nt 109 (2007) 249–259www.elsevier.com/locate/rse

Remote Sensing of Environme

Remote sensing of water clarity in Tampa Bay

Zhiqiang Chen ⁎, Frank E. Muller-Karger, Chuanmin Hu

Institute for Marine Remote Sensing (IMaRS), College of Marine Science, University of South Florida,140 Seventh Ave. S., St. Petersburg, FL 33701, USA

Received 30 August 2006; received in revised form 5 January 2007; accepted 6 January 2007

Abstract

We examined the spatial and temporal variability of the Secchi Disk Depth (SDD) within Tampa Bay, Florida, using the Sea-viewing WideField-of-View Sensor (SeaWiFS) satellite imagery collected from September 1997 to December 2005. SDD was computed using a two-stepprocess, first estimating the diffuse light attenuation coefficient at 490 nm, Kd(490), using a semi-analytical algorithm and then SDD using anempirical relationship with Kd(490). The empirical SDD algorithm (SDD=1.04×Kd(490)

−0.82, 0.9bSDDb8.0 m, r2=0.67, n=80) is based onhistorical SDD observations collected by the Environmental Protection Commission of Hillsborough County (EPCHC) in Tampa Bay. SeaWiFSderived SDD showed distinctive seasonal variability, attributed primarily to chlorophyll concentrations and color in the rainy season and toturbidity in the dry season, which are in turn controlled by river runoff and winds or wind-induced sediment resuspension, respectively. The Bayalso experienced strong interannual variability, mainly related to river runoff variability. As compared to in situ single measurements, the SeaWiFSdata provide improved estimates of the “mean” water clarity conditions in this estuary because of the robust, frequent, and synoptic coverage.Therefore we recommend incorporation of this technique for routine monitoring of water quality in coastal and large estuarine waters likeTampa Bay.© 2007 Elsevier Inc. All rights reserved.

Keywords: Remote sensing; Light attenuation coefficient; Water clarity; Secchi disk depth; Bio-optical algorithm; SeaWiFS; Tampa Bay estuary

1. Introduction

TampaBay is the largest open-water estuary in Florida, U.S.A.,with a surface area of ∼1000 km2. It is traditionally divided intofour main sub-segments, namely Old Tampa Bay (OTB),Hillsborough Bay (HB), Middle Tampa Bay (MTB), and LowerTampaBay (LTM) (Fig. 1). TampaBay is a diverse and productivenatural system that provides a vital habitat for crustaceans, fish,shellfish and a variety of marine mammals, reptiles and birds(Harwell et al., 1995), which contributes over $5 billion annuallyto the economy of the state from trade, tourism, development, andfishing (FDCA, 1996). It is therefore critical that the developmentof the Bay area be conducted in an environmentally sound way tosustain a healthy system.

In the decades prior to the 1980's, Tampa Bay was heavilypolluted by nutrient loadings from sources like sewage and other

⁎ Corresponding author. Tel.: +1 727 553 1186.E-mail address: zchen@marine.usf.edu (Z. Chen).

0034-4257/$ - see front matter © 2007 Elsevier Inc. All rights reserved.doi:10.1016/j.rse.2007.01.002

wastewater. This led to sustained phytoplankton blooms thatreduced the water clarity, which in turn is considered as the causeof the substantial losses of seagrass coverage (Lewis et al., 1998;Tomasko et al., 2005). Since then, significant ecosystem resto-ration efforts have been under way. In 1990, the Tampa BayNational Estuary Program (TBNEP) was established to coor-dinate and integrate efforts to restore and protect the Bay. In1996, the TBNEP developed a Comprehensive ConservationandManagement Plan (CCMP), which focused on restoration ofseagrass to levels similar to those observed in the 1950s byreducing nutrient (primary nitrogen) inputs into the Bay (Janicki& Wade, 1996). Water quality of the Bay has gradually im-proved, and some of the seagrass has recovered (Johansson,2000; Tomasko et al., 2005). Thus water clarity in Tampa Bayhas been monitored as a measure of the impact of nutrient loadon phytoplankton concentrations, and is a key parameter used inriverine nutrient input management (Janicki et al., 2001).

Water clarity in Tampa Bay has been measured with a Secchidisk once per month at established stations (Janicki & Wade,

Fig. 2. Light attenuation coefficient at 490 nm (Kd(490), m−1) derived from

SeaWiFS vs. in situ Secchi disk depth (SDD, m). Kd(490) was estimated using(1) the empirical band-ratio algorithm (EA) (triangles, updated Mueller's (2000)algorithm) and (2) the semi-analytical algorithm(SA) (circles, Lee et al., 2005a).

