Application of watershed deposition tool to estimate from CMAQ simulations the atmospheric deposition of nitrogen to Tampa Bay and its watershed

Abstract

The U.S. Environmental Protection Agency (EPA) has developed the Watershed Deposition Tool (WDT) to calculate from the Community Multiscale Air Quality (CMAQ) model output the nitrogen, sulfur, and mercury deposition rates to watersheds and their sub-basins. The CMAQ model simulates from first principles the transport, transformation, and removal of atmospheric pollutants. We applied WDT to estimate the atmospheric deposition of reactive nitrogen (N) to Tampa Bay and its watershed. For 2002 and within the boundaries of Tampa Bay's watershed, modeled atmospheric deposition rates averaged 13.3 kg N ha⁻¹ yr⁻¹ and ranged from 6.24 kg N ha⁻¹ yr⁻¹ at the bay's boundary with Gulf of Mexico to 21.4 kg N ha⁻¹ yr⁻¹ near Tampa's urban core, based on a 12-km × 12-km grid cell size. CMAQ-predicted loading rates were 1,080 metric tons N yr⁻¹ to Tampa Bay and 8,280 metric tons N yr⁻¹ to the land portion of its watershed. If we assume a watershed-to-bay transfer rate of 18% for indirect loading, our estimates of the 2002 direct and indirect loading rates to Tampa Bay were 1,080 metric tons N and 1,490 metric tons N, respectively, for an atmospheric loading of 2,570 metric tons N or 71% of the total N loading to Tampa Bay. To evaluate the potential impact of the U.S. EPA Clean Air Interstate Rule (CAIR, replaced with Cross-State Air Pollution Rule), Tier 2 Vehicle and Gasoline Sulfur Rules, Heavy Duty Highway Rule, and Non-Road Diesel Rule, we compared CMAQ outputs between 2020 and 2002 simulations, with only the emissions inventories changed. The CMAQ-projected change in atmospheric loading rates between these emissions inventories was 857 metric tons N to Tampa Bay, or about 24% of the 2002 loading of 3,640 metric tons N to Tampa Bay from all sources.

Implications:

Air quality modeling reveals that atmospheric deposition of reactive nitrogen (N) contributes a significant fraction to Tampa Bay's total N loading from external sources. Regulatory drivers that lower nitrogen oxide emissions from power plants and motor vehicles are important to bay management strategies, which seek to improve water quality through N load reduction.

Introduction

Tampa Bay is a 100,000-ha subtropical, shallow, tidal estuary on the western coast of central Florida and eastern side of Gulf of Mexico (Figure 1). Land use in Tampa Bay's 600,000-ha watershed is predominantly urban in its western and northern sectors, agricultural in its southern and eastern sectors, and industrial in ports on the bay's coastline. Fluxes of reactive nitrogen (N) to the bay and recycling of accumulated nutrients within the bay contribute to algal blooms, which shade the water column and create conditions that are unfavorable for sea grasses (Conley et al., 2009; Greening et al., 2011; Wang et al., 1999). Sea grasses are important to the estuary as a nursery for fish and shellfish and for their ability to retain sediment. In 2002, sea-grass coverage within the bay was ∼11,500 ha, or 75% of the coverage seen in the 1950s (Greening et al., 2011). The bay management strategy seeks to reduce N inputs and to reestablish lost sea-grass beds (Greening and Janicki, 2006).

Figure 1. Tampa Bay and its watershed. Black outlines indicate watershed basin and bay segments boundaries, and darker shading shows municipal boundaries for major cities.

In the mid-1990s, the Tampa Bay Estuary Program and its stakeholders set as a goal reductions in loading of ∼15 metric tons N yr⁻¹ to maintain N loading at levels conducive to sea-grass growth even as the human population grew within the watershed (Greening et al., 2011; Greening and Janicki, 2006). Through management actions, N inputs to Tampa Bay were reduced. In apparent response and concurrent to these reductions, bay water clarity improved, chlorophyll α concentrations decreased, and sea-grass coverage increased (Conley et al., 2009; Greening and Janicki, 2006; Sherwood, 2011).

New N enters Tampa Bay from nonpoint sources (62%), direct atmospheric deposition (21%), point sources (12%), groundwater discharge (4%), and accidental fertilizer losses (1%) (Greening and Janicki, 2006). Atmospheric N that deposits first to land surface within the watershed and then moves either with stormwater runoff or via groundwater to the bay is referred to as indirect atmospheric deposition and is a nonpoint source. From the preceding percentages, between 21% and 83% of the new N reaching Tampa Bay is from atmospheric N deposition. Reductions of atmospheric contributions to N loading are important to achieving bay water quality goals.

Measurements of atmospheric N deposition near Tampa Bay are scarce. Therefore, air quality modeling is an important source of deposition estimates and for assessments of the change in deposition due to changes in emissions inventories. The Community Multiscale Air Quality (CMAQ) model simulates from first principles the transport, transformation, and removal of atmospheric pollutants (Byun and Schere, 2006). The U.S. Environmental Protection Agency (EPA) has developed the Watershed Deposition Tool (WDT) to calculate from the CMAQ model output N, sulfur, and mercury deposition rates to watersheds and their basins (Schwede et al., 2009). Figure 2, for example, shows WDT's display for annual atmospheric N deposition by basins and bay segments within Tampa Bay's watershed for 2002.

Figure 2. Watershed Deposition Tool (WDT) display for 2002 atmospheric reactive nitrogen (N) deposition by basins and bay segments within Tampa Bay's watershed.

We applied WDT to estimate the atmospheric deposition of N to Tampa Bay and its watershed, to look at temporal trends in N deposition, and to evaluate the aggregate impact to Tampa Bay of the U.S. EPA health-based regulatory drivers associated with the 1990 Clean Air Act Amendments. Monthly CMAQ-modeled wet N deposition rates were compared with wet N deposition rates measured at National Atmospheric Deposition Program (NADP) sites in and near Tampa Bay's watershed.

