• Ahn, M.-H., , E.-H. Shon, , and B.-J. Hwang, 2003: A new algorithm for sea fog/stratus detection using GMS-5 IR data. Adv. Atmos. Sci., 20, 899913, doi:10.1007/BF02915513.

    • Search Google Scholar
    • Export Citation
  • Albrecht, B. A., , D. A. Randall, , and S. Nicholls, 1988: Observations of marine stratocumulus cloud during FIRE. Bull. Amer. Meteor. Soc., 69, 618626, doi:10.1175/1520-0477(1988)069<0618:OOMSCD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Betts, A. K., , and R. Boers, 1990: A cloudiness transition in a marine boundary layer. J. Atmos. Sci., 47, 14801497, doi:10.1175/1520-0469(1990)047<1480:ACTIAM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Betts, A. K., , C. S. Bretherton, , and E. Kliner, 1995: Relation between boundary-layer structure and cloudiness at the R/V Valdivia during ASTEX. J. Atmos. Sci., 52, 27522762, doi:10.1175/1520-0469(1995)052<2752:RBMBLS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Cermak, J., , and J. Bendix, 2008: A novel approach to fog/low stratus detection using Meteosat 8 data. Atmos. Res., 87, 279292, doi:10.1016/j.atmosres.2007.11.009.

    • Search Google Scholar
    • Export Citation
  • Ellrod, G. P., 1995: Advances in the detection and analysis of fog at night using GOES multispectral infrared imagery. Wea. Forecasting, 10, 606619, doi:10.1175/1520-0434(1995)010<0606:AITDAA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Ellrod, G. P., , and I. Gultepe, 2007: Inferring low cloud base heights at night for aviation using satellite infrared and surface temperature data. Pure Appl. Geophys., 164, 11931205, doi:10.1007/s00024-007-0214-7.

    • Search Google Scholar
    • Export Citation
  • Gultepe, I., , M. Pagowski, , and J. Reid, 2007a: A satellite-based fog detection scheme using screen air temperature. Wea. Forecasting, 22, 444456, doi:10.1175/WAF1011.1.

    • Search Google Scholar
    • Export Citation
  • Gultepe, I., and Coauthors, 2007b: Fog research: A review of past achievements and future perspectives. Pure Appl. Geophys., 164, 11211159, doi:10.1007/s00024-007-0211-x.

    • Search Google Scholar
    • Export Citation
  • Gultepe, I., and Coauthors, 2009: The Fog Remote Sensing and Modeling field project. Bull. Amer. Meteor. Soc., 90, 341359, doi:10.1175/2008BAMS2354.1.

    • Search Google Scholar
    • Export Citation
  • Ishida, H., , and T. Y. Nakajima, 2009: Development of an unbiased cloud detection algorithm for a spaceborne multispectral imager. J. Geophys. Res., 114, D07206, doi:10.1029/2008JD010710.

    • Search Google Scholar
    • Export Citation
  • Ishida, H., , K. Miura, , T. Matsuda, , K. Ogawara, , A. Goto, , K. Matsuura, , Y. Sato, , and T. Y. Nakajima, 2014: Scheme for detection of low clouds from geostationary weather satellite imagery. Atmos. Res., 143, 250264, doi:10.1016/j.atmosres.2014.02.015.

    • Search Google Scholar
    • Export Citation
  • Kodama, Y., 1997: Airmass transformation of the yamase air-flow in the summer of 1993. J. Meteor. Soc. Japan, 75, 737751. [Available online at http://www.st.hirosaki-u.ac.jp/~kodama/JMSJkodama1997.pdf.]

    • Search Google Scholar
    • Export Citation
  • Kodama, Y., , Y. Tomita, , and S. Asano, 2009: Air mass transformation along trajectories of airflow and its relation to vertical structures of the maritime atmosphere and clouds in yamase events. J. Meteor. Soc. Japan, 87, 665685, doi:10.2151/jmsj.87.665.

    • Search Google Scholar
    • Export Citation
  • Koračin, D., , J. M. Lewis, , W. T. Thompson, , C. E. Dorman, , and J. A. Businger, 2001: Transition of stratus into fog along the California coast: Observations and modeling. J. Atmos. Sci., 58, 17141731, doi:10.1175/1520-0469(2001)058<1714:TOSIFA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Koračin, D., , C. E. Dorman, , J. M. Lewis, , J. G. Hudson, , E. M. Wilcox, , and A. Torregrosa, 2014: Marine fog: A review. Atmos. Res., 143, 142175, doi:10.1016/j.atmosres.2013.12.012.

    • Search Google Scholar
    • Export Citation
  • Lee, T. F., , F. J. Turk, , and K. Richardson, 1997: Stratus and fog products using GOES-89 3.9-μm data. Wea. Forecasting, 12, 664677, doi:10.1175/1520-0434(1997)012<0664:SAFPUG>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Lewis, J., , D. Koracin, , R. Rabin, , and J. Businger, 2003: Sea fog off the California coast: Viewed in the context of transient weather systems. J. Geophys. Res., 108, 4457, doi:10.1029/2002JD002833.

    • Search Google Scholar
    • Export Citation
  • Lewis, J., , D. Koracin, , and K. T. Redmond, 2004: Sea fog research in the United Kingdom and United States: A historical essay including outlook. Bull. Amer. Meteor. Soc., 85, 395408, doi:10.1175/BAMS-85-3-395.

    • Search Google Scholar
    • Export Citation
  • Muroi, C., , T. Fujita, , and Y. Ishikawa, 2008: Hourly analysis at the Japan Meteorological Agency (in Japanese).Tenki, 55, 401408.

  • Norris, J. R., 1998a: Low cloud type over the ocean from surface observations. Part I: Relationship to surface meteorology and the vertical distribution of temperature and moisture. J. Climate, 11, 369382, doi:10.1175/1520-0442(1998)011<0369:LCTOTO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Norris, J. R., 1998b: Low cloud type over the ocean from surface observations. Part II: Geographical and seasonal variations. J. Climate, 11, 383403, doi:10.1175/1520-0442(1998)011<0383:LCTOTO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Norris, J. R., , Y. Zhang, , and J. M. Wallace, 1998: Role of low clouds in summertime atmosphere–ocean interactions over the North Pacific. J. Climate, 11, 24822490, doi:10.1175/1520-0442(1998)011<2482:ROLCIS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Randall, D. A., , J. A. Coakley Jr., , C. W. Fairall, , R. A. Kropfli, , and D. H. Lenschow, 1984: Outlook for research on subtropical marine stratiform clouds. Bull. Amer. Meteor. Soc., 65, 12901301, doi:10.1175/1520-0477(1984)065<1290:OFROSM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Rogers, D. P., , and D. Koračin, 1992: Radiative transfer and turbulence in the cloud-topped marine atmospheric boundary layer. J. Atmos. Sci., 49, 14731486, doi:10.1175/1520-0469(1992)049<1473:RTATIT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Saito, K., , T. Fujita, , J. Ishida, , Y. Kumagai, , K. Aranami, , S. Ohmori, , R. Nagasawa, , and S. Kumagai, 2006: The operational JMA nonhydrostatic mesoscale model. Mon. Wea. Rev., 134, 12661298, doi:10.1175/MWR3120.1.

    • Search Google Scholar
    • Export Citation
  • Saito, K., , J. Ishida, , K. Aranami, , T. Hara, , T. Segawa, , M. Narita, , and Y. Honda, 2007: Nonhydrostatic atmospheric models and operational development at JMA. J. Meteor. Soc. Japan, 85B, 271304, doi:10.2151/jmsj.85B.271.

    • Search Google Scholar
    • Export Citation
  • Underwood, S. J., , G. P. Ellrod, , and A. L. Kuhnert, 2004: A multiple-case analysis of nocturnal radiation-fog development off California utilizing GOES nighttime fog product. J. Appl. Meteor., 43, 297311, doi:10.1175/1520-0450(2004)043<0297:AMAONR>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Wang, J., , W. B. Rossow, , T. Uttal, , and M. Rozendaal, 1999: Variability of cloud vertical structure during ASTEX observed from a combination of rawinsonde, radar, ceilometer, and satellite. Mon. Wea. Rev., 127, 24842502, doi:10.1175/1520-0493(1999)127<2484:VOCVSD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Flowchart for the selection of pixels containing low water clouds for this study. Here, CCL is the clear confidence level, Tb(ch1) is the channel-1 brightness temperature, T(700 hPa) is the GPV temperature at 700 hPa, RH(any) is the relative humidity at an arbitrary pressure level (below 700 hPa), and RH(surface) is the relative humidity at the surface.

  • View in gallery

    Relative frequency of low cloud that is touching the surface (TS cloud) in the (a) spring, (b) summer, (c) autumn, and (d) winter samples. Values are the ratio of total days of low-cloud occurrence to all 30 days of data in each season sample.

