Observed response of marine boundary layer cloud to the interannual variations of summertime Oyashio extension SST front

Abstract

Active roles of both sea surface temperature (SST) and its frontal characteristics to the atmosphere in the mid-latitudes have been investigated around the western boundary current regions, and most studies have focused on winter season. The present study investigated the influence of the variation of the summertime Oyashio extension SST front (SSTF) in modulating low-level cloud properties (i.e., low-level cloud cover [LCC], cloud optical thickness [COT], and shortwave cloud radiative effect [SWCRE]) on inter-annual timescales, based on available satellite and Argo float datasets during 2003–2016. First, we examined the mechanism of summertime SSTF variability itself. The strength of the SSTF (S_SSTF), defined as the maximum horizontal gradient of SST, has clear inter-annual variations. Frontogenesis equation analysis and regression analysis for subsurface temperature indicated that the inter-annual variations of the summertime S_SSTF in the western North Pacific are closely related to the variations of not surface heat flux, but western boundary currents, particularly the Oyashio Extensions. The response of low-level cloud to intensified S_SSTF is that negative SWCRE with positive COT anomaly in the northern flank of the SSTF can be induced by cold SST anomalies. The spatial scale of the low-level cloud response was larger than the SST frontal scale, and the spatial distribution of the response was mainly constrained by the pathways of Kuroshio and Oyashio Extensions. Multi-linear regression analysis revealed that the local SST anomaly played largest role in modulating the SWCRE and COT anomalies among the cloud controlling factors (e.g., estimated inversion strength, air-temperature advection) accounting for more than 50% of the variation. This study provides an observational evidence of the active role of local SST anomalies in summertime associated with the western boundary currents to the oceanic low-level cloud.

1 Introduction

Marine boundary layer (MBL) clouds (e.g., fog, stratus, and stratocumulus), referred to as low-level clouds, tend to occur near the sea surface in the lower troposphere, whose top height is generally lower than 3 km in altitude. They interact strongly with sea surface temperature (SST) via two-way physical processes. Due to their high albedo, they reflect large amounts of shortwave (SW) radiation from the sun back to space, and therefore have a strong cooling effect on the surface of the land and ocean (Hartmann and Short 1980). On the other hand, low-level cloud properties are modulated by SST anomalies through marine atmospheric boundary layer processes. For example, low-level cloud cover (LCC) increases with decreasing SST due to the suppression of the entrainment of dry air from the free troposphere in the boundary layer (Bretherton et al. 2013; Qu et al. 2015; Kawai et al. 2017). Thus, more humidity can be trapped within the boundary layer if this entrainment is weakened, and LCC increases with decreasing SST itself. As a result, a positive feedback loop exists between variations in both LCC and SST in the North Pacific (NP), especially in summertime (e.g. Norris and Leovy 1994). In the boreal summer (June, July, and August [JJA]) climatology, the LCC and cloud optical thickness (COT) are large over the NP poleward from 40° N and over the south-east of the NP near the Californian coast where the SST is low (contours in Fig. 1a, shaded area and contours in Fig. 1b). In particular, LCC and COT in the NP drastically increase poleward of the SST frontal region around 40° N where the horizontal gradient of SST (∇SST) is larger than 1 °C/100 km (shaded area in Fig. 1a) known as SST front (SSTF). Although a clear correlation has been observed between low-level cloud properties and SST on inter-annual timescales (e.g. Norris and Leovy 1994; Kubar et al. 2012; Koshiro et al. 2017; Myers et al. 2018), their causal relationship is still under debate; this is due to the complex processes in low-level cloud evolution and the difficulty in distinguishing the two-way process within the LCC–SST feedback loop based on casual observational data analysis.

Fig. 1

Climatological conditions in the North Pacific in summertime (June, July, and August [JJA]) from 2003 to 2016. a Sea surface temperature (SST; contours, unit: °C, contour interval [CI]: 2 K) and its horizontal gradient (∇SST; colored areas, unit: °C/100 km) from OISST dataset. b Cloud optical thickness (COT) of low-level cloud (colored areas, unit: none) and low-level cloud cover (LCC) from a MODIS dataset (contours, unit: %, CI: 5%). Red line indicates the climatological mean position of SST front. Climatological SST front properties in JJA: c position (P_SSTF, unit: °N) and d strength of the SST front (S_SSTF, unit: °C/100 km). Error bars in (c, d) indicate the inter-annual variability

Previous studies have generally found that the mid-latitude ocean plays a passive role in the air-sea interaction process, meaning that SST tends to be forced by the atmospheric variability. In particular, in the summertime mid-latitudes, the mixed-layer depth is so shallow that the SST is strongly forced by surface heat flux variability (e.g. Tanimoto et al. 2003; Cronin et al. 2013). However, recent studies have revealed the active role of the ocean around the western boundary current regions (i.e., the Kuroshio Extension and Gulf Stream) using high-resolution satellite datasets and coupled ocean–atmosphere models (e.g. Nonaka and Xie 2003; Xie 2004; Kwon et al. 2010; Masunaga et al. 2015), meaning that SST can force the atmospheric variability. One of key elements in the SST impacts on the atmosphere is the large ∇SST formed in the confluence zone of the warm and cold western boundary currents. Two mechanisms have been proposed and widely accepted to explain the influences of SST front on the atmosphere. One is the vertical mixing mechanism (Hayes et al. 1989; Wallace et al. 1989; Chelton 2004). Near sea surface atmosphere destabilizes over the warm flank of the SST front, resulting in promoted vertical momentum mixing and enhanced near surface wind speed. The other is the pressure adjustment mechanism (Lindzen and Nigam 1987; Shimada and Minobe 2011). Surface wind convergence (divergence) occurs over the warm (cold) flank due to high- and low-pressure anomalies generated over the low and high SST anomalies across the front, respectively. These impacts are seasonally dependent (e.g. Minobe et al. 2010) and it is known that heat and moisture supplies are enhanced the most in wintertime because the difference between SST and overlaying atmospheric temperature is maximized (Nakamura and Yamagata 1999; Tanimoto et al. 2003; Tomita et al. 2011).

