The Atmospheric Radiation Measurement (ARM) program operates three climate observation stations in the tropical western Pacific region. One of these sites, located on Manus Island in Papua New Guinea, has been operating since 1996. The Manus ARM site includes an extensive array of instruments chosen to observe cloud properties, water vapor and temperature profiles, and the surface radiation budget. This dataset provides an opportunity to examine variability in tropical cloudiness on a wide range of time scales. The focus of this study is on the annual cycle. Analysis of cloud distribution and radiation data from Manus reveals a clear annual cycle in clouds associated with convective activity. The most convectively active period is found to be the Northern Hemisphere summer, while the least active period is the Northern Hemisphere autumn. Outgoing longwave radiation (OLR) data are also examined in order to relate observations at Manus with the surrounding region. Significant differences are found between the annual cycle at Manus and adjacent large islands within the Maritime Continent. Analysis of the combined ARM–OLR data suggests that during the Northern Hemisphere winter, a significant amount of the high clouds observed over Manus are associated with continental convection over the large Maritime Continent islands.
The tropical western Pacific (TWP), the eastern Indian Ocean, and the Maritime Continent (MC) are regions that are characterized by strong solar heating, high ocean temperatures, and frequent deep convection. Regional convection-driven upwelling plays an important part in controlling atmospheric circulations on a global scale (Ramage 1968; Webster and Lucas 1992; Neale and Slingo 2003). Because this region plays an important role in global circulation, the Atmospheric Radiation Measurement (ARM) program (Stokes and Schwartz 1994) selected the TWP to deploy three climate observation stations. The three stations are located at Manus (Papua New Guinea), the island nation of Nauru, and Darwin (Australia; Fig. 1). Instruments at these sites collect detailed observations of cloud properties, the surface radiation budget, and a variety of meteorological parameters (Mather et al. 1998a). One of the main goals of the ARM program is to use these observations to improve the representation of clouds and radiation in climate models. To effectively make the connection between ARM data and climate models, it is important to understand how cloud properties and variability in these properties at the ARM sites relate to the larger TWP region.
The distribution of convection within the TWP–MC region is illustrated in Fig. 1, which depicts the average and standard deviation of satellite-derived outgoing longwave radiation (OLR; Chelliah and Arkin 1992). Low values of OLR represent cold cloud tops and, therefore, regions of active convection. Active regions include the Maritime Continent, the intertropical convergence zone (ITCZ; Wang and Wang 1999), the South Pacific convergence zone (SPCZ; Vincent 1994), Southeast Asia, and northern Australia. Manus is located at the intersection of the Maritime Continent, the ITCZ, and the SPCZ.
A broad spectrum of phenomena modulate the basic state of convection depicted in Fig. 1 (Murakami et al. 1986; Sui and Lau 1992). These phenomena span time scales ranging from diurnal to decadal. Diurnal cycles are particularly strong over and adjacent to the large Maritime Continent islands where differential land–sea solar heating results in sea-breeze-related convection (Yang and Slingo 2001; Liberti et al. 2001). There are a wealth of tropical waves that propagate through the equatorial Pacific region. These waves couple with convection to produce cycles of active and suppressed convection. Kelvin, Rossby, and inertial gravity waves span periods from several days to a few weeks and are predicted by equatorial wave theory (Matsuno 1966; Lindzen 1967; Wheeler et al. 2000). The eastward-propagating Madden–Julian oscillation (MJO) has a period of approximately 30–60 days and is also strongly coupled with convection (Madden and Julian 1994; Hendon and Salby 1994). The MJO is an important component of convective variablility in the TWP–MC region; however, GCMs have had a difficult time accurately simulating this phenomenon. In particular, models do not tend to capture the passage of the MJO through the Maritime Continent. The complex distribution of islands in this region is not well resolved by GCMs. These islands lead to convective characteristics such as a strong diurnal cycle that are quite different than the open ocean (Inness and Slingo 2003; Neale and Slingo 2003).