  • View in gallery

    As in Fig. 2, but for low cloud that is not touching the surface (NTS cloud).

  • View in gallery

    Relative frequency of the number of layers for (a) TS clouds and (b) NTS clouds in the summer sample.

  • View in gallery

    Relative frequency of the top height of TS clouds in the (a) spring, (b) summer, (c) autumn, and (d) winter samples. The relative frequencies are derived through division by the total number of pixels of TS clouds in each season sample. The histogram bin width is 250 m.

  • View in gallery

    As in Fig. 5, but for NTS clouds.

  • View in gallery

    Two-dimensional histogram of NTS clouds for the cloud-top height and the cloud thickness in the (a) summer and (b) winter samples. Both histogram bin widths are 250 m.

  • View in gallery

    Two-dimensional histogram of (a) TS clouds and (b) NTS clouds for the cloud-top height and the bottom height of the Γ > Γm layer in the summer sample. Both histogram bin widths are 250 m.

  • View in gallery

    Relative frequency of the normalized PT gap across the bottom of the Γ > Γm layer for (a) TS clouds and (b) NTS clouds in the spring sample. If the Γ > Γm layer attaches to the surface, the normalized PT gap corresponds to the vertical gradient of the potential temperature at the layer just above the surface. Histogram bin width is 0.002 K m−1.

  • View in gallery

    (a) Schematic of a hypothetical vertical structure of θe and a corresponding ideal cloud layer. Also shown are the θe profiles for categorization as listed in Table 7: (b) a Ge > 0 layer exists in the lower atmosphere or (c) it does not; when a Ge > 0 layer exists, the Ge > 0 layer is included within the lower atmosphere and (d) a Ge < 0 layer exists between the 700-hPa level and the lowest Ge > 0 layer or (e) it does not; (f) a Ge ≤ 0 layer exists under the lowest Ge > 0 layer or (g) it does not.

  • View in gallery

    Relative frequency of the normalized difference in θe between the bottom of the lowest Ge > 0 layer and the surface for (a) TS clouds and (b) NTS clouds in the spring sample. The relative frequencies are derived through division by the number of pixels that contain the Ge ≤ 0 layer under the lowest Ge > 0 layer for each case. The histogram bin width is 0.002 K m−1.

  • View in gallery

    Relative frequency of the difference between the forecast Tsfc and Tb(ch1) for (a) TS cloud and (b) NTS cloud in the autumn sample. The histogram bin width is 1 K.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 5 5 1
PDF Downloads 2 2 0

Investigation of Low-Cloud Characteristics Using Mesoscale Numerical Model Data for Improvement of Fog-Detection Performance by Satellite Remote Sensing

View More View Less
  • 1 Graduate School of Science and Engineering, Yamaguchi University, Ube, Japan
  • 2 Japan Weather Association, Tokyo, Japan
  • 3 Research and Information Center, Tokai University, Tokyo, Japan
© Get Permissions
Full access

Abstract

The comprehensive relationship between meteorological conditions and whether low water cloud touches the surface, particularly at sea, is examined with the goal of improving low-cloud detection by satellite. Gridpoint-value data provided by an operational mesoscale model with integration of Multifunction Transport Satellite-2 data can provide sufficient data for statistical analyses to find general parameters that can discern whether low clouds touch the surface, compensating for uncertainty due to the scarcity of observation sites at sea and the infrequent incidence of fog. The analyses reveal that surface-touching low clouds tend to have lower cloud-top heights than those not touching the surface, although the frequency distribution of cloud-top height differs by season. The bottom of the Γ > Γm layer (where Γ and Γm are the vertical gradient and the moist-adiabatic lapse rate of the potential temperature, respectively) with surface-touching low-cloud layers tends to be very low or almost attached to the surface. In contrast, the tops of low-cloud layers not touching the surface tend to occur near the bottom of the Γ > Γm layer. Mechanisms to correlate these meteorological conditions with whether low clouds touch the surface are inferred from investigations into the vertical structure of equivalent potential temperature. These results indicate that the temperature difference between cloud-top height and the surface can be an appropriate parameter to infer whether low clouds touch the surface. It is also suggested that only a little addition of meteorological ancillary data, such as the forecast sea surface temperature, to satellite data allows successful performance of the discrimination.

Corresponding author address: H. Ishida, Graduate School of Science and Engineering, Yamaguchi University, 2-16-1, Tokiwa-dai, Ube, Yamaguchi 755-8611, Japan. E-mail: ishidah@yamaguchi-u.ac.jp

Abstract

The comprehensive relationship between meteorological conditions and whether low water cloud touches the surface, particularly at sea, is examined with the goal of improving low-cloud detection by satellite. Gridpoint-value data provided by an operational mesoscale model with integration of Multifunction Transport Satellite-2 data can provide sufficient data for statistical analyses to find general parameters that can discern whether low clouds touch the surface, compensating for uncertainty due to the scarcity of observation sites at sea and the infrequent incidence of fog. The analyses reveal that surface-touching low clouds tend to have lower cloud-top heights than those not touching the surface, although the frequency distribution of cloud-top height differs by season. The bottom of the Γ > Γm layer (where Γ and Γm are the vertical gradient and the moist-adiabatic lapse rate of the potential temperature, respectively) with surface-touching low-cloud layers tends to be very low or almost attached to the surface. In contrast, the tops of low-cloud layers not touching the surface tend to occur near the bottom of the Γ > Γm layer. Mechanisms to correlate these meteorological conditions with whether low clouds touch the surface are inferred from investigations into the vertical structure of equivalent potential temperature. These results indicate that the temperature difference between cloud-top height and the surface can be an appropriate parameter to infer whether low clouds touch the surface. It is also suggested that only a little addition of meteorological ancillary data, such as the forecast sea surface temperature, to satellite data allows successful performance of the discrimination.

Corresponding author address: H. Ishida, Graduate School of Science and Engineering, Yamaguchi University, 2-16-1, Tokiwa-dai, Ube, Yamaguchi 755-8611, Japan. E-mail: ishidah@yamaguchi-u.ac.jp

1. Introduction

Fogs, which are clouds that occur very near the surface (with visibility of less than 1 km for the strict definition; e.g., Gultepe et al. 2007b), have various impacts on human activities. In particular, a reduction in visibility near the surface as a result of fog affects land, sea, and air traffic (e.g., Gultepe et al. 2009), occasionally causing weather hazards. Fogs are sometimes optically thick and of sufficient duration to effectively reflect solar radiance, resulting in substantial changes to the radiation budget and Earth’s climate (e.g., Randall et al. 1984; Norris et al. 1998). Fogs influence meteorological conditions, especially in the boundary layer, through radiative cooling and heating within a fog layer. Changes induced in the thermal profile by fog modify turbulent transfer; this, in turn, influences fog development and dissipation through changes in moisture and heat flux (e.g., Rogers and Koračin 1992; Koračin et al. 2001). It is therefore important to observe fog generation, development, advection, and dissipation.

Remote sensing by geostationary weather satellites is useful for observing fogs in a wide region and monitoring fog continuously in near–real time. Because the temperature contrast between the top of a fog layer and the underlying surface is usually less distinct than that for other cloud types, fog detection by satellite requires care to distinguish between clouds and clear sky. This is especially true during nighttime, when reflectance in the visible- to near-infrared wavelengths is unavailable. Many remote sensing schemes have been developed for this purpose (e.g., Ellrod 1995; Lee et al. 1997; Ahn et al. 2003; Underwood et al. 2004; Gultepe et al. 2007a; Cermak and Bendix 2008). A general fog-detection scheme using current geostationary weather satellites at night is constructed by considering the radiative properties of water particles: the emissivity of small water particles in the midinfrared region is less than that in the window region (Ellrod 1995). In contrast to this scheme, fog detection in the daytime typically utilizes the reflectance of solar radiance. All fog-detection schemes that use a passive sensor on a satellite (e.g., the current geostationary weather satellites) share a common problem, however: because passive sensors cannot easily observe the environment below the cloud top, it is difficult to retrieve the cloud-bottom height and then to distinguish whether a cloud layer touches the surface. In other words, geostationary weather satellites are in principle able to detect low-cloud occurrence only. The difference between touching and not touching the surface, however, is crucial for human activity. Therefore, observation with passive sensors requires an additional procedure to infer whether a low-cloud layer touches the surface.

