Influence of precipitation events on phytoplankton biomass in coastal waters of the eastern United States

Authors

  • Tae-Wook Kim,

    1. School of Environmental Science and Engineering, Pohang University of Science and Technology, Pohang, South Korea
    2. Now at Ocean Circulation and Climate Research Division, Korea Institute of Ocean Science and Technology, Ansan, South Korea
    Search for more papers by this author
  • Raymond G. Najjar,

    1. Department of Meteorology, The Pennsylvania State University, University Park, Pennsylvania, USA
    Search for more papers by this author
  • Kitack Lee

    Corresponding author
    1. School of Environmental Science and Engineering, Pohang University of Science and Technology, Pohang, South Korea
    Search for more papers by this author

Abstract

Precipitation over the ocean surface in the vicinity of industrialized and populated coastlines can increase the ocean nitrate concentration and consequently enhance ocean primary productivity. Using satellite data and a meteorological reanalysis product, we evaluated the impact of precipitation events on the chlorophyll a concentration in coastal and offshore waters located downwind of the eastern United States. We found that in low-nutrient areas (defined as having nitrate concentrations < 1 μM) precipitation events were associated with increased levels of chlorophyll a (up to approximately 15%), but in high-nutrient areas (nitrate concentrations > 1 μM) they were associated with decreased levels. These contrasting responses of chlorophyll a concentration to precipitation were attributed to the correlation of precipitation with wind speed and to other factors (nutrients and light) limiting phytoplankton growth. Increases in wind speed accompanied by precipitation events typically deepen the mixed layer, which can entrain additional nutrients into the mixed layer but simultaneously reduce light availability. We suggest that in nutrient-depleted areas (south of 36°N) the added nutrients were a dominant factor increasing the chlorophyll a concentration, whereas in the nutrient-replete areas (north of 36°N), where phytoplankton growth was light limited, reduced light availability was the dominant factor determining reduced chlorophyll a concentration. Our results indicate that an increase in wind speed accompanied by precipitation events was a major contributor to the observed changes in chlorophyll a concentration during wet days, whereas the wet deposition of pollutant nitrogen slightly increased the chlorophyll a concentration (< 5%) only in nutrient-depleted areas.

1 Introduction

Precipitation is potentially a large driver of change in the ocean because the chemical and physical characteristics of rain are changing markedly. In particular, the chemical composition of rainwater has changed considerably because of the accumulation in the atmosphere of pollutants originating from the increasing use of fossil fuels and agricultural fertilizers. These pollutants (including sulfur, nitrogen, and heavy metals) ultimately reach the marine environment via precipitation (i.e., wet deposition) [Rodhe et al., 2002; Galloway et al., 2004, 2008; Doney et al., 2007]. As a result, the impact of precipitation on ocean biogeochemistry probably increased during the twentieth century, although our knowledge of this process is only now developing. Recent studies have shown that deposition of pollutant nitrogen (N) species (NOx and NHy) has substantially increased N availability in lakes and ocean waters [Elser et al., 2009; Kim et al., 2011]. The impacts of precipitation on marine N budgets may be substantial; estimates from modeling and observational studies indicate that atmospheric N deposition is equivalent to 4–50% of upwelled N (i.e., new production) over large areas downwind of populated regions [Guerzoni et al., 1999; Krishnamurthy et al., 2007; Duce et al., 2008; Onitsuka et al., 2009; Zhang et al., 2010]. In addition, anthropogenic global warming has increased the water-holding capacity of the atmosphere, which has altered precipitation regimes in many ways, including an increase in the frequency of extreme events [Allan and Soden, 2008; Min et al., 2011]. These human-induced changes in the chemical and physical characteristics of precipitation have probably modified the availability of nutrients and light (through associated changes in winds and clouds during these events) in the upper mixed layer, thereby influencing ocean primary productivity and the numerous elemental cycles to which it is coupled. Therefore, to predict future changes in ocean biogeochemistry, it is critical to determine how precipitation influences phytoplankton in the modern ocean.

Several studies have investigated the relationship between precipitation and ocean productivity. However, these have been limited to either in situ or manipulated bioassay experiments that only investigated the response of phytoplankton to the addition of rainwater containing pollutant N [Paerl, 1985; Paerl et al., 1990, 1999; Willey and Cahoon, 1991; Willey and Paerl, 1993; Paerl and Fogel, 1994; Zou et al., 2000; Baker et al., 2007]. More importantly, the studies published to date have not considered how other processes associated with precipitation events (including the fluxes of solar radiation, freshwater-associated buoyancy, and momentum from enhanced winds) influence phytoplankton biomass and productivity. Therefore, the impact of precipitation on ocean phytoplankton biomass and productivity is likely to differ from that estimated from bioassay experiments. Unfortunately, there is little direct evidence linking precipitation events to changes in phytoplankton biomass or productivity. Indeed, we are aware of only one study that links time series of precipitation to either metric [Paerl, 1985].