Some studies have investigated the summertime imprint of the SST front on the atmospheric boundary layer (Tanimoto et al. 2009; Tokinaga et al. 2009; Liu et al. 2014; Kawai et al. 2015, 2019), convective cloud and precipitation (Minobe and Takebayashi 2015; Sasaki and Yamada 2018), and basin-scale atmospheric circulation (Kwon et al. 2010; Frankignoul et al. 2011; Okajima et al. 2014; O’Reilly and Czaja 2015; Yao et al. 2017). For example, Tokinaga et al. (2009) showed seasonal differences in the impact of the Kuroshio Extension on the vertical development of clouds using both satellite observations and ship-based measurements as a climatological mean, revealing the enhancement of marine fog over the cold flank of the Kuroshio Extension SST front. However, previous studies only focused on the mean state of the summertime low-level cloud properties, paying much less attention to their inter-annual variability. In addition, they also tend to emphasize the importance of warm SST anomalies in wintertime, while cold SST anomalies in summertime can also be important to form the low-level cloud in the North Pacific via modulation of lower atmospheric stability. Kawai et al. (2019) and Jiang et al. (2019) provided the importance of cold SST anomalies to modulate overlaying atmosphere via the pressure adjustment mechanism using ship-based observations and regional atmospheric numerical simulation, suggesting that cold SST anomalies can significantly influence the marine boundary layer clouds in the western North Pacific.

Oceanic dynamics have the potential to drive both the climatological mean state and the inter-annual variation of low-level clouds in summertime through ocean-induced SST anomalies. Recently, Hosoda et al. (2016) investigated the inter-annual variation of the summertime oceanic subsurface temperature in the western, central, and eastern NP based on Argo float observations. They observed a large temperature variability both in the near-surface and in the subsurface up to 600 m depth in the western NP, that is affected by western boundary currents (i.e., the Kuroshio and Oyashio Currents). The findings of Hosoda et al. (2016) also imply the importance of oceanic dynamics for summertime SST despite the strong atmospheric forcing: this motivates us to investigate not only SST but also subsurface oceanic temperature simultaneously in order to reveal the active role of the ocean. We hypothesized that inter-annual variation of SST and the SST front (SSTF) driven by oceanic dynamics has an impact on the local cloud properties despite the strong summertime atmospheric forcing. There are some challenges that must be overcome to reveal the impact of oceanic dynamics on the summertime marine boundary layer and low-level cloud based on observational data analysis. Oceanic forcing to SST is hard to retrieve because the summertime SST is highly sensitive to surface heat flux due to the shallow mixed-layer depth. In this study, to consider oceanic dynamics associated with the western boundary currents, the vertical structure of subsurface temperature determined by Argo float observations was used to diagnose the variability in the Kuroshio or Oyashio Extension. Additionally, we focused on the “frontal” characteristics of SST because the SSTF is a prominent structure that is mainly formed by the western boundary currents (i.e. the oceanic dynamics). For example, the SSTF in the western NP is formed in the confluence region of the Kuroshio and Oyashio Extensions (Nakamura and Kazmin 2003; Kazmin 2016). The objective of the present study was to reveal the impact of the inter-annual variation of oceanic dynamics on low-level cloud properties—that is, the response of low-level cloud properties to the SSTF variations—based on an observational dataset.

This paper is organized as follows. In Sect. 2, the observational datasets and methods that were used to investigate the properties of the SSTF are described. The properties of the SSTF are also defined. In Sect. 3, we describe the inter-annual variation of the strength of the SSTF and related variability in ocean subsurface temperature (T_sub) in the western NP, suggesting that variation in the SSTF is driven by oceanic dynamics. Section 4 describes our investigation into the response of low-level cloud properties to the inter-annual variation of the SSTF. We quantify the contribution of SST to the modulation of low-level cloud properties compared with other controlling factors for low-level cloud using a multi-linear regression model. Section 5 summarizes and discusses our findings.

2 Data and methods

• a. Data

In this study, we used daily mean datasets of low-level cloud properties, SST, surface heat flux, and meteorological parameters to investigate the impact of the ocean-driven SSTF around the Oyashio extension on low-level clouds on inter-annual timescales. In addition to these datasets, we also used the monthly-mean datasets of cloud radiative effect (CRE) and oceanic subsurface properties. The period from 2003 to 2016 was analyzed because of the availability of the datasets. This study basically focuses on the boreal summer from June to August, and the monthly-mean states are estimated from the daily data except for CRE and the subsurface properties. Details of each observational dataset are described below.

• 1. Cloud properties and surface heat flux

To estimate the LCC and the COT of low-level cloud, we used the L3 gridded cloud product (MYD08 D3) from the latest Moderate Resolution Imaging Spectroradiometer (MODIS) release (version 6) from the MODIS instruments on board the Aqua satellite platforms (Platnick 2015) with a horizontal resolution of 1°. Note that this horizontal resolution is not adequate to fully resolve the SST frontal scale impact, such as vertical mixing and pressure adjustment listed before (e.g. Smirnov et al. 2015; Ma et al. 2017). However, the SST impact along the front can be identified as shown in the summertime-mean map of SST and low-level cloud properties (Fig. 1a, b). In addition, the MODIS product is a common one with long-term record with moderate spatial resolution used in the previous studies (e.g. Masunaga et al. 2016; Miyamoto et al. 2018). Further discussion about spatial–temporal scale dependence of the low-level cloud response to local SST anomaly will be discussed in Sect. 4. Due to the difficulty in estimating the true LCC from passive sensors, the cloud cover was calculated under the random overlapping assumption. Comparison of the cloud cover obtained from the radar and lidar observations with active sensors indicated that the random overlap assumption was reasonable over the mid-latitude oceanic regions compared with other assumptions (Li et al. 2015). Thus, the true LCC was estimated as follows:

$$\begin{array}{*{20}c} {LCC \left[ { \% } \right] = \frac{{f_{L} }}{{1 - f_{M} - f_{H} }},} \\ \end{array}$$