The additional procedure should be simple and should use only the least additional data so that discrimination is convenient and suitable for near-real-time release of data. To construct a simple and convenient procedure, we need to know what meteorological conditions determine and/or are related to whether low clouds touch the surface and their applicability to a given scheme for low-cloud detection. In general, thermodynamic and dynamic processes govern the development and dissipation of low cloud (including fog). Observations and model simulations indicated that low clouds tend to occur under certain meteorological conditions. For example, a comprehensive analysis of sounding data and synoptic cloud-type observations taken at several weather stations at sea over a long period (Norris 1998a) revealed that, at the observation locations, fogs (and fair-weather stratus) typically occur in a surface-based inversion layer with larger stability and that stratocumulus typically occurs in a shallow marine boundary layer. The incidence and duration of sea fogs off the California coast depended not only on the strength of large-scale subsidence but also on the strength and height of the marine-layer inversion (Lewis et al. 2003). A model simulation suggested that there is likely to be an optimal inversion strength that is favorable to fog occurrence (Koračin et al. 2001). Kodama et al. (2009) showed from ship observations that sea fogs in the northeastern Pacific Ocean during summer also tended to appear in a shallow stable layer attached to the surface. Furthermore, several studies suggested that some parameters measured by satellite might be related to the height of low-cloud layers. For example, a comparison by Ellrod and Gultepe (2007) of visibility observations at the surface with satellite data indicated that the difference between the surface temperature and the infrared brightness temperature of the cloud top was correlated with the cloud-base height. There has been little research that has focused on the relationship of meteorological conditions with whether a low cloud touches the surface and its applicability to fog detection by satellite, however.

In this study, we statistically analyze meteorological and satellite data to find general parameters that not only are able to discriminate whether low clouds touch the surface, especially at sea, but also are useful for fog detection from geostationary-weather-satellite data. In particular, we focus on the cloud-top height and the vertical structure of the potential temperature, because previous studies imply that these things are likely related to whether low clouds touch the surface. In addition, we try to infer the mechanisms relating the parameters to surface touching by low clouds. We use Multifunction Transport Satellite-2 (MTSAT-2) data in low-cloud detection by employing a particular retrieval algorithm. This means that this study could develop a procedure that is suitable for the employed low-cloud-detection algorithm as an example of an improvement in fog detection by satellite. We also use gridpoint-value (GPV) data provided from the operational mesoscale model (MSM) of the Japan Meteorological Agency (JMA) (Saito et al. 2006, 2007). We combine the GPV data with the coincident MTSAT-2 data to select appropriate data for the study purpose and to examine the meteorological conditions when low clouds occur. GPV data might be less accurate than in situ observation data, but the size of the dataset allows statistical analyses to extract general and comprehensive tendencies in the meteorological conditions that occur coincident with low cloud: this mitigates uncertainty that is due to both the scarcity of observation sites at sea and the low incidence of fog.

This study is restricted to water cloud, excluding ice and mixed-phase fogs from the analysis, so as to make the study simple. The strict definition of fog requires visibility to be less than 1 km, as mentioned above, but it is generally difficult to accurately retrieve visibility and distinguish cloud type from GPV and satellite data alone. We therefore ignore visibility (i.e., optical thickness) and cloud type and only take into account the vertical location of cloud. In this paper, we define “lower atmosphere” as the atmosphere below the 700-hPa pressure level (generally corresponding to an altitude of ~3000 m). We also define “low cloud” as (water) cloud within the lower atmosphere, implying stratus, stratocumulus, and cumulus, and we consider surface-touching low cloud to be fog.

2. Data description

a. MTSAT-2 data

MTSAT-2 is a geostationary weather satellite located at a longitude of ~145°E and operated by JMA since 2009. MTSAT-2 has one visible channel and four thermal infrared channels (listed in Table 1) and usually observes with an interval of 30 min. We apply MTSAT-2 data not only to low-cloud detection but also to the selection of appropriate data for this study. We extracted the MTSAT-2 data within an area from 22.4° to 47.6°N and from 120.0° to 150.0°E, which corresponds to the domain of the MSM GPV data. The area covered by the data includes the seas around Japan, where dense fog sometimes occurs (e.g., Lewis et al. 2004). In particular, cold air is often transported from the Okhotsk Sea to the northern Pacific near Japan in summer (an air movement known as yamase), inducing fog formation and advection (Kodama 1997; Kodama et al. 2009). The MTSAT-2 data are arranged on a grid with a resolution of 0.05° longitude × 0.05° latitude. This resolution is reasonably close to the original resolution of the thermal channels (Table 1).

Table 1.

Characteristics of the MTSAT-2 channels.

Table 1.

b. Low-cloud detection

We employ an algorithm of Ishida et al. (2014) for detection of low water cloud with MTSAT-2. Here, we briefly explain this algorithm and refer the interested reader to that paper for details. The algorithm contains the following steps: 1) exclude cirrus, 2) exclude middle and high clouds by inferring the cloud-top height, and 3) distinguish between clear sky and (low) clouds by using all the channels of MTSAT-2. The brightness temperature difference (BTD) between channels 1 (10.8 μm) and 2 (12.0 μm), referred to as the split window, is applied to cirrus exclusion, which is required for avoiding a mistake in discrimination. (Hereinafter channels will be referred to as ch#, where # indicates the channel number.) The cloud-top height can be roughly estimated from the BTD between ch1 and ch3 (6.7 μm) (referred to as VBTD) under the assumption that ch3 radiance is only minimally affected by conditions in the lower atmosphere because of strong absorption by water vapor. VBTD is applied to exclude high and middle clouds, with clear sky or low clouds remaining. The scheme to discriminate between clear sky and (low) clouds depends on the presence of insolation. At night, the brightness temperature difference between ch4 and ch1 (referred to as MBTD) of low clouds with smaller water droplets (an effective radius smaller than ~8 μm) tends to be a lower (often negative) value than that for clear sky, and the split-window value of low clouds with an effective radius larger than about 16 μm is usually lower than that of clear sky. In contrast, the solar reflectance of the visible channel for general clouds (low or high) is usually greater than the solar reflectance of the visible channel on the ocean surface (i.e., clear sky). MBTD, which includes the solar radiance as well as the thermal radiance, is usually lower for clear sky than for low clouds. These radiative properties are utilized for low-cloud detection, depending on whether it is nighttime or daytime, which is determined from the solar zenith angle at each pixel. Because it is usually difficult to detect low clouds when the solar zenith angle is large, we do not carry out low-cloud detection, and we exclude these pixels from the data. This means that low clouds occurring at dawn and dusk are not included in the analysis of this study. This water low-cloud detection algorithm finally yields a clear confidence level (CCL), which represents the certainty of the clear-sky/cloud discrimination (Ishida and Nakajima 2009). CCL of 0 and 1 represent (low) cloud and clear sky, respectively, and a CCL between 0 and 1 represents vague or indistinct conditions. The accuracy of the fog-detection results is discussed in Ishida et al. (2014) with some verification. Table 2 summarizes the threshold tests used for low-cloud detection with MTSAT-2.

Table 2.

Threshold tests used for low-cloud detection; Tb and R indicate brightness temperature and reflectance, respectively.

Table 2.

c. GPV data

We need meteorological data in order to select appropriate data for this study as well as to examine the meteorological conditions that are related to whether low clouds touch the surface. The most reliable data for obtaining the vertical location of the cloud bottom are in situ observations at the surface, such as the visibility. In general, however, observation sites at sea are sparse and fog or low-cloud occurrence is infrequent. These factors lead to anomalies and local characteristics in the data, making such data inappropriate for statistically determining general and comprehensive parameters to infer whether clouds touch the surface. Therefore, we use MSM GPV data provided by JMA to ensure a sufficiently large dataset. The MSM GPV data include the initial and forecast values of fundamental meteorological parameters in the region from 22.4° to 47.6°N and from 120.0° to 150.0°E, at the surface and at several pressure levels. The spatial resolution of the data grid is 0.0625° longitude × 0.05° latitude for the surface and 0.125° longitude × 0.1° latitude for the pressure levels: this resolution is roughly comparable to the grid-arranged MTSAT-2 data. We used the pressure, temperature, and relative humidity at the surface and at pressure levels of 1000, 975, 950, 925, 900, 850, 800, and 700 hPa. Table 3 lists the resolution of the pressure levels in the GPV data.

Table 3.

Vertical resolution of the pressure levels in the JMA MSM GPV data.

Table 3.

For validation of the MSM GPV data, we estimated the bias and root-mean-square error (RMSE) between the GPV data and radiosonde measurement data. We use the radiosonde data at Hachijyo-jima (33.11°N, 139.78°E) and Chichi-jima (27.08°N, 142.18°E) for a month of summer (1–30 June 2012) and winter (15 February–15 March 2012). Because the radiosonde measurement is performed every 12 h, the number of compared data is 60 for a site. Table 4 lists the bias and RMSE for temperature and relative humidity at each pressure level, season sample, and site. This error analysis shows that the RMSE of temperature is not dependent on the altitude (pressure level), season, or site. On the other hand, the RMSE of humidity tends to increase with altitude. At Hachijyo-jima, the RMSE of humidity in the winter sample is larger than that in the summer sample for every pressure level. The overall bias and RMSE including all of the cases for temperature are 0.1 and 0.8 K, respectively, and those for humidity are 0% and 8%, respectively. It is suggested that we generally must use the GPV data in the lower atmosphere, considering the error of approximately ±1 K and ±10% for temperature (and also potential temperature) and humidity, respectively. For example, the equivalent potential temperature difference caused by a relative humidity change from 90% to 100% at 1000 hPa results in about 3 and 1 K for 290 and 270 K of the ambient temperature, respectively. We also performed the error analysis for the 6- and 12-h forecast values of the GPV data with the same data and procedure as was used for the initial values. This analysis showed that the forecast values, especially of humidity, have a larger error than the initial values for almost all of the cases (i.e., with a different season, a different pressure level, and a different radiosonde measurement site). We therefore use only the initial values in this study.