250 Z. Chen et al. / Remote Sensing of Environment 109 (2007) 249–259

1996; Fig. 1), as done in many coastal environmentalmonitoring programs. A white disk (sometimes also painted inblack/white quarters), usually ∼20 cm in diameter, is loweredinto the water and the depth at which the disk is no longervisible is recorded as the Secchi Disk Depth (SDD, in units ofmeters). SDD provides a simple, inexpensive measurement ofthe rate at which light is attenuated with depth. Due to the size ofTampa Bay and logistical limitations, it usually takes threeweeks to conduct bay-wide SDD observations at a series of pre-established historical stations (Fig. 1). Clearly this samplingprogram is not synoptic, so a natural question is whether thismonitoring strategy is sufficient to help assess the mean andvariability in water clarity in time over the extent of Tampa Bay,the extent and changes in eutrophication due to anthropogenicnutrient inputs, and the impact on areas where seagrass re-growth is desired.

Water clarity or specifically light attenuation coefficientshave also been estimated for coastal and open oceanwaters usingsatellite ocean color measurements drawing upon the repeatedand synoptic sampling capability. However, previous applica-tions frequently used site- and/or time-specific empiricalalgorithms (Austin & Petzold, 1981; Mueller, 2000; Prasad

Fig. 1. SeaWiFS true-color (Red–Green–Blue) satellite image of Tampa Bay,U.S.A., with an inset showing the location of this estuary in the state of Florida.The four main sub-segments of the bay are Old Tampa Bay (OTB),Hillsborough Bay (HB), Middle Tampa Bay (MTB), and Lower Tampa Bay(LTB). Major rivers are the Hillsborough River (HR), Alafia River (AR), LittleManatee River (LMR), and Manatee River (MR). Environmental ProtectionCommission of Hillsborough County's (EPCHC) water quality monitoringstations are shown using different symbols to indicate monthly sampling times:diamonds (OTB) are generally sampled the first week of the month, triangles(HB) the second week, and squares (MTB and LTB) in the third or fourthweeks. Time-series data were extracted for this study at four stations (92, 23,14, and 40, crosses).

et al., 1998; Stumpf & Pennock, 1991). For example, thestandard empirical algorithm used to estimate the diffuse lightattenuation coefficient at 490 nm, Kd(490) (m

−1), from the Sea-viewing Wide Field-of-View Sensor (SeaWiFS) data wasdeveloped based largely on open ocean observations wheremost Kd(490)b0.15 m−1 (Mueller, 2000). Thus the previousempirical algorithms are prone to generate large errors whenapplied to coastal and estuarine waters where the opticalproperties are different from those waters for algorithmdevelopment (Lee et al., 2005b). More recently, improvedlight attenuation coefficient estimates for coastal waters havebeen possible by application of a semi-analytical algorithm to insitu collected data (Lee et al., 2005a,b). However the accuracy ofestimates based on space-based observations remains unknown,because satellite-derived reflectance contains uncertainties overcoastal and estuarine zones due to non-zero remote sensingreflectance (Rrs) in the near-infrared in some turbid waters, and/or due to high concentrations of blue-absorbing aerosols in somenear-shore atmosphere (Harding et al., 2005; Hu et al., 2000).

Here, we used a two-step transformation of SeaWiFS ima-gery to estimate water clarity in Tampa Bay between September1997 and December 2005. We first we estimated Kd(490) basedon the semi-analytical method of Lee et al. (2005a). The Kd

(490) estimates were closely related to SDD observations col-lected by the Environmental Protection Commission ofHillsborough County (EPCHC). This allowed us to examinethe temporal and spatial variability of SDD in Tampa Bay usingthe satellite data, by computing SDD based on the remotely-sensed Kd(490). Based on our results, we further providerecommendations for improving the water quality monitoringprogram of Tampa Bay and similar programs in other estuaries.

2. Methods and materials

2.1. Satellite data

SeaWiFS “merged local area coverage” (MLAC) Level-1Adata (nominal spatial resolution of ∼1 km) were downloadedfrom the NASA Goddard Space Flight Center (GSFC, http://

Fig. 3. Overall cross-year monthly composites (September 1997–December 2005) of SeaWiFS SDD (m). Color legend also shows the corresponding Kd(490) values(m−1). White color within Tampa Bay represents shallow water (bottom depthb2.0 m). Grey color represents land.

251Z. Chen et al. / Remote Sensing of Environment 109 (2007) 249–259

252 Z. Chen et al. / Remote Sensing of Environment 109 (2007) 249–259

oceancolor.gsfc.nasa.gov/), and processed using the SeaWiFSData Analysis System software (SeaDAS, Version 4.9). Twomethods were used to estimate Kd(490). One was the defaultSeaDAS empirical band-ratio algorithm (Mueller, 2000). Thisalgorithm is based on the relationship between Kd(490) andthe blue-to-green ratio of normalized water leaving radiance,Lnw(490)/Lnw(555). The second was the semi-analyticalalgorithm proposed by Lee et al. (2005a), which first derivesabsorption and backscattering coefficients from Rrs and thenuses these coefficients to estimate Kd. During SeaWiFS imageprocessing, pixels were masked for conditions of atmosphericcorrection failure, land, clouds, large sun glint, and large solar/sensor angles. Processing software flags for stray light,shallow water, negative water-leaving radiance, and turbidcase 2 water were disabled to increase the data coverage intime and space over Tampa Bay.