(1)

where f_L, f_M, and f_H are the fractions of the scene covered by low-level (680 hPa ≤ cloud top pressure [CTP] < 1000 hPa), mid-level (440 hPa ≤ CTP < 680 hPa), and high-level (CTP < 440 hPa) cloud types, respectively. The cloud cover of each of these types was calculated using the histogram of CTP at each grid point within 1 month. Additionally, the “conditional-mean” COT for low-level cloud was calculated using the CTP-COT histograms in each grid with a criterion that CTP is greater than 680 hPa.

Two observational heat flux datasets (SW radiation, longwave [LW] radiation, sensible heat [SH], and latent heat [LH]) at sea surface were used in this study. To estimate monthly mean radiative fluxes (SW and LW), we used the Clouds and the Earth’s Radiant Energy System (CERES) Energy Balanced and Filled (EBAF) product Edition 4.0 (Loeb et al. 2018; Kato et al. 2018). The variables were calculated using a radiative transfer model initialized using satellite-based cloud and aerosol data and meteorological assimilation data obtained from reanalysis. The variables were also constrained by the observed top-of-atmosphere radiative fluxes. The CERES product was also used to obtain the CRE for SW and LW at the top-of-atmosphere and sea surface, which is defined as the difference between the all-sky radiative flux and clear-sky radiative flux using the radiative transfer model. In the following analysis, we analyzed the SWCRE as one of low-level cloud properties because SWCRE variability depends on the LCC and COT of low-level cloud rather than those of high-level cloud (not shown). The results for LWCRE are also not shown here because its response and variability are smaller than those of SWCRE by one order of magnitude. Global ocean-surface heat flux products were obtained from the daily mean turbulent fluxes dataset of the Objectively Analyzed air-sea Heat Fluxes (OAFlux) project of the Woods Hole Oceanographic Institution (Yu et al. 2008). The OAFlux products are constructed not from a single data source, but from an optimal blending of satellite retrievals and three atmospheric reanalysis datasets. The horizontal resolutions of the CERES and OAFlux products are 1° in longitude and latitude.

• 2. Oceanic properties

The objectively analyzed daily mean SST of the National Oceanic and Atmospheric Administration (NOAA) Optimum Interpolation version 2 (OISST; Reynolds et al. 2002) was used as the SST dataset. Monthly mean SST was calculated by averaging from the interpolated daily dataset with a horizontal resolution of 0.25°. To assess the interannual variability of oceanic subsurface structures associated with the strength of the SST front, we used a monthly mean three-dimensional temperature dataset based on Argo profiling float observations (Argo 2000). The Roemmich–Gilson (RG) Argo Climatology dataset provided by the Scripps Institution of Oceanography (Roemmich and Gilson 2009) was used. The horizontal resolution of the RG product is 1°, and the vertical resolution varies with depth (e.g., 25 m resolution from a depth of 10 to 200 m), and analysis was conducted using a two-dimensional optimal interpolation method on pressure surfaces for temperature and salinity. To calculate the mixed layer depth (MLD), we used the Monthly Isopycnal/Mixed‐layer Ocean Climatology (MIMOC; Schmidtko et al. 2013) dataset estimated mostly by Argo float observation, supplemented by ship observations and conductivity-temperature-depth profile data.

• 3. Meteorological properties

To estimate the variability of the meteorological field associated with the inter-annual variations of low-level cloud and SSTF, we used ERA5 reanalysis hourly data (Hersbach et al. 2020) with a horizontal resolution of 0.25° and 137° vertical levels whose resolution is finer than that of ERA-Interim (Dee et al. 2011). Monthly mean data of air temperature at 700 and 1000 hPa, horizontal wind at 1000 hPa, and sea-level pressure were used to calculate the EIS and T_adv as a proxy for LCC controlling factors. EIS was calculated following the method of Wood and Bretherton (2006), showing the inversion strength at the top of planetary boundary layer given the air temperature at 700 hPa and at the sea surface. T_adv was calculated by:$T_{{adv}} \left[ {K/day} \right] = - u_{{1000hPa}} \partial T_{{1000hPa}} /\partial x - v_{{1000hPa}} \partial T_{{1000hPa}} /\partial y$, where T_1000hPa, u_1000hPa and v_1000hPa are the temperature, zonal wind speed, and meridional wind speed at 1000 hPa, respectively. Instead of T_1000hPa and wind at 1000 hPa, the SST and wind at 10 m could be used, respectively. However, the computed temperature advections were not significantly different and did not depend on the variables used (Zelinka et al. 2018). Additionally, the vertical pressure velocity and relative humidity at 700 hPa (ω₇₀₀ and RH₇₀₀, respectively) were also used as the controlling factors for the low-level cloud properties. To obtain the monthly-mean states of EIS and T_adv, we first calculated the hourly values at each grid point, and then calculated monthly-mean values. Compared with the calculation method using monthly-mean datasets, the estimated monthly-mean states of EIS and T_adv are significantly different due to the large sub-monthly variations in air-sea temperature and near sea surface winds around the oceanic frontal zone (e.g. Miyamoto et al. 2018; Sasaki and Yamada 2018; Ogawa and Spengler 2019; Masunaga et al. 2020).