Table 4.

Bias and RMSE of temperature and humidity between radiosonde data and MSM GPV data at Chichi-jima (27.08°N, 142.18°E) and Hachijyo-jima (33.11°N, 139.78°E) in the summer and winter samples.

Table 4.

The relative humidity of the GPV data is applied to estimate the vertical location of cloud layers under the threshold that a relative humidity greater than 90% is considered to be in a cloud layer. Whether a cloud layer touches the surface is determined from the humidity at the surface by the same decision process. The GPV temperature of the cloud top determined from the humidity with the procedure mentioned above and the ch1 brightness temperature of pixels in the low-cloud datasets exhibits a strong linear correlation, although the ch1 brightness temperature is generally about 2–5 K higher than the determined cloud-top temperature because of molecular absorption. This comparison demonstrates the adequacy of the humidity of 90% for the threshold of cloud occurrence. The potential temperature and equivalent potential temperature are derived from the GPV values for each pressure level and for the surface to examine meteorological conditions.

Note that MTSAT-2 data are assimilated into the MSM GPV data for retrieval of only wind direction and water vapor amount above the middle atmosphere (Muroi et al. 2008). This means that parameters in the lower atmosphere of the MSM data are determined almost independently of the MTSAT-2 data. Therefore, the MSM GPV data are not likely to be correlated with the fog-detection results from MTSAT-2 data.

d. Data preparation for analysis

The MTSAT-2 data (including results for low-cloud detection) and the GPV data are merged to prepare datasets for statistical analysis. Because these analyses will lead to construction of a procedure for discriminating whether detected low clouds touch the surface after performing the low-cloud-detection scheme, which should be irrelevant to cirrus or higher clouds as mentioned above, the datasets for the analyses should consist of only those pixels containing low water clouds. First, we extracted pixels that contained low clouds, as determined by using both the humidity of the GPV data and the fog-detection result from MTSAT-2 data (here pixels with CCL < 0.5 were considered to contain low clouds). A comparison of low-cloud detection results with the Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) data indicated that it was sometimes difficult for MTSAT-2 with the algorithm explained above to estimate cloud-top height correctly, especially during daytime (Ishida et al. 2014). To perform a stricter exclusion of high clouds, we compared the ch1 brightness temperature of MTSAT-2 with the air temperature at 700 hPa of the GPV data, and we excluded pixels with a ch1 brightness temperature that was lower than the air temperature at 700 hPa. In addition, we intended to exclude ice clouds to focus on only water cloud. Ice fog tends to occur with a temperature below 243 K, but freezing fog sometimes occurs when temperature gradually decreases below 273 K (e.g., Gultepe et al. 2009). We therefore excluded pixels with a ch1 brightness temperature of lower than 273 K to conservatively exclude ice clouds. Multilayered low clouds, diagnosed from the humidity of the GPV data, were also excluded to simplify the analysis. The extracted pixels should ultimately be more likely than not to contain a water low-cloud layer only, which may or may not touch the surface. The process of the pixel selection is illustrated in Fig. 1. The GPV data are provided at 3-h intervals, and therefore eight merged scenes with the MTSAT-2 data can usually be obtained per day. We prepared the datasets for 1 month in every season for 2012: 15 February–15 March for winter, April for spring, June for summer, and September for autumn, referred to as “season samples.” The total numbers of pixels of low water cloud that is touching the surface (hereinafter referred to as “TS cloud”) and that is not touching the surface (“NTS cloud”) in the prepared datasets are listed in Table 5. The number of extracted pixels exhibits seasonal variability. In particular, the number of pixels of TS cloud in the winter sample is considerably smaller than not only the pixel count for the other season samples but also the pixel count of NTS cloud in the winter sample. There are, however, enough data to carry out statistical analysis.

Fig. 1.
Fig. 1.

Flowchart for the selection of pixels containing low water clouds for this study. Here, CCL is the clear confidence level, Tb(ch1) is the channel-1 brightness temperature, T(700 hPa) is the GPV temperature at 700 hPa, RH(any) is the relative humidity at an arbitrary pressure level (below 700 hPa), and RH(surface) is the relative humidity at the surface.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-13-0363.1

Table 5.

Total numbers of pixels of TS cloud and NTS cloud in the prepared datasets for each season sample.

Table 5.

We summarized the prepared datasets by calculating the incidences of low cloud that is touching and that is not touching the surface in each season, as depicted in Figs. 2 and 3, respectively. In these figures, for example, 0.1 means that cloud occurred for a total of 3 days during a month (i.e., 0.1 × 30 days) for each season sample at a location. These figures indicate that there were some regions where low cloud occurred more frequently, depending not only on the season but also on varying between TS and NTS clouds. The TS clouds occurred primarily in narrow regions, especially in the Pacific Ocean near Hokkaido, which experienced a much higher incidence of TS cloud than other areas for all of the season samples except the winter sample. The incidence of NTS cloud was more uniform among regions than the incidence of TS cloud, but there are some regions where NTS cloud occurred with a higher frequency. The north part of the domain apparently tends not to have low clouds in the winter sample; this may be due to the exclusion of low clouds with a cloud-top temperature lower than 273 K, however. We also examined the incidences of low cloud in nighttime and daytime, respectively (the results are not shown), but no distinct difference was found. Note that these frequencies are different from actual water low-cloud incidence even if the results of low-cloud detection with MTSAT-2 are correct. Low clouds under cirrus or high-cloud layers as well as multilayered low-cloud layers must be excluded from the frequency estimation. Low clouds occurring during dawn and dusk are also not included. In addition, low clouds with supercooled water particles, which are likely to have a cloud-top temperature that is lower than 273 K, are neglected. Nevertheless, the frequency distribution of TS clouds in the summer sample is qualitatively similar to the dense-fog frequency obtained from ship reports over a long period (Lewis et al. 2004). One of the distinct differences in the characteristics between TS and NTS cloud in the northern Pacific near Japan in summer is that the area of the highest incidence of TS clouds was located to the north of the area with the highest incidence of NTS clouds. This difference also qualitatively appeared in the “climatology” of the low-cloud frequencies in Norris (1998b), if we view “sky-obscuring fog” and “fair-weather stratus” as defined in Norris (1998b) as corresponding to TS cloud and NTS cloud, respectively.

Fig. 2.
Fig. 2.

Relative frequency of low cloud that is touching the surface (TS cloud) in the (a) spring, (b) summer, (c) autumn, and (d) winter samples. Values are the ratio of total days of low-cloud occurrence to all 30 days of data in each season sample.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-13-0363.1

Fig. 3.
Fig. 3.

As in Fig. 2, but for low cloud that is not touching the surface (NTS cloud).

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-13-0363.1

We checked the effects of the exclusion of multilayered low clouds on the analysis. If multilayered low clouds are predominant in nature, this exclusion would reduce the real-world applicability of this study. Figure 4 shows the relative frequencies of the number of cloud layers, as determined from the humidity of the GPV data, in the summer sample. Here, multilayer clouds with higher clouds (altitude higher than ~3000 m) or cirrus were previously excluded by the low-cloud detection with MTSAT-2 data, as mentioned above, and low clouds with a ch1 brightness temperature of lower than 273 K were included so as not to systematically ignore multilayered clouds with a lower cloud-top temperature. The relative frequencies of multilayered TS and NTS clouds in the summer sample were about 20% and 10%, respectively. The relative frequencies of multilayered clouds in the other season samples were almost the same, except in the winter sample for which the relative frequency of multilayered TS clouds reached about 30%. This suggests that single-layer cloud is dominant among low cloud, whether touching the surface or not. Note that the actual incidence of multilayered low clouds might be larger than this estimation even if the GPV data are accurate, because multilayered low clouds with a smaller interval than the vertical resolution of GPV data were neglected. This estimation is consistent, however, with observations of the incidences of multilayered low clouds by surface-based instruments in the northeastern Atlantic Ocean, where incidences during winter and summer were in the range from ~5% to ~20% (Wang et al. 1999). Therefore, these prepared datasets should be reasonably representative of low-cloud incidence and can be applied to examination of differences in conditions between TS and NTS clouds.

