Critical Ecosystem Studies Initiative Final Project Report

I. Introduction

This report describes the activities of the second year of Critical Ecosystem Studies Initiative (CESI) research into the use of statistical models to simulate salinity in Florida Bay. The activities of the first year of CESI work are presented in Salinity Simulation Models for North Florida Bay Everglades National Park (Marshall, et al 2003a). Because this report describes follow-on activities, reference to the first year report may be needed to understand all of the background for the second year of work. During the first year of this investigation, two types of statistical modeling procedures were found to be well-suited for use with time series data - SARIMA models (seasonal autoregressive integrated moving average models) and multivariate linear regression models (MLR models). SARIMA models were found to be useful for one-step forward predictions, but for other simulation purposes, MLR models were found to be much easier to use and almost as robust to the idiosyncrasies of time series data. Since the models are intended for use with the output from the South Florida Water Management District (SFWMD) Everglades watershed model (South Florida Water Management Model, or 2X2 model), which simulates hydrologic conditions in south Florida beginning in 1965, MLR models were selected for further development.

Tasks in the original Project Description for the Year 2 work are summarized as follows:

1. Coordinate with SFWMD to obtain 2X2 model data and evaluate the uncertainty in the 2X2 model simulations;
2. Obtain any other data that may be needed to use MLR models with 2X2 model data;
3. In conjunction with ENP staff, prepare MLR models of salinity, water level, or flow;
4. Simulate salinity, water level, or flow using the 2X2 Natural System Model and other appropriate input parameters;
5. Decode a previously prepared SARIMA model; and
6. Prepare draft and final reports.

As the project progressed considerable experience and feedback were gained using MLR salinity models. Additionally, when the uncertainty in the 2X2 model output was evaluated, the structure of the MLR salinity models was modified, as described further in the following sections of this report.

From the onset of the CESI work on statistical models, it was hoped that they would prove useful for simulating salinity in Florida Bay for the Initial CERP (Comprehensive Everglades Restoration Plan) Update (ICU) evaluations. However, when the Year 2 work began, it did not appear that the MLR models would be ready for this purpose, and the scope of the Year 2 project was made intentionally broad to investigate the use of MLR models to simulate other parameters besides salinity, such as water level or flow. As the project progressed it became clear that MLR models were capable of making acceptable simulations of salinity such that different water management schemes could be evaluated, and that the revised schedule for the ICU evaluations was going to make it possible for MLR models to be used for the evaluations. Therefore, the work being done was concentrated on the development of MLR models for that purpose (ICU evaluations).

The first tasks were completed as scheduled. Residual plots were examined and the updated dataset was assembled. In the midst of completing the project tasks, a need for the MLR models developed at Everglades National Park for use with the Interim Operational Plan (IOP) evaluation Congressional Report. ENP was tasked with analyzing the water management regimes that had been modified to lessen the impact of flow diversions on the Cape Sable seaside sparrow. MLR models were developed for use with these evaluations, and valuable experience was gained that has benefited the CESI project. In order to complete the IOP evaluations, a six-month extension of this CESI project was requested and granted. The IOP evaluation model development procedure also allowed the project dataset used for model development to be lengthened.

One committee that is charged with completing the ICU evaluations is the Southern Estuaries Sub-team of RECOVER. Beginning in the spring of 2003, the Principal Investigator has been coordinating with the Sub-team, preparing to use the models developed by this CESI project for evaluating the established salinity performance measures for Florida Bay and the southwest coast. At the time of preparation of this report, the models presented herein are intended to be used by the Sub-team in this manner.

When work re-started on this CESI project following the IOP evaluation activities, the tasks to be completed were officially modified to take into account the focus on modeling for ICU evaluation purposes. The revised Project Description included the following revised list of tasks:

1. Contact / meet with SFWMD staff to coordinate the acquisition of 2X2 model output and additional information about the modeling procedure.
2. Acquire the other data needed to create a complete input data set for running the multivariate linear regression models with 2X2 model output, including the historical record for wind at Key West and Miami weather stations.
3. Eliminate flow parameters from MLR salinity models.
4. Meet with the Southern Estuaries Sub-team to obtain their needs for MLR salinity models for ICU performance measure evaluations.
5. Prepare MLR salinity models for Joe Bay, Little Madeira Bay, Terrapin Bay, Garfield Bight and North River using 2X2 model output to calibrate the MLR salinity models. This task was completed. However, because the SFWMD updated the 2X2 model output subsequent to their development, these models were rendered obsolete. All future MLR salinity models will be developed from real (observed) data.
6. Adapt the IOP models prepared from observed data for use with the Southern Estuaries Sub-team performance measure evaluations by expanding the data used for model development, where possible for Little Madeira Bay, Terrapin Bay, Whipray Basin, Butternut Key, and Duck Key.
7. Prepare new MLR models using the longest period of record available for Taylor River, Little Blackwater Sound, Highway Creek, and Bob Allen Key.
8. Run simulations at all stations using the following 2X2 model runs: NSM 4.5, NSM 4.6, 95 Restudy, and 2000 CERP. These are the same runs being made by the Southern Estuaries Sub-team at other stations.
9. Evaluate the level of uncertainty in the models and in the simulations. Some statistical tests that may be used include the mean error, mean absolute error, root mean square error, maximum absolute error, relative mean error and relative absolute mean error.
10. Prepare draft and final Project Reports describing the activities that were completed and present the findings.

Details on the activities of these revised tasks are presented in this report.

II. Study Area and Data Set

Figure 1. The Everglades and Florida Bay Study Area Showing Monitoring Stations [larger image]

The study area for this CESI project encompasses northeastern, north, and central Florida Bay; the extreme southwestern coast of the Florida; and the Everglades watershed within Everglades National Park. This modeling effort utilized data that have been collected at 15 to 60 minute increments and averaged to daily and monthly values. Salinity data is taken from the ENP Marine Monitoring Network (MMN) data base. The stage data are ENP Physical Monitoring Network Everglades water levels. Details about these data can be found in Everglades National Park (1997a and 1997b), and Smith (1997, 1998, 1999, and 2001). To these data other time series data were added, including wind data from the National Weather Service (Southeast Regional Climate Center), and water level data collected at Key West from the National Ocean Service. Wind data from Key West and Miami were used as these locations had the longest continuous records for wind and were considered to be representative of the regional wind patterns. Sea level data from Key West were considered to be representative of the average effect of oceanic water level influences, and, to some extent, the average water level patterns within Florida Bay.

The locations of each of the monitoring stations where water level and salinity data were collected are presented in Figure 1. The salinity monitoring stations for which MLR salinity models were prepared as part of this CESI study or the IOP evaluation (shown on Figure 1) are as follows:

1. Joe Bay
2. Little Madeira Bay
3. Terrapin Bay
4. North River
5. Whipray Basin
6. Duck Key
7. Butternut Key
8. Taylor River
9. Highway Creek
10. Little Blackwater
11. Bob Allen Key
12. Long Sound.

Continuous water level records in the Everglades begin in the 1950's in some locations, but most stage records date from the 1990's. Continuous salinity data extend back to 1988 at several locations in northeast Florida Bay. Because the shortest data record (for E146) begins on March 24, 1994, the period of data used for most of these modeling activities begins on this date. The period of record extends through October 31, 2002, which means that there are 3143 daily values in a record with no missing data. In reality, most data sets contained some missing values. Information on the parameters that were used for the modeling activities is presented in Table 1.

View the PDF version (2.1 MB) of this publication. (Please note: you will need the free Adobe Acrobat Reader in order to view this file.) Appendices are available separately and can be obtained by contacting Frank E. Marshall, III