Freshwater Inflows to Northeastern Florida Bay (Mark Zucker, Clinton Hittle), Estimation of Critical Parameters in Conjunction with Monitoring the Florida Snail Kite Population (Wiley Kitchens), and Southern Inland and Coastal Systems (Eric Swain). In addition, other projects were identified where data mining techniques can be applied during Years 2 and 3 of the project. In the second year of the Data Mining Analysis Project, additional projects included working with PI's involved with the Everglades Depth Estimation Network (Pamela Telis, Roy Sonenshein, John Jones, and Leonard Pearlstine) and beginning work on the analysis and modeling of the water level and water quality of the A.R.M. Loxahatchee National Wildlife Refuge (Laura Brandt and Mike Waldon).

Status: The hindcasted hydrology for 17 vegetation transects of the Snail Kite study in WCA-3a developed during Year 1 were incorporated into a Decision Support System (DSS) in Year 2. The Excel application incorporates the historical database, ANN hindcast models (including decorrelation models), model simulation controls, statistical output summaries, streaming graphics, and model output into a easily disseminated spreadsheet package. Output from regional hydrologic routing models (for example the “2 by 2” model) can be used as input to the DSS system to analyze water depth response at the Snail Kite vegetation transects. In Year 1, methodologies for estimating water levels and water depths at ungaged areas using ANN models were developed using a data set from WCA- 3A. The approach utilizes static variables of location and percent vegetation and dynamic variables of water levels at known locations. During Year 2, the application of the methodology to the EDEN network of the Greater Everglades was undertaken. The first step in scaling up the methodology was to cluster the water level data of the EDEN data into classes of similar behavior. The estimation methodology can then be applied to classes within compartments in the Everglades. The 6-year EDEN data set was clustered and preliminary models for estimating at ungaged sites have been developed for WCA-3a. ANN models of salinity response for the 5 USGS gaged tributaries to Florida bay were finalized using the 1996-2004 data. The models were used to analyze the influence of control releases and natural streamflow on the salinity dynamics of 5 tributaries to Florida Bay. In Year 2, we have begun working with water-level and water-quality datasets from the A.R.M. Loxahatchee National Wildlife Refuge to analyze the influence of operations schedule and control of high conductivity water intrusion into the Refuge. Data from 1996 to 2005 has been incorporated into a database for analysis and ANN model building.

Recent & Planned Products: Major products for Year 2 include (1) an Excel based DSS application for of the hindcasted hydrology for the Snail Kite study; (2); cluster analysis of 6-year EDEN database (3) ANN models used to analyze freshwater inflows for 5 tributaries into Florida Bay for natural and anthropogenic components; and (4) 10-year database of water level and water quality database for A.R.M. Loxahatchee National Wildlife Refuge.

Year 2 conference papers (peer reviewed), posters, and presentations:

Papers:

Conrads, P.A. and Roehl, E.A., 2006, Estimating water depths using artificial neural networks, Hydroinformatics 2006, edited by Philippe Gourbesville, Jean Cunge, Vincent Guinot, Shie-Yui Liong, Vol. 3, p.1643-1650

Conrads, P.A. and Roehl, E.A., Daamen, R.C., and Kitchens, W.M., 2006, Using artificial neural network models to integrate hydrologic and ecological studies of the snail kite in the Everglades, USA, Hydroinformatics 2006, edited by Philippe Gourbesville, Jean Cunge, Vincent Guinot, Shie-Yui Liong, Vol. 3, p.1651-1658

Posters:

Conrads, P.A. and Roehl, E.A., 2006, Analysis of the Process Physics of Tributaries to Florida Bay Using Artificial Neural Networks and Three-Dimensional Response Surfaces, Florida Bay and Adjacent Marine Systems Science Conference, Duck Key, Florida, Dec. 11-14, 2005

Conrads, P.A. and Roehl, E.A., 2006, Application of a Dynamic Clustering Algorithm to the Water-Level Hydrographs of the EDEN Hydrologic Network, Greater Everglades Ecosystem Restoration Conference, Orlando, Florida, June 5-9, 2006

Conrads, P.A. and Roehl, E.A., 2006, Estimating Water Depths at Ungaged Locations in the Florida Everglades Using Artificial Neural Networks, Greater Everglades Ecosystem Restoration Conference, Orlando, Florida, June 5-9, 2006

Presentations:

Paul A. Conrads, Ruby Daamen, Edwin A. Roehl, Wiley M. Kitchens, and Christa Zweig, 2006, Using Artificial Neural Network Models to Integrate Hydrologic and Ecological Studies of the Snail Kite Falcon in the Everglades, Greater Everglades Ecosystem Restoration Conference, Orlando, Florida, June 5-9, 2006

Specific Relevance to Information Needs Identified in DOI's Science Plan in Support of Ecosystem Restoration, Preservation, and Protection in South Florida (DOI's Everglades Science Plan): [Page numbers listed below are from the DOI Everglades Science Plan. The Science Plan is posted on SOFIA's Web site: http://sofia.usgs.gov/publications/reports/doi-science-plan/]:

• The study supports the ecological studies of impacts of hydrologic change on the Everglade snail kite habitat and the Combined Structural and Operational Plan project (CSOP) by addressing the needed science for “…refinement of hydrologic targets and operating protocols (p. 63).”
• The study supports the Water Conservation Area 3 Decompartmentalization and Sheetflow Enhancement Project (DECOMP) by addressing the science needed for “...additional research to understand the effects of different hydrologic regimes and ecological processes on restoring and maintaining ecosystem function…” (p.64, 66) by hindcasting hydrologic records.
• The study supports the water project of the Loxahatchee National Wildlife Refuge by addressing links between hydrology, water quality, and ecology in the refuge (p.40, 43).
• The study supports the science needs related to the Combined Structural and Operational Plan (CSOP), including the C111 Spreader Canal, by separating the impacts of controlled flows and tidal dynamics on salinity on the five tributaries in Northeast Florida Bay (p. 66).

Key Findings:

1. The DSS developed for the Snail Kite study shows how disparate data networks (long-term USGS water-level dataset and short-term Snail Kite water-depth dataset) can be integrated to maximize expensive data collection efforts. The Excel based DSS allows ecologist to interrogate hindcasted hydrology to improve the predictive capabilities of their vegetation models and how output for regional models can be incorporated for additional analysis. The development of the DSS shows how complex models can be packaged to meet the needs of end users with a variety of technical backgrounds and abilities.
2. The cluster analysis of the EDEN 6-year dataset grouped gages by similar behaviors within each compartment. Within some compartments there were seemingly anomalous cluster assignments. The water-level behavior at these gages, either due to the local topography, proximity to the canal-levee system, or some other hydrologic or geologic factor, is more similar to the mean behavior of more distant gages than the proximal gages of a different cluster. These gages were presented to hydrologists familiar with the particular compartment and the apparent anomalies were explained by geomorphologic or vegetation characteristics of the area. These anomalies also show the power of the clustering approach to objectively partition data by similar behaviors.
3. Using the ANN models of the salinity response of the five tributaries to Florida Bay, we were able to evaluate the relative contribution of control flows and Florida Bay dynamics using all the historical data. In the analysis of the data, it was determined that the time delays for the daily mean salinity to respond to controlled releases are on the order of 90 to 150 days, depending on the site. Using visualization techniques, the historical interaction between two explanatory variables (for example, flows and water levels) are shown with the response variable (salinity). The approach is an alternative method to analyzing, understanding, and visualizing long-term data of complex systems. The length of these time delays may be of significance for determining the planning horizon for controlling salinity intrusion from Florida Bay.