Fig. 4.
Fig. 4.

Relative frequency of the number of layers for (a) TS clouds and (b) NTS clouds in the summer sample.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-13-0363.1

3. Data analysis

a. Cloud-top height

First, we focus on the relationship between surface touching by low cloud and cloud-top height. We determine the cloud-top and cloud-bottom heights in the prepared datasets from the GPV humidity, as discussed above. If the pressure level just above a cloud layer (i.e., humidity > 90%) is clear (i.e., a humidity ≤ 90%), we take the mean of the heights of the two pressure levels as the cloud-top height. Figures 5 and 6 show the frequencies of the cloud-top height of TS and NTS clouds, respectively, in each season sample. The mean, standard deviation, median, and mode in each case are listed in Table 6. Note that, as listed in Table 2, the height resolution for the GPV data between 1000 and 2000 m (roughly corresponding to 900 and 800 hPa, respectively) is twofold lower than the resolution from 0 to 1000 m and that resolution between 2000 and 3000 m (roughly corresponding to 800 and 700 hPa, respectively) is fourfold lower than that between 0 and 1000 m. A bin for a height greater than 1000 m therefore tends to have a larger frequency (e.g., 2500–2750 m in Fig. 6b), balanced by another bin that has a correspondingly smaller frequency (e.g., 2000–2250 m in Fig. 6b). The mode value, especially when greater than 1000 m, is therefore less meaningful and is only for reference. The mean and the median of NTS clouds are about 2 and 3 times, respectively, those of TS clouds, except for the winter sample. The significance test (such as a t test) is not applicable, however, because the frequency distribution of the cloud-top height seems not to follow a normal distribution. Figure 5 reveals that the cumulative frequency of the cloud-top heights lower than 1000 m for TS clouds is about 70% in all seasons except the winter sample. On the other hand, Fig. 6 reveals that the cumulative frequencies of pixels of NTS clouds with cloud-top heights lower than 1000 m in the spring, summer, autumn, and winter sample are about 25%, 45%, 20%, and 10%, respectively. We estimated the cloud-top-height frequency by dividing the data into nighttime and daytime, but the tendencies of these frequency distributions are similar to each other except in the winter sample. These results suggest that the cloud-top height is related to whether low clouds touch the surface, with the relationship depending on season. This phenomenon (TS clouds with a lower cloud top) may be associated with mechanisms for fog formation and maintenance. Previous studies [e.g., summarized in Lewis et al. (2003) and Koračin et al. (2014)] indicated that fog occurrences are sometimes initiated by stratus lowering due to subsidence and that trapping the moisture in a shallower atmospheric layer just above the surface is important for fog maintenance.

Fig. 5.
Fig. 5.

Relative frequency of the top height of TS clouds in the (a) spring, (b) summer, (c) autumn, and (d) winter samples. The relative frequencies are derived through division by the total number of pixels of TS clouds in each season sample. The histogram bin width is 250 m.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-13-0363.1

Fig. 6.
Fig. 6.

As in Fig. 5, but for NTS clouds.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-13-0363.1

Table 6.

Mean, standard deviation, median, and mode of the cloud-top height of TS cloud and NTS cloud in each season sample. Mode values greater than 1000 m are only for reference.

Table 6.

NTS-cloud layers with a higher top can result from two phenomena: a thicker cloud layer with a low bottom and a higher cloud layer itself. The two-dimensional histograms of the thickness and the cloud-top height of NTS clouds in the summer and winter samples are shown in Fig. 7 to investigate which is more predominant. In the summer sample (Fig. 7a), the highest frequency of the cloud thickness for each cloud-top height usually occurred at the same altitude as or 250 m lower than the cloud-top height, and most cloud thickness frequencies fell within a narrow range. This means that during summer a higher cloud top often resulted from a thicker cloud with the bottom near the surface. In the winter sample (Fig. 7b), the highest frequency was not distinctly highlighted, and cloud thickness incidence was distributed widely from the smallest thickness (i.e., the bin width) to the largest thickness (i.e., cloud-top height). This means that both of the causes of a higher cloud top may be comparable in winter. The tendencies of the frequency distributions in the spring and autumn samples were qualitatively similar to those observed in the winter sample. We could make a hypothesis from these results: in summer the occurrence of NTS clouds may be mainly associated with cloud vertical development while maintaining a lower bottom height (but without touching the surface), whereas in the other seasons the occurrence of NTS clouds may be associated with cloud lifting as well as vertical development.

Fig. 7.
Fig. 7.

Two-dimensional histogram of NTS clouds for the cloud-top height and the cloud thickness in the (a) summer and (b) winter samples. Both histogram bin widths are 250 m.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-13-0363.1

b. Relation to the vertical structure of the potential temperature

We examined the relationship between surface touching of cloud and the vertical structure of the potential temperature. In this paper, we define a layer in which the vertical gradient Γ of the potential temperature, determined from the GPV data, is greater than the moist-adiabatic lapse rate as the “Γ > Γm layer”; this layer is expected to be stable and to suppress upward motions. When [θ(1000 hPa) − θ(surface)]/z(1000 hPa) is greater than the moist-adiabatic lapse rate, we consider that the Γ > Γm layer attaches to the surface, where θ(1000 hPa) and θ(surface) are the potential temperature at 1000 hPa and the surface, respectively, and z(1000 hPa) is the altitude of the 1000-hPa level. Whether or not low clouds touch the surface, more than 98% of low-cloud pixels are coincident with a Γ > Γm layer, except for NTS clouds in the autumn sample (about 89%) and TS clouds in the winter sample (about 92%). We examined the relationship between the cloud-top height and the bottom height of the Γ > Γm layer. Figure 8 displays the frequencies of cloud-top heights and the bottom height of the lowest Γ > Γm layer for TS and NTS clouds in the summer sample. For TS clouds (Fig. 8a), the bottom height of the Γ > Γm layer occurred in the range of 0–250 m for about 80% of pixels regardless of the cloud-top height, which means that a TS-cloud layer is more likely to be coincident with a very low or (almost) surface-attached Γ > Γm layer, and a low-cloud layer often penetrated into the Γ > Γm layer. This is consistent with observations about the occurrence of the surface-based inversion layer with fog (e.g., Norris 1998a; Lewis et al. 2003). On the other hand, the top of NTS clouds (Fig. 8b) was often coincident with the bottom of the Γ > Γm layer for cases of a cloud-top height lower than 1500 m. This is partly consistent with previous observations (e.g., Albrecht et al. 1988; Betts and Boers 1990; Norris 1998a), which indicated that the tops of low stratiform cloud usually occur near the base of the temperature inversion layer, but this study suggests that the bottom of the Γ > Γm layer may be more appropriate for the occurrence of the cloud top. A similar statistical tendency was found in the other seasons except the winter sample.

Fig. 8.
Fig. 8.

Two-dimensional histogram of (a) TS clouds and (b) NTS clouds for the cloud-top height and the bottom height of the Γ > Γm layer in the summer sample. Both histogram bin widths are 250 m.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-13-0363.1

Some previous studies suggested that the difference in temperature or the potential temperature (PT) across the bottom of the inversion layer, referred to as the inversion strength, can be related to whether a low cloud touches the surface. We defined a similar index value, the normalized PT gap, which is the vertical gradient of the potential temperature between just below and just above the bottom of the Γ > Γm layer, considering that the interval of the altitude in the GPV data is not constant. When the Γ > Γm layer attached to the surface, [θ(1000 hPa) − θ(surface)]/z(1000 hPa) was considered to be the normalized PT gap. Figure 9 shows the frequencies of the normalized PT gap for TS and NTS cloud in the spring sample to investigate the relation between the gap of potential temperature across the Γ > Γm layer and whether low cloud touches the surface. The highest frequency occurred at 0.006–0.008 K m−1 for both conditions, but the distribution tendency was different. For NTS cloud (Fig. 9b), the highest frequency was greater than for TS cloud (Fig. 9a) and the frequency was distributed across a narrower range, reaching a maximum value of 0.022 K m−1. In contrast, the frequency for the case of TS clouds was widely distributed, reaching a maximum value of 0.06 K m−1. This indicates that TS clouds tend to have a slightly greater gap of the potential temperature than NTS clouds do. The same statistical tendency was found in the other season samples, appearing in the spring sample more clearly. This is consistent with a simulation study of Koračin et al. (2001), which suggested that a stronger inversion strength promotes cloud lowering (but also reduces the duration of fog). In the winter sample, however, the maximum of the normalized PT gap for TS cloud was only 0.01 greater than that for NTS cloud. These results would have suggested that the mechanism for formation and maintenance of TS cloud in winter might differ from that in the other seasons in the study area.

Fig. 9.
Fig. 9.