The OLR standard deviation in Fig. 1 indicates several regions with particularly large seasonal and interannual variability: Southeast Asia, northern Australia and the adjacent oceanic region, and the central equatorial Pacific. Variability in the central Pacific is associated with the El Niño–Southern Oscillation. Convection in this region is most active during El Niño events (e.g., Philander 1985; Neelin et al. 1998). The second TWP ARM site at Nauru is located within this region of high variability.
Variability in northern Australia and southern Asia is associated with the annual monsoon cycle (e.g., Webster et al. 1998; Hung et al. 2004). Within the MC–TWP region, convection exhibits an annual cycle in which the region of most active convection tracks along a northwest to southeast line. During the Northern Hemisphere (NH) summer, convection associated with the India–Asia monsoon is concentrated in Southeast Asia. During the NH winter, convection associated with the Australian phase of the monsoon is found in northern Australia and the Maritime Continent. During the transition seasons between the Asia and Australia monsoons, convection tends to be concentrated over the large islands of the Maritime Continent: New Guinea, Borneo, and Sumatra (Meehl 1987). Figure 2 illustrates the southeastward progression of convection through the Maritime Continent during the NH summer and fall. To produce the OLR composites in Fig. 2, 21-yr time series (1982–2002) of OLR over four Maritime Continent locations were smoothed using 20- and 120-day half-width Gaussian filters. The 120-day filtered data were subtracted from the 20-day filtered data to remove long-term trends. Finally, the 21 yr were averaged for each site to produce annual composites. Negative excursions in this figure indicate active convection while positive values indicate relatively suppressed conditions. The same technique will be used in several subsequent plots to examine the annual cycle in other parameters related to convection.
An OLR composite is included for Manus in Fig. 2. Manus is located approximately 300 km north of the island of New Guinea but the Manus OLR composite does not fit the southeastward progression of convection during the NH autumn. OLR is particularly sensitive to cirrus due to their very low temperatures. Cirrus, in turn, are often used as an indicator of convective activity because, in the Tropics, cirrus are often produced in the outflow of deep convective cells. However, cirrus may persist for many hours. In a satellite image of OLR, it may be difficult to determine the source of cirrus over a given location. If Manus were in a spatially homogeneous region, one might expect that cirrus observed over the site would have originated from oceanic convection and that over time, one would observe clouds with randomly distributed ages. But Manus is located adjacent to the Maritime Continent and lies near several large islands over which convection is strongly influenced by the annual solar cycle.
With vertically pointing remote sensing instruments such as those used at the TWP ARM sites, one cannot determine the source of a cirrus layer without auxiliary information. The purpose of this paper is to examine the annual cycle of convection and associated cloudiness at Manus and to determine the source of cirrus observed at Manus throughout the annual cycle. To study cirrus over Manus and understand the source of these clouds, a variety of observations will be used. Most of these observations are from instruments associated with the ARM site. First, the annual cycle in vertical cloud distributions will be examined using observations from active remote sensors at the ARM site. The impact of these clouds on the surface radiation budget and the relationship between cloudiness and local convective activity will be examined through analysis of radiation and precipitation data obtained at Manus. In addition to the ARM observations, OLR measurements and the National Centers for Environmental Prediction (NCEP) reanalysis data will be used to provide a top-of-the-atmosphere perspective and to relate the observations at Manus to the surrounding region.
2. Cloud observations at Manus
The two primary instruments for detecting clouds over the ARM sites are the millimeter cloud radar (Moran et al. 1998) and the micropulse lidar (Spinhirne 1993). Figure 3 plots the frequency with which these instruments observe clouds in a vertical column over Manus as a function of time and altitude for the period August 1999–November 2000. This period represents the longest period of concurrent lidar and radar data currently available for the Manus site. To produce this plot, the ARM Active Remotely-Sensed Cloud Locations (ARSCL) data product was used. ARSCL combines cloud radar and lidar data to determine the vertical distribution of clouds as a function of time (Clothiaux et al. 2000). One of the features that is most obvious in this figure is the annual cycle of the maximum observed cloud-top altitude. To examine the possible impact of variations in the tropopause height on cloud top, a time series of the cold-point altitude has been superposed on the radar data in Fig. 3. The cold-point data were derived from radiosondes launched from the Manus site. These data have been smoothed using a 10-day running mean and are presented as an estimate of the tropopause altitude. As noted previously by others (Reid and Gage 1981; Zhang 1993) there is an annual cycle in the tropopause height, but over Manus, the amplitude of this cycle is considerably less than the oscillation in the cloud-top heights.