In the present study we used satellite-derived data sets and meteorological products to investigate the links between surface ocean chlorophyll a (a proxy for phytoplankton biomass) concentration and two meteorological drivers: precipitation and wind. We assessed whether there are consistent responses of phytoplankton biomass to physical and chemical changes in the ocean associated with precipitation. Wind speed was included because of the expectation of a correlation with precipitation as well as prior research demonstrating a link between satellite data sets of wind speed and chlorophyll a [Kahru et al., 2010]. The study area involved coastal and offshore waters of the eastern U.S., where atmospheric N deposition has increased by approximately fivefold during the last two centuries, and the atmospheric load of reactive N is estimated to account for up to 40% of the new N introduced to this area during recent decades [Prospero et al., 1996; Castro and Driscoll, 2002]. Because of these large N inputs, we hypothesized that surface chlorophyll a concentrations would increase in response to precipitation events.

2 Study Area

The study area encompasses the ocean surface from 28°N to 44°N and from the East Coast of the United States to 60–70°W and is divided into two major regions and seven subregions (Figure 1a). The first major region includes the productive coastal shelf (CS) areas and comprises three smaller subregions (the Gulf of Maine, GOM; the Mid-Atlantic Bight, MAB; and the South Atlantic Bight, SAB). The second major region is characterized by low concentration of chlorophyll a (LC) and comprises four subregions: two south of 36°N (LCs1 and LCs2) and two north of 36°N (LCn1 and LCn2). Annual mean chlorophyll a concentrations estimated using satellite-derived ocean color measurements were used to determine subregional boundaries: 0.125 mg m−3 for LCs1/LCs2, 0.25 mg m−3 for LCs2/SAB and LCn1/LCn2, and 0.5 mg m−3 for LCn2/MAB and LCn2/GOM.

Figure 1.

(a) A map of the study area, showing the seven subregions (see text), which are indicated by colors and separated, in part, by contours of the long-term-mean (1997–2010) surface chlorophyll a concentration (black lines) based on the SeaWiFS level 3 product (0.083° × 0.083°). The colored dots in the Atlantic waters indicate the grid points of the NARR (North American Regional Reanalysis) precipitation product. The gray squares along the coastline of the U.S. indicate the locations of the precipitation gauge measurements used for evaluating precipitation products. Colored lines with arrows indicate the flow pathways of warm (red) and cold (blue) currents, respectively. The white circles and triangles indicate the locations of the NADP stations along the U.S. East Coast. (b) Mean wet N deposition (mmol m−2 yr−1) of all weekly measurements during the 1980s to 2000s versus latitude. Inset in Figure 1b indicates the variation in mean weekly wet N deposition (mmol m−2) with weekly precipitation (1 mm interval) observed at three NADP stations found in Figure 1a (FL99: upright triangle, NC03: inverted triangle, NY99: triangle facing left). Error bars in Figure 1b, including inset, represent the 95% confidence interval.

3 Data Sources

We used data sets of daily chlorophyll a (available on a 0.083° × 0.083° grid) for the period of 4 September 1997 to 11 December 2010, acquired from the level 3 product of the Sea-viewing Wide Field-of-view Sensor (SeaWiFS) [O'Reilly et al., 1998]. For precipitation we tested the accuracy of four data sets of daily accumulation, specifically the Climate Prediction Center morphing technique, Tropical Rainfall Measuring Mission Multi-satellite Precipitation Analysis, Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN), and North American Regional Reanalysis (NARR). The former three were binned into an area of 0.25° latitude × 0.25° longitude, and were derived from similar geostationary infrared and polar-orbiting passive microwave data sets using different merging algorithms; details of the merging and generation of these data sets are provided in Sapiano and Arkin [2009]. In contrast, the NARR product (32 km resolution), which is an extension of the National Centers for Environmental Prediction Global Reanalysis, is a numerical model that assimilates meteorological data, including precipitation gauge data over land and the Climate Prediction Center Merged Analysis of Precipitation product over the ocean (a merged data set of satellite-based and gauge-based measurements on a 2.5° latitude × 2.5° longitude grid) [Mesinger et al., 2006]. Evaluation of the four precipitation data sets using National Climatic Data Center rain gauge data indicated that the NARR product was the most suitable for our study and that the PERSIANN product was a reasonable second choice (rationale discussed in detail in the supporting information).

Several other data sets were also used to facilitate interpretation of our results. Daily wind speed data available on a 0.25° × 0.25° grid were acquired from the level 3 Quick Scatterometer product [Perry, 2001] and used to evaluate the effect of wind speed on chlorophyll a concentration. Monthly mean climatological nitrate (NO3) data on a 1° × 1° grid were acquired from the 2009 World Ocean Atlas [Garcia et al., 2010] and used in interpreting the response of chlorophyll a to precipitation. To quantify the contribution of the wet deposition of atmospheric nitrogen (ANDWET) to new production in the study area, ANDWET data (NO3 and NH4+ for the period 1980–2010) were obtained from 20 U.S. East Coast stations of the National Atmospheric Deposition Program (NADP, http://nadp.sws.uiuc.edu/). New production was calculated using published f-ratio values [Eppley and Peterson, 1979; Buesseler, 1998; Townsend, 1998; Lee, 2001; Bisagni, 2003] and satellite-derived annual mean primary productivity data from the Vertically Generalized Production Model [Behrenfeld and Falkowski, 1997], provided by the Oregon State University (OSU, http://www.science.oregonstate.edu/ocean.productivity/).

