Flow over heated islands

Cumulus cloud plumes extending up to 200 km off the island of Nauru are often seen in GMS images (see, e.g., Fig. 4). Nordeen et al. (2001) found that while the island and ocean temperatures are the same at midnight, the island surface air temperature is roughly 4.5°C higher by 1030 local standard time (LST). Because the maximum elevation of the island is only 60 m, the heating of the island surface relative to the surrounding ocean is assumed to be the primary formation mechanism for the cloud plumes. The possible role of orographic lifting is discussed further in section 4b.

Flow over heated islands—both urban heat islands and oceanic islands—has been studied extensively. Observational and theoretical studies of similar cloud plumes forming downwind of Nantucket Island, Massachusetts, in the 1950s (Malkus and Stern 1953; Stern and Malkus 1953) indicated that the key factors in the formation of the island-induced cloud plumes were the heating of the island relative to the ocean, a conditionally unstable atmosphere, and the development of a deep mixed layer. Later observational and theoretical studies, summarized by Garstang et al. (1975), found that uniform flow over heat islands results in a region of descending air over the heat source and a region of ascending air downwind of the heat source.

Lin and Smith (1986) analytically solved the linearized time-dependent problem of a heat source in a uniform flow. Their theoretical results indicated that a region of upward displacement was generated at the origin of the heat source and was then advected downstream by the mean wind. Baik (1992) solved a similar problem for a constant shear flow, using both a linear analytical model and a nonlinear numerical model. His results indicated that the velocity perturbations in a shear flow had a similar pattern to that in a uniform flow, but were larger in magnitude as a result of the wind shear being a source of perturbation energy. He also found that in the nonlinear solution, as the amplitude of the island heating increased, the updraft cell downwind of the heat source strengthened and moved further downwind.

More recently, Baik and Chun (2001) studied moist convection forced by an urban heat island using a two-dimensional nonhydrostatic compressible model with cloud microphysics. This study explicitly showed that the location of the initial cloud formation in the model coincided with the center of the updraft cell that was induced downwind of the heat source. Additionally, they found that all of the updraft cells generated by the heat source continued to propagate downwind for all wind speeds, although they weaken as they move downstream.

Matthews (2003) extensively analyzed the surface and aircraft data taken during the Nauru99 experiment. He described the daytime flow over Nauru in terms of a thermal internal boundary layer that was generated by the sea–island contrast in surface temperature and roughness length, and a thermal heat island circulation. The island heating sets up roll vortices that the subsequent convective lifting maintains to keep the cloud plume moving downstream. Savijärvi and Matthews (2004) modeled the flow over Nauru with a two-dimensional nonlinear model, using observed wind speed and diurnal temperature variations. In their study, the model’s afternoon wind field displayed rising motion downwind of the island and descending motion upwind of the island, as in previous studies. With the large-scale wind flow removed, the model developed a weak sea-breeze pattern. The transition between the sea-breeze pattern and the heat island perturbation occurred at wind speeds of about 5 m s⁻¹ in the model.

Effect of El Niño on plume occurrence

In the analysis of the GMS-5 visible images from June 1999 through June 2000 by Nordeen et al. (2001), the Nauru cloud plume was seen frequently, almost always extending from the island toward the west (plume headings ranged from 220° to 320°, with an average directional heading from north of 265°). During this time period, the region was in a phase of suppressed convection, and the wind direction was primarily from the east in the form of the prevailing easterly trade winds. This combination resulted in highly favorable conditions for the formation of the Nauru cloud plume. However, the tropical Pacific experiences large interannual variability associated with ENSO. In mid-2001, El Niño conditions began developing in the tropical Pacific, which affected the Nauru cloud plume in several ways. During El Niño conditions, convection in the tropical western Pacific shifts eastward into the central Pacific (Philander 1985), which increases the total cloudiness at Nauru and reduces the impact of the cloud plume on the cloud and radiation measurements at the surface. Also, one of the fundamental characteristics of the observed El Niño cycle is a breakdown of the easterly trade winds as air pressure rises over the western Pacific and falls over the eastern Pacific (Battisti and Sarachik 1995). The weakening of the trade winds at Nauru can clearly be seen in Fig. 5, which shows the frequency of the surface wind direction measured at the ARCS site from January 1999 through June 2003. The dominance of the easterly trade wind regime is evident in the wind direction histograms for 1999 through 2001. With the relaxation of the prevailing trade winds resulting from El Niño in 2002 and early 2003, local and regional circulations become more important and there is no dominant surface wind direction.