• b. Methods

The position of the SSTF (P_SSTF) was defined as the latitude where the horizontal gradient of the SST has a maximum value for each month and each longitude bin of 0.25° width. To extract the properties of the SSTF associated with the Oyashio extension (Isoguchi et al. 2006), we searched for the P_SSTF in the latitude range of 35–45°N. The strength of the SST front (S_SSTF) was defined as the maximum horizontal gradient of the SST. Mathematically, S_SSTF can be expressed as:

$$\begin{array}{*{20}c} {S_{SSTF} \left( {x,t} \right) \left[ { {}^{ \circ }C/100\,km } \right] = \left[ {{\text{max}}\left( {\nabla {\text{SST}} \left( {x,y,t} \right)} \right)} \right]_{x,t} } \\ \end{array}$$

(2)

where x, y, and t are the longitude, latitude, and time, respectively; and []_x,t denotes searching for the maximum value of the horizontal gradient of the SST for the longitude bin (x) and time (t). As an example, we plotted the climatological P_SSTF and S_SSTF in JJA (Fig. 1c and d).

To confirm that the SSTF properties, in particular S_SSTF, are determined by oceanic forcing rather than atmospheric forcing (i.e., the impact of SST/SSTF on the atmosphere), we investigated the frontogenesis/frontolysis process (i.e., strengthen/weaken in S_SSTF) in the western NP using the following frontogenesis equation (Tozuka and Cronin 2014; Ohishi et al. 2017; Tozuka et al. 2017):

$$\begin{array}{*{20}c} { \frac{\partial }{\partial t}\left( {\frac{{\partial {\text{SST}}}}{\partial y}} \right) = \frac{\partial }{\partial y}\left( {\frac{Q}{{\rho_{0} C_{p} H_{mix} }}} \right) + \frac{\partial }{\partial y}\left( {oceanic} \right),} \\ \end{array}$$

(3)

where Q is the net sea-surface heat flux, $\rho_{0}$ is the typical sea water density, $C_{p}$ is the specific heat at a constant pressure, $H_{mix}$ is the MLD derived from the MIMOC dataset, oceanic is the SST tendency term derived from the oceanic dynamics (i.e., other than the surface heat flux term; horizontal advection, vertical mixing, etc.). The $\frac{\partial }{\partial y}\left( {oceanic} \right)$ term in Eq. 3 is calculated as the residual of the equation. This equation was based on a simplified mixed-layer temperature budget (Moisan and Niiler 1998) and assumes that SST is equivalent to the temperature of the mixed layer. In the SSTF region in the western NP, the horizontal gradient of SST is dominated by the meridional component rather than the zonal component. Thus, we only considered the meridional component of the SST gradient in Eq. 3, and investigated the variation of S_SSTF and the underlying physical process that modulate it using the above frontogenesis equation.

3 Mechanism of SSTF variations in the summertime western NP

• a. Climatology of SSTF Properties in Summertime from 2003 to 2016

Figure 1c, d show the climatological zonal distributions of P_SSTF and S_SSTF in the boreal summer (JJA) NP and their standard deviations (SD) of the inter-annual variability. The figure shows some zonal differences in SSTF properties between the parts to the west and east of 170° E, which are respectively called the western NP (WNP) and central NP (CNP) hereafter. The zonal differences in the climatological P_SSTF are found to be not as large as those for S_SSTF; however, the SD of the P_SSTF was much smaller in the WNP than in the CNP (Fig. 1c), implying the stable and unstable SSTF in the WNP and CNP, respectively. Second, the climatological S_SSTF is stronger in the WNP than in the CNP (i.e., 3–5 °C/100 km in the WNP, < 3 °C/100 km in the CNP; Fig. 1d). Additionally, there are two peaks for S_SSTF at around 150 and 170°E, respectively. These correspond to the Oyashio extension SSTFs trapped by bottom topography, called the Isoguchi Jet (IJ) 1 and IJ2 regions (Isoguchi et al. 2006). Qiu et al. (2017) suggested that these SSTFs vary independently on decadal timescales, and other studies showed their close relationship with the variability of the Kuroshio Extension (e.g. Sugimoto et al. 2014), which is one of the strongest warm western boundary currents in the world and has a deep vertical structure with a thickness of about 600 m (Qiu 2001). Additionally, observational studies have examined the relative contributions of atmospheric forcing (i.e., surface heat flux) and oceanic forcing (i.e., subsurface temperature advection, vertical mixing across the base of mixed layer, etc.) to the SST tendency in summertime; for example, oceanic forcing has been shown to be dominant in the WNP (Wu and Kinter 2010; Hosoda et al. 2016). Figure 2 displays a longitude–depth cross-section of the SD of T_sub in JJA along the SST frontal region (35–45° N) estimated from the RG product. Here, the large temperature variability in the WNP reaches to a deeper depth of 400 m than that in the CNP of 100 m, suggesting that the oceanic forcing and the surface heat flux forcing are dominant cause of the temperature variability in the WNP and CNP, respectively (Hosoda et al. 2016). In this study, we set our target region in WNP as the IJ2 region from 160° to 170° E because the summertime SSTF in this region is likely driven by oceanic dynamics rather than atmospheric forcing.

• b. Inter-annual variation of subsurface temperature and SSTF properties

Fig. 2

The standard deviation (SD) of the inter-annual variation of subsurface temperature obtained from Argo observation at latitudes of 35–45° N for June, July, and August (JJA) in 2003–2016 (colored areas, unit: °C). The climatological mean of subsurface temperature is also plotted (contours, unit: °C, CI: 1 °C)

