Study Area
Tampa Bay is located on the west-central GOM coast of the Florida Peninsula. Its watershed is among the most highly developed regions in Florida (Fig. 1). More than 60% of land use within 15 km of the Bay shoreline is urban or suburban (SWFWMD 2018). The Bay has been a focal point of economic activity since the 1950s and currently supports a mix of industrial, private, and recreational activities. The watershed includes one of the largest phosphate production regions in the country, which is supported by port operations primarily in the northeast portion of the Bay (Greening et al. 2014).
Current water quality in Tampa Bay is dramatically improved from the degraded historical condition. Nitrogen loads into the Bay in the mid-1970s have been estimated as 8.9 × 106 kg/year, largely from wastewater effluent (Greening et al. 2014). In addition to reduced esthetics, hypereutrophic environmental conditions were common and included elevated chlorophyll-a and harmful algae, and reduced bottom water–dissolved oxygen, water clarity, and seagrass coverage. Yields for some commercial and recreational species were also depressed, although emergent tidal wetland loss and fisheries practices (i.e., widespread Bay trawling and gill-netting) likely also contributed to declines (Comp 1985; Lombardo and Lewis 1985).
A long-term monitoring program in Tampa Bay has been instrumental in assessing and tracking restoration efforts. In the early 1970s, the initial baywide ambient monitoring program was established by a local environmental leader (Roger Stewart), which was subsequently institutionalized through State legislation by the creation of the Environmental Protection Commission of Hillsborough County. This occurred largely in response to citizen outcry of the Bay’s deteriorating ecology (Greening et al. 2014). Ongoing local support for this program has remained since 1972, and other local, municipal governments have created complementary water quality monitoring programs, all of which now support water quality assessments and management efforts spearheaded by the Tampa Bay Estuary Program (Sherwood et al. 2016). Some of the key attributes supporting the maintenance of this long-term monitoring program are summarized in Schiff et al. (2016) and Gross and Hagy (2017).
Nearly 900 public and private projects to improve water quality have also been completed in Tampa Bay and its watershed over the past four decades. These projects represent numerous voluntary (e.g., coastal habitat acquisition, restoration, preservation, etc.) and compliance-driven (e.g., stormwater retrofits, process water treatment upgrades, site-level permitting, power plant scrubber upgrades, improved agricultural practices, residential fertilizer use ordinances, etc.) activities. Linking data from the long-term monitoring program with data on these projects will provide an understanding of how the cumulative effects of small-scale restoration have contributed to water quality relative to the historical water infrastructure upgrades.
Data Sources
Several databases were combined to document restoration projects in Tampa Bay and its watershed. Each database was unique and no overlap in documented projects was observed. Data from the Tampa Bay Water Atlas (version 2.3, http://maps.wateratlas.usf.edu/tampabay/; TBEP 2017) documented 253 projects from 1971 to 2007 that were primarily focused on habitat establishment, enhancement, or protection along the Bay’s immediate shoreline or within the larger watershed area. Examples include restoration of salt marshes and mangroves, exotic vegetation control, and conversion of agricultural lands to natural habitats. Information on an additional 265 recent (2008–2017) projects was acquired from the US EPA’s National Estuary Program Mapper (https://gispub2.epa.gov/NEPmap/). This database provides only basic information, such as year of completion, geographic coordinates, general activities, and areal coverage. Data from the TBEP Action Plan Database Portal (https://apdb.tbeptech.org/index.php) documented locations of infrastructure improvement projects, structural best management practices, and policy-driven stormwater or wastewater management actions. This database included 368 projects from 1992 to 2016 for county, municipal, or industrial activities, such as implementation of best management practices at treatment plants, creation of stormwater retention or treatment controls, or site-specific controls of industrial and municipal point sources.
For all restoration datasets, shared information included the project location, year of completion, and classification of the restoration activity. We developed and applied a two-level classification scheme that described each restoration project as (1) a habitat or water infrastructure improvement and (2) more specifically as enhancement, establishment, or protection for habitat or as nonpoint or point source controls for water infrastructure. These categories were used to provide a broad characterization of restoration activities that were considered to contribute to improvements in water quality over time. The final combined dataset included 887 projects from 1971 to 2017 (Fig. 2). Projects with incomplete information were not included in the final dataset.