Relative frequency of the normalized PT gap across the bottom of the Γ > Γm layer for (a) TS clouds and (b) NTS clouds in the spring sample. If the Γ > Γm layer attaches to the surface, the normalized PT gap corresponds to the vertical gradient of the potential temperature at the layer just above the surface. Histogram bin width is 0.002 K m−1.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-13-0363.1

c. Relation to the vertical structure of the equivalent potential temperature

In this section, we focus on the vertical profile of the equivalent potential temperature θe, because θe is a suitable representation of the vertical structure in a wet atmosphere with clouds. Several studies (e.g., Betts et al. 1995; Kodama 1997; Norris 1998a) observed that low cloud tends to be coincident with a layer in which the vertical gradient Ge of θe is positive (hereinafter referred to as the “Ge > 0 layer”), which sometimes overlies the cloud layer and sometimes (partly) embeds it. A Ge > 0 layer sometimes attaches to the surface and sometimes overlaps a layer with a negative vertical gradient of θe (hereinafter referred to as the “Ge < 0 layer”) or approximately 0 (“Ge = 0 layer”). A hypothetical profile of θe in the lower atmosphere for a complicated case is illustrated in Fig. 10a. We estimated the statistics of not only the existence of the Ge > 0 layer in the lower atmosphere but also the structure under and above the Ge > 0 layer. Although the error of θe should vary according to the ambient temperature and humidity, a typical error of the vertical gradient can be ±0.001–0.003 K m−1, because a typical error of θe in a moist atmosphere can be about ±1–3 K, as mentioned above, and the largest interval of the altitude is about 1000 m (corresponding to the pressure level between 800 and 700 hPa). In the estimation, we therefore consider a layer with Ge between 0.002 and −0.002 K m−1 as a Ge = 0 layer, and the altitude of the lowest level with Ge greater than 0.002 K m−1 is defined as the bottom of the lowest Ge > 0 layer. In a similar way, the altitude of the lowest level with Ge less than −0.002 K m−1 and above the bottom of the lowest Ge > 0 layer is defined as the top of the layer. If the bottom of a Ge > 0 layer occurs below the 700-hPa level, we say that a Ge > 0 layer exists in the lower atmosphere (Fig. 10b; see Fig. 10c for the contrasting condition). If the top of the lowest Ge > 0 layer occurs below the 700-hPa level, we say that a Ge < 0 layer exists between the 700-hPa level and the top of the lowest Ge > 0 layer (Fig. 10d; see Fig. 10e for the contrasting condition). In addition, if the bottom of the lowest Ge > 0 layer occurs apart from the surface, we say that the Ge < 0 or Ge = 0 layer attaching to the surface exists under the lowest Ge > 0 layer (Fig. 10f), whereas if the bottom of the lowest Ge > 0 layer occurs just above the surface (i.e., attaches to the surface), we say that the Ge < 0 or Ge = 0 layer does not exist under the lowest Ge > 0 layer (Fig. 10g). Note that this estimation implicitly neglects a finer vertical structure than the vertical resolution of the GPV data, especially above 900 hPa.

Fig. 10.
Fig. 10.

(a) Schematic of a hypothetical vertical structure of θe and a corresponding ideal cloud layer. Also shown are the θe profiles for categorization as listed in Table 7: (b) a Ge > 0 layer exists in the lower atmosphere or (c) it does not; when a Ge > 0 layer exists, the Ge > 0 layer is included within the lower atmosphere and (d) a Ge < 0 layer exists between the 700-hPa level and the lowest Ge > 0 layer or (e) it does not; (f) a Ge ≤ 0 layer exists under the lowest Ge > 0 layer or (g) it does not.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-13-0363.1

Table 7 lists the relative frequencies for TS and NTS clouds in each season sample. The statistics show that a Ge > 0 layer often existed in the lower atmosphere, with a frequency higher than 80% except for the NTS-cloud case in the autumn sample and the TS-cloud case in the winter sample. The incidence of Ge > 0 layers for TS clouds was usually greater than for NTS clouds in every season sample except winter. For TS clouds, a Ge < 0 layer existed between 700 hPa and the Ge > 0 layer with a frequency from 68% to 78%, except in the winter sample. The corresponding frequency for NTS clouds was less than 50% in every season sample. These statistics imply that a low-cloud layer is usually coincident with a Ge > 0 layer and that a TS-cloud layer is more likely to be coincident with a Ge < 0 layer above the lowest Ge > 0 layer than an NTS low-cloud layer is. One of the mechanisms leading to such vertical structures of θe may be subsidence of dry air, which is expected not only to decrease water vapor amount and hence θe in the lower atmosphere above the cloud layer but also to force the cloud layer to descend and eventually attach to the surface (e.g., Koračin et al. 2001; Lewis et al. 2003). For TS clouds in winter, the occurrence of this vertical structure was relatively infrequent, implying the prevalence of other mechanisms to produce TS clouds.

Table 7.

Incidence of a Ge > 0 layer in the lower atmosphere (i.e., below the 700-hPa level), as illustrated in Fig. 10b, a Ge < 0 layer between the 700-hPa level and the top of the lowest Ge > 0 layer (Fig. 10d), and a Ge ≤ 0 layer under the lowest Ge > 0 layer (Fig. 10f). Determination is based on the equivalent potential temperature derived from GPV data for the TS-cloud and NTS-cloud cases in each season sample. The relative frequencies are derived through division by the total number of pixels for each case.

Table 7.

The incidence of a Ge ≤ 0 layer under the lowest Ge > 0 layer for TS clouds was less than 50% and was less than that for NTS clouds in every season sample. When the Ge ≤ 0 layer existed under the lowest Ge > 0 layer, we also estimated the normalized difference in θe between the bottom of the lowest Ge > 0 layer and the surface; that is, [θe(bottom) − θe(surface)]/h(bottom), where θe(bottom) and θe(surface) are the equivalent potential temperature at the bottom of the lowest Ge > 0 layer and at the surface, respectively, and h(bottom) is the bottom height of the lowest Ge > 0 layer. This value is expected to represent the degree of mixing under the lowest Ge > 0 layer. Figure 11 shows the normalized difference in θe for the TS-and NTS-cloud cases in the spring sample. This value was less than 0.002 K m−1 in both the TS- and NTS-cloud cases. The normalized difference in θe for NTS clouds slightly tended to be more negative than that for TS clouds: the incidence of the normalized difference in θe greater than −0.006 K m−1 for TS clouds was about 60% (Fig. 11a), whereas that for NTS clouds was about 28% (Fig. 11b). This means that the difference in θe between the bottom of the lowest Ge > 0 layer and the surface tends to be enhanced with an NTS-cloud layer. In contrast, θe under the lowest Ge > 0 layer with a TS cloud tends to be relatively vertically constant. This tendency might be related to turbulent transfer below a cloud layer. Some simulation studies (e.g., Rogers and Koračin 1992) suggested that less turbulence flux under a cloud layer reduces transfer of water vapor and heat, resulting in a vertically inhomogeneous θe and suppressing condensation at the cloud bottom. This tendency was also found in the summer and autumn samples. The statistical analyses also indicate that the difference between the vertical structure of θe for TS and NTS clouds is not distinct, however; that is, a particular vertical structure of θe can occur with either a TS cloud or an NTS cloud.

Fig. 11.
Fig. 11.

Relative frequency of the normalized difference in θe between the bottom of the lowest Ge > 0 layer and the surface for (a) TS clouds and (b) NTS clouds in the spring sample. The relative frequencies are derived through division by the number of pixels that contain the Ge ≤ 0 layer under the lowest Ge > 0 layer for each case. The histogram bin width is 0.002 K m−1.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-13-0363.1

d. Differences by season

The analysis indicates that the predominant conditions coincident with TS or NTS clouds differed slightly but significantly by season. In particular, TS clouds in the winter sample are distinctive in the following ways. They did not tend to have a lower cloud-top height, but the distribution of the cloud-top height frequency was mostly uniform: the frequencies of the bottom height of the Γ > Γm layer in the range of 0–250 m at each cloud-top height were comparable to those near the cloud-top height: the incidence of a Ge > 0 layer within the lower atmosphere was small in comparison with other seasons. We analyzed the frequency of the cloud-top height of TS clouds by dividing the prepared datasets into nighttime and daytime. The highest frequency of cloud-top heights for the nighttime data in the winter sample occurred at the bin for 250–500 m and the incidence of cloud-top heights of less than 1000 m was ~48%, whereas the frequency of cloud-top heights for the daytime data in the winter sample was distributed almost uniformly and the incidence of cloud-top heights lower than 1000 m was about 37%, which means that the distribution of cloud-top heights in daytime was more different from the typical distribution (i.e., higher incidences for lower cloud-top heights) than was that during nighttime. It might be hypothesized from these analyses that during winter in the study area even a thicker atmospheric layer between the surface and a stable layer can accumulate sufficient moisture for the formation or maintenance of TS clouds through some mechanisms, which may be more effective during the daytime in winter, although more observation or simulation is required to prove this hypothesis. Therefore, it is better that a criterion for distinguishing between TS and NTS clouds is defined for each season.