4 Methods of Data Analysis

Most of our results are presented using the NARR product and, for comparison, we show selected results using PERSIANN. We therefore describe our methodology using the NARR; the only methodological difference using PERSIANN is the resolution.

To minimize the large temporal and spatial variations in chlorophyll a concentration, prior to data analysis the daily SeaWiFS chlorophyll a (dChla) data were normalized to monthly mean climatological values (ClimChla) for the 1997–2010 period using the equation:

display math(1)

where nChla indicates normalized dChla. For example, dChla data for 1–30 November of all 14 years were normalized by monthly mean climatological value for November. The grid size of the daily nChla data (approximately 9 km in the study area) was matched with that of the daily NARR precipitation data (32 km) by averaging all daily nChla data available within each NARR grid box (Figure 1a). Using the averaged daily nChla data (referred to here as NChla where “N” indicates “NARR”) we evaluated the individual impacts of precipitation (ΔNChlaprcp), wind speed (ΔNChlawind), and precipitation with constant wind (ΔNChlaprcp|ws) on chlorophyll a concentration at every NARR grid point for each of the 12 calendar months.

4.1 ΔNChlaprcp

All daily NChla data were categorized as NChlaDRY and NChlaWET, which represent data collected on dry and wet days, respectively. Wet-day chlorophyll a data were not actually measured during precipitation, because the presence of cloud cover prevents satellite collection of data for the ocean. Therefore, wet-day chlorophyll a data were probably retrieved either immediately prior to or following precipitation events. As a result, some of the NChlaWET data obtained prior to precipitation events were not influenced by precipitation. In section 6 we assess possible biases caused by inclusion of such data in our calculations. As the mean absolute error of the NARR precipitation data was approximately 1 mm d−1 in precipitation ≤ 10 mm d−1 where approximately 92% of available chlorophyll a data during wet days were found, we defined wet days as those on which precipitation was > 1 mm d−1 and dry days as those on which precipitation was ≤ 1 mm d−1.

For each NARR grid box and calendar month, ΔNChlaprcp was calculated as the difference between the means of all dry-day NChla (math formula) and wet-day NChla (math formula):

display math(2)

Every NARR grid box had one ΔNChlaprcp value for each of the 12 calendar months. Finally, the subregional mean ΔNChlaprcp for each calendar month was calculated from individual ΔNChlaprcp values for all the grid boxes belonging to each subregion (Figure 1a), and the confidence intervals of the mean ΔNChlaprcp were calculated at a significance level of p = 0.05, using bootstrapping [Efron and Tibshirani, 1993].

The possible lag effect of precipitation events on chlorophyll a concentration was also estimated by matching NChla on a particular day with precipitation 1, 2, or 3 days earlier. This had the effect of altering the values of NChlaWET, NChlaDRY, and thus of math formula in equation (2).

We also tested whether the NARR-based ΔNChlaprcp could be reproduced using the PERSIANN product. Thus, we calculated ΔPChlaprcp (the prefix “P” indicates “PERSIANN”) in the same manner as we calculated ΔNChlaprcp, except that the nChla values were averaged within a given PERSIANN grid box (0.25° resolution).

4.2 ΔNChlawind

The effect of wind speed on chlorophyll a concentration was evaluated using only the chlorophyll a data collected during dry days; hence, no effect of precipitation was considered. Wind speed at each NARR grid box was estimated in the same way to calculating NChla without normalization. We calculated the differences (ΔNChlawind) in mean NChla concentrations observed during days with strong and weak wind speed as follows:

display math(3)

where days with strong and weak wind speed refer to days with wind speeds greater or less than, respectively, the mean wind speed over all dry days (math formula) at every NARR grid point and for each of the 12 calendar months. The subregional means were then estimated. For direct comparison with ΔNChlaprcp, we defined a new term, ΔNChlawind|ws as follows:

display math(4)

where ΔWSprcp and ΔWSwind are the differences in mean wind speeds representing wet and dry days and the differences in mean wind speeds representing days with strong and weak wind speed, respectively.

4.3 ΔNChlaprcp|ws

Under the condition of constant wind speed, the effect of precipitation on chlorophyll a concentration is expected to be associated with ANDWET and solar radiation change accompanied by precipitation events. The differences in NChla values between dry and wet days with constant wind were estimated from

display math(5)

In this calculation the effect of the wind speed difference between wet and dry days on chlorophyll a concentration is minimized. To achieve this, dry days on which wind speeds were within two standard deviations (±2σWS) from math formula were initially chosen (DRY*). If the difference between math formula and math formula was > 1 m s−1 (which was usually due to a higher value for math formula), the initial window (math formula ± 2σWS) for selecting dry-day (DRY*) data was incrementally shifted by 0.1 m s−1 (math formula ± 2σWS + 0.1 m s−1) until the difference between math formula and math formula was < 1 m s−1. Using this approach, the effect of precipitation with constant wind on chlorophyll a concentration could be more accurately assessed.