Water quality data in Tampa Bay have been collected consistently since 1974 by the Environmental Protection Commission of Hillsborough County (Sherwood et al. 2016; TBEP 2017). Data were collected monthly at 45 stations using a water sample or monitoring sonde at bottom, mid-, or surface depths, depending on parameter. The locations of monitoring stations were fixed and cover the entire Bay from the uppermost mesohaline sections to the lowermost euhaline portions that have direct interaction with the GOM (Fig. 1). Mid-depth water samples at each station are laboratory processed immediately after collection. Chlorophyll-a (μg/L) and total nitrogen (mg/L) measurements at each site were used for analysis, totaling up to 515 discrete observations for each station. Total nitrogen concentrations were included in initial data assessments, as this nutrient is considered limiting in Tampa Bay (Greening et al. 2014).
Data Synthesis and Analysis Framework
The five subcategories for each project (habitat enhancement, establishment, and protection; nonpoint and point source controls) were separately evaluated to describe the likelihood of changes in water quality associated with each type. Water quality monitoring sites were matched to the closest selected restoration projects, and changes in the water quality data were evaluated relative to the completion dates of the selected projects. Spatial and temporal matching can be accomplished using several methods that vary in complexity. For example, hydrologic distances or other non-Euclidean distance weightings by watershed topology can be used to link measurements to modeled locations in space (Curriero 2006; Gardner et al. 2011). However, we adopted a relatively simple approach with limited data requirements to maximize potential applications in other regions (e.g., no hydrology data are needed, only spatial location). The matchings began with a spatial joint wherein the Euclidean distances between each water quality station and each restoration project were quantified. The restoration projects closest to each water quality station were identified using the distances between projects and water quality stations. The distances were also grouped by the five restoration project types (i.e., habitat protection, nonpoint source control, etc.) such that the closest n sites of a given project type could be identified for any water quality station (Fig. 3).
For each spatial match, temporal matching between water quality stations and restoration projects was obtained by subsetting the water quality data within a time window before and after the completion date of each restoration project (Fig. 4). For the closest n restoration sites for each of the five project types, two summarized water quality estimates were obtained to quantify a before and after estimate of chlorophyll-a associated with each project. The before estimate was the average of observations for the year preceding the completion of a project and the after estimate was the average of observations for a selected window of time (e.g., 5 years) that occurred after completion of a project. The before estimate for the year prior established the basis of comparison for the water quality estimates in the selected window of time after project completion, where the latter could be manually changed to characterize a potential duration of time within which water quality could improve after project completion. The final two estimates of the before and after values of the five types of restoration projects at each water quality station were based on an average of the n closest restoration sites, weighted inversely by distance from the monitoring station. Lastly, no data were available on project duration and we assumed that the year associated with each project was generally inclusive of project implementation and completion. Time windows that overlapped the start and end date of the water quality time series were discarded.
Change in water quality relative to each type of restoration project was estimated as:
$$ \Delta \mathrm{WQ}=\frac{\sum_{i=1}^n\hat{\mathrm{wq}}\in \mathrm{win}+\mathrm{pro}{\mathrm{j}}_{i,\mathrm{dt}}}{n\times \mathrm{dis}{\mathrm{t}}_{i\in n}}-\frac{\sum_{i=1}^n\hat{\mathrm{wq}}\in \mathrm{pro}{\mathrm{j}}_{i,\mathrm{dt}}-\mathrm{win}}{n\times \mathrm{dis}{\mathrm{t}}_{i\in n}} $$
(1)
where ΔWQ was the difference between the after and before averages for each of n spatially matched restoration projects. For each i of n projects (proj), the average water quality (\( \hat{\mathrm{wq}} \)) within the window (win) either before (proji, dt − win) or after (win + proji, dt) the completion date (dt) for project i was summed. The summations of water quality before and after each project were then divided by the total number of n matched projects, multiplied by the distance of the projects from a water quality station (disti ∈ n). This created a weighted average of the before–after estimates for each project that was inversely related to the distance from a water quality station. A weighted average by distance (or parametric distance weights; Sickle and Johnson 2008) was used based on the assumption that restoration projects farther from a water quality station will have a weaker association with potential changes in chlorophyll-a. The total change in water quality for a project type was simply the difference in weighted averages. This process was repeated for every station (Fig. 5). Overall differences between project types were evaluated by ANOVA F tests, whereas pairwise differences between project types were evaluated by t tests with corrected probability values for multiple comparisons.