During some periods when the prevailing easterlies were weak, the measurements at the NIES site showed evidence of a sea-breeze circulation [as predicted by the modeling of Savijärvi and Matthews (2004)], with the winds shifting from westerly (from land to ocean) during the nighttime to northeasterly (from the ocean to the land) during the daytime, and the wind speeds increasing during the daytime. Other sources of variability in the surface winds during El Niño conditions are the westerly wind bursts associated with mesoscale deep convective systems (Kiladis et al. 1994).

Because of the relaxation of the trade winds in the tropical Pacific in 2002–03, the frequency of times when the wind blew consistently from the east was greatly reduced during the NIES experiment relative to the time period examined by Nordeen et al. (2001). Thus, the frequency of the cloud plume developing downwind of the ARCS site was greatly reduced. However, a cloud plume may instead develop downwind of the NIES site if the wind blows consistently from the west for a period of several hours, such that the NIES site becomes the “downwind” site. In their analysis Nordeen et al. (2001) cite an “anomalous” case in which the cloud plume heading was 130°, which is consistent with the possibility of an island effect at the NIES site.

Following Nordeen et al. (2001), we manually analyze the hourly visible GMS 1.25-km images between 0730 and 1630 LST over the period of 1 September 2002 through 11 May 2003. The images are classified into one of the following three categories: the cloud plume was detected, the island was obscured (generally by optically thick clouds that covered much of the region), and no plume observed (but other low clouds might exist in the region and over the island). In the top panel of Fig. 6, the hourly mean occurrence of each category is shown. Although the analysis is subjective and, therefore, care must be taken with direct comparisons with the Nordeen et al. (2001) results, there is clearly a much lower frequency of occurrence of the cloud plume during the NIES experiment. Nordeen et al. (2001) found that the mean plume frequency was 50% for all observations and 63% during the afternoon, while for the NIES period the plume frequency was 17% for all observations and 23% in the afternoon. The directional heading (from north) of the cloud plume in each image was also determined. The frequency distribution of the plume headings is shown in the bottom panel of Fig. 6. A much wider range of headings is seen than in the Nordeen et al. study, with plumes with headings less than 180° occurring 38% of the time. The existence of plumes with headings less than 180° indicates that the measurements at the NIES site may also be affected by island-induced clouds during certain conditions.

Characteristics of the island-induced clouds

The clouds induced by the island of Nauru are generally shallow, nonprecipitating cumulus. Matthews (2003) reports on several aircraft flights along the cloud plume made during the Nauru99 experiment. During these flights, the cloud plume was generally composed of individual cloud cells, the largest having a horizontal width of ∼600 m. The clouds tended to be distributed in clusters, with individual clouds within the clusters separated by <100 m and the clusters separated by up to 2 km. Liquid water contents in the clouds ranged from 0.1 g m⁻³ near the cloud base to 0.6 g m⁻³ near the cloud top.

Analysis of radar and microwave radiometer observations at the ARCS site over the period of 1 June 1999 through 31 May 2000 indicated that the shallow cumulus clouds observed over the island generally have bases around 850 m and are ∼500 m deep (McFarlane and Evans 2003). Over 90% of the observed, nonprecipitating water clouds had optical depths less than 16 and liquid water paths less than 100 g m⁻².

Wind direction

There is a very strong relationship between the plume heading and the surface wind direction, as seen in Fig. 7. For each satellite image in which a plume was detected, the vector-averaged wind direction was calculated at each site for the 1-h period surrounding the image. There is a very good correlation (correlation coefficient greater than 0.9) between the plume headings and the average surface wind directions at both sites.

Thus, the primary indicator of periods when the surface measurements might be affected by the island plume is wind direction. If the wind consistently blows across the width of the island before reaching the surface measurement site, the air mass might be warmed or lifted enough for convection to be initiated. We classify each 1-min data point as being possibly island influenced or non–island influenced at each site, based on a 60-min average of the wind direction θ, centered on that point. (Analysis was also done using 30-min averages, and the results were similar.) In the analysis that follows, only data points in which good surface radiation, meteorological, and ceilometer data existed at both sites are used. The period is classified as being possibly island influenced at the ARCS site if 50° < θ < 180° at both sites, and being possibly island influenced at the NIES site if θ < 10° or 250° < θ < 360° at both sites. If neither of the above conditions is true, the period is non–island influenced. Out of 1567 data periods, 22% are classified as island influenced at ARCS, 29% as island influenced at NIES, and 49% as non–island influenced.