5 Results

5.1 Chlorophyll a Data Availability

Intense and extended precipitation events tend to decrease the availability of chlorophyll a data during wet days. As this issue is critical for assessing the validity of our analysis methods and results, we address it prior to presenting our main results (section 5.2).

The SeaWiFS chlorophyll a data were available only for approximately 28% of all dry (≤ 1 mm) days during the past 14 years (Figure 2a). This proportion would be the upper limit of data availability for wet days (>1 mm), which was only ~10%, presumably reflecting the greater cloud cover during wet days. The data availability decreased considerably in number with increasing precipitation (Figure 2b), probably because greater precipitation is associated with longer events, which lower the probability of the satellite seeing the surface on a given day. As a result, approximately 92% of available wet-day chlorophyll a data have daily precipitation less than 10 mm d−1. Therefore, the various data sets used in the present study were only adequate for investigating the effects of relatively weak precipitation events.

Figure 2.

Percentage (%) of available daily SeaWiFS chlorophyll a data as a function of (a) daily precipitation threshold (mm) and (b) daily precipitation interval (mm). Intervals X–Y on the horizontal-axis in Figure 2b indicates X < daily precipitation ≤ Y. (c) Ratio (%) of the number of successfully retrieved chlorophyll a data during wet days (> 1 mm) over that during dry days (≤ 1 mm) by subregion and season. (d) The number of wet-day (> 1 mm) chlorophyll a data per NARR grid box (32 km resolution) by subregion and season.

The ratio of the number of wet-day chlorophyll a data over the number of dry-day chlorophyll a data was mostly between 10 and 25%, and a mean value for all subregions and seasons was 17% when a value of 1 mm d−1 was used as the precipitation threshold (Figure 2c). The mean ratios over seven subregions were approximately 43, 11, and 8% when the thresholds were 0, 2, and 3 mm d−1, respectively (not shown). The number of wet-day chlorophyll a data points (threshold: 1 mm d−1) in the subregions were mostly between 20 and 80 (average 42) per NARR grid box (32 km resolution) with the highest values occurring in summer and the lowest in winter (Figure 2d). The mean numbers of wet-day chlorophyll a data points over the seven subregions were approximately 110, 27, and 19 per NARR grid box when the precipitation thresholds were 0, 2, and 3 mm d−1, respectively (not shown). The total number of wet-day chlorophyll a data points differed among subregions, ranging from tens of thousands (at a threshold of 0 mm d−1) to thousands (at a threshold of 3 mm d−1) (not shown). At a threshold of 1 mm d−1, the lowest number of wet-day chlorophyll a data was approximately 1,100 within LCn1 subregions and during winter season (November–January) while the highest number was approximately 17,000 within the LCs1 subregion and during summer season (June–August). In summary, although satellite-derived chlorophyll a data during wet days are limited in space and time, the high spatial resolution and long record lengths (14 years) of the SeaWiFS product has resulted in data set that is large enough for the execution of the present analysis.

5.2 Effect of Precipitation on Chlorophyll a Concentration

The ΔNChlaprcp is a measure of the change in chlorophyll a as a consequence of increased wind speed, solar radiation change, and the presence of ANDWET associated with precipitation events. The distributions of ΔNChlaprcp were patchy, but an increase in chlorophyll a during wet days was found throughout the year over most of the southern region (south of 36°N) and mainly from June to September in the northern region (north of 36°N) (Figure 3). In contrast, a decrease in chlorophyll a during wet days was evident in the northern region from October to April. ΔNChlaprcp ranged from −20% to 15% as a subregional mean value (Figure 4). The contrasting responses of chlorophyll a concentration to precipitation were related to seawater NO3 availability: Generally, ΔNChlaprcp was < 0 when NO3 concentrations were high (> 1 μM), and ΔNChlaprcp was > 0 when NO3 concentrations were low (< 1 μM) (Figure 3). The former trend was particularly evident in March, April, October, and November in the northern region.

Figure 3.

Change in monthly mean chlorophyll a concentration resulting from precipitation events (ΔNChlaprcp, %) for January to December. The red and yellow colors indicate the enhancement of chlorophyll a concentration, whereas the blue indicates the depression of chlorophyll a. The contour lines indicate the climatological monthly mean NO3 concentration (μM).

Figure 4.

(a) A composite regression fit (red line) between subregional means of all ΔNChlaprcp and NO3 data used in (b) GOM, (c) MAB, (d) SAB, (e) LCn2, (f) LCn1, (g) LCs2, and (h) LCs1. Monthly mean ΔNChlaprcp (%, red circles) and NO3 concentration (μM, blue circles) for subregions shown in Figures 4b–4h. The three NO3 values that are higher than 5 μM were forced to 5 μM for visualization. Vertical lines are 95% confidence intervals.