Experimental Methods

CMAQ modeling

The CMAQ v4.7 chemical transport model simulated horizontal and vertical advection, horizontal and vertical diffusion, gas-phase and aqueous-phase reactions, cloud mixing, aerosol dynamics, size distribution and chemistry, and deposition of gases and aerosols to Earth's surface (Byun and Schere, 2006; Schwede et al., 2009). CMAQ v4.7 included revisions to the cloud model, improvements to treatment of secondary organic aerosols and dinitrogen pentoxide (N₂O₅) parameterization, and added dynamic mass transfer for coarse-mode aerosols, as examples (Foley et al., 2010). Computations for wet deposition included in-cloud scavenging and below-cloud washout of air pollutants based on Henry's Law partitioning for gases and absorption of aerosols into cloud or rain water; dry deposition estimates assumed turbulent transfer of gases and aerosols to the surface and resistance to gas transfer at the surface (Byun and Schere 2006; Schwede et al., 2009).

We used CMAQ v4.7 model simulations for 2002–2006 from Appel et al. (2011) that had been postprocessed for the WDT. For these simulations, the modeling domain covered the eastern two-thirds of the United States, southern Canada, and northern Mexico, with a 12-km × 12-km grid cell size and where boundary conditions were provided by a separate 36-km × 36-km continental U.S. simulation (Appel et al., 2011; Foley et al., 2010). Year-specific meteorology was modeled with the Fifth Generation Penn State University/National Center for Atmospheric Research (NCAR) mesoscale model (MM5; Grell et al., 1994). Sparse Matrix Operator Kernel Emission (SMOKE; Houyoux et al., 2000) gridded hourly emissions were built from the EPA 2002 National Emission Inventory (NEI) v3 with year-specific adjustments for emissions from major point sources, mobile sources, and fires. Details of the meteorological modeling, emissions processing, CMAQ inputs, and CMAQ model configuration can be found in Appel et al. (2011). CMAQ-modeled species included reduced N: ammonia (NH₃) and ammonium (NH₄ ⁺); and oxidized N: nitrogen oxide (NO), nitrogen dioxide (NO₂), nitrite (NO₂ ⁻), nitrous acid (HNO₂), nitrate (NO₃ ⁻), nitric acid (HNO₃), dinitrogen pentoxide (N₂O₅), organic nitrate, peroxyacetyl nitrate (PAN), and higher PANs.

Additional files representing 2020 projected emissions based on expected growth and emissions reductions due to the Clean Air Interstate Rule (CAIR, replaced with Cross-State Air Pollution Rule), Tier 2 Vehicle and Gasoline Sulfur Rules, Heavy Duty Highway Rule, and Non-Road Diesel Rule (Houyoux, 2005; U.S. EPA, 2011a; U.S. EPA, 2011b) were used to evaluate the effects on N deposition of proposed air quality regulations. These 2020 CMAQ files were made available by the EPA and were based on the same MM5 meteorology as the 2002 CMAQ runs from Appel et al. (2011) with only the emissions inventories changed.

The Watershed Deposition Tool (WDT)

The EPA has developed the WDT to facilitate the extraction and analysis of CMAQ gridded output data for irregular spatial delineations such as basins and watersheds (Schwede et al., 2009). We downloaded the WDT (U.S. EPA, 2011c) and available CMAQ v4.7 output files (U.S. EPA, 2011d). Watershed basin polygons obtained from the Tampa Bay Estuary Program (TBEP) online technical resources were modified to better define water and land polygons and to achieve a finer spatial resolution (TBEP, 2011). In WDT, CMAQ's grid cells were superimposed over the watershed basins and bay segments as shown in Figure 3. WDT calculates N deposition for each watershed basin or bay segment as the summed product of deposition for each grid cell and the fraction of the overlay area between grid cell and polygon. Average N deposition rates are calculated as basin or bay segment N deposition divided by its area (Widing, 2006).

Figure 3. Reactive nitrogen (N) atmospheric deposition rates for Tampa Bay and its watershed from 2002 CMAQ model simulations: (a) gridded N deposition rates (kg N ha⁻¹); (b) average N deposition rates by watershed basin and bay segment (kg N ha⁻¹).

Nitrogen export from the watershed

Calculation of indirect N loading is challenging, and several approaches for determining N export from watersheds have been used, including (1) a watershed-level mass balance (Castro et al., 2004; Groffman et al., 2004; Van Breemen et al., 2002), (2) land use-specific N export modeling (Poe et al., 2005; Whitall et al., 2003), (3) process-based modeling (Pan et al., 2004), and (4) a conserved tracer study (Pollman and Poor, 2003). For this analysis, we used the export coefficient of 18% determined by Pollman and Poor (2003).

Pollman and Poor (2003) used ion ratios obtained from multiyear stream water quality measurements to infer retention and export of atmospheric inputs of N from Tampa Bay's watershed to Tampa Bay, after screening out streams influenced by calcium-rich groundwater or sodium-rich seawater. Chloride was assumed to be a conservative tracer for atmospheric deposition, such that the expected concentration of N ions in stream water was calculated from the product of the observed stream water chloride ion concentration and the ratio of atmospheric N flux to chloride flux. The ratio of observed to expected N ion concentrations in stream water yielded a minimum N retention of 82% or maximum export of 18% for the watershed, which is lower than the estimate of 26% N export for northeastern U.S. watersheds (Howarth et al., 2006) and higher than the estimate of 6% N export for North Carolina's Neuse River Estuary (Whitall et al., 2003).