During much of the period with frequent high clouds (October–April) there are relatively few low- and midaltitude clouds. A high frequency of cirrus clouds in the absence of lower clouds suggests that those cirrus were not formed locally (i.e., within 10–20 km). In Fig. 4, the occurrence of clouds above 14 km is compared to a time series of OLR over the large island of New Guinea. These data show a strong correlation between convection over New Guinea with high cirrus over Manus. The annual cycle in convection over New Guinea, and other large islands in the Maritime Continent, is closely synchronized with the annual solar cycle. Figure 2 illustrates this phenomenon by plotting the annual OLR cycle for pixels over selected locations in Maritime Continent locations ranging from Sumatra in the northwest to New Guinea in the southeast. Solar insolation over Papua New Guinea (PNG) is highest during the period when convection at that site is active. It seems, therefore, that a likely source for the high-level clouds over Manus is convection over the larger islands of PNG: New Guinea, New Ireland, and New Britain. New Guinea lies approximately 300 km to the south of Manus while New Ireland and New Britain lie approximately 250–500 km to the southeast.
Figure 5 shows the 150-hPa zonal and meridional winds from sounding data at Manus. These data show that during the NH winter, the upper-level winds are from the east with speeds of approximately 10 m s−1 for much of the period. To corroborate these observations, streamlines of the 150-hPa NCEP climatological reanalysis winds are plotted in Fig. 6. In the wind analysis for January, the streamlines in the vicinity of Manus originate from the east. Given that the wind direction during the [December–March (DJFM)] period is easterly, it is likely that rather than the main island of New Guinea, New Britain or New Ireland are more likely sources for upper-level clouds. Given a wind speed of 10 m s−1, the time for a cirrus cloud to propagate from these islands to Manus would be approximately 7–14 h. Analysis of tropical cirrus persistence indicates that the lifetime for cirrus directly associated with detrainment is 30 ± 16 h (Luo and Rossow 2004).
3. Cloud radiative forcing at Manus
The daily averaged surface shortwave (0.3–4.0 μm) solar radiation exhibits an annual cycle due to the progression of the solar declination angle and Earth’s distance to the Sun. To focus on the effect of clouds on solar radiation, Fig. 7 displays a time series of the surface shortwave cloud radiative forcing (CRF) defined here as the (observed downwelling solar flux minus derived clear downwelling solar flux)/(derived clear downwelling solar flux). Clear-sky fluxes were estimated following the technique of Chou and Zhao (1997) in which periods are found when the solar direct beam is maximized at the same time the diffuse beam is minimized. The resulting shortwave observations are then used to construct a relationship between solar zenith angle and clear-sky flux (Mather et al. 1998b). By normalizing the shortwave data to the theoretical clear-sky value, the dependence on solar position is removed. Thus, the quantity CRF illustrates the effect of clouds on the solar radiation. Periods when CRF approaches zero tend to indicate suppressed convection while large negative values indicate active conditions.
The time series in Fig. 7 was smoothed using a Gaussian filter with a half-width of 7 days. In much of the time series, an intraseasonal oscillation is evident in the data. To evaluate the period of this oscillation, the power spectrum of this time series is plotted in Fig. 8. This spectrum indicates that the activity exhibits a maximum in variability for time scales in the range of 40–70 days. This range corresponds to the range associated with the MJO (Madden and Julian 1994; Hendon and Salby 1994). The spectrum also shows peaks at longer wavelengths, particularly at 120 days. In some years, particularly 1999 and 2001, a semiannual cycle is evident in Fig. 7. In this cycle, relatively suppressed conditions are found during the NH spring and fall.