Subregional means more clearly show the relationship between ΔNChlaprcp and NO3 concentration (red and blue circles in Figure 4, respectively). Chlorophyll a concentration either decreased or remained unchanged during wet days in the GOM and the MAB except during summer, when an increase was observed. During the transition from spring to summer in the GOM there was a sharp increase in ΔNChlaprcp, which was accompanied by a corresponding decrease in the NO3 concentration. A similar but more gradual spring-to-summer increase in ΔNChlaprcp was observed in the MAB. Positive values of ΔNChlaprcp persisted from May to July in the GOM, and from June to August in the MAB, and decreased thereafter in each of these subregions. In contrast, little seasonality was observed in the SAB, where ΔNChlaprcp was generally positive and the NO3 concentration was low. In each of the four LC subregions (shown in the right column in Figure 4), positive values of ΔNChlaprcp persisted throughout most of the year, except in some winter months in the northern region (LCn1 and LCn2) when the NO3 concentration was elevated. The relationship between subregional means of ΔNChlaprcp and NO3 concentration was significant (p < 0.001) and showed that the transition from enhanced to reduced chlorophyll a concentration as a result of a precipitation event occurred at a NO3 concentration of 1–2 μM (Figure 4a). The insignificant effect of lag days on ΔNChlaprcp was obvious in all subregions (SOM Figure S2). The correlation (r = 0.71, p < 0.01) between NARR-based ΔNChlaprcp and PERSIANN-based ΔPChlaprcp was significant (SOM Figure S3). The fact that two independent precipitation data products (NARR and PERSIANN) yielded similar results increases our confidence that the observed chlorophyll a changes associated with precipitation were not artifacts arising from errors resulting from the precipitation data products or errors associated with analyzing those data products.

5.3 Effect of Wind Speed on Chlorophyll a Concentration

Figures 5 shows subregional means of ΔNChlawind and their relationship to NO3 concentration. There was a significant correlation between subregional means of ΔNChlawind and the NO3 concentration (p < 0.001) (Figure 5a), and the spatial and temporal patterns in ΔNChlawind were similar to those of ΔNChlaprcp (Figure 4). However, the magnitude of the ΔNChlawind was generally greater than that of ΔNChlaprcp, which can be explained based on the observation that the wind speed difference between wet and dry days (ΔWSprcp) was substantially less than the wind speed differences between strong and weak wind speed groups during dry days (ΔWSwind, Figure 6a). To show this more clearly, we present in Figure 6c regional means of ΔNChlaprcp, ΔNChlawind, and ΔNChlawind|ws when ΔNChlaprcp > 0 (generally low-nitrate conditions). Recall that ΔNChlawind|ws is a normalized form of ΔNChlawind, which takes into account the difference between ΔWSprcp and ΔWSwind. Figure 6c clearly shows that for ΔNChlaprcp > 0, ΔNChlawind|ws was either similar to or a few percent lower than ΔNChlaprcp in all subregions.

Figure 5.

(a) A composite regression fit (red line) between subregional means of all ΔNChlawind and NO3 data used in (b) GOM, (c) MAB, (d) SAB, (e) LCn2, (f) LCn1, (g) LCs2, and (h) LCs1. Monthly mean ΔNChlawind (%, red circles) and NO3 concentration (μM, blue circles) for subregions shown in Figures 5b–5h. The three NO3 values that are higher than 5 μM were forced to 5 μM for visualization. Vertical lines are 95% confidence intervals.

Figure 6.

(a) Differences in wind speed (ΔWSprcp and ΔWSwind) between wet and dry days, and between strong and weak wind speed groups during dry days, respectively. (b) Relationship between wind speed and daily precipitation in all subregions. Mean wind speeds were estimated at 1 mm daily precipitation intervals (0–1 mm, 1–2 mm, etc.). (c) ΔNChlaprcp, ΔNChlawind, and ΔNChlawind|ws for the seven subregions. Means (gray and white bars) and standard deviations (error bars) from the means were calculated only for ΔNChlaprcp > 0 at each subregion.

To isolate the impact of precipitation only on chlorophyll a concentration, we present in Figure 7 ΔNChlaprcp|ws, which minimizes the effect of wind speed change on ΔNChlaprcp. Note that the magnitude of ΔNChlaprcp|ws is less than that of ΔNChlaprcp or ΔNChlawind, indicating that factors other than wind speed change have small effects on chlorophyll a concentration during precipitation events. However, for at least half of the year we found positive values of ΔNChlaprcp|ws (up to 10%) in all subregions except MAB and GOM. Positive ΔNChlaprcp|ws values were particularly pronounced in the summer months in the SAB, LCn1, and LCn2, where seawater NO3 concentrations were low (Figure 7).

Figure 7.