One of the key assumptions of our approach is that restoration projects will benefit water quality through a reduction in chlorophyll-a. We make no assumptions about the expected magnitude of an association given that the model does not describe a specific mechanism of change, nor do we make any explicit assumption about the direction of change (i.e., two-tailed hypothesis tests were used), although a general assumption was that chlorophyll-a would decrease over time in agreement with known changes in water quality. However, we hypothesized that the magnitude of chlorophyll-a changes varies by project type and number of projects or length of time window evaluated. An expected outcome is that explicit, quantitative conclusions can be made about the relative differences between projects types, particularly regarding how additional projects of a particular type could benefit water quality and within what general time windows a change might be expected (Diefenderfer et al. 2011).
The model was also designed to quantify cumulative relationships of restoration projects with water quality at different spatial scales. In Eq. (1), the association of a restoration type with chlorophyll-a is estimated for one water quality station, whereas estimates from several water quality stations can be combined to develop an overall description of a particular restoration type as it applies to an areal unit of interest, potentially over broad regional scales. For example, estimated associations of point source control projects with each water quality station in the Bay can be combined to develop an overall narrative of how these projects could (assuming a causal relationship) influence environmental change across the entire Bay. Estimates across stations were evaluated to describe associations in baywide improvements from various restoration project types throughout the watershed. Estimates were also evaluated by individual Bay segments that have specific management targets for chlorophyll-a concentration (Florida Statute 62-302.532; Janicki et al. 1999). Stratification by Bay segments provided an alternative context for interpreting the results based on areal differences between segments and how restoration projects varied in space and time. Evaluating the results at different scales can also provide insights into potential (or lack of) stressors and processes controlling the impacts, which can help prioritize management actions by location (Diefenderfer et al. 2009; Thom et al. 2011).
The analysis of each project type was bounded by two key parameters in Eq. (1). These include n, the number of spatially matched restoration projects used to average the cumulative estimate of each project type, and win, the time windows before and after a project completion date that were used to subset a station’s water quality time series. These boundaries affected our ability to characterize each restoration project type with water quality changes. Identifying values that maximized the difference between before and after water quality measurements was necessary to quantify how many projects were most strongly associated with a change in water quality, the time within which a change is expected, and the magnitude of an expected change between project types. For simplicity, we evaluated different combinations of 5- or 10-year time windows from the date of each project completion and the 5 or 10 closest projects to each water quality station. All analyses were conducted with customized scripts created for the R statistical programming language (RDCT 2018).
Testing Effects of Restoration Dates and Location
Because of the documented improvements in water quality in Tampa Bay, a concern with our approach is that any association between restoration projects and chlorophyll-a may result from correlations between the two parameters, confounding a true demonstration of water quality improvements in relation to restoration activities. To address this challenge, estimated changes in chlorophyll-a were evaluated in response to temporal and spatial matching with restoration projects, as above, but with random date and location assignments for each restoration project that were then compared to the actual results. An expected outcome of randomization is that no differences are observed between project types and that all associations between projects and chlorophyll-a changes should reflect the continuous decline of chlorophyll-a over time, as observed in the independent water quality record. In other words, the randomization creates a null model where the estimated effects of restoration projects would not differ from a simple evaluation of trends in the raw data—slicing the observed time series by random dates and evaluating before/after averages with random projects is expected to reflect the known decline of chlorophyll-a in the raw data. Alternatively, evidence that our framework provides meaningful results would be supported by differences in chlorophyll-a changes between project types and the timing associated with the changes.