The time series of the SSTF properties for all months and only JJA-mean state in the IJ2 region are shown in Fig. 3. The inter-annual variation of the SSTF properties in the IJ2 region was relatively small (SD of P_SSTF: 0.83°N for all months and 0.62°N for JJA only), and cannot be resolved by the coarse grid of the low-level cloud dataset (i.e. horizontal resolution of 1°). The seasonal variation of P_SSTF was unclear. In contrast, S_SSTF in the IJ2 region had a clear seasonality; S_SSTF had minimum and maximum values in summertime and wintertime, respectively (not shown), consistent with the results of previous studies (Kazmin 2016; Tozuka et al. 2017, 2018). Interestingly, clear inter-annual variation of S_SSTF was also observed in the summertime-mean state, e.g., a negative anomaly from 2005 to 2008 and a positive anomaly from 2009 to 2011 (Fig. 2b). This kind of low-frequency variation of the Oyashio extension SSTF is possibly modulated by oceanic process (Frankignoul et al. 2011; Sugimoto et al. 2014; Qiu et al. 2017). Correlation coefficient with the indices of P_SSTF and S_SSTF are almost zero (r = 0.02), thus the variations of these indices are considered to be independent. Hereinafter, we focus on only the inter-annual variation of S_SSTF and the response of low-level cloud to this parameter, rather than the P_SSTF. Note that the P_SSTF anomaly in summertime of 2007 has large negative values (i.e. located further south from the climatological mean position) due to an influence of oceanic eddies (not shown). Oceanic eddies around the front possibly induce large biases in the low-level cloud response to the front, however, results shown later do not change whether the period of 2007 is included or not.

Fig. 3

Time series of mean a position of the sea surface temperature front (P_SSTF) and b strength of the sea surface temperature front (S_SSTF) in the Isoguchi Jet 2 region (160–170° E). Thin and thick lines indicate the original time series and June, July, and August (JJA)-mean values, respectively. Horizontal dashed lines show the climatological mean values of JJA-mean sea surface temperature front properties

Figure 4 displays the regressions of SST and T_sub onto the summertime S_SSTF in the IJ2 region (160–170°E; Fig. 4a, c) and a part of the CNP (155–165°W; Fig. 4b, d), indicating the oceanic state at the sea surface and in the subsurface when S_SSTF is increased by one standard deviation (1σ = 0.23 °C/100 km). In Fig. 4, the x-axis is modified to relative latitude from the JJA-mean P_SSTF in each part of the SSTF region. As expected, the response of SST to the strengthen of S_SSTF (ΔSST) was negative (colder) in the northern flank of the SSTF (Fig. 4a). ΔSST in the southern flank is positive but not significant Although the spatial patterns of ΔSST were similar in the IJ2 region and the CNP (Fig. 4b), the spatial patterns of the response to T_sub (ΔT_sub) were notably different in these two regions. In the IJ2 region, the meridional contrast of ΔT_sub across the position of SSTF was found both at the near sea surface and in the deeper part (Fig. 4c). Additionally, a warm ΔT_sub appeared from a depth of 100–600 m in the southern flank around – 10° (Fig. 4c), corresponding to the relative position of Kuroshio Extension. In contrast, the significant meridional contrast of ΔT_sub in the CNP does not appear in the subsurface, implying that changes in SST tend to be strongly controlled by the surface heat flux forcing (Hosoda et al. 2016). In other words, the meridional contrast of ΔT_sub in the deeper part of the IJ2 region presents the observational evidence for the ocean-induced SST and SSTF properties even in the summertime. Figure 5 shows the spatial distribution of summertime ΔSST regressed onto S_SSTF of the IJ2 region. As shown in Fig. 4a, a warm (cold) ΔSST appears in the southern (northern) flank of the SSTF when S_SSTF is increased (Fig. 5). Additionally, the ΔSSTs spread widely away from the IJ2 region. For example, cold ΔSSTs appear in the northern flank of the IJ2 SSTF with a zonal expansion along the path of the Oyashio Extension. On the other side, warm ΔSSTs appear from 140° to 170° E, which is along the path of the Kuroshio Extension. These spatial characteristics imply that the ΔSSTs and the strengthening process of S_SSTF in the summertime IJ2 region are closely related to the Kuroshio and Oyashio Extensions.

• iii. Mechanisms of frontogenesis in the IJ2 region; atmospheric forcing vs oceanic forcing

Fig. 4

a Meridional distribution of the response of the sea surface temperature (SST) to the increase of the strength of the sea surface temperature front by one standard deviation (bar corresponding with the left y-axis) and the climatological mean value of the SST (line corresponding with the right y-axis) in the Isoguchi Jet 2 region. The x-axis is the shifted latitude based on the position of the sea surface temperature front (P_SSTF). c Latitude–depth cross-section of the response of the subsurface temperature (T_sub) in the Isoguchi Jet 2 region obtained from the Roemmich–Gilson product of Argo float observation from 2004 to 2016. Responses within the 95% confidence interval are displayed in color. Contour lines are climatological values of subsurface temperature (CI: 2 °C). b, d are the same as (a, c), respectively, but for the central North Pacific (CNP; 155–165° W)

Fig. 5

The responses of the sea surface temperature (SST) (°C) at each grid point to the increased strength of the sea surface temperature front in the Isoguchi Jet 2 region (at 160–170° E) by one standard deviation. The black line (red points) shows the climatological mean position of the sea surface temperature front (in the Isoguchi Jet 2 region). Grid boxes with a black cross indicate responses within the 95% confidence interval

We investigated how the summertime SSTF properties were determined, based on the frontogenesis equation (Eq. 3). Figure 6 shows the seasonal evolution of each term in the frontogenesis equation from May to September. Note that, in Fig. 6, a positive (negative) rate indicates frontogenesis (frontolysis) and the meridional gradient of SST (∂SST/∂y) is always negative in the IJ2 region. As described above, S_SSTF in the IJ2 region is the smallest in summertime and strongest in wintertime (e.g., Tozuka et al. 2018). Frontolysis occurs in JJA due to the surface net heat flux (HF) term (red line), which weakens the S_SSTF (Fig. 6). In contrast, the oceanic term (OCN; blue line) always induces frontogenesis in the IJ2 region. The results shown in Fig. 6 imply that the HF term is a dominant factor modulating summertime frontolysis in the climatological seasonal cycle.