(a) A composite regression fit (red line) between subregional means of all ΔNChlaprcp|ws and NO3 data used in (b) GOM, (c) MAB, (d) SAB, (e) LCn2, (f) LCn1, (g) LCs2, and (h) LCs1. Monthly mean ΔNChlaprcp|ws (%, red circles) and NO3 concentration (μM, blue circles) for subregions shown in Figures 7b–7h. The three NO3 values that are higher than 5 μM were forced to 5 μM for visualization. Vertical lines are 95% confidence intervals.

6 Discussion

6.1 Possible Mechanisms for the Observed Change in Chlorophyll a Associated With Precipitation

The finding of positive values of ΔNChlaprcp in low-nutrient areas (Figures 4e–4h) was consistent with our hypothesis that precipitation events increased chlorophyll a levels through the stimulatory influence of deposited N. However, the negative values of ΔNChlaprcp in winter months in the nutrient-replete areas (GOM and MAB) must have been due to other factors because additional N would have been expected to have no influence. This led us to consider changes in other meteorological variables that might be correlated with precipitation, particularly wind speed. Figure 6b shows that in all subregions higher wind speeds were associated with higher levels of precipitation. Therefore, it is possible that some of the reduction in chlorophyll a concentration that was evident during wet days was a result of stronger winds, which would deepen the mixed layer, decrease the light available to phytoplankton, and consequently decrease chlorophyll a concentration. In a global analysis of monthly averages of remotely sensed chlorophyll a and wind speed, Kahru et al. [2010] also came to the conclusion that higher wind speeds significantly depressed surface chlorophyll a levels in nutrient-replete areas and enhanced chlorophyll a concentration in oligotrophic regions of the ocean, the latter effect presumably as a result of wind-induced entrainment of nutrients below the mixed layer. Thus, it is probable that some of the increase in chlorophyll a concentration we observed during wet days in LC subregions (Figures 3 and 4) was a consequence of an increase in wind-driven mixing. This is further supported by the overall similarity in the patterns of the impacts of precipitation (ΔNChlaprcp) and wind speed (ΔNChlawind) on chlorophyll a concentration (Figure 4 versus Figure 5) and of the comparable signals of ΔNChlaprcp and ΔNChlawind|ws (Figure 6c). Therefore, the major portion of the variability in ΔNChlaprcp was a result of the relationship of wind speed to precipitation, and thus, the wind speed is a major driver in determining the impact of precipitation events on chlorophyll a concentration.

The influence of deposited N (ANDWET) on chlorophyll a concentration may also be included in ΔNChlaprcp but cannot be isolated from ΔNChlaprcp values. We suggest that the stimulatory effect of ANDWET on chlorophyll a concentration is reflected in ΔNChlaprcp|ws (Figure 7), based on the following rationale. Two key factors in determining ΔNChlaprcp|ws are ANDWET and decreased solar radiation, which have opposite effects on chlorophyll a concentration. In theory, ANDWET should have increased the chlorophyll a concentration (i.e., ΔNChlaprcp|ws always ≥ 0), whereas decreased solar radiation should have acted to decrease the chlorophyll a concentration (i.e., ΔNChlaprcp|ws always < 0). Either factor could determine ΔNChlaprcp|ws because our subregions show the typical seasonal alternation of nutrient and light limitation of phytoplankton growth [O'Reilly and Zetlin, 1998]. One possibility is that some subregions are colimited and also showed positive ΔNChlaprcp|ws values in some months. In such rare cases the positive ΔNChlaprcp|ws values would indicate that the stimulatory effect of ANDWET is greater than that of reduced solar radiation, giving a lower bound of the ANDWET effect on chlorophyll a concentration. In addition, one characteristic of phytoplankton that conflicts with our rationale above is the light dependence of variations in chlorophyll a to biomass (e.g., as carbon) ratio. Phytoplankton are known to increase their ratio of chlorophyll a to biomass as they move to greater depths in order to harvest light more efficiently [Le Bouteiller et al., 2003]. Along the same line of reasoning, the chlorophyll a concentration under the reduced radiation during wet days may be increased. However, the wet-day enhancement in chlorophyll a is unlikely because the use of additional energy to increase chlorophyll a without adequate nutrient supply can be disadvantageous to phytoplankton survival. Therefore, in the subregions where seawater NO3 concentrations were low and ΔNChlaprcp|ws was positive (relatively evident in SAB, LCn1, and LCn2), the positive ΔNChlaprcp|ws values were attributed to ANDWET because this factor tended to reduce the nutrient deficiency.

We found the largest negative ΔNChlaprcp|ws values in the GOM and MAB. In the coastal shelf areas (GOM, MAB, and SAB), ANDWET increased from 25°N to 42°N and then rapidly decreased northward of 42°N (Figure 1b). This meridional distribution of ANDWET suggested that the GOM and MAB would show the greatest values of ΔNChlaprcp|ws, but the SAB, and other LC subregions where the NO3 and chlorophyll a concentrations were low, showed greater fractional chlorophyll a increases. Such a counter-intuitive response (negative ΔNChlaprcp|ws) of chlorophyll a concentration to ANDWET indicates that the addition to the GOM and MAB of rainwater containing dissolved N was not effective in increasing phytoplankton biomass, and the negative ΔNChlaprcp|ws values were probably a result of reduced mixed-layer light levels due to less surface solar radiation accompanying precipitation events.