4. Applicability to fog detection by satellite

The statistical analyses in section 3 suggest that the cloud-top height and the structure of the Γ > Γm layer are broadly related to whether a low cloud touches the surface and may be incorporated in fog detection by satellite. In particular, the temperature difference between the cloud-top height and the sea surface temperature (SST) is expected to be applicable to low-cloud-detection schemes, because not only a lower cloud-top height but also a temperature lapse rate in the Γ > Γm layer, in which the top of TS clouds tends to exist, may make the temperature difference (here defined as SST minus cloud-top temperature) for TS clouds lower than that for NTS clouds. In addition, it is relatively easy to estimate the cloud-top temperature from weather-satellite data, because the cloud-top temperature roughly coincides with the brightness temperature of a channel in the window region. If the SST can be obtained from some ancillary data, the temperature difference between the surface and the cloud top can be estimated. Although it is difficult to measure SST through cloud layers with a geostationary weather satellite, the range of SST fluctuations over a few hours is generally narrow, and forecast data can be used to assist in our purpose. SST itself is also strongly related to low-cloud occurrence. Many past observations revealed that both warmer and colder SST than the air temperature just above the surface can initiate fog formation (e.g., Lewis et al. 2004; Koračin et al. 2014). Cold sea fog may result in a lower or sometimes negative temperature difference, whereas warm sea fog may increase the temperature difference and cancel the feature of TS clouds. It is therefore necessary to investigate whether the temperature differences of TS cloud and NTS cloud are distinct from each other.

Figure 12 shows the distribution of differences between the forecast SST Tsfc and the ch1 brightness temperature Tb(ch1) of MTSAT-2 for the cases of TS and NTS clouds in the autumn sample. It reveals that the distributions of the temperature-difference frequency for the two cases seemed to follow a normal distribution but have individual mean and standard deviations. This tendency is also found in the samples for other seasons. A few pixels had a negative value that was due in part to a strong inversion or a colder SST. Table 8 summarizes the means, the medians, the modes, and the standard deviations of TsfcTb(ch1) of TS and NTS clouds in each season sample. For all of the season samples, the standard deviation of TS clouds is near that of NTS clouds, but the two means are different by a factor of ~2 except for the winter sample. The results of significance testing by t test (not presented) indicate that the means of TS and NTS cloud are different with sufficient confidence. This result suggests that the forecast SST is appropriate as a reference for estimation of the cloud-top height and that TsfcTb(ch1) is a suitable index value for discriminating between TS and NTS clouds.

Fig. 12.
Fig. 12.

Relative frequency of the difference between the forecast Tsfc and Tb(ch1) for (a) TS cloud and (b) NTS cloud in the autumn sample. The histogram bin width is 1 K.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-13-0363.1

Table 8.

The means, medians, modes, and standard deviations of TsfcTb(ch1) of MTSAT-2 of TS cloud and NTS cloud in each season sample. The threshold value and Pr derived with the threshold values to discriminate TS and NTS cloud are also shown.

Table 8.

We tried to determine a threshold value for actually distinguishing between TS and NTS clouds. As illustrated in Fig. 12, the temperature-difference frequency for NTS clouds in the autumn sample become greater than that for TS clouds at 8 K, which would be appropriate for the threshold to distinguish between TS and NTS clouds. In the case of TS clouds, the incidence of temperature difference smaller than 8 K was about 76% (Fig. 12a). In the case of NTS clouds, the incidence of temperature difference smaller than 8 K was about 14% (Fig. 12b). This suggests that if we want to carry out fog detection (ignoring multilayered clouds) with MTSAT-2 then with application of a temperature-difference threshold of 8 K ~76% of such clouds would have been correctly detected as “fog.” The detected TS clouds, however, would include incorrect pixels, which are provided from about 14% of NTS-cloud pixels. This incorrect detection may be due to a higher cloud top or a warmer SST. This method is applicable throughout the year, but the threshold value for surface-touching detection must be determined according to season, as listed in Table 8. For example, the incidence of temperature difference smaller than 7 K in the summer sample was ~74% for TS clouds and ~18% for NTS clouds, indicating that a summer threshold of 7 K is comparable to the autumn threshold of 8 K. This result is consistent with the results in the previous section that the cloud-top heights for NTS clouds in the summer sample tended to be lower than those in the other season samples. A reasonable threshold for discrimination in winter is 12 K, although the accuracy of the method will be worse than for the other seasons because differences in the distributions of TS and NTS clouds in the winter sample are less distinct than those for the other seasons.

Table 8 also lists the cumulative probability Pr with the assumption that the probability of TsfcTb(ch1) in each season obeys a normal distribution with the obtained sample mean and standard deviation, defined as
e1a
e1b
for TS and NTS clouds, respectively, where x = TsfcTb(ch1) and xd is the threshold value. This indicates that about 70%–80% of low clouds might be correctly distinguished, regardless of whether they touch the surface. The Pr of NTS clouds is larger than that of TS clouds except for winter, suggesting that NTS clouds are distinguished slightly more correctly than are TS clouds. It is also suggested, as mentioned above, that distinguishing NTS clouds in winter would be more difficult than for the other seasons.

The difference of the vertical structure of θe between TS and NTS clouds might also be applicable to fog detection by satellite. As mentioned in the previous section, TS clouds tend to be coincident with the Ge < 0 layer between the cloud layer and 700 hPa, which may result from subsidence of dry air. A possibility of the application of this condition is to estimate the amount of water vapor above the cloud layer by satellite, as indicated by Lewis et al. (2003). Retrieval of the amount of water vapor only above a cloud layer with passive sensors is more difficult than is estimation of the cloud-top temperature. Furthermore, the vertical structures of θe for TS and NTS clouds do not distinctly differ from each other. More investigation is therefore required for the application of the vertical structure of θe.

5. Summary and conclusions

Although geostationary weather satellites are useful for continuous near-real-time fog monitoring, it is difficult for passive sensors to detect whether a low cloud touches the surface, which is critical information for human activities. To find parameters that enable simple and convenient discrimination between low clouds that touch the surface and those that do not, particularly at sea, we statistically examined meteorological GPV data provided from MSM by combining low-cloud-detection results with MTSAT-2 data. These datasets allowed obtainment of general and comprehensive understanding of the relationship between meteorological conditions and low-cloud occurrence and mitigated the uncertainty that is due to both the scarcity of observation sites at sea and the low incidence of fog. We carefully selected the MTSAT-2 data pixels that contained only a single low-cloud layer, using the GPV and fog-detection results from MTSAT-2 data to exclude cirrus or other high clouds, ice fogs, or multilayered clouds.

The statistical analyses revealed the following tendencies of the meteorological conditions.

  1. Low clouds that are touching the surface (TS clouds) tend to have a lower cloud-top height than low clouds that are not touching the surface (NTS clouds); this is true in all seasons, although the distribution of the occurrence frequency of the cloud-top height differed by season. This result indicates that cloud-top height is related to whether a low cloud touches the surface, although the quantitative relationship may depend on season.
  2. The Γ > Γm layer (i.e., a layer with a vertical gradient of the potential temperature that is greater than the moist-adiabatic lapse rate) with a TS cloud tends to be very low or almost attached to the surface. TS cloud also tends to have a slightly greater gap of the potential temperature across the bottom of the Γ > Γm layer than NTS cloud does. On the other hand, the top of NTS clouds tends to occur near the bottom of the Γ > Γm layer, especially when the cloud-top height is below 1500 m.

The reasons why these conditions are related to touching or not touching the surface can be partially inferred from the vertical profile of the equivalent potential temperature θe. A Ge > 0 layer (where Ge is the vertical gradient of θe) often exists with a low-cloud layer. The TS clouds are more likely to be coincident with a Ge < 0 layer above the Ge > 0 layer than are NTS clouds. One presumed mechanism for forming such a vertical structure of θe is subsidence of dry air, which should not only force the water vapor amount above the Ge > 0 layer to decrease but also should attach the cloud layer to the surface. In addition, the Ge ≤ 0 layer sometimes exists below the lowest Ge > 0 layer when a low cloud occurs. The Ge ≤ 0 layer below the lowest Ge > 0 layer with a TS cloud tends to be more vertically homogeneous than that with an NTS cloud, implying that the magnitude of turbulent transfer under the cloud layer may be related to whether a cloud touches the surface.

This study suggests that a small addition of ancillary meteorological data allows for inferring whether low clouds touch the surface by satellite. For example, we indicated that the temperature difference between the forecast SST and the brightness temperature at the window region is an effective and convenient parameter to discriminate whether low clouds touch the surface. We also determined a threshold value for the discrimination from the statistical analysis, revealing that the threshold value depends on season. For future work, this scheme should be used to conduct fog detection on a large set of satellite data and to validate detection by comparison with other observations, such as radiosonde-profile data.