2.2. Field and ancillary data

We did not find published values for in situ Kd(490) obser-vations in Tampa Bay. Therefore, historical in situ SDDmeasurements from the EPCHC's Tampa Bay water quality

Fig. 4. Overall cross-year monthly means of SeaWiFS (filled circles) and in situ (openThe number of SeaWiFS observation days in eachmonth is shown on the right-hand sidthe y-axis scale for Sta. 92 is different from those for other stations. The vertical error bHB are shown due to lack of sufficient points for the relationship shown in Fig. 2.

monitoring program (Boler et al., 1991) were used as a surrogateof Kd(490). Indeed SDD has been empirically related to lightattenuation coefficients in some previous studies (Giesen et al.,1990; Jean-Franc & Giuseppe, 2004; Kratzer et al., 2003). TheEPCHC program also collects chlorophyll concentrations,turbidity (reported in nephelometric turbidity units or NTU), andcolor (Pt-units) measurements. These parameters were used in thisstudy to examine the factors affecting water clarity in the Bay.

Daily averaged river flow rates (1997–2005) of the AlafiaRiver and the Hillsborough River, the two largest tributariesdischarging fresh water into Tampa Bay, were obtained from theUnited States Geological Survey National Water InformationSystem (USGS NWIS). We derived calendar monthly means ofriver flow, as well as overall cross-year monthly flow averages(i.e. averages of the same month across all years). Wind data(1997–2005) were obtained from one of the National OceanicAtmospheric Administration (NOAA) Tampa Bay physicaloceanographic real-time (PORT) stations located in Tampa Baynear Saint Petersburg (27 °45.6′N, 82 °37.6′W). Wind speedmeasurements were binned into daily means, and the number ofdays when daily wind speed was N4.0 m s−1 was calculated toidentify conditions for wind-induced sediment resuspension

circles) Secchi disk depth (SDD, m) at several stations across Tampa Bay (Fig. 1).e, while there are only 8 in situ observations in eachmonth (not shown). Note thatars indicate the standard deviation of satellite or in situ observations. No data from

Fig. 5. Overall cross-year monthly means of in situ (A) chlorophyll (mg m−3), (B) color (Pt-units), and (C) turbidity (nephelometric turbidity units, NTU)measurements from the EPCHC's monitoring program at several stations in various bay-segments (see Fig. 1 for the station locations) collected between September1997 and December 2005. The horizontal lines show the means of each measurement over the 8-year period at each station.

253Z. Chen et al. / Remote Sensing of Environment 109 (2007) 249–259

Fig. 5 (continued ).

254 Z. Chen et al. / Remote Sensing of Environment 109 (2007) 249–259

events. Thiswind speed thresholdwas based on visual inspectionsof time series of wind speed and sediment resuspension events inTampa Bay (Chen et al., in press-b).

2.3. Satellite-in situ comparison

A narrow window of 2 h was used to find matching in situand SeaWiFS satellite observations (Bailey &Werdell, 2006). Amedian value from a 3×3 pixel box centered at an in situmeasurement site was used to filter sensor and algorithm noise(Hu et al., 2001). To ensure spatial homogeneity in the satellitedata, the median value was used only when the number of validpixels identified with the satellite data processing flags was N4and the coefficient of variation (CV) was b0.4. Since theaverage depth of Tampa Bay is ∼4.0 m, all pixels shallowerthan 2.0 m were masked using a digital bathymetric database(Gesch & Wilson, 2001) and excluded from the match-upcomparison to minimize possible interference due to bottomreflectance. The depth of 2.0 m was chosen by trial-and-error,seeking a depth threshold that provided the highest correlationcoefficient between in situ SDD and satellite Kd(490).

3. Results

3.1. Comparison between in situ SDD and satellite Kd(490)

A total of 80 matching pairs were found from 1997 to 2005after applying the various data quality control criteria described

above. The matching pairs were spatially distributed as follows:Old Tampa Bay (OTB) had seven matching station pairs, MiddleTampa Bay (MTB) had eight, and Lower Tampa Bay (LTB) hadfive stations. Only one match-up station was found forHillsborough Bay (HB) due to the small area of that segment ofthe Bay, where there are also several small islands. The matchingpairs were relatively evenly distributed in time across four seasons(24 in spring, 20 in summer, 9 in fall, and 27 in winter).

The statistical relationships between in situ SDD and Kd(490)derived empirically (EA) and the Kd(490) derived semi-analytically (SA) were significantly different (Fig. 2). Kd

EA(490)showed no significant relationship with in situ SDD (r2=0.14,n=80), while Kd

SA(490) showed a strong correlation with in situSDD (SDD=1.04×Kd

SA(490)−0.84, 0.9bSDDb8.0 m, r2=0.67,n=80). The median ratio (a measure of overall bias) between thepredicted and in situ measured SDD from the 80 matching pairswas 1.00 (mean ratio was 1.02). The median absolute percentagedifference (MPD, a measure of uncertainty, Bailey & Werdell,2006) was 14% (mean absolute percentage difference was∼16%). The Kd

SA(490) root mean square error (RMSE) was0.55 m over the observed SDD range from 0.9 to 8.0 m.