A possible, but probably less likely effect is that salinity stratification as a result of rain events acted to reduce mixed layer depth and increase light availability. In the nutrient-limited areas, rather than light and salinity stratification, nutrient entrainment due to mixed layer deepening and atmospheric deposition of nitrogen are primary factors for causing the observed increase in chlorophyll a. In the light-limited and nutrient-rich subregions, the chlorophyll a concentration in response to precipitation was decreased, which indicates that light availability was actually reduced and the effect of wind speed increase exceeded that of salinity stratification.

Another result from our analysis that deserves discussion is that we found no lag effect of precipitation on chlorophyll a concentration (ΔNChlaprcp), possibly because the nutrients added by precipitation were not sufficient to support phytoplankton growth for more than 1 day (SOM Figure S2). In the study area ANDWET was generally < 1 mmol N m−2 per 10 mm precipitation (inset in Figure 1b), which corresponds to a rainwater N concentration of 100 μM, which would have been diluted to less than 0.1 μM because of mixing within a few hours, about the time it takes for upper 10 m to mix [Michaels et al., 1993]. An immediate response to precipitation of phytoplankton without lag periods indicates that our results were not probably influenced by the elevated riverine discharge from possibly simultaneous precipitation in upstream watersheds, which takes much longer to influence phytoplankton in the coastal ocean. For example, Acker et al. [2005] found a significant correlation between freshwater discharge and chlorophyll a in the Chesapeake Bay with a lag of 1 month.

6.2 Contribution of Wet N Deposition to New Production and Chlorophyll a Concentration

Our results from the analysis of precipitation products and ground-based measurements of ANDWET can provide information about the impact of precipitation on ocean productivity and chlorophyll a concentration. Based on direct measurements of ANDWET along the East Coast of the U.S., the N supply through wet deposition was estimated to be 25–45 mmol N m−2 yr−1 (Figure 1b). This amount of ANDWET was estimated to contribute approximately 1–2% of new production in the coastal shelf areas (GOM, MAB, and SAB), which was calculated to be roughly 2000 mmol N m−2 yr−1 using an f-ratio of 0.3 [Townsend, 1998; Bisagni, 2003], primary production of ~500 g C m−2 yr−1 (estimated from the OSU productivity data), and the C:N ratio of ~7.3 [Anderson and Sarmiento, 1994]. The amount of ANDWET rapidly decreased in an eastward direction toward the shoreline. For example, there was an approximately 40% reduction from New York (NY68) to the eastern tip of Massachusetts (MA01), which is located approximately on the 42°N latitudinal line (Figure 1). Therefore, in estimating ANDWET for the LC subregions we used direct measurements made in Bermuda (32.3°N, 64.7°W; in the LC area) rather than using measurements obtained from shoreline stations. Bates and Peters [2007] reported that the annual mean wet deposition of NO3 in Bermuda was approximately 5.2 mmol N m−2 yr−1. On the East Coast of the U.S. the wet deposition ratio of NH4+ to NO3 was estimated to be approximately 0.7. Based on this ratio, the estimated ANDWET at the Bermuda site is approximately 8.8 mmol N m−2 yr−1, which is 25% of the value observed at a coastal station located in North Carolina (NC03; 36.13°N, 77.17°W). Knapp et al. [2010] also reported the total ANDWET at Bermuda is 10–19 mmol N m−2 yr−1 which includes organic and inorganic nitrogen species (NH4+ and NO3). As the Bermuda site is approximately 85 km from the eastern boundary of the study area, the wet deposition measured at this site can serve as the lower limit of ANDWET for the LC subregion. New production in the oligotrophic ocean near Bermuda was estimated to be approximately 130 mmol N m−2 yr−1, based on an f-ratio of 0.1 [Eppley and Peterson, 1979; Buesseler, 1998; Lee, 2001], primary production of ~110 g C m−2 yr−1 at LCs1 (estimated from the OSU data) and the C:N Redfield ratio [Anderson and Sarmiento, 1994]. Therefore, ANDWET in the LC subregions was responsible for approximately 7–15% of new production based on the ANDWET estimates of Bates and Peters [2007] and Knapp et al. [2010].