Acknowledgments

This work was partly supported by the Nippon Foundation of Tokyo, Japan, and by the Japan Weather Association (JWA) of Tokyo. The MTSAT-2 data were provided by JWA. Japan Meteorological Agency GPV data were provided by the GPV/JMA Archive of the Center for Computational Science, Tsukuba University. We thank Mr. Ryoji Nagasawa of JMA for providing important information about the JMA mesoscale model and the GPV data.

REFERENCES

  • Ahn, M.-H., , E.-H. Shon, , and B.-J. Hwang, 2003: A new algorithm for sea fog/stratus detection using GMS-5 IR data. Adv. Atmos. Sci., 20, 899913, doi:10.1007/BF02915513.

    • Search Google Scholar
    • Export Citation
  • Albrecht, B. A., , D. A. Randall, , and S. Nicholls, 1988: Observations of marine stratocumulus cloud during FIRE. Bull. Amer. Meteor. Soc., 69, 618626, doi:10.1175/1520-0477(1988)069<0618:OOMSCD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Betts, A. K., , and R. Boers, 1990: A cloudiness transition in a marine boundary layer. J. Atmos. Sci., 47, 14801497, doi:10.1175/1520-0469(1990)047<1480:ACTIAM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Betts, A. K., , C. S. Bretherton, , and E. Kliner, 1995: Relation between boundary-layer structure and cloudiness at the R/V Valdivia during ASTEX. J. Atmos. Sci., 52, 27522762, doi:10.1175/1520-0469(1995)052<2752:RBMBLS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Cermak, J., , and J. Bendix, 2008: A novel approach to fog/low stratus detection using Meteosat 8 data. Atmos. Res., 87, 279292, doi:10.1016/j.atmosres.2007.11.009.

    • Search Google Scholar
    • Export Citation
  • Ellrod, G. P., 1995: Advances in the detection and analysis of fog at night using GOES multispectral infrared imagery. Wea. Forecasting, 10, 606619, doi:10.1175/1520-0434(1995)010<0606:AITDAA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Ellrod, G. P., , and I. Gultepe, 2007: Inferring low cloud base heights at night for aviation using satellite infrared and surface temperature data. Pure Appl. Geophys., 164, 11931205, doi:10.1007/s00024-007-0214-7.

    • Search Google Scholar
    • Export Citation
  • Gultepe, I., , M. Pagowski, , and J. Reid, 2007a: A satellite-based fog detection scheme using screen air temperature. Wea. Forecasting, 22, 444456, doi:10.1175/WAF1011.1.

    • Search Google Scholar
    • Export Citation
  • Gultepe, I., and Coauthors, 2007b: Fog research: A review of past achievements and future perspectives. Pure Appl. Geophys., 164, 11211159, doi:10.1007/s00024-007-0211-x.

    • Search Google Scholar
    • Export Citation
  • Gultepe, I., and Coauthors, 2009: The Fog Remote Sensing and Modeling field project. Bull. Amer. Meteor. Soc., 90, 341359, doi:10.1175/2008BAMS2354.1.

    • Search Google Scholar
    • Export Citation
  • Ishida, H., , and T. Y. Nakajima, 2009: Development of an unbiased cloud detection algorithm for a spaceborne multispectral imager. J. Geophys. Res., 114, D07206, doi:10.1029/2008JD010710.

    • Search Google Scholar
    • Export Citation
  • Ishida, H., , K. Miura, , T. Matsuda, , K. Ogawara, , A. Goto, , K. Matsuura, , Y. Sato, , and T. Y. Nakajima, 2014: Scheme for detection of low clouds from geostationary weather satellite imagery. Atmos. Res., 143, 250264, doi:10.1016/j.atmosres.2014.02.015.

    • Search Google Scholar
    • Export Citation
  • Kodama, Y., 1997: Airmass transformation of the yamase air-flow in the summer of 1993. J. Meteor. Soc. Japan, 75, 737751. [Available online at http://www.st.hirosaki-u.ac.jp/~kodama/JMSJkodama1997.pdf.]

    • Search Google Scholar
    • Export Citation
  • Kodama, Y., , Y. Tomita, , and S. Asano, 2009: Air mass transformation along trajectories of airflow and its relation to vertical structures of the maritime atmosphere and clouds in yamase events. J. Meteor. Soc. Japan, 87, 665685, doi:10.2151/jmsj.87.665.

    • Search Google Scholar
    • Export Citation
  • Koračin, D., , J. M. Lewis, , W. T. Thompson, , C. E. Dorman, , and J. A. Businger, 2001: Transition of stratus into fog along the California coast: Observations and modeling. J. Atmos. Sci., 58, 17141731, doi:10.1175/1520-0469(2001)058<1714:TOSIFA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Koračin, D., , C. E. Dorman, , J. M. Lewis, , J. G. Hudson, , E. M. Wilcox, , and A. Torregrosa, 2014: Marine fog: A review. Atmos. Res., 143, 142175, doi:10.1016/j.atmosres.2013.12.012.

    • Search Google Scholar
    • Export Citation
  • Lee, T. F., , F. J. Turk, , and K. Richardson, 1997: Stratus and fog products using GOES-89 3.9-μm data. Wea. Forecasting, 12, 664677, doi:10.1175/1520-0434(1997)012<0664:SAFPUG>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Lewis, J., , D. Koracin, , R. Rabin, , and J. Businger, 2003: Sea fog off the California coast: Viewed in the context of transient weather systems. J. Geophys. Res., 108, 4457, doi:10.1029/2002JD002833.

    • Search Google Scholar
    • Export Citation
  • Lewis, J., , D. Koracin, , and K. T. Redmond, 2004: Sea fog research in the United Kingdom and United States: A historical essay including outlook. Bull. Amer. Meteor. Soc., 85, 395408, doi:10.1175/BAMS-85-3-395.

    • Search Google Scholar
    • Export Citation
  • Muroi, C., , T. Fujita, , and Y. Ishikawa, 2008: Hourly analysis at the Japan Meteorological Agency (in Japanese).Tenki, 55, 401408.

  • Norris, J. R., 1998a: Low cloud type over the ocean from surface observations. Part I: Relationship to surface meteorology and the vertical distribution of temperature and moisture. J. Climate, 11, 369382, doi:10.1175/1520-0442(1998)011<0369:LCTOTO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Norris, J. R., 1998b: Low cloud type over the ocean from surface observations. Part II: Geographical and seasonal variations. J. Climate, 11, 383403, doi:10.1175/1520-0442(1998)011<0383:LCTOTO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Norris, J. R., , Y. Zhang, , and J. M. Wallace, 1998: Role of low clouds in summertime atmosphere–ocean interactions over the North Pacific. J. Climate, 11, 24822490, doi:10.1175/1520-0442(1998)011<2482:ROLCIS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Randall, D. A., , J. A. Coakley Jr., , C. W. Fairall, , R. A. Kropfli, , and D. H. Lenschow, 1984: Outlook for research on subtropical marine stratiform clouds. Bull. Amer. Meteor. Soc., 65, 12901301, doi:10.1175/1520-0477(1984)065<1290:OFROSM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Rogers, D. P., , and D. Koračin, 1992: Radiative transfer and turbulence in the cloud-topped marine atmospheric boundary layer. J. Atmos. Sci., 49, 14731486, doi:10.1175/1520-0469(1992)049<1473:RTATIT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Saito, K., , T. Fujita, , J. Ishida, , Y. Kumagai, , K. Aranami, , S. Ohmori, , R. Nagasawa, , and S. Kumagai, 2006: The operational JMA nonhydrostatic mesoscale model. Mon. Wea. Rev., 134, 12661298, doi:10.1175/MWR3120.1.

    • Search Google Scholar
    • Export Citation
  • Saito, K., , J. Ishida, , K. Aranami, , T. Hara, , T. Segawa, , M. Narita, , and Y. Honda, 2007: Nonhydrostatic atmospheric models and operational development at JMA. J. Meteor. Soc. Japan, 85B, 271304, doi:10.2151/jmsj.85B.271.

    • Search Google Scholar
    • Export Citation
  • Underwood, S. J., , G. P. Ellrod, , and A. L. Kuhnert, 2004: A multiple-case analysis of nocturnal radiation-fog development off California utilizing GOES nighttime fog product. J. Appl. Meteor., 43, 297311, doi:10.1175/1520-0450(2004)043<0297:AMAONR>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Wang, J., , W. B. Rossow, , T. Uttal, , and M. Rozendaal, 1999: Variability of cloud vertical structure during ASTEX observed from a combination of rawinsonde, radar, ceilometer, and satellite. Mon. Wea. Rev., 127, 24842502, doi:10.1175/1520-0493(1999)127<2484:VOCVSD>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
Save