Fig. 6

The seasonal variation from May to September of each term of the frontogenesis equation in the Isoguchi Jet 2 region, calculated according to Tozuka and Cronin (2014). The surface heat flux term calculated based on the meridional gradient of net heat flux (HF; red line), the oceanic term (OCN; solid line), and the total frontogenesis rate (Rate; black line) are plotted. A positive (negative) value indicates frontogenesis (frontolysis), meaning that the strength of the sea surface temperature front is increased (decreased) by the frontogenesis (frontolysis) process in each month

Figure 7a displays the mean meridional distribution of ∂SST/∂y and that of the mean meridional gradient of heat flux (∂Q/∂y) in the IJ2 region. The minimum ∂SST/∂y and maximum ∂Q/∂y occur close to the position of the SSTF. Note that a positive ∂Q/∂y means more heat release or less heat input in the south of SSTF as compared to that in the north; frontolysis process. To further investigate the relationship between the HF term and S_SSTF on inter-annual timescales, Fig. 7b displays a scatter plot of ∂SST/∂y vs ∂Q/∂y on the SSTF in JJA from 2003 to 2016, indicating the relationship between the strength of the SSTF (~ ∂SST/∂y) and the frontolysis by surface heat flux forcing (~ ∂Q/∂y) on inter-annual timescales. Note that the meridional gradient of MLD has a negligible effect on the frontolysis in summertime (not shown). A significant negative correlation (~ − 0.85) can be seen between ∂SST/∂y and ∂Q/∂y, suggesting that more frontolysis (i.e., positive values of ∂Q/∂y) occurs when the SSTF is strengthened (i.e., negative values of ∂SST/∂y). This indicates that a frontolysis does not decrease S_SSTF, but that an increased S_SSTF induces more frontolysis by modulating the heat release from the sea surface, reflecting the active role of the ocean in the modulation of surface heat flux. The zonal distribution of the correlation coefficients in Fig. 8 is similar to that shown in Fig. 7b, and the correlation coefficients with each component of heat flux are also displayed. The correlation between ∂Q/∂y and ∂SST/∂y was significantly negative in both the IJ2 region and in the western part from 170° W, where the climatological mean S_SSTF is greater than 1.5 °C/100 km (Fig. 1a). Furthermore, the correlation coefficient between SH and LH is also significantly negative, indicating that S_SSTF is effectively damped by the turbulent flux due to the SST anomaly itself. Unlike the western part including the IJ2 region, there was no significant correlation between ∂Q/∂y and ∂SST/∂y in the CNP; the eastern part from 170°W. The relationship between HF and SST (or ∂Q/∂y and ∂SST/∂y) that acts to damp the SST (SSTF) has been recognized as an active role in SST (Tanimoto et al. 2003), which has been confirmed in the western part from the date line. Thus, we can assume that, in the IJ2 region, the inter-annual variation of S_SSTF is mainly driven by the oceanic forcing, rather than the heat flux forcing. Hereinafter, we regarded the regressions of low-level cloud properties onto S_SSTF as the impact of ocean-driven SST anomalies on the low-level cloud.

Fig. 7

a Mean meridional distribution of the meridional gradient of the sea surface temperature (∂SST/∂y; unit: °C/100 km; left y-axis) and that of net heat flux (∂Q/∂y; unit: W/m²/100 km; right y-axis) close to the position of the sea surface temperature front (P_SSTF) in the Isoguchi Jet 2 (IJ2) region in June, July, and August (JJA). Downward heat flux is defined as positive; thus, a positive ∂Q/∂y indicates that a small heat input (or large heat release) occurs in the south of the IJ2 region, suggesting frontolysis. b Scatter plots of ∂SST/∂y vs ∂Q/∂y from 2003 to 2016 in JJA. The correlation coefficient between ∂SST/∂y and ∂Q/∂y is displayed in the upper right

Fig. 8

Correlation coefficients between ∂SST/∂y and the meridional gradients of each flux term (shortwave [SW] radiation: blue, longwave [LW] radiation: red, sensible heat [SH]: green, and latent heat [LH]: orange) and the correlation coefficients between ∂SST/∂y and the meridional gradient of net heat flux (∂Q/∂y) (gray bars) close to the position of the sea surface temperature front (P_SSTF) in the Isoguchi Jet 2 region in June, July, and August (JJA). Correlation coefficients were calculated for every 5° longitudinal bin from 145° E eastward. For all colored markers and gray bars, the correlation coefficients are within the 95% confidence interval

4 Response of summertime low-level cloud properties to S_SSTF in the IJ2 region

SST frontal characteristics (i.e., S_SSTF and P_SSTF) have been recognized as the ocean-induced factors modulating atmospheric conditions in the mid-latitudes, particularly for wintertime, including the storm tracks (Nakamura et al. 2004; Ogawa et al. 2012, 2016; Masunaga et al. 2016; Parfitt and Seo 2018) and vertical cloud distributions (Tanimoto et al. 2009; Tokinaga et al. 2009; Liu et al. 2014; Kawai et al. 2015, 2019). In the previous section, we confirmed that the inter-annual variation of the summertime S_SSTF in the IJ2 region is likely driven by oceanic forcing from the Kuroshio and Oyashio Extensions. In this section, we estimate the response of low-level cloud properties to the increased S_SSTF in the IJ2 region. To do this, we calculated the linear regression coefficient of each cloud variable and the controlling factor at every grid point to the inter-annual variation of the mean S_SSTF in the IJ2 region. We recorded the meridional distributions of the responses of low-level cloud properties (i.e., LCC, COT of low-level cloud, and SWCRE at the sea surface) to the S_SSTF index in JJA (Fig. 9). We found a distinct pattern in these responses, particularly for SWCRE, in the northern flank of the SSTF. In the northern flank, ΔSWCRE was negative over the cold ΔSST (Fig. 9a), mainly due to the positive ΔCOT (Fig. 9c), rather than the ΔLCC. The opposite was apparent for the negative ΔLCC in the northern flank; however, the ΔLCC in this region showed lower statistical significance. While the responses were not significant in the southern flank, ΔSWCRE was positive over the warm ΔSST (Fig. 9a) mainly due to the negative ΔLCC (Fig. 9b). Similar results are displayed in Fig. 10, except for their horizontal distributions. A common feature of all three low-level cloud properties is that the responses to S_SSTF appear in a wide region not confined to around the IJ2 region. This spatial feature is similar to the results for SST (Fig. 5) associated with the western boundary currents, that is, the paths of Kuroshio and Oyashio Extensions. The positive response of ΔLCC to ΔSST in the southern flank of the SSTF is similar to the observations of previous studies described in Sect. 1 (Norris and Leovy 1994; Norris et al. 1998; Bretherton et al. 2013). Similar to the response of LCC to SST, COT can also be modulated by changes in the entrainment process of dry air from the free troposphere into the boundary layer associated with an SST anomaly; COT increases with decreasing SST (Terai et al. 2016, 2019), which is consistent with the positive ΔCOT in the northern flank of the SSTF observed in the present study (Figs. 9c and 10c).