These comparisons suggested that KdSA(490) can be used to

estimate SDD over a large dynamic range (about one order ofmagnitude in both parameters) (Fig. 2), in most areas of TampaBay, and in all seasons across years. Therefore we applied theregression relationship to the SeaWiFS series of images(September 1997–December 2005) to obtain a time-series ofSDD images for Tampa Bay. Calendar monthly means and

Fig. 6. (A) Overall cross-year monthly means of river flow from the HillsboroughRiver (open circles) and the Alafia River (filled circles) from 1997 to 2005. (B) Thetotal number of days when the daily averaged wind speed was N4.0 m s−1 from1997 to 2005 at the PORTstation near Saint Petersburg (27 °45.6′N, 82 °37.6′W).

255Z. Chen et al. / Remote Sensing of Environment 109 (2007) 249–259

overall cross-year monthly means of SDD were derived fromthe nearly 8-year time-series of daily SDD imagery.

3.2. SDD image series

The overall cross-year monthly composites of SDD showeddistinct spatial and temporal patterns across Tampa Bay (Fig. 3).A seasonal cycle was apparent, with smaller Kd(490) (largerSDD) from May to August, and larger Kd(490) (smaller SDD)from November to March. Relatively larger SDD values wereconsistently found in LTB and MTB in all months (e.g., N4.0 min May), primarily along the deep channel in the central portionof the Bay. OTB and HB consistently showed smaller SDDvalues except in HB during July, August, and September, whenSDD in these areas appeared to be overestimated (e.g., N4.0 m).This is likely the result of an erroneous atmospheric correction(see discussion below). Due to these artifacts and insufficientvalidation points for the relationship shown in Fig. 2 (only onestation in HB), we omitted HB from subsequent analyses.

The spatial and seasonal variability can be further charac-terized by extracting time-series data at selected stations(Fig. 4). In general, chlorophyll was highest in the upper Bayand decreased toward the lower Bay, while turbidity was lowestin the intermediate reaches of the Bay and higher in the upperand lower Bay. Yet, SDD was relatively uniform across the Bayin November–March, but increased from the upper (Sta. 40 andSta. 14) to lower bay portions (Sta. 23 and Sta. 92) in late spring(May to June). These patterns are visible both in the satellite andin situ data (e.g., within 1 standard deviation; Fig. 4).

The seasonal patterns are consistent with the time series ofchlorophyll, color, and turbidity measurements (Fig. 5). Thelargest SDD occurred in May or June, when the lowestchlorophyll concentration, color, and turbidity were measured.Then SDD gradually decreased with increasing chlorophyll andcolor, while turbidity remained relatively stable at low levels. Thissuggested that during summer season (corresponding to the rainyseason from mid-June to mid-September; Fig. 6) SDD wasprimarily controlled by phytoplankton concentration and factorsthat affect “color” measurements. Chlorophyll and color reachedmaxima between July and October depending on the locations,after which they decreased through the fall andwinter and reachedminima in the late spring or until the start of a new rainy season.However SDD showed a different variation pattern fromphytoplankton and color. SDD did not show a would-beincrease as a response of reduced phytoplankton and colorafter the summer. In contrast it continued decreasing throughthe fall and winter. So SDD was generally lower in the dryseason than in the rainy season, particularly in Middle andLower Tampa Bay (Fig. 3). This lower SDD is coincident withincreased turbidity observed in those areas and these times.SDD remained at the lower levels until in early spring (e.g.,April) when it began to increase with continuously decreasingphytoplankton, color, and slightly reduced turbidity. SDDrapidly reached maxima in May or June when turbiditysignificantly decreased from April to May and all three factorswere nearly the lowest levels over a year (Figs. 4 and 5).After that, a new seasonal cycle of SDD started. These patternsindicate that in the dry season SDD is controlled mainly byturbidity rather than by phytoplankton or color. Indeed, in theupper bay (e.g., station 40), SDD slightly increased in the dryseason when turbidity decreased, while phytoplankton andcolor varied little (Figs. 4 and 5).

The seasonality in water clarity can be explained by seasonalvariations of river runoff and wind forcing or wind-drivensediment resuspension (Fig. 6). Larger river runoff in mid-Junethrough mid-September (Fig. 6A) delivers higher nutrient anddissolved organic matter fluxes into the Bay, contributingphytoplankton growth and increased color in the Bay. In com-parison, stronger winds lead to sediment resuspension in the dryseason, especially in March–April (Fig. 6B). The sedimentresuspension events lead to higher turbidity in the dry seasonthan in the rainy season.