Wet deposition

Daily rainfall amount and N concentration data were available from a NADP Atmospheric Integrated Research Monitoring Network (AIRMoN) site (NADP, 2012) that was operated by the Environmental Protection Commission of Hillsborough County (EPCHC) from August 1996 to June 2006 in urban Tampa at the eastern end of the Gandy Bridge (FL 18, Figure 1) (Strayer et al., 2007). From the Gandy Bridge data, wet deposition rates were calculated according to eq 1, where C_rain is the daily rainfall concentration of either NO₃ ⁻ -N or NH₄ ⁺ -N and has units of mg L⁻¹, D is the rainfall depth and has units of cm d⁻¹, and F_wet is the deposition rate (i.e., flux) and has units of kg N ha⁻¹ d⁻¹. Daily fluxes were summed to obtain monthly fluxes.

1

Long-term monthly rainfall amount and N concentration data were available for Verna Wellfield (FL 41), a NADP National Trends Network (NTN) site (NADP, 2011), which is located in Sarasota County and is proximate to the southern boundary of Tampa Bay's watershed (Figure 1). From the Verna Wellfield data, wet deposition rates were calculated according to eq 1, except that C is the monthly volume-weighted rainfall concentration of either NO₃ ⁻ -N or NH₄ ⁺ -N and has units of mg l⁻¹, D has units of cm mo⁻¹, and F is the deposition rate (i.e., flux) and has corresponding units of kg N ha⁻¹ mo⁻¹.

From the simulations of Appel et al. (2011), we obtained CMAQ-modeled precipitation amounts and HNO₃, NO₃ ⁻, NH₃, and NH₄ ⁺ wet N deposition rates, as well as oxidized N and reduced N wet deposition rates, for the 12-km × 12-km grid cells that enclosed NADP's Gandy Bridge (FL 18) and Verna Wellfield (FL 41) sites. Pearson correlation coefficient (r), normalized mean bias (NMB, eq 2), normalized mean error (NME, eq 3), and root mean squared error (RMSE, eq 4) were computed and plots prepared to assess agreement between modeled and observed wet N deposition rates (Eder and Yu, 2006).

2

3

4

where n is the number of observations.

Dry deposition

Measurements of ambient air HNO₃, NO₃ ⁻, NH₃, and NH₄ ⁺ concentrations were collocated with wet deposition monitoring at FL 18 (Figure 1) from August 1996 to April 2003, sponsored in the latter years by the Bay Region Atmospheric Chemistry Experiment (BRACE) (Poor et al., 2006). Twenty-four-hour integrated samples were collected on a 1-in-6-day schedule with duplicate annular denuder systems (Poor et al., 2006). For the same time period, subhourly measurements of water temperature and wind speed, wind direction, air temperature, and relative humidity were available from a midbay weather station, which was one of the Tampa Bay Physical Oceanographic Real-Time System sites (PORTS t03010) (Sopkin et al., 2007). These measurements were used in an integrated National Oceanic and Atmospheric Administration (NOAA) buoy gas deposition and Williams's particle deposition model (Evans et al., 2004; Valigura, 1995; Williams, 1982). Deposition rates were calculated according to eq 5, where C_air is the monthly average N concentration ( μg N m⁻³), v is the monthly average deposition velocity (m s⁻¹), t is the number of days in a month, and F_dry is the dry deposition flux (kg N ha⁻¹ mo⁻¹) and represents the dry N deposition rate directly to Tampa Bay. NH₃ deposition rates were corrected for a bidirectional flux at the bay surface (Poor et al., 2001).

5

Dry deposition rates so obtained for January 2002 to April 2003 were presented as annual averages to compare with CMAQ-modeled dry deposition rates for the grid cell that enclosed the midbay meteorological tower. This comparison serves as a springboard for discussion of the sources of uncertainty in dry deposition rates.

Nitrogen loading rates

The spatial resolution of watershed basin polygons imported into the WDT allowed us to calculate separately from CMAQ model output the N deposition to land and water (Table 1). For direct N loading to Tampa Bay, we integrated the bay segment-specific loading rates; for indirect N loading, we integrated basin-specific rates and assumed that 18% of atmospherically deposited N was transferred from land to bay (Pollman and Poor, 2003). Atmospheric loading rates were calculated separately for dry- and wet-deposited N and for oxidized and reduced N.

Table 1. Estimated 2002 direct and indirect atmospheric N loading rates within Tampa Bay's watershed. Water area includes embayment, canals, and major tributaries apportioned by bay segment; land area includes watershed drainage surface by bay segment

Bay segment	Water area, ha	Land area, ha	Direct atmospheric loading, metric tons N	Indirect atmospheric loading,* metric tons N	Atmospheric loading, metric tons N
Old Tampa	22,451	67,326	261	181	441
Hillsborough	10,852	330,364	211	843	1,050
Middle Tampa	29,450	79,924	286	194	480
Lower Tampa	25,265	8,437	218	19.2	237
Boca Ciega	6,188	21,353	43.6	52.5	96.1
Terra Ceia	1,650	2,851	20.6	6.48	27.1
Manatee River	3,460	92,416	42.1	196	238
Totals	99,317	602,669	1,080	1,490	2,570
Notes: Assumes an 18% transfer rate from watershed to bay. Note: Arithmetic operations were performed prior to rounding numbers for presentation in the table and as such the sums of columns and rows may not match exactly the sums shown.

Atmospheric N loading to Tampa Bay (LR_AD ) was computed as the summed contributions from direct N deposition to the bay surface and indirect N deposition to the land surface of the watershed, where indirect deposition was modified by the fraction of N exported from land to bay or export coefficient (EC), as shown in eq 6, where F has units of kg N ha⁻¹ yr⁻¹, Area has units of ha, and LR_AD has units of metric tons N yr⁻¹:

6

We compared N loading from atmospheric deposition with total N loading to Tampa Bay (Table 2). For 2002–2006, Poe et al. (2005) and JEI (2008) estimated total N loading to Tampa Bay from direct atmospheric deposition, nonpoint sources, domestic and industrial point sources, springs, groundwater, and material losses. They calculated direct atmospheric deposition from monthly bay segment-specific rainfall rates, rainfall N concentrations, and seasonal dry:wet deposition ratios. Monthly rainfall N concentrations and seasonal dry:wet deposition ratios came from measurements made at an atmospheric deposition site at the eastern end of the Gandy Bridge (Figure 1). Contributions from all other source categories were based in general on basin- or bay segment-specific measurements of stream, spring, and groundwater or wastewater discharges and their respective N concentrations; for sub-basins without either stream flow gauges or water quality measurements, an empirical model was applied to determine water discharge rates and N concentrations from monthly rainfall amount and seasonal runoff coefficients based on land use and soil type. TBNMC (2010) revised the point source and materials handling N loading rates used by Poe et al. (2005) and JEI (2008); their updated total and direct N atmospheric deposition rates are given in Table 2.

