ASTER thermal infrared (TIR) satellite data was searched in the US Geological Survey Global Visualization Viewer (GLOVIS) (http://glovis.usgs.gov/). Exploration of the image archive focused the imagery search on nighttime ASTER observations, since cloud-free images were more abundant and often higher thermal contrast between land and water features than daylight imagery. Although Landsat-7 ETM+ TIR imagery would be applicable, prior searches of available archives only uncovered limited and rather extensively cloud-covered data during the focal study period. For ASTER data, cloud-free flags and Quicklook images were examined for over 200 images, with tidal calculations used to select the maximum flood tide penetration dates. Quicklook images of TIR bands 13 and 14 also provided qualitative, initial judgments of thermal contrast between land, water, and among estuarine and ocean thermal water bodies. Images that highlighted strongly contrasting thermal differences between lagoonal and coastal marine waters were sought and visually compared for selection for subsequent analyses (tidal footprint delineation and spatial analysis.) Since continuous tidal station and water level monitoring observations were not available for Chincoteague Inlet, predicted tidal levels from WXTide32 were used, a free tide and current prediction software (http://www.wxtide32.com/) and tidal predictions for National Oceanographic and Atmospheric Administration (NOAA) stations at Ocean City Inlet (station 8570283), 40 km to the north and Wachapreague (station 8631044), approximately 30km to the south (http://tidesandcurrents.noaa.gov/). Figures 2 and 3 present the final set of selected images and their approximate tidal state at time (adjusted to synchronize image and tidal prediction to local standard time). Quicklooks provided for the analysis of relative thermal contrast using band 13 in scene selection. While ultimately these eight scenes from 2004 and 2005 were selected, over 80 other scenes were similar visually assessed but were not chosen owing to incompatible tidal stage at time of acquisition or extreme local mixing (e.g., wind and/or inter-lagoonal water mass exchanges between the adjoining bays). Table 1 summarizes the key characteristics of the images relative to tidal level and timing.

The availability of nighttime ASTER images offered the advantage of shorter revisit times as compared with other satellite sensors. ASTER’s high spatial resolution in the TIR bands (90 m) provides for discerning finer patterns in small coastal lagoons such as our study area. The ASTER sensor was launched in December 1999 onboard the NASA Terra satellite. Its unique data are collected in five wavebands of the thermal infrared (TIR) region of the spectrum, 8.30, 8.65, 9.10, and 11.3 m wavelengths with a nominal ground resolution of 90 m, allowing image characterization of water pixels smaller than 250 m in diameter [21]. Since ASTER’s satellite platform operates on a near-polar sun-synchronous orbit that allows day or nighttime thermal IR observing, the potential for more frequent image acquisition, up to every 8 days or less at higher latitudes, is possible.

Dynamic ocean, coastal, and inland water features are well suited to ASTER’s thermal remote sensing capabilities. Validation studies of the ASTER and MODIS instruments by NASA Jet Propulsion Laboratory affirmed the performance of airborne MODIS/ASTER (MASTER) thermal emissivity and skin surface temperature retrieval in comparison to buoy and float-tethered sensors [22]. ASTER has also been shown capable of retrieving temperature for land surfaces [23], lake surface temperatures [10], and detecting wind-induced coastal upwelling and surface circulation patterns. Handcock et al. [15] noted limitations of spatial resolution of ASTER TIR images (90 m × 90 m) might still restrict water surface temperature applications, or suffer overprediction of temperature in comparison to field measurements for stratified waters or anomalous surface skin effects. Atmospheric corrections are also important to the conversion from ASTER emissivity radiance to temperature and usually involve empirical (most often regression-based) corrections to ground measurements (e.g., “split window” or two-channel techniques or physics-based transformation with a radiative transfer model.) For land-surface temperature derivation, the accuracy of these two methods was shown to be comparable [21]. The ASTER science team has developed these derivatives into “on-demand” products, including surface temperature (AST08) or surface emissivity (AST05) digital image data products. Using these products, the thermal band 13 brightness temperature image received by the sensor is suited to the identification of thermally distinct water masses [21]. Following from these advancing capabilities, ASTER TIR data are applied to improve understanding of lagoonal and estuarine water dynamics.