To better examine this seasonal cycle, an annual composite, constructed from the time series in Fig. 7 is plotted in Fig. 9. To generate this composite, a bandpass filter (20–120 days) was applied to the shortwave cloud forcing time series (as in Fig. 2). The bandpass filter was important for removing interannual variability from the data. This step also resulted in removing the mean from the time series so that the data oscillate about zero. Positive values represent weaker-than-average cloud forcing values (relatively clear conditions) while negative values correspond to stronger-than-average forcing (relatively cloudy conditions). The pattern in the composite is very similar to the pattern referred to in Fig. 7 for 1999 and 2001. In the annual composite, clear conditions are found in April–June and August–October with the most suppressed conditions found during the latter period. Active conditions are found in November–March and in July. The long active period during NH winter is coincident with the Australian monsoon. This period also coincides with the period of frequent high clouds in Fig. 3. There is little evidence of enhanced convection during July in the radar data but this is difficult to evaluate with only 1 yr of data.
In Fig. 2, the annual composite of OLR over Manus was shown along with composites for several Maritime Continent sites. As with the shortwave cloud forcing composite in Fig. 9, positive values in the OLR composite correspond to relatively clear conditions while negative values correspond to relatively cloudy conditions. The data in Fig. 2 span the period 1982–2002. This relatively long period can be used to test the robustness of the annual cycle pattern. In Fig. 10, composites are calculated for the entire 1982–2002 period as well as for three 7-yr subperiods. For each subperiod, the pattern is remarkably similar to that observed in the shortwave time series. Again, suppressed conditions are found peaking in May and August–September. The precise timing of the convective maxima and minima varies in the subperiods but the pattern is consistent for all periods.
4. Analysis and discussion
In the previous two sections, the effects of convection have been seen through the distribution of clouds and on the impact of these clouds on solar radiation at the surface and infrared radiation at the top of the atmosphere. In the following discussion, the relationship between these observations will be further studied to better understand the physical basis for the annual cycle observed at Manus.
In Fig. 11, the shortwave cloud radiative forcing at Manus is plotted as a function of OLR. A linear regression of these two variables shows a strong relationship between the two variables with the OLR over Manus explaining 70% of the variance in CRF. To further examine the relationship between these two indicators of convective activity, their annual composites are compared in Fig. 12. In this figure, the cloud radiative forcing is scaled to the OLR using the regression in Fig. 11. The residual of these two composites (i.e., OLR-scaled CRF) is also plotted. Interestingly, this residual exhibits an annual cycle. Strong negative residual values indicate periods when the actual OLR is less than predicted from the surface shortwave forcing. This would occur if the predominant clouds over Manus are thin cirrus, which have a strong impact on the OLR but significantly less impact on the surface shortwave. Conversely, positive residuals would indicate a high frequency of optically thick clouds that could include convective cells or anvil outflow in close proximity to convection.
The OLR residual reaches its minimum values (frequent cirrus relative to local convection) during the NH winter. This timing is consistent with the observations of OLR and radar-derived cloud observations in the previous sections for the NH winter. Thus, like the radar observations, this view of convection at Manus suggests that while cirrus is frequent during the NH winter, much of the cirrus was not formed through local convection.
The OLR residual maximum occurs in late July–early August shortly after the NH summer convection peak at Manus (Figs. 9 and 2) at a time when convection is suppressed over the Maritime Continent islands (Fig. 2). During the NH summer, solar insolation over the Tropics is at a minimum. The reduction of island-based convection combined with a weaker sensitivity of the ocean surface to the annual solar cycle (in comparison to the adjacent land) increases the likelihood of maritime convection during this period.