Table 2. Modeled atmospheric deposition (AD) and total N loading to Tampa Bay, 2002–2006

	Previous estimates*			Current estimates
Year	Direct AD N loading, metric tons	Total N loading, metric tons	^§Adjusted Total N loading, metric tons	Direct AD N loading, metric tons	Indirect AD N loading, metric tons	AD N loading, metric tons	Total N loading that is AD loading, %
2002	840	3,400	3,640	1,080	1,490	2,570	70.6
2003	1,150	6,690	6,440	903	1,340	2,240	34.8
2004	801	5,930	6,010	885	1,240	2,120	35.4
2005	603	3,230	3,570	950	1,310	2,260	63.3
2006	552	2,980	3,230	803	1,150	1,950	60.5
Notes: *Poe et al., 2005; JEI, 2008; TBNMC, 2010. ^§Previous estimate of total N loading – previous estimate of direct AD N loading + current estimate of direct AD N loading.

Uncertainty in atmospheric N loading rates

To get an approximation of the combined uncertainty on LR_AD , we applied a spreadsheet technique as described by Kragten (1994). We assumed that the standard deviation for each of deposition rates in eq 6 was equal to the estimate of the loading rate and that the standard deviation for the export coefficient in eq 6 was equal to the estimate of the export coefficient. These assumptions were not unreasonable based on the RMSE values given in Tables 3 and 4 for wet N deposition rates, comparisons of predicted dry S deposition rates for 15 watersheds reported by Mitchell et al. (2011), and the range of watershed N export rates across 43 watersheds studied by Castro et al. (2003). Uncertainties on land and water areas were assumed to be ∼0, and areas and the unit conversion were multiplied through eq 6 such that units on deposition rates were metric tons N yr⁻¹.

Table 3. Statistics to assess agreement between modeled versus observed monthly rainfall and wet N deposition rates for observations made at NADP's Gandy Bridge site (FL18) for 2002 to 2006. Units are cm on mean modeled, mean observed, and RMSE precipitation; units are kg N ha⁻¹ mo⁻¹ on mean modeled, mean observed, and RMSE deposition rates

Species	Mean modeled	Mean observed	r, %	NMB, %	NME, %	RMSE
Precipitation	12.5	10.0	58.4	26.8	71.1	10.6
NH4 ^+ -N	0.127	0.158	48.9	–18.5	55.8	0.144
NO3 ^− -N	0.232	0.171	80.1	38.0	59.8	0.157
Total N	0.365	0.329	69.1	10.8	53.9	0.265

Table 4. Statistics to assess agreement between modeled versus observed monthly rainfall and wet N deposition rates for observations made at NADP's Verna Wellfield site (FL41) for 2002 to 2006. Units are cm on mean modeled, mean observed, and RMSE precipitation; units are kg N ha⁻¹ mo⁻¹ on mean modeled, mean observed, and RMSE deposition rates

Species	Mean modeled	Mean observed	r, %	NMB, %	NME, %	RMSE
Precipitation	16.3	12.3	73.7	32.6	66.8	12.6
NH4 ^+ -N	0.113	0.108	50.1	2.91	62.1	0.103
NO3 ^− -N	0.173	0.166	78.6	0.416	43.4	0.107
Total N	0.287	0.274	73.7	1.39	43.3	0.186

Results and Discussion

Nitrogen loading to Tampa Bay and its watershed

For 2002, the CMAQ model yielded loading rates of 1,080 metric tons N yr⁻¹ to Tampa Bay and 8,280 metric tons N yr⁻¹ to the land portion of its watershed. If we assume a watershed-to-bay transfer rate of 18% (Poor and Pollman, 2003), atmospheric direct and indirect loading rates to Tampa Bay were 1,080 metric tons N yr⁻¹ and 1,490 metric tons N yr⁻¹, respectively, or an atmospheric loading rate to the bay of 2,570 metric tons N yr⁻¹. Table 1 shows N loading rates by bay segment.

Poe et al. (2005) estimated direct atmospheric and total loading rates for 2002 to Tampa Bay of 840 metric tons N and 3,400 metric tons N, respectively. To determine the relative fraction of CMAQ-modeled atmospheric loading to total loading, we first adjusted the 2002 total N loading rate from Poe et al. (2005) for the CMAQ-modeled direct atmospheric N loading rate (3,400 – 840 + 1,080 = 3,640 metric tons) and second computed the fraction of atmospheric to total loading rates (2,570 metric tons N/3,640 metric tons N), which was 71% and a significant fraction of Tampa Bay's total N loading rate. JEI (2008) estimated direct atmospheric and total N loading rates to Tampa Bay for 2003 through 2006 (Table 2). Using the approach just described, we calculated from the CMAQ-model output the direct and indirect atmospheric N loading and fraction of total N loading that is atmospheric for 2003 through 2006 (Table 2). Atmospheric loading rates ranged from 1,950 to 2,260 metric tons N yr⁻¹, with significant fractional contributions that ranged from 35% to 63%. In 2003 and 2004, higher rainfall rates increased the hydrological loading within Tampa Bay's watershed, which increased N loading from nonpoint sources and in 2004 led to emergency and accidental discharges of N from industrial point sources.