Figure 8 shows the frequency distribution of the 1-min minimum cloud-base heights measured by the ceilometer at the two sites. Over the entire dataset, the shape of the ARCS and NIES distributions are similar, with the peak in cloud base at 675 m for the ARCS site and at 645 m for the NIES site. Because the cloud plume is a low-cloud phenomenon, we are most interested in the low-cloud frequency of occurrence. For this study, we define low clouds as clouds with bases less than 1 km. Over the entire dataset, the NIES site has a higher frequency of low clouds than does the ARCS site, with a total low-cloud frequency of occurrence of 0.26 at the NIES site, as compared with 0.20 at the ARCS site. This modest difference might be accounted for by sensitivity differences between the two instruments as much as real cross-island low-cloud occurrence differences. However, for the periods classified as being island influenced, the windward site has a significantly higher frequency of low-cloud occurrence than the leeward site.

Clearly the surface wind direction can be used to determine if it is possible that an island effect is occurring at a given site. This is especially important during El Niño conditions, when the local conditions are much less uniform than they are during the suppressed phase of ENSO. However, wind direction alone is not sufficient to identify times when the plume is actually affecting the surface cloud and radiation measurements on the island itself, because other conditions may exist, which inhibit the formation of the plume, cause it to form offshore instead of directly over the island, or mitigate its effects on the surface observations.

Surface air temperature

The satellite analysis presented in this paper, as well as that of Nordeen et al. (2001), both show that cloud plume frequency is highest in the afternoon, with few plumes seen before 0930 LST. This is due primarily to the fact that the island surface needs time to heat, relative to the ocean, for the island to be able to induce convection. Figure 9 shows the average diurnal cycle of surface air temperature at the two sites. On average, the surface air temperature is 3.1°C higher at 1200 than at 0600 LST at the NIES site (which is surrounded by mostly bare ground) and is 2.4°C higher at 1200 than at 0600 LST at the ARCS site (which is surrounded by mostly vegetated ground). Individual days show larger diurnal variation, with maximum surface air temperature reaching up to 305 K at each site. During the day the surface air temperature measured at the NIES site is generally higher than that measured at the ARCS site, while they are similar at night. This temperature difference may be a result of the differing surface characteristics of the two sites, or of the fact that the ARCS temperature system is ventilated while the NIES system is not.

The island surface begins heating at sunrise and continues until the maximum temperature is reached around 1300 LST. However, there are times when the island surface might not be warm enough, even in the afternoon, to induce convection because of factors such as recent precipitation or shadowing by more cloudy conditions. To examine the relationship between the cloud plume existence and surface temperature, we looked at the maximum surface temperature at each site during the hour surrounding each satellite image. No plumes were detected in the satellite imagery during periods where the maximum temperature at either site was less than 302 K.

In section 3a, we stated that island heating was assumed to be the primary mechanism for plume formation because of the low elevation of the island. To investigate this assumption, we examined the frequency of ceilometer-detected low clouds at the two sites during daytime (0800–1600 LST) and nighttime (2000–0400 LST) periods. As seen in Fig. 10, during the nighttime periods the low-cloud frequency at the two sites is fairly similar, ranging from 0.15 to 0.21. These frequencies are similar to the values seen during the daytime at the non-island-influenced sites (0.15 and 0.19 at ARCS and NIES, respectively), and are much less than the frequencies seen during the day at the island-influenced sites (0.29 and 0.32 at ARCS and NIES, respectively). These data indicate that the island does not significantly enhance low-cloud formation at night, so orographic lifting is not a primary factor in the cloud plume formation.

Correlation and variability in radiation measurements

Not all of the low-level cloud observed at Nauru is induced by the island. Low-level cloud may form over the open ocean and then be advected over the island. Additionally, mesoscale systems may dominate the region, either inhibiting local convection, or simply affecting the cloud and radiation measurements more than the shallow cumulus produced by the island. Because the separation between the ARCS and NIES sites is so small (<5 km), major differences in the magnitude and variability of the cloud amount and surface feature are assumed to be a result of the island-induced clouds. Barnett et al. (1998) found that points within a 30-km radius of a central station in the Oklahoma Mesonet had average correlations >0.9 for normalized global shortwave (GSW) radiation under both sunny and cloudy conditions. Although these results are only representative of midlatitude climate regimes, with only 5-km separation between the ARCS and NIES sites, the surface radiation that is observed at the two sites should be well correlated in the absence of local island effects.