The range of ANDWET per a relatively weak (i.e., < 10 mm) precipitation event is approximately 0.2–0.6 mmol N m−2 in the CS subregions (inset in Figure 1b), which is equivalent to chlorophyll a concentration increase by 0.5–1.5 × 10−2 mg m−3 for a mixed layer depth of ~40 m (reasonable, except for winter) [Steinberg et al., 2001] and a phytoplanktonic chlorophyll-to-N ratio of 1 mg chlorophyll a to 1 mmol N [Doney et al., 1996] were used. Therefore, an independent estimate of the contribution of ANDWET to chlorophyll a concentration is less than 0.5–1.5% in the CS subregions where the annual mean chlorophyll a concentration is typically higher than 1 mg m−3 (Figure 1a). Around Bermuda, ANDWET for an individual precipitation event is expected to be ~40% of that for the CS regions, using the ratio of measured annual mean ANDWET in the two regions (25–45 mmol N m−2 yr−1 for CS versus 9–19 mmol N m−2 yr−1 for Bermuda). Given the annual mean chlorophyll a concentration of ~0.125 mg m−3 near Bermuda, the contribution of ANDWET to chlorophyll a concentration is approximately 1.6–5% via relatively weak precipitation events. These values are consistent with our estimates of ΔNChlaprcp|ws particularly in the LCn1 and LCn2 subregions. Although these are approximations and sensitive to the choice of the chlorophyll-to-nitrogen ratio and the mixed layer depth, the consistency of the results obtained from two independent methods indicates that our analyses based on ΔNChlaprcp and ΔNChlaprcp|ws reasonably reflected the response of ocean phytoplankton biomass to precipitation events.

6.3 Limitations of Our Study

There are caveats associated with the methods employed to evaluate the individual impacts of precipitation (ΔNChlaprcp) and ANDWET (positive ΔNChlaprcp|ws) on chlorophyll a concentration using satellite-derived data sets of chlorophyll a concentration. Two caveats include the use of wet-day chlorophyll a data obtained prior to precipitation events and the incorrect separation of chlorophyll a data into wet- and dry-day categories as a consequence of the erroneous predictions of the NARR precipitation. These unavoidable errors could lead to underestimation of ΔNChla associated with precipitation events because the errors tend to reduce the difference in chlorophyll a values between wet and dry days (ΔNChlaprcp). However, it is likely that only a small proportion of the wet-day chlorophyll a data were associated with such biases. Had the opposite been the case, the significant and consistent changes in chlorophyll a associated with precipitation would not have been found in most of the study areas.

Our analysis of satellite-derived chlorophyll a data cannot account for the presence of a subsurface chlorophyll a maximum and colored dissolved organic matter (CDOM). The subsurface chlorophyll a maximum typically found in stratified oceans is often invisible to satellite sensors [Hyde et al., 2007]. In addition, CDOM is lower in surface than subsurface layers because of photo bleaching at the surface, and the standard chlorophyll a algorithm cannot distinguish CDOM from chlorophyll a [Coble, 2007]. As a result, wind-driven entrainment of subsurface chlorophyll a and CDOM can result in an erroneous increase in the satellite-derived chlorophyll a concentration. In areas where these factors are considerable the NChlaWET may be overestimated, and thus, ΔNChlaprcp can be biased toward more positive values. Thus, ΔNChlaprcp can be overestimated when it is positive, and vice versa. However, U.S. coastal regions or open ocean regions in winter (mostly ΔNChlaprcp < 0) are not affected by such biases because of shallow subsurface chlorophyll a maxima (~15 m [Hyde et al., 2007]) and strong mixing. Such biases are less likely to occur in summer in the LC subregions (mostly ΔNChlaprcp > 0) because it is unlikely that weak precipitation extends from the mixed layer (~20 m) to the subsurface maximum layer (60–100 m) [Steinberg et al., 2001; Nelson et al., 2004].

7 Conclusion

We found that precipitation events in coastal waters of the eastern United States increased the chlorophyll a concentration up to 15% in low-nutrient areas (< 1 μM NO3) but decreased the chlorophyll a concentration in nutrient-replete areas (> 1 μM NO3). The underlying cause of these contrasting results was probably related to the factors limiting phytoplankton growth and the correlation between wind speed and precipitation. The new N input via precipitation and the enhanced entrainment of N from below the mixed layer increased the in situ chlorophyll a concentration in nutrient-depleted environments. In contrast, it is likely that wind-induced deepening of the mixed layer associated with precipitation events lowered the chlorophyll a concentration in nutrient-rich environments, probably because of reduced light availability. Despite the errors inherent in satellite-based measurements, our study has shed light on the role of precipitation in ocean productivity, which is a phenomenon that is yet to be fully appreciated. As precipitation can either increase or decrease ocean phytoplankton productivity, any change in the pattern of precipitation will modify the natural state of the ocean under future climatic and environmental conditions.

Acknowledgments

This work was supported by Mid-career Researcher Program (No. 2012R1A2A1A01004631) funded by the National Research Foundation (NRF) of Ministry of Education, Science and Technology and by the project titled “Long-term change of structure and function in marine ecosystems of Korea” funded by the Ministry of Land, Transport and Maritime Affair. Support for T.-W.K was provided by Basic Science Research Program through the NRF (2012R1A6A3A04038883). Support for R.G.N. was provided by NASA's Ocean Biology and Biogeochemistry Program and NASA's Interdisciplinary Science Program.

Ancillary