Contemporary published values for atmospheric loading rates to Tampa Bay vary from 1,100 metric tons N yr⁻¹ to 3,030 metric tons N yr⁻¹; the broad range in estimates suggests significant differences in measurement techniques, computational tools, and methodical assumptions (Poor et al., in review).

The WDT's graphical results for atmospheric N deposition rates are shown Figures 3, 4, and 5. For 2002, the CMAQ-modeled average atmospheric deposition rate across the bay and its watershed was 13.3 kg N ha⁻¹ yr⁻¹. Based on the 12-km × 12-km grid cell size, deposition rates ranged from 6.24 kg N ha⁻¹ yr⁻¹ to 21.4 kg N ha⁻¹ yr⁻¹, with the lowest rates predicted over the Gulf of Mexico and highest rates centered on Tampa's urban core, which reflected emissions from the underlying human settlements (Figure 3). These N deposition rates are characteristic of an estuary and watershed with moderate N pollution (Dentener et al., 2006; Holland et al., 2005) Predictions with GEOS-Chem global chemical transport model at a horizontal resolution of 1/2˚ × 2/3 ˚ (∼56 km × 65 km) for west central Florida were in range of ∼8–10 kg N ha⁻¹ yr⁻¹ (Zhang et al., 2012), which illustrates the potential impact of grid cell size on modeled N deposition rates.

Figure 4. Atmospheric average (a) dry and (b) wet N deposition rates (kg N ha⁻¹) for Tampa Bay and its watershed, by watershed basin and bay segment, from 2002 CMAQ model simulations.

Figure 5. Atmospheric average (a) oxidized and (b) reduced N deposition rates (kg N ha⁻¹) for Tampa Bay and its watershed, by watershed basin and bay segment, from 2002 CMAQ model simulations.

For dry- and wet-deposited N, the 2002 modeled average atmospheric deposition rates were 7.70 kg N ha⁻¹ yr⁻¹ and 5.64 kg N ha⁻¹ yr⁻¹, respectively (Figure 4). Over land, the dry and wet loading rates were 4,800 metric tons N yr⁻¹ and 3,490 metric tons N yr⁻¹, respectively; over water, the dry and wet loading rates were 608 metric tons N yr⁻¹ and 474 metric tons N yr⁻¹, respectively. Assuming an 18% transfer rate of N from land to water, atmospheric loading rates to Tampa Bay were 1,470 metric tons (57%) dry N and 1,100 metric tons (43%) wet N deposition. Dry N deposition rates were higher in sectors with major traffic corridors and power generation plants near Tampa Bay. For a topographical map of major roadways and industrial and power plant sources near Tampa Bay, refer to Pancras et al. (2006) and Park et al. (2005). Modeled spatial gradients of dry N deposition were steeper than corresponding wet deposition N gradients (Figure 4) due to the localized influence of sources or source regions on ambient air concentrations of N at the watershed's surface. Wet N deposition was higher east of the bay over a region of industrial NH₃ emissions (Figure 4).

For oxidized and reduced N, the 2002 modeled average atmospheric deposition rates were 9.73 kg N ha⁻¹ yr⁻¹ and 3.61 kg N ha⁻¹ yr⁻¹, respectively (Figure 5). Over land, the oxidized and reduced loading rates were 5,990 metric tons N yr⁻¹ and 2,300 metric tons N yr⁻¹, respectively; over water, the oxidized and reduced loading rates were 842 metric tons N yr⁻¹ and 240 metric tons N yr⁻¹, respectively. Assuming an 18% transfer rate of N from land to water, atmospheric loading rates to Tampa Bay were 1,920 metric tons (75%) oxidized N and 650 metric tons (25%) reduced N. The highest rates of oxidized N deposition were associated with urban centers (Figure 5).

CMAQ-modeled temporal trends in N deposition

Between 2005 and 2010, Florida's annual and ozone season oxidized N emissions from power plants had dropped by 0.11 million metric tons (63%) and 0.054 million metric tons (65%), respectively (U.S. EPA, 2011b), and continued a decade-long downward trend in both power-plant NO_x emissions and NO_x concentrations measured at urban monitors near Tampa Bay (Poor et al., in review). The observed downward trend in atmospheric NO_x concentrations may have important consequences for N deposition to Tampa Bay. According to CMAQ modeling results, dry N deposition dominates both spatial and temporal N deposition rates within Tampa Bay and its watershed. These results suggest that within a watershed monitoring dry N deposition may be as important as monitoring wet N deposition.

By comparison, between 2002 and 2006 the CMAQ-modeled annual average N deposition rates to Tampa Bay and its watershed suggested a modest decline of 3.10 kg N ha⁻¹ yr⁻¹. Of this decline, 0.66 kg N ha⁻¹ yr⁻¹ was from dry deposition of oxidized N, 0.70 kg N ha⁻¹ yr⁻¹ from wet deposition of reduced N, and 1.77 kg N ha⁻¹ yr⁻¹ from wet deposition of oxidized N; dry deposition of reduced N increased by 0.03 kg N ha⁻¹ yr⁻¹ (Figure 6).

Figure 6. CMAQ-modeled 2002–2006 annual average N deposition rates for Tampa Bay and its watershed.

We investigated whether a decline in wet N deposition could be discerned at NADP's Gandy Bridge (FL 18) or Verna Wellfield (FL 41) sites (Figure 1). For the Gandy Bridge site, we applied an unpaired, two-sided Student's t-test to test the hypothesis that between the first 27 mo and the second 27 mo of the 4.5-yr period the means of log-transformed monthly NO₃ ⁻-N or NH₄ ⁺-N deposition rates were equal. This hypothesis was not rejected for either NO₃ ⁻-N or NH₄ ⁺-N deposition rates (p = 0.72 and p = 0.43, respectively). For the Verna Wellfield site, we applied an unpaired, two-sided Student's t-test to test the hypothesis that between the first 30 mo and the second 30 mo of the 5-yr period the means of log-transformed monthly NO₃ ⁻-N or NH₄ ⁺-N deposition rates were equal. This hypothesis was not rejected for either NO₃ ⁻-N or NH₄ ⁺-N deposition rates (p = 0.67 and p = 0.17, respectively). Our approach was repeated for rainfall rates and for NO₃ ⁻ or NH₄ ⁺ concentrations with similar results. Thus, no significant downward trend in N wet deposition was evident either at the Gandy Bridge or at the Verna Wellfield sites. Small changes in N deposition are likely difficult to discern over short time periods and among the large swings in rainfall and thus wet deposition rates (Figure 7).