The SDD images show that the Bay experienced significantinterannual variation in water clarity. Fig. 7 shows the timeseries of calendar monthly mean SDD at several stations invarious bay segments. To study interannual variations, weextracted SDD anomalies (Fig. 8) by computing the differencebetween the calendar monthly means and the overall cross-yearmonthly means. Lower SDD occurred in winter–spring of 1998and summer–fall of 2001, 2003, and 2004 at all four stations.SDD was higher in the middle to lower bay (stations 23 and 92)between late 2002 and early 2003, and in late 2005. Inparticular, at station 92, higher SDD was observed after 2002. AStudent's t-test indicated that the mean SDD (2.80 m) from July2002 to December 2005 was significantly larger than that seenbetween January 1999 and June 2002 (2.42 m; pb0.05, n=42,

Fig. 7. Monthly means of SeaWiFS and in situ SDD at selected stations (Fig. 1) from September 1997 to December 2005. The number of SeaWiFS observation days ineach month is shown on the right-hand side. When this number is N1, standard deviation is shown. Note that there is only one in situ observation in each month. J, M,and S stand for January, May, and September.

256 Z. Chen et al. / Remote Sensing of Environment 109 (2007) 249–259

data before 1999 were not used due to strong El Niño effects onwater quality in 1998). Whether or not this is an indication of atrend toward improved water quality needs to be furtherinvestigated by extending the SDD series beyond 2005.

The observed interannual SDD variations are generally inagreement with the variability in river discharge patterns(Fig. 9). Lower SDD coincided with higher river flow, andvice-versa. For example, the lowest SDD observed in early1998 is associated with abnormally higher river flow in winterof 1997–1998, triggered by the 1997–1998 El Niño event(Schmidt & Luther, 2002). The high river flow in August andSeptember 2004, a result of anomalous rainfall caused byseveral hurricanes (Hu et al., 2006), led to lower SDDparticularly in the middle and lower Bay (also partly causedby higher winds). There were, however, some exceptions.Lower SDD values were found during the lower river flow of2000. Higher SDD between 2002 and early 2003 occurredduring a period of increased river flow, while wind speed during2002 was slightly below the long-term cross-year wind speedmean of that month (Fig. 10). Clearly, river flow and wind arethe two major factors that affect SDD, but other factors such astides and estuarine circulation patterns (Weisberg & Zheng,2006) may also contribute to modulate water clarity in the Bay.

Overall, SeaWiFS SDD estimates are consistent with in situSDD within one standard deviation (Figs. 4 and 7). However,SeaWiFS SDD showed more pronounced seasonal andinterannual variability that is not captured by the in situ SDD

sampled once per month. For example, abnormally high in situSDD values (N5.0 m) were observed at Sta. 92 in February andMarch 2003, but the monthly SeaWiFS means (N4 observationsper month) showed lower SDD values (b3.0 m), comparable tothose of other years. Although the higher in situ SDD valuesmight capture actual events, they clearly suffer from the aliasingeffect to represent monthly means of the SDD. Thosedifferences of N1 or 2 m between the two data sets are likelydue to the mismatch in sampling frequencies between the in situprogram and the satellite collections, because the uncertaintiesof the SeaWiFS SDD (RMS error of 0.55 m in Fig. 2) aresmaller than the differences observed. SeaWiFS typically hasN4 quality-controlled observations per month (about once perweek), and sometimes N10 observations per month betweenJanuary and May (Fig. 7). This helps reduce aliasing oftemporal variations and construct more realistic monthly meansthan simply using a single in situ measurement per month, atdifferent times of the month for different areas of the bay.

4. Discussion

4.1. Algorithm issues

The Kd(490) derived from satellite images using a semi-analytical algorithm provides an excellent estimate of in situSDD values, particularly relative to the empirical band-ratio Kd

(490). The improvement can be attributed to two unique

Fig. 8. SeaWiFS SDD anomaly between September 1997 and December 2005 at several stations in Tampa Bay (Fig. 1). The anomaly is defined as the difference ofSDD between the current month and the overall cross-year monthly mean (red lines). Some gaps at Sta. 14 and Sta. 40 are due to missing data in those months. Theboxes indicate the four negative anomalies that occurred in winter–spring of 1998, and summer–fall of 2001, 2003, and 2004, respectively. The filled arrows in Sta. 92show positive anomalies in late 2002–early 2003 and in late 2005.

257Z. Chen et al. / Remote Sensing of Environment 109 (2007) 249–259

features of the semi-analytical algorithm. First, the semi-analytical algorithm explicitly estimates absorption and back-scattering coefficients (two major factors that determine Kd)based on radiative transfer theory. Various semi-analyticalalgorithms have demonstrated that the absorption and back-scattering coefficients can be retrieved from ocean color remote

Fig. 9. Monthly means of river flow from the Hillsborough River (open circles)and the Alafia River (filled circles) from September 1997 to December 2005.

sensing within the accuracy of b15% in various water types(Lee et al., 2005a, and references therein). Consequently, asemi-analytical algorithm can be accurately applied to a varietyof regions even without concurrent in situ measurements withsatellites. This general applicability contributes to extend ocean

Fig. 10. The number of days when the daily averaged wind speed was N4.0 m s−1

in each month from September 1997 to December 2005 at the PORTstation nearSaint Petersburg (27 °45.6′N, 82 °37.6′W). The red line shows the long-termcross-year means of number of the days when the daily averaged wind speed wasN4.0 m s−1of 12 months.