Fig. 9

As in Fig. 4a, but for a the shortwave cloud radiative effect (SWCRE) (unit: W/m²), b cloud optical thickness (COT; unit: none), and c low-level cloud cover (LCC; unit: %). The standard deviation of the strength of the sea surface temperature front is shown in the upper right

Fig. 10

As in Fig. 5, but for a the shortwave cloud radiative effect (SWCRE; unit: W/m²), b low-level cloud cover (LCC; unit: %), and c cloud optical thickness (COT; unit: none)

Low-level cloud properties are modulated by not only the local SST but also other atmospheric variables, as reported in many previous studies (Qu et al. 2015; Seethala et al. 2015; Myers and Norris 2016; McCoy et al. 2017; Klein et al. 2017; Miyamoto et al. 2018; Zelinka et al. 2018). For example, they investigated the contribution from the various controlling factors listed above to assess the response of low-level cloud to global warming (Qu et al. 2015; Zhai et al. 2015; Myers and Norris 2016; Brient and Schneider 2016; Terai et al. 2016; McCoy et al. 2017). The consensus is that the formation of a large amount of low-level cloud is favored by strong temperature inversions at the top of cloud (Wood and Bretherton 2006), cool SST (Bretherton et al. 2013), weak subsidence (Myers and Norris 2013), and enhanced turbulence with upward heat flux induced by cold-air advection near the sea surface (Norris and Iacobellis 2005). Thus, we attempted to quantify the contributions of the different controlling factors separately using the multi-linear regression analysis method that was employed in the studies referenced above. Figure 11 shows the responses of four other LCC controlling factors—namely EIS, T_adv, RH₇₀₀, and ω₇₀₀—to S_SSTF; all of these four factors have been widely used. When S_SSTF is increased, ΔEIS is negative in the southern flank of the SSTF, and the opposite in the northern flank (Fig. 11a). Negative ΔEIS caused by a positive ΔSST indicates unstable conditions in the lower troposphere, which is unfavorable for the formation of low-level cloud (Klein and Hartmann 1993; Wood and Bretherton 2006; Kawai et al. 2017). Although the low-level cloud tends to form with cold T_adv through a destabilization process in the atmospheric boundary layer (Wood 2012), ΔT_adv was observed to be negative around the SSTF (Fig. 11b). For other atmospheric factors that control LCC, ΔRH₇₀₀ was negative (positive) in the southern (northern) flank (Fig. 11c) and Δω₇₀₀ was positive in the southern flank (Fig. 11d) indicating strong subsidence or weak updraft. Strong subsidence and a dry free-troposphere are also unfavorable conditions for low-level cloud formation (Bretherton et al. 2013; Myers and Norris 2013, 2015). In conclusion, almost all of the LCC controlling factors responded simultaneously to the inter-annual variation of S_SSTF and acted to change the low-level cloud properties toward the same sign around the IJ2 SSTF region, except for T_adv.

Fig. 11

As in Fig. 5, but for a estimated inversion strength (EIS) (unit: K), b horizontal air-temperature advection (T_adv) (unit: K/day), c relative humidity at 700 hPa (RH₇₀₀; unit: %), and d vertical pressure velocity at 700 hPa (ω₇₀₀; unit: hPa/day)

To quantify the relative contributions of the controlling factors to the properties of low-level cloud (Figs. 9 and 10), a multi-linear regression (MLR) analysis was conducted. All variables were spatially interpolated to the coarsest common grid of 1° for the low-level cloud products. The response of low-level cloud properties (referred to as C; i.e. LCC, COT, or SWCRE) to the increased S_SSTF is expressed as follows:

$${\Delta }C = \frac{\partial C}{{\partial S_{SSTF} }} = \mathop \sum \limits_{i} \frac{\partial C}{{\partial x_{i} }} \frac{{\partial x_{i} }}{{\partial S_{SSTF} }} ,$$