Figure 7. CMAQ-modeled 2002–2006 monthly average wet and dry N deposition rates for Tampa Bay and its watershed.

In Figure 7 we show for Tampa Bay and its watershed the temporal variation in modeled monthly dry and wet N deposition rates, which for the 5-yr period shown had means (± standard deviation) of 0.605 (±0.081) and 0.365 (±0.340) kg N ha⁻¹ mo⁻¹, respectively. Given their relatively low coefficient of variation (CV), dry N deposition rates were well predicted by an average value. This was clearly not the case for wet N deposition rates, as temporal gradients were steep with the highest rates seen in midsummer. Dry-to-wet N deposition ratios were 5.99 (±5.17) when averaged over the drier months September through May, and 0.801 (±0.172) when averaged over the wetter months June through August. Annual watershed dry-to-wet N deposition ratios varied from 3.03 in 2002 to 6.48 in 2006 (a dry year) if calculated as the annual average of monthly dry-to-wet ratios, and from 1.36 in 2002 to 2.22 in 2006 if calculated as the average annual dry deposition to annual wet deposition. One can see the challenges in using dry-to-wet N deposition ratios to predict dry N deposition from wet N deposition to the watershed.

Comparison of modeled and observed wet N deposition rates

Wet deposition monitoring at the NADP AIRMoN and NTN stations in and near Tampa Bay's watershed, respectively, offered us an opportunity to assess the agreement between CMAQ modeling and observations of rainfall and wet N deposition rates. Each comparison was of a predicted grid cell average and a point measurement of rainfall amount or N deposition rate, which is by itself a source of error (Davis and Swall, 2006).

Gases of HNO₃, HONO, and NH₃ are water soluble and are thus scavenged by rainfall. Precipitation acidity in the range of pH 4–5 (Strayer et al., 2007) favors the disassociation of these compounds into NO₃ ⁻, NO₂ ⁻, and NH₄ ⁺ in rainfall samples. Water-dissolved NO₂ ⁻ slowly oxidizes to NO₃ ⁻ (Finlayson-Pitts and Pitts, 2000) during the collection, transport, and storage of rainfall samples prior to their analysis at NADP's central laboratory. Aerosols containing NO₃ ⁻ and NH₄ ⁺ are also removed by rainfall, albeit with differing size-related removal efficiencies (Finlayson-Pitts and Pitts, 2000). Gases of NO and NO₂ are poorly water soluble but may interact with an aerosol surface to form HONO; N₂O₅ readily dissolves in water to form NO₃ ⁻ and NO₂ ⁻(Finlayson-Pitts and Pitts, 2000). Thus, NADP-reported concentrations of NO₃ ⁻ and NH₄ ⁺ include both gas and particle contributions.

In CMAQ-modeled wet N deposition, error is present in the emissions inventory, meteorology, and mechanistic representations of atmospheric chemistry and physics. Such errors are discussed by Gilliland et al. (2006), who proposed corrections to the NEI for seasonality of NH₃ emissions; by Davis and Swall (2006), based on their examination of the effect of MM5’s simulation of precipitation on CMAQ-modeled NH₄ ⁺ wet deposition; and by Chemel et al. (2011) from their investigation of CMAQ's ability to predict a power plant's contribution to N deposition, as examples.

Error is also present in observed wet N deposition rates, so care must be taken not to assign all of the disagreement to modeling. According to Wetherbee et al. (2006), for example, “atmospheric wet deposition is slightly underestimated in 48 to 68 percent of the NADP/NTN samples, depending on the analyte.” This underestimate is a consequence of analyte loss during field exposure, sampling handling, and/or shipping. Atmospheric wet deposition is overestimated if the sample is marginally contaminated; gross contamination, however, leads to sample rejection and loss of data, which if uncorrected causes an underestimate in cumulative atmospheric wet deposition.

For the eastern United States, MM5 -simulated precipitation rates were underestimated for synoptic-scale storm systems and overestimated for subgrid-scale convective storm systems (Appel et al., 2011; Davis and Swall, 2006). A similar trend was evident at both NADP's Gandy Bridge (FL 18) and Verna Wellfield (FL 41) sites, where reasonable agreement was seen between modeled and observed rainfall rates for cooler and drier months when rainfall was more likely produced by a synoptic system but a moderate positive bias was seen for summertime (wetter) months when rainfall was more likely generated in convective storms (Figure 8).

Figure 8. Comparison of CMAQ-modeled and observed precipitation rates for NADP's (a) AIRMoN site at Gandy Bridge (FL18) and (b) NTN site at Verna Wellfield (FL41).

Over a 4.5-yr period at the Gandy Bridge site, the modeled and measured precipitation rates were moderately correlated, with averages of 12.5 cm and 10.0 cm, respectively (Table 3); 53% of the modeled monthly rainfall rates were within a factor of 2 of observed rainfall rates. Over a 5-yr period at the Verna Wellfield site, the modeled and measured precipitation rates were moderately correlated, with averages of 16.3 cm and 12.3 cm, respectively (Table 4); 62% of the modeled monthly rainfall rates were within a factor of 2 of observed rainfall rates (Figure 9). Since deposition rates are the product of N concentration and rainfall rate ( eq 1), a bias in rainfall rate will contribute to a bias in N wet deposition.

Figure 9. Scatter plots of CMAQ-modeled and observed (a) precipitation rates, (b) oxidized N wet deposition rates, and (c) reduced N wet deposition rates for NADP's NTN Verna Wellfield (FL41) site.