258 Z. Chen et al. / Remote Sensing of Environment 109 (2007) 249–259

color remote sensing from the open ocean to coastal waters.Second, semi-analytical algorithms use more than two bands toestimate the light attenuation coefficient (Lee et al., 2005b).Therefore even though atmospheric correction over coastal andestuarine waters contains uncertainties over coastal waters (e.g.,Hu et al., 2000), some water quality parameters may still beestimated with some accuracy and applicable for coastal watermonitoring programs.

The present study focused on SeaWiFS data, but the sameapproach can be used with the Moderate Resolution ImagingSpectroradiometer (MODIS) data. The MODIS has similarspectral bands and spatial resolution as the SeaWiFS, and theatmospheric correction strategy and algorithms are comparableor identical. Since MODIS and SeaWiFS have different satelliteoverpass times, combining data from these sensors actuallyimproves spatial and temporal coverage (e.g., Hu & Muller-Karger, 2003). However, a careful comparison of remotesensing reflectance data between the two sensors would berequired beforehand.

SeaWiFS-derived SDD or Kd estimates may suffer fromsome uncertainties in certain regions and times. The under-estimates of light attenuation in Hillsborough Bay (HB) duringthe summer or rainy season are artifacts likely due to theperformance of the atmospheric correction algorithm. Duringsummer in this region, remote sensing reflectance (Rrs) in theblue and green wavelengths is very low (the maximum Rrs atvisible domain is b0.002 sr−1, unpublished data) due to the highinput of terrigenous colored dissolved organic matter (CDOM)(Chen et al., in press-a). A slight error in the atmosphericcorrection may cause large relative errors in Rrs (Harding et al.,2005; Hu et al., 2001) and therefore lead to large errors in theestimates of light attenuation. For example, overestimation ofRrs will result in underestimation of attenuation (thus overes-timation of SDD), as Rrs is inversely proportional to theattenuation coefficient. Indeed, a detailed examination of thesatellite derived Rrs reveals that satellite derived Rrs(412) inthose invalid pixels was often larger than Rrs(443), which ishighly unlikely for this type of water. The exact reason for thisdiscrepancy was not clear. However, these effects appear to belimited to HB and are negligible for other bay segments and foroverall agreement between satellite and in situ observations(Figs. 2 and 4).

4.2. Implications for Tampa Bay water quality monitoring andecosystem restoration

SeaWiFS SDD imagery clearly shows that the seasonal cycleof SDD is influenced by river runoff because river waterscontain high nutrients and dissolved organic matter, thusincrease phytoplankton and color concentrations (e.g., Chenet al., in press-a). The result is consistent with previous obser-vations that Tampa Bay water quality is closely related toprecipitation or river runoff (Lipp et al., 2001; Schmidt &Luther, 2002) and that SDD variation in Tampa Bay is largelythe result of changes in chlorophyll concentrations (Janicki &Wade, 1996). These previous findings led to an implementationof Nitrogen Management Strategy by the Tampa Bay National

Estuary Program (TBNEP), which seeks to increase waterclarity for seagrass restoration through reducing phytoplanktonconcentration by controlling riverine nitrogen loadings into thebay (Janicki et al., 2001). However, our results also clearlyshow that SDD is controlled by turbidity as well, which is inturn related to winds or wind-induced sediment resuspension,particularly in the dry season. The effects of turbidity on waterclarity have been observed in previous studies in Tampa Bay(McPherson & Miller, 1994) and in other estuaries (e.g.,Christian & Sheng, 2003). Thus, to fully understand waterclarity variations, both sediment load from runoff and wind-driven resuspension events need to be considered for an optimalmanagement plan in Tampa Bay.

Because of the ability to provide more frequent and synopticcoverage, satellite sensors such as the SeaWiFS can helpimprove estimates of the monthly “mean” patterns of SDD andmonitoring long-term trends. It is difficult to determine shortand long-term trends from the in situ data from the EPCHCmonitoring program because they are collected only once permonth and therefore are easily biased by “extreme” events.However, complementing these in situ data with satelliteobservations is ideal because together they help detect bothevents and trends, such as the water clarity improvement in thelower bay after 2002. We recommend that future monitoringplans include satellite ocean color remote sensing to helpinterpret the spatial-temporal patterns of some important waterquality indices.