Four images were selected for detailed analysis whose acquisition times fell within approximately 1 h of predicted peak flood tide at Chincoteague Inlet. A series of display enhancements using histogram contrast enhancement and color ramps were used to visually highlight and examine relative circulation patterns prior to statistical and predictive analyses. Following techniques employed by the ASTER science team, surface temperature grids were produced for these images following calculation of emissivity and land mask removal by a GIS overlay operation using ASTER bands 13 and 14 and split-window method demonstrated by Trunk and Bernard [10]. This mask sought to eliminate terrestrial, supratidal areas that would be above the mean high water line at flood tide, but this mask would necessarily include intertidal areas, including marshes and mudflats and lower portions of backbarrier estuarine beaches. Resulting water surface temperature images were qualitatively compared against in situ water temperatures in order to approximate peak flood tide (within 1 h) and provide for comparison of the flood tide penetration and pattern through Chincoteague Inlet and Bay.

Images were also masked and subsetted to water features. Using iterative seed pixel extraction techniques, the repletion footprints of flood-tide images were delineated (determining the tidal stage from astronomic predictions coincident with image acquisitions.) After evaluating the nearshore marine surface temperature, a representative pixel was selected in the inlet throat and used to expand a homogeneous polygonal feature boundary using the Erdas IMAGINE seed pixel tool. Iterative analysis and selection of a Euclidean distance temperature threshold (T) and visual inspection determined the flood tide extent distal boundary. Thresholds were used in consideration of observations of seasonal coastal water temperatures as well as lagoonal temperatures (Table 2).

To assess the spatial pattern and variation of tidal footprints and surface temperature, Principal components analysis (PCA) and multiple write function memory overlays were used to visually and statistically evaluate dispersion of the flood tide, mixing, and repletion footprints. Each calculated repletion footprint was converted to a vector polygon Area of Interest and then translated to polygon shapefile for spatial analysis. Repletion footprint vector files were used to mask each image and create a multitemporal stack of repletion footprint pixels. These images were analyzed by PCA for determination of pattern and persistence over time and extent of the repletion zone. Derived PCA images offered the potential to reveal zonal spatial concentration variation of tidal flushing, reducing the volume of the time series for flood tide images to highlight areas of flushing versus more variable tidal penetration and residual waters. Eigenvalue and eigenvector analysis would reveal correlation of images and potential distinct residual vs. tidal waters vs. zones of mixing. Descriptive statistics and comparative analysis by correlation to the State of Maryland’s “in situ Eyes on the Bay” Coastal Bays Monitoring Program (http://www.eyesonthebay.net/) were also used to distinguish representative versus anomalous values in the imagery. In order to distinguish ocean, estuarine and gradational mixing of waters, boxplots comparing tidal water temperatures and historical bay temperatures and paired T-tests for difference of means between tidal flushing waters and residual, non-flushed waters were calculated.

Initial results illustrated in Figure 4 revealed the seasonal trends and pattern of water surface temperatures across a generally north-south gradient from Chincoteague Inlet, through the main channels and around Chincoteague Island into the southern bay. Although covering multiple years, the series shows the predominant seasonal trend of increasing water temperature into summertime and decreasing bay and marine temperature contrast. Two exceptions to this trend are revealed in Figure 4(b), which shows marked cooler temperatures in intertidal marsh and shallows and Figure 4(c), which indicates relatively strong thermal contrast between the flood tide waters and the interior lagoon temperature. Since the image in Figure 4(b) is late May 2004, this season is representative of the onset of more rapid estuarine water heating versus the slower heating of flood tide coastal marine waters. A reversal of this pattern may be expected during the autumn, when the lagoon cools faster than the adjoining ocean arising from the lower volume and loss of heat in the lagoon.

Utilizing the seed pixel homogeneity feature extraction method, each derived flood tide boundary was mapped for visual analysis and comparison (Figure 5). Table 2 summarizes the seed pixel temperature, threshold, and each scene surface water temperature range. Seed pixels were chosen in the throat of Chincoteague Inlet with exclusion of separate internal polygons. Values were iteratively adjusted by increasing the Euclidean threshold until the zone expanded to include the visually discernible extent of the flood tide in the scene. These thresholds are later compared with the distribution of scene temperature ranges, flood repletion zone temperatures, and in situ observations (Table 2 and Figures 6–8.) The flood tide penetrates an average of approximately 10 km from the inlet throat to the distal interior boundary in the lagoon. Subject to estuarine circulation and winds, the bays and intertidal areas closest to the inlet are most well flushed. An exception shown in Figure 5(a) is to the west of the inlet, where only slight penetration is evident as compared with the cove to the east. For all images, the flood tide penetrates 4–5 km or more past the Town of Chincoteague, highlighted as a point symbol in Figure 5. A potential limitation of this discrete delineation is also found in Figure 5(d), which only shows slight tidal penetration into the coves adjoining the inlet. It is likely in these very shallow, clear waters that summer time heating of adjoining surfaces could rapidly heat the incoming tidal waters, even for nighttime images such as these. Given the modest thermal contrast between ocean and lagoonal waters in mid-summer, the tidal water delineation such as Figure 5(d) could be underestimated by this technique. Similarly, low thermal differentiation in the February scene (Figure 5(a)) also lends to less contrast and more potential error. In addition, these two scenes (mid-winter and mid-summer) exhibited the lowest range of temperatures (Table 2.)