The difference in convective activity at Manus in the NH winter and summer seasons can also be seen in a composite of daily precipitation observed at the Manus ARM site (Fig. 13). The rainfall data show a sharp peak, corresponding to local convection, during July. Moderate amounts of rainfall are found throughout the NH winter and spring, corresponding to a period of widespread convergence associated with the Australian monsoon. A pronounced period of dry conditions is found during September–October. During this period of transition between the Asia and Australia monsoons, convection over the large Maritime Continent islands sets in well in advance of convection over Manus (Fig. 2). The combination of concentrated convective activity over the large islands with suppressed convection at Manus suggests that convection during this transition is dominated by the surface heating of the islands.
The difference in the annual cycles of island and maritime convection within the TWP–MC region is important because these two classes of convection have different characteristics (Keenan and Carbone 1992; Zipser 1994; Carbone et al. 2000). Island convection is characterized by strong updrafts and very high cloud tops resulting from strong surface heating and the dynamical effects of sea breezes. Maritime convection tends to be associated with weaker updrafts. While the properties of these two classes of convective clouds have not been explored here, it is likely that island and maritime convection have different microphysical and radiative characteristics. In the context of capturing TWP–MC convective clouds and their radiative impact in large-scale models, it will be important to correctly attribute cirrus layers to land- or ocean-based convection so that the radiative properties of these clouds are accurate.
Located on a small island adjacent to the Maritime Continent region, the Manus ARM site experiences convection throughout the year. Time series of cloud occurrence frequency or solar radiation are complex but when the influences of the MJO and other relatively high-frequency waves are removed, an annual pattern is evident. This pattern is quite different than one finds in convection over nearby large islands. New Guinea exhibits a strong annual cycle, in phase with the annual solar cycle, with a maximum in convective activity during the NH winter. Local convection at Manus is relatively suppressed during this period of maximum activity over the islands although cirrus occurs with greater frequency than during any other season. The period that exhibits the most consistently active convection at Manus is the NH summer when convection over the large islands is relatively suppressed. Thus, over the course of a year, the relative likelihood of maritime and island convection within and near the Maritime Continent shifts toward the islands as solar forcing of the islands increases and toward maritime convection as solar forcing decreases in the region.
Thus, the presence of large islands within the Maritime Continent has a significant impact on the distribution of convection and associated cirrus in this region. Neale and Slingo (2003) have noted that the removal of these islands from a global climate model has far-reaching consequences for global dynamics but that global-scale models are currently unable to capture island-scale dynamics. From the discussions in this paper, it is clear that important processes are occurring on spatial scales smaller than typically resolved by global climate models (the separation of Manus and New Guinea is approximately 300 km). These processes strongly influence the distribution, timing, and radiative/microphysical characteristics of convection within the TWP–MC region.
The Manus site, located in the core of the Pacific oceanic warm pool, was chosen to represent the TWP–MC region, a region of high convective activity. Because of its proximity to New Guinea, New Britain, and New Ireland, remote sensing instruments at Manus will at times observe cirrus generated locally and at times from convection over these large islands. Given that both types of convection are found within the TWP–MC region, the Manus site provides an opportunity to study the broad range of cirrus types found in the Maritime Continent. However, to use these data, it will be important to properly associate observations of cloud properties observed at Manus to the convective source of those clouds. Combining cloud property retrievals from the ARM active remote sensors with sources of spatial information, including the satellite and reanalysis datasets used in this study, will provide a means to apply results obtained at the Manus ARM site throughout the region.
This work was supported by the DOE Atmospheric Radiation Measurement program. Daily outgoing longwave radiation and NCEP reanalysis data were obtained from the NOAA–CIRES Climate Diagnostics Center (available online at http://www.cdc.noaa.gov/). Coastline data for maps of the tropical Pacific region were obtained from the National Geophysical Data Center (available online at http://www.ngdc.noaa.gov/).
Corresponding author address: Dr. James H. Mather, Pacific Northwest National Laboratory, P.O. Box 999, MS K9-24, Richland, WA 99354. Email: firstname.lastname@example.org