5. Conclusions

A major issue in the application of satellite data to study andmonitor estuarine water quality has been the lack of accuracyand uncertainties introduced by both the atmospheric correctionand bio-optical inversion algorithms. Using a semi-analyticalalgorithm and a rigorous data quality control step, we haveshown that a critical parameter measured routinely in estuarinewater quality monitoring programs, the Secchi Disk Depth(SDD), can be effectively derived from satellite data. Wederived a time series (1997–2005) of SDD from SeaWiFSimagery for most of Tampa Bay (standard error of 0.55 m forSDD ranging from 0.9 to 8.0 m). Although artifacts remain dueto imperfect atmospheric correction and interference of bottomreflectance (e.g., depth b2.0 m), the frequent and synopticcoverage achieved with the satellite provides a better charac-terization of the spatial–temporal patterns as well as long-termtrends of SDD. Because of the synoptic, robust, and the repeatedglobal coverage, satellite sensors such as the SeaWiFS andMODIS provide important complements to traditional in situsampling methods. Thus, we strongly recommend incorporatingsatellite observations of light attenuation or water clarity tovarious coastal monitoring programs to help better manageestuarine and coastal resources.

These satellite data provide new insights that can greatlybenefit long-term monitoring efforts of estuarine waters, most orall of which are based solely on field surveys. The distinct spatialand temporal viability revealed in SeaWiFS SDD imageryhighlights the importance of river runoff (phytoplankton and

259Z. Chen et al. / Remote Sensing of Environment 109 (2007) 249–259

color) in the rainy season and of wind or wind-induced sedimentresuspension (turbidity) in the dry season in controlling the lightattenuation (SDD) across Tampa Bay, particularly in the middleand lower portions of the Bay.

The results support the concept of ecosystem-based manage-ment, in the sense that optimal estuarine water quality manage-ment plans should focus on control of several critical parameters,i.e. not simply on nutrient input control but also on understandingand mitigating the inputs of colored dissolved organic matter andtotal sediment load from various sources such as stormwater, riverrunoff, dredging, and transportation activities.

Acknowledgements

This work was funded and logistically supported by theTampa Bay Integrated Science Project of the U.S. GeologicalSurvey (USGS), Coastal and Marine Geology Program.Additional financial support was provided by the USF-USGSCooperative Graduate Assistantship Program, and by theNational Aeronautics and Space Administration (NASANAG5-10738). GeoEye (formerly Orbimage) and the SeaWiFSProject at NASA's Goddard Space Flight Center are acknowl-edged for providing the SeaWiFS level-1A data. We particularlythank the Environmental Protection Commission of Hillsbor-ough County (EPCHC) for their willingness to share TampaBay water quality monitoring data. The comments and sug-gestions of two anonymous reviewers significantly improvedthis manuscript. IMaRS contribution # 110.

References

Austin, R.W., & Petzold, T. J. (1981). The determination of the diffuse attenuationcoefficients of sea water using the coastal zone color scanner. In J. F. R. Gower(Ed.), Oceanography from space (pp. 239−256). New York: Plenum.

Bailey, S. W., & Werdell, P. J. (2006). A multi-sensor approach for the on-orbitvalidation of ocean color satellite data products. Remote Sensing ofEnvironment, 102(1–2), 12−23.

Boler, R.N.,Molloy,R.C.,&Lesnett, E.M. (1991). Surfacewater qualitymonitoringby the Environmental Protection Commission of Hillsborough Country. In S. F.Treat & P.A. Clark (Eds.), Proceedings from BASIS II Tampa, FL, 528 pp.

Chen, Z., Hu, C., Conmy, R. N., Swarzenski, P., Mullerr-Karger, F. (in press-a).Colored Dissolved Organic Matter in Tampa Bay, Florida.Marine Chemistry.doi:10.1016/j.marchem.2006.12.007

Chen, Z., Hu, C., Muller-Karger, F., & Luther, M. (in press-b). Physical forcingof short bio-optical variability in Tampa Bay: Observations from a coastaltower. Limnology and Oceanography.

Christian, D., & Sheng, Y. P. (2003). Relative influence of various water qualityparameters on light attenuation in Indian River Lagoon. Estuarine, Coastaland Shelf Science, 57(5–6), 961−971.

Florida Department of Community Affaires. (1996). State of the coastal report:preparing a sustainable future. Prepared by the Florida Center for PublicManagement. Tallahassee, FL: Florida State University.

Gesch, D., & Wilson, R. (2001). Development of a seamless multisourcetopographic/bathymetric elevation model of Tampa Bay.Marine TechnologySociety Journal, 35(4), 58−64.

Giesen, W. B. J. T., Van Katwijk, M. M., & Den Hartog, C. (1990). Eelgrasscondition and turbidity in the Dutch Wadden Sea. Aquatic Botany, 37, 71−85.

Harding, L. W., Magnusona, A., & Mallonee, M. E. (2005). SeaWiFS retrievalsof chlorophyll in Chesapeake Bay and the mid-Atlantic bight. Estuarine,Coastal and Shelf Science, 62, 75−94.

Harwell, M. A., Ault, J. S., & Gentile, J. H. (1995). Comparison of the EcologicalRisks to the Tampa Bay Ecosystem from Spills of Fuel #6 and Orimulsion.

Comparative Ecological Risk Assessment. Center for Marine and Environ-mental Analyses, Vol. 1. Miami, Florida: University of Miami.