Prior to evaluating further spatial pattern and variation, the measured water temperatures of flood tide and “residual” lagoonal waters (water mass believed to be resident and not-recirculated in the tidal flushing) were evaluated using exploratory statistics and visual analysis. Figure 6 shows the resulting distributions of residual and flushing tidal waters for each scene, underscoring the prior interpretation of low thermal contrast between coastal and lagoon waters during mid-late winter (February and March scenes, Figure 6(a,b)) and gradually increasing contrast as lagoonal waters more rapidly warmed relative to the ocean by late May (Figure 6(c)). However, one potential discrepancy was revealed in the plot for July 2004 (Figure 6(d)), which exhibited a greater degree of contrast between the tidal and residual water in the lagoon. Since these plots portray the distribution for all image pixels, it can be inferred that this likely masks the fact that some flushing tidal waters could rapidly warm in extreme shallows and amidst warm air advection in this breezy coastal location. Thus, it is likely that there exists a combination of mixing and shallow tidal water heating that confuses the higher temperature residual water mass with mixed and rapidly-warmed tidal influx. Comparison of results to circulation or detailed in situ measurements would be necessary to further affirm this interpretation.

Results of this investigation reveal strong potential for satellite data such as ASTER TIR response to delineate tidal flooding and repletion in coastal lagoons and estuaries. Synoptic assessments of tidal flushing patterns can improve water level zoning and hydrographic surveying for the siting of tide stations and correction algorithms for bathymetry. However, there is room for improvement in the delineation of discrete tidal flushing zones, such as better integration of astronomical and harmonic components in hydrodynamic models and networks of water level and water quality monitoring. Overall, this investigation provides a foundation of first-order estimates using thermal satellite earth observation to predict hydraulic turn-over in lagoons and estuaries for managers seeking more information on watershed, estuary, and eutrophication vulnerability assessment. Subsequent investigations might explore composite repletion footprints to estimate tidal volume and hydraulic turn-over time, including other tidal stages, extreme events, or tidal progression with stereo observations. A longer time series of thermal satellite data might also be able to develop a coastal or estuarine sea surface temperature climatology and fine-scale data on surface energy fluxes, which would benefit ecological monitoring and management. The relatively high-resolution of ASTER TIR data shown here has evident potential for extensive and numerous coastal lagoons and bays with inlets and tidal creek channels. A longer time series of surface temperatures over such waters could also provide information on energy fluxes and micrometeorological processes that might improve forecasting (e.g., sea breezes and wind power) as compared with coarser satellite data. Observed tidal flushing from satellite also holds value in for assessing the range of uncertainty and variability of tidal repletion volumes under different meteorological forcing, particularly in small and/or distant coastal bays and lagoons where model data are limited and in situ observations expensive or infeasible. Management applications of these data include coastal ecological monitoring and restoration, such as seagrass restoration, benthic habitat assessment such as oysters, clam and aquaculture uses, and water quality protection. The availability of repetitive satellite thermal imagery and the synoptic approach presented thus provides an efficient means for resource monitoring as well as scientific investigation of estuarine processes.

The availability of TIR data on Landsat-7 ETM+ sensor with 60 m spatial resolution and sensitivity to 10.4–12.5 mm has not been overlooked in this study and remains highly applicable. Even with the onset of the Scan-Line Corrector (SLC) anomaly, Landsat-7 continued to collect data, which has since been provided free as well as reliably with interpolation and SLC-off post-processing [23,24]. However, preliminary searches on the Landsat-7 archive discovered frequent obscuration of data collected during daytime observing and a paucity of imagery available during appropriate near flood-tide conditions in the relatively short timespan focused upon by this study. Other studies have applied Landsat-7 TIR data successfully to regional oceanographic sea-surface temperature over longer timespans [12,13] or in relatively more arid environments than the coastal investigation herein [14,15]. Nonetheless, Landsat TIR data have a long history and near-global data availability, providing for possible historical change analysis or systematic coastal lagoon comparative research. In addition, the Landsat Data Continuity Mission (LDCM), scheduled to launch in 2013, will contain a Thermal Infrared Sensor (TIRS) package sensitive in two longwave bands (10.3–11.3 mm and 11.5–12.5 mm) with 100 m resolution, 12-bit data depth, and improved signal-to-noise ratio [25].