For each satellite image, the correlation in the GSW at the two sites was calculated for the 1-h period centered on the time of the satellite image. Figure 11 shows the frequency distribution of correlation in normalized GSW between the two sites as a function of the plume classification. For periods where plumes were seen in the satellite images, there is a peak at low correlation and very few points with a correlation >0.8. For no-plume conditions the frequency distribution is flat, while for overcast conditions there is a peak in correlation >0.8. Therefore, we add a correlation threshold as a criterion for plume existence, assuming that periods with a correlation >0.8 are not affected by the cloud plume and are, instead, dominated by the larger-scale circulation.

The variability in the GSW and IRT measurements is strongly related to the low-cloud frequency differences because of the size of the island-induced clouds. The clouds that are induced by the island form as small individual cells separated by clear air, and then advect downwind (Matthews 2003). Because of the clear air separation between the induced clouds, they will produce considerable variability in the downwelling GSW and IRT measurements as they pass over the measurement sites. Periods with larger variability in GSW and IRT temperature at the downwind site are also likely to have a higher frequency of low clouds at the downwind site, as shown in Fig. 12. The magnitudes of the IRT measurements and the downwelling GSW also show similar correlations with cloud frequency differences. Because the magnitudes are much more dependent than the variability on the precise calibration of the instruments, and because we want to examine the effect of the cloud plume on the magnitude of the radiation at the affected sites, we do not use the magnitude of the GSW and IRT measurements to identify plume-affected periods in the later analysis.

Frequency of plume-affected periods

Based on the above analysis we determine for which periods the surface measurements are most likely affected by the island cloud plume. For a period to be identified as being plume affected at a given site, we require the average wind direction to indicate that it is possible that the site could be island influenced (as defined in section 4a), the average surface temperature to be greater than 302 K, the correlation in GSW at the two sites to be less than 0.8, and the variability in the downwelling GSW measurements to be at least 10% larger at the downwind site than at the upwind site. Figure 13 compares the frequency of our plume-affected periods with the frequency of plumes seen in the satellite images. The frequencies are similar in the afternoon, although the frequency of affected periods is higher than the frequency of observed satellite plumes in the morning. The difference in the frequency of identified plumes in the morning may be the result of an incomplete specification of criteria for identifying plume-affected periods in the surface data or to the inability to detect early stages of plume formation in the satellite images. During the initial formation of the cloud plumes the clouds are too small to be detected in the 1.25-km imagery, although they can still affect the surface measurements. Nordeen et al. (2001) report that the average plume length when a plume is first observed in the satellite imagery is 46.6 km. Based on the analysis of the radiosondes launched during the NIES period, the mean wind speed at a height of 850 m is 6.4 m s⁻¹. If the plume advects at this speed in the downwind direction, it would take 2 h for the plume to reach 46 km in length. The peaks in the satellite cloud frequency and in the plume frequency determined from the surface observations occur around 1130 LST. However, the diurnal variation in surface air temperature reaches its peak around 1300 LST (Fig. 9). Thus, the island begins to initiate cloud plumes well before the surface has reached its maximum temperature.

Impact on low-cloud frequency

Figure 14 shows the distribution of differences in the ceilometer low-cloud frequency of occurrence at the two sites for periods identified as being plume affected at the ARCS and NIES sites. During these times, the frequency of occurrence of low clouds at the downwind site is greater than or equal to the low-cloud frequency at the upwind site over 75% of the time. During some of the times in which the surface measurements are identified as being affected by the cloud plume, the low-cloud frequency of occurrence at the upwind site is equal to or greater than that at the downwind site. One explanation is the exact location of the cloud plume. Matthews (2003) found that in the morning and afternoon, when surface heating is less than that during the middle of the day, cloud formation often occurred over the ocean, slightly downwind of the island. During the middle of the day, cloud production moved upwind, over the island. If the cloud forms downwind of the site, over the ocean, then it may affect the magnitude and variability of the downwelling GSW that is measured at the downwind site (especially in the afternoon at the ARCS site, when clouds forming downwind of the site would be in the line of the direct solar beam) without affecting the ceilometer low-cloud frequency, because the ceilometer is a vertically pointing, narrow field-of-view instrument.