Temporal trends in CMAQ model predictions of precipitation (Figure 8) and atmospheric N deposition (Figure 10) are shown for NADP's Gandy Bridge and Verna Wellfield sites. Rainfall rates were lower but N deposition rates higher at the urban Gandy Bridge site when compared to the rural Verna Wellfield site (Figure 1). This is a consequence of higher ambient air concentrations of N near the urban core, as is suggested by Figure 3.

Figure 10. Comparison of CMAQ-modeled and observed atmospheric N wet deposition rates for NADP's (a) AIRMoN site at Gandy Bridge (FL18) and (b) NTN site at Verna Wellfield (FL41).

For the Gandy Bridge site, CMAQ model predictions of wet deposition rates were in fair agreement for NO₃ ⁻ -N and NH₄ ⁺ -N, with moderate positive bias relative to observations seen for NO₃ ⁻-N and a moderate negative bias seen for NH₄ ⁺ -N (Table 3); 59% of the data points for wet deposition rates of both NO₃ ⁻ -N and NH₄ ⁺ -N were within a factor of 2, respectively. Thus, CMAQ modeled appeared to overpredict NO₃ ⁻ concentration and underpredict NH₄ ⁺ concentration in rainwater for this site.

For the Verna Wellfield site, CMAQ model predictions of wet deposition rates agreed remarkably well for NO₃ ⁻ -N and NH₄ ⁺ -N, with a only small positive bias relative to observations (Table 4); 75% and 54% of the data points for wet deposition rates of NO₃ ⁻ -N and NH₄ ⁺ -N were within a factor of 2, respectively (Figure 9). For this site, N wet deposition biases were not as large as the precipitation bias, which suggested that errors in precipitation rates were offset by errors in N concentrations; in other words, both NO₃ ⁻ and NH₄ ⁺ concentrations were biased low.

Appel et al. (2011) found that for the eastern United States, CMAQ-modeled wet deposition of both NO₃ ⁻-N and NH₄ ⁺-N were underestimated but that these biases were diminished when model simulations were corrected for lightning-generated NO and bidirectional NH₃ exchange. The better model performance for NO₃ ⁻-N relative to NH₄ ⁺-N wet deposition reflected, perhaps, differences in spatial and temporal resolution of NO_x and NH₃ emissions inventories (Appel et al., 2011; Gilliland et al., 2006) and the lower uncertainty in the NO_x emissions compared to NH₃ emissions.

Comparison of dry N deposition rates

Eighteen months of direct dry N deposition rates prepared for Tampa Bay as part of the BRACE were compared with the CMAQ model estimates (Table 5) for a grid cell that enclosed a midbay meteorological station (Figure 1). The annual average BRACE dry deposition rates were similar to those published by Poor et al. (2001) except for a lower NH₄ ⁺-N deposition rate, which reflected an improvement in particle deposition modeling. Note that BRACE estimates had lower HNO₃ and higher NO₃ ⁻, NH₃, and NH₄ ⁺ N deposition rates when compared to CMAQ-modeled values, although the summed direct dry N deposition rates were within 10% at 3.18 and 2.91 kg N ha⁻¹ yr⁻¹, respectively (Table 5).

Table 5. Comparison of annual average CMAQ-modeled and measurement-based direct dry deposition rates to Tampa Bay for 2002–2003

Species	CMAQ, kg N ha^−1 yr^−1	BRACE, kg N ha^−1 yr^−1
HNO3	1.42	0.81
NH3	0.38	2.29
NO3 ^−	0.48	0.07
NH4 ^+	0.10	0.01
Other Ox N	0.54	—
Total	2.91	3.18

Monthly CMAQ-modeled and BRACE dry N deposition rates were not correlated, nor were they similar in magnitude for individual species. BRACE dry deposition rates were based on N concentrations made over land and assumed to be the same over water, which likely biased estimates high; significant losses of HNO₃ and NO₃ ⁻ at the inlet of the annular denuder system biased estimates low; infrequent sampling attenuated temporal variability and obscured covariance between concentrations and deposition velocities (Poor et al., 2006; Poor et al, 2001). Atmospheric measurements revealed a vertical HNO₃ concentration gradient with higher concentrations aloft as a consequence of rapid deposition of HNO₃ at the surface (Luke et al., 2007). BRACE concentration measurements were made at a height of 4.2 m asl and CMAQ model concentrations were based on a midpoint of 18 m asl in the lowest model grid cell. Thus, dry N deposition rates may be overestimated if HNO₃ concentrations are either measured or modeled well above the surface. CMAQ v4.7 did not include representation of a bidirectional NH₃ flux nor the NO_x emissions associated with lightning, which may have biased estimates low (Foley et al., 2010). These results highlighted the importance for continued development of robust sensors that are capable of detecting N deposition of gases and particles, not only to monitor temporal and spatial trends of N dry deposition within a watershed but to help establish watershed N budgets and validate air quality model predictions.

Uncertainty in atmospheric N loading rates

Table 6 summarizes a spreadsheet calculation of combined uncertainty on N loading rates. Given our assumptions on standard deviations of N wet and dry deposition rates and the watershed export coefficient, the overall uncertainty on the 2002 CMAQ-modeled atmospheric loading rate (LR_AD ) to Tampa Bay was 2,570 ± 1,990 metric tons N yr⁻¹. The largest and second largest sources of error came from the uncertainty on the N export coefficient and on the estimate of dry N deposition to land, which suggested a need to better understand or to reduce these uncertainties within the Tampa Bay watershed. To address uncertainty in N emissions, CMAQ v5.0 accounts for NO_x production from lightning and for bidirectional flux of NH₃ at the earth's surface.