Hu, C., & Muller-Karger, F. E. (2003). MODIS monitors Florida's oceandispersal of the Piney Point phosphate treated wastewater. The EarthObserver (NASA), 15(6), 21−23.

Hu, C., Carder, K. L., & Muller-Karger, F. E. (2000). Atmospheric correction ofSeaWiFS imagery over turbid coastal waters: A practical method. RemoteSensing of Environment, 74, 195−206.

Hu, C., Carder, K. L., & Muller-Karger, F. E. (2001). How precise are SeaWiFSocean color estimates? Implication of digitization-noise errors. RemoteSensing of Environment, 76, 239−249.

Hu, C., Muller-Karger, F. E., & Swarzenski, P. W. (2006). Hurricanes,submarine groundwater discharge, and Florida's red tides. GeophysicalResearch Letters, 33, L11601. doi:10.1029/2005GL025449

Janicki, A., &Wade, D. (1996). Estimating critical nitrogen loads for the TampaBay Estuary: An empirically based approach to setting management targets.Report to the Tampa Bay national estuary program # 94-444-14.

Janicki, A., Pribble, R., Janicki, S., & Winowitch, M. (2001). An analysis oflong-term trends in Tampa Bay Water Quality. Tampa Bay estuary technicalreport #04-01, 48 pp.

Jean-Franc, O. B., & Giuseppe, Z. (2004). Bio-optical relationships for thenorthern Adriatic Sea. International Journal of Remote Sensing, 25(7–8),1527−1532.

Johansson, J. O. R. (2000). Historical overview of Tampa Bay water quality andseagrass: Issues and trends. In H. Greening (Ed.), Seagrass Management:It's Not Just Nutrients! Tampa Bay Estuary Program Symposium Pro-ceedings, 22–24 August, St. Petersburg, FL (pp. 1−10).

Kratzer, S., Hakansson, B., & Sahlin, C. (2003). Assessing Secchi and photiczone depth in the Baltic Sea from satellite data. Journal of the Human-Environment System, 32(8), 577−585.

Lee, Z. P., Du, K. P., & Arnone, R. (2005a). A model for the diffuse attenuationcoefficient of downwelling irradiance. Journal of Geophysical Research,110, C02016. doi:10.1029/2004JC002275.1

Lee, Z. P., Darecki, M., Carder, K. L, Davis, C. O., Stramski, D., & Rhea, W.J., (2005b). Diffuse attenuation coefficient of downwelling irradiance: Anevaluation of remote sensing methods. Journal of Geophysical Research, 110,C02017. doi:10.1029/2004JC002573

Lewis, R. R., Clark, P. A., Fehring, W. K., Greening, H. S., Johansson, R. O., &Paul, R. T. (1998). The rehabilitation of the Tampa Bay Estuary, Florida,USA, as an example of successful integrated coastal management. MarinePollution Bulletin, 37(8–12), 468−473.

Lipp, E. K., Schmidt, N., Luther, M., & Rose, J. B. (2001). Determining theeffects of El Niño-Southern Oscillation events on coastal water quality.Estuaries, 24, 491−497.

McPherson, B. F., & Miller, R. L. (1994). Cause of light attenuation in TampaBay and Charlotte Harbor, southwestern Florida. Water Research Bulletin,30, 43−53.

Mueller, J. L. (2000). SeaWiFS algorithm for the diffuse attenuation coefficient,K(490), using water-leaving radiances at 490 and 555 nm. In J.E. O VReilly,and co-authors: SeaWiFS Postlaunch Calibration and Validation Analyses,Part 3. S. B. Hooker & E. R. Firestone, Eds., NASA/TM-2000-206892,Vol. 11, NASA Goddard Space Flight Center, Greenbelt, Maryland, 24–27.

Prasad, K. S., Bernstein, R. L., Kahru, M., & Mitchell, B. G. (1998). Ocean coloralgorithms for estimating water clarity (Secchi depth) from SeaWiFS. Journalof Advanced Marine Science and Technology Society, 4(2), 301−306.

Schmidt, N., & Luther, N. E. (2002). ENSO impacts on salinity in Tampa Bay,Florida. Estuaries, 25(5), 976−984.

Stumpf, R. P., & Pennock, J. R. (1991). Remote estimation of the diffuseattenuation coefficient in a moderately turbid estuary. Remote Sensing ofEnvironment, 38, 183−191.

Tomasko,D.A., Corbett, C.A.,Greening, H. S., &Raulerson,G. E. (2005). Spatialand temporal variation in seagrass coverage in Southwest Florida: assessing therelative effects of anthropogenic nutrient load reductions and rainfall in fourcontiguous estuaries. Marine Pollution Bulletin, 50, 797−805.

Weisberg, R. H., & Zheng, Y. (2006). Circulation of Tampa Bay driven bybuoyancy, tides, and winds, as simulated using a finite volume coastal oceanmodel. Journal of Geophysical Research, 111, C01005. doi:10.1029/2005JC003067