By comparing the ceilometer data for times when plumes affect the site with times when plumes do not affect the site, we can assess the magnitude of the plume effect on the low-cloud frequency at the downwind sites. In Fig. 15, the distribution of the hourly averaged ceilometer low-cloud frequency is shown at each site for periods classified as being plume affected (as established by the previous criteria) and for all other periods. The average frequency of low-cloud occurrence at the ARCS site is 0.34 for affected periods and 0.18 for nonaffected periods. Similarly, at the NIES site, the average low-cloud frequency is 0.35 for affected periods and 0.23 for nonaffected periods. Thus, the existence of the low-cloud plume increases the average low-cloud frequency of occurrence from roughly 20% to 35% at both sites during the NIES study period.

Impact on shortwave surface radiation

To examine the impact of the island-induced plume on the surface radiation, we must compare equivalent time periods because of the solar zenith angle influence on the magnitude of downwelling shortwave radiation. Instead of comparing the surface radiation at one site with and without plumes, we compare the ratio of the downwelling surface radiation at the two sites, as shown in Fig. 16. The average ratio of the ARCS downwelling GSW to the NIES downwelling GSW between 0800 and 1600 LST for periods with no land influence is 1.02. For cases that are plume affected at the ARCS site, the average ratio is 0.92, with a minimum in the late afternoon. This minimum is a result of the geometrical coincidence of the cloud plume at the ARCS site being situated in the west and the direct solar beam being from the west in the late afternoon. The average ratio for plume-affected periods at the NIES site is 1.10, and it is largest between 0900 and 1200 LST. Again, this maximum is probably because of geometry, with the NIES plume being to the east and the sun from the east before noon. Over the entire dataset the average GSW at the surface between 0800 and 1600 LST is 598 W m⁻² at both sites. The average effect of the cloud plume on the total daily surface radiation is around 8%–10%, or 50–60 W m⁻². The impact of the cloud plume on the shortwave radiation can also be seen in Fig. 17, which shows the average downwelling GSW as a function of time at the two sites.

For consistency between the two sites, the analysis in this paper has compared the GSW measured by the Eppley pyranometers at the NIES and ARCS sites. However, when the permanent instruments are reinstalled on an upwind site at Nauru, the downwelling GSW will be measured by a LI-COR pyranometer—a decision influenced by operational constraints. Analysis of the coincident Eppley and LI-COR pyranometers at the NIES site shows that the LI-COR instrument is fairly stable and should provide reasonable data to identify the cloud plume and to estimate the magnitude of the radiative effect at the ARCS site. Figure 18 shows a comparison of the 60-min-averaged GSW and standard deviation in GSW from the LI-COR and Eppley pyranometers at the NIES sites. The correlation between the mean and standard deviation of the GSW measured by the two pyranometers is 0.99, although the LI-COR tends to underestimate the downwelling shortwave at the highest values (under clear skies). Using the LI-COR instead of the Eppley instrument in the analysis that is presented in this paper did not significantly change the results.

Impact on longwave surface radiation

To estimate the impact of the island cloud plume on the downwelling longwave radiation, we again look at the ratio of measured radiation at the two sites, which is shown in Fig. 19 for the daylight hours. The average ratio of the ARCS downwelling longwave radiation to the downwelling longwave radiation at the NIES site from 0800 to 1600 LST is 0.98, with an average value of 425.6 W m⁻² at the ARCS site and 432.4 W m⁻² at the NIES site. For ARCS plume-affected periods, the average ratio is 1.00, with the ratio at its largest in the afternoon. For NIES plume-affected periods, the average ratio is 0.97. The difference in the hourly averages of downwelling longwave radiation between plume-affected and non-plume-affected periods is generally 1%–2%, or less than 10 W m⁻², which is not surprising given the moist nature of the tropical atmosphere.

Aerosol data

As part of the temporary instrumentation, a MFRSR was installed at the NIES site to determine if the aerosol measurements at the ARCS site are affected by the phosphate dust that is stirred up during topside mining on the island. Unfortunately, the MFRSR at the NIES site was not correctly leveled during the initial installation so there were only 2 months of data for comparison. In general, the aerosol optical depths at the two sites were found to track well and agreed within the typical uncertainty of 0.02. Figure 20 shows a time series of retrieved aerosol optical depths at the two sites on 3 May 2003, which is typical of the results. Although there was only a limited dataset, most of the periods had the wind consistently from the east, as would be expected during suppressed conditions at Nauru; and there was no indication that the island was affecting the retrieved optical depths. Because of the typically low values of aerosol optical depths at Nauru (<0.2), the retrieved Ångström parameters have large uncertainties, but, again, no island effect was seen in the retrieved values.