Table 6. Spreadsheet calculation of combined uncertainty for atmospheric N loading to Tampa Bay using eq 6, where

			Standard deviations, metric tons N yr^−1
Term ID	N loading term	Estimate, metric tons N yr^−1	s1	s2	s3	s4	s5
1	Fwet_land	3,490	6,980	3,490	3,490	3,490	3,490
2	Fdry_land	4,800	4,800	9,600	4,800	4,800	4,800
3	Fwet_bay	474	474	474	948	474	474
4	Fdry_bay	608	608	608	608	1,216	608
5	EC	0.18	0.18	0.18	0.18	0.18	0.36
	LRAD+	—	3,202	3,438	3,048	3,182	4,066
	− LRAD	2,574	2,574	2,574	2,574	2,574	2,574
		—	628	864	474	608	1,492

2002 and 2020 N emission scenarios

Through the Clean Air Interstate Rule (CAIR, replaced with Cross-State Air Pollution Rule), Tier 2 Vehicle and Gasoline Sulfur Rules, Heavy Duty Highway Rule, and Non-Road Diesel Rule (U.S. EPA, 2011a, b), U.S. EPA seeks to improve air quality by phased reductions in air pollutant emissions from both fixed and mobile sources. A fully implemented CAIR, for example, removes 6.7 million metric tons of NO_x emissions from all states in the continental United States, of which 5.2 million metric tons is from states in the eastern United States (Houyoux, 2005). By 2020, NH₃ emissions are expected to grow by 0.44 million metric tons for all states in the continental United States, of which 0.22 million metric tons of NH₃ emissions are from states in the eastern United States (Houyoux, 2005). Moreover, slated reductions in sulfur dioxide (SO₂) emissions will shift aerosol NH₄ ⁺ to gaseous NH₃ and thus favor localized dry deposition of reduced N (Pinder et al., 2008).

For 2002, the CMAQ-modeled direct plus indirect atmospheric loading rate to Tampa Bay was 2,570 metric tons N. Using WDT, we calculated the potential impact of fully implemented CAIR (now Cross-State Air Pollution Rule), Tier 2 Vehicle and Gasoline Sulfur Rules, Heavy Duty Highway Rule, and Non-Road Diesel Rule. Between the 2002 and 2020 emission scenarios, the projected change in atmospheric N loading was 857 metric tons N, which represented 33% less atmospheric loading and 24% less total loading to Tampa Bay for a total loading of 3,640 metric tons N. Such a change if implemented over 18 yr represents a decrease of ∼48 metric tons N yr⁻¹, well above a target reduction rate of 15 metric tons N yr⁻¹ (Greening et al., 2011; Greening and Janicki, 2006). A change in N loading on this order of magnitude bodes well for water clarity goals and expansion of sea grass beds.

Changes in atmospheric N deposition rates across Tampa Bay and its watershed between the 2002 and 2020 emissions scenarios are shown in Figure 11 and revealed a broad decrease of 1 to 2 kg N ha⁻¹ over the entire watershed but as high as 6 to 7 kg N ha⁻¹ in the vicinity of major power plant and industrial sources. Thus, both regional and local emissions controls explain reductions in atmospheric N deposition under the future emissions scenario.

Figure 11. CMAQ-modeled changes in watershed N deposition for 2002 base case and 2020 future scenario emissions for Tampa Bay and its watershed, processed with WDT from CMAQ 2002 12-km gridded scale output: (a) change in gridded N deposition rates (kg N ha⁻¹); (b) change in average N deposition rates by watershed basin and bay segment (kg N ha⁻¹).

Conclusions

The Water Deposition Tool (WDT) offers a user-friendly interface with CMAQ model output to calculate atmospheric deposition of reactive nitrogen (N) to watersheds and their basins. With WDT we were able to estimate N deposition to both land and bay portions of Tampa Bay's watershed and thereby include direct and indirect atmospheric deposition components. For 2002 and within the boundaries of Tampa Bay's watershed, modeled atmospheric deposition rates averaged 13.3 kg N ha⁻¹ yr⁻¹ and ranged from 6.24 kg N ha⁻¹ yr⁻¹ at the bay's boundary with Gulf of Mexico to 21.4 kg N ha⁻¹ yr⁻¹ near Tampa's urban core, based on a 12-km × 12-km grid cell size. CMAQ-predicted loading rates were 1,080 metric tons N yr⁻¹ to Tampa Bay and 8,280 metric tons N yr⁻¹ to the land portion of its watershed. If we assume a watershed-to-bay transfer rate of 18% for indirect loading, our estimates of the 2002 direct and indirect loading rates to Tampa Bay were 1,080 metric tons N and 1,490 metric tons N, respectively, for an atmospheric loading of 2,570 metric tons N or 71% and a substantial fraction of the total N loading to Tampa Bay.

WDT also opened the door for trend analysis of CMAQ model output. As examples, CMAQ modeling indicated that during an extended dry season, dry N deposition dominated N deposition rates to Tampa Bay and its watershed. Moreover, oxidized N deposition rates greatly exceeded reduced N deposition rates. With WDT we had the opportunity to compare past and future emissions scenarios to assess the impact of emissions controls on N deposition to Tampa Bay and its watershed. To evaluate the potential impact of the U.S. EPA Clean Air Interstate Rule (CAIR, replaced with Cross-State Air Pollution Rule), Tier 2 Vehicle and Gasoline Sulfur Rules, Heavy Duty Highway Rule, and Non-Road Diesel Rule, we compared CMAQ outputs between 2020 and 2002 simulations, with only the emissions inventories changed. CMAQ-projected change in atmospheric loading rates between these emissions inventories was significant at 857 metric tons N to Tampa Bay, or about 24% of the 2002 loading of 3,640 metric tons N to Tampa Bay from all sources.

Acknowledgments

This research was funded by Florida Department of Environmental Protection (FDEP) through the Tampa Bay Estuary Program (TBEP) under FDEP contract AQ206. Noreen Poor is affiliated with the University of South Florida College of Public Health. The U.S. Environmental Protection Agency through its Office of Research and Development collaborated in the research described here. It has been subjected to agency review and approved for publication.