where $x_{i}$ is a controlling factor, namely, SST, EIS, T_adv, RH₇₀₀, or ω₇₀₀. The partial derivatives of the low-level cloud properties were calculated using a multi-linear regression approach similar to that used in previous studies (summarized in Klein et al. 2017). To construct the MLR model to estimate the response of low-level cloud to the five LCC controlling factors, we used the anomalies of all cloud parameters and predictors from each seasonal mean value in JJA at each grid point in the western NP region (130–170° E, 30–55° N). Due to multicollinearity between the SST and EIS, this model possibly underestimates the contributions from the SST and EIS. Then, from original EIS variability, we subtracted EIS regressed onto SST before calculating the slopes in the MLR model. Table 1 shows the regression slopes calculated using the constructed MLR model for LCC, COT, and SWCRE. Except for the removal of multicollinearity between the SST and the EIS, this process is similar to the procedures used in the previous studies (e.g., Klein et al. 2017). The signs of the slopes are also consistent with the physical mechanisms confirmed in the previous observational and numerical simulation studies using CMIP5 model outputs (e.g. Klein et al. 2017). For example, the negative slopes of LCC and COT to SST suggest that, when SST is low, low-level cloud appears more frequently and becomes more dense due to the suppression of the entrainment process of dry air from the free troposphere into the moist boundary layer (Bretherton et al. 2013; Qu et al. 2015). The variances of the three MLR models to the actual response of each low-level cloud variable are 66% for LCC, 40% for COT, and 68% for SWCRE. We also calculated the confidence interval from 5 to 95% of each slope, but the interval for each slope was smaller than the slope value itself by one order of magnitude.The coefficients for LCC and COT are calculated from the daily-mean states, but the results are not significantly different from those using the monthly-mean states (not shown). These are also consistent with Miyamoto et al. (2018), which focused on a different western boundary current region, the Agulhas Return Current. We recorded comparisons between the responses of low-level cloud to S_SSTF calculated directly by linear regression (shown in Fig. 9) and those estimated using the MLR model (“Total”) around the IJ2 SSTF (Fig. 12). The figure also shows the contributions of each controlling factor calculated based on the slopes of each low-level cloud variable output by the MLR model (Table 1). These contributions show that SST is a primary controlling factor for all cloud responses. In particular, the ratio of the actual responses to the SST contribution was 50% for ΔCOT and 103% for ΔSWCRE in the northern flank of the IJ2 SSTF (i.e. Meridional mean from Δlatitude of 0˚ to + 5˚). In contrast, the responses of low-level cloud properties in the southern flank, where the cloud responses are not significant (Figs. 9 and 10), could not be estimated well. Takahashi and Hayasaka (2020) suggested that T_adv is likely to be a main driver of the LCC variability in the WNP on the intra-seasonal timescales (about 20–100 days), however, in the present study, we observed SST to be a main driver of LCC and T_adv a secondary contributor on inter-annual timescales. Additionally, they also suggest that SST and EIS are followers of LCC variation on such a short timescale. This implies that the SST has a potential to actively modulate the LCC variation on inter-annual timescales, rather than on intra-seasonal timescales. We conclude that, even in the summertime IJ2 region, S_SSTF variation related to the Oyashio Extension can actively control the monthly-mean anomalies of COT and SWCRE via, not the anomalous atmospheric circulation (i.e. T_adv), but rather via direct SST anomalies induced by a change in the western boundary currents. They indicate similar results to those obtained in Fig. 12, while the MLR model could not adequately capture the spatial characteristics of low-level cloud responses on a small spatial scale (i.e., a frontal scale at a Δlatitude of around 0°).

Table 1 Calculated regression slopes for each cloud variable for each predictor in the multi-linear regression model of LCC, COT, and SWCRE

Fig. 12

As in Fig. 9, but for the total response estimated using the multi-linear regression model (Total), and the contributions of each controlling factor of the local low-level cloud cover (sea surface temperature [SST], estimated inversion strength [EIS], horizontal air-temperature advection [T_adv], relative humidity at 700 hPa [RH700], and vertical pressure velocity at 700 hPa [ω₇₀₀]) are also plotted

5 Summary and discussion

This study investigated the response of the properties of summertime low-level cloud to the properties of the SSTF (i.e., the strength of SST front; S_SSTF) associated with the Oyashio Extension in the WNP on inter-annual timescales, using satellite observational data, an atmospheric reanalysis dataset, and Argo float observations. The mechanisms responsible for the inter-annual variation of S_SSTF was investigated in detail before calculating the low-level cloud response to this variable. The results indicate that there is a strong contribution from the western boundary currents (i.e., the Kuroshio and Oyashio Extensions) to changes in the SSTF properties in the summertime WNP, while atmospheric forcing is the dominant contributing factor across a wide area of the CNP. The response of low-level cloud to the ocean-induced variation in the S_SSTF showed that negative SWCRE with more COT in the northern flank of the IJ2 SSTF can be induced by cold SST anomalies along with Oyashio Extension. Furthermore, the low-level cloud responses to the S_SSTF were examined using a multi-linear regression model to quantify the contributions from SST and other atmospheric factors (e.g., EIS and T_adv). The results of this model suggest that the local SST anomaly is a primary factor controlling the COT and SWCRE responses to the S_SSTF in summertime. However, other atmospheric controlling factors make only minor contributions.

Although the variation in the S_SSTF might be related to oceanic forcing, the SST anomaly also seems to be forced by the SWCRE of low-level cloud due to the positive feedback loop between the SST and low-level cloud properties (Norris and Leovy 1994). Using the HF term in the mixed-layer temperature budget equation (Moisan and Niiler 1998), we estimated how SST can be forced by ΔSWCRE; the ratio of the estimated ΔSST to the actual ΔSST shown in Fig. 13. As shown in Figs. 9a and 10a, positive (negative) ΔSWCRE appears in the southern (northern) flank of the SSTF, which induces warming (cooling) of the SST when the S_SSTF is high. Thermodynamically, the meridional distribution of ΔSWCRE induced by ΔSST can enhance S_SSTF within the positive feedback loop. However, we found no significant correlation between S_SSTF in the meridional direction and the meridional gradient of the SW directly over the SSTF (the IJ2 region in Fig. 8). This implies that there is no robust observational evidence for the feedback between S_SSTF and the SWCRE induced by low-level cloud just around the SSTF. However, far away from the SSTF in both flanks of the IJ2 SSTF (around 35°N or 45°N), ΔSWCRE-induced ΔSST can explain about half of the actual ΔSST. The area-mean ratios of the SST response in the northern and southern flanks of the IJ2 SSTF are 51% and 47%, respectively (Fig. 13b). In other words, the large-scale characteristics of the SST can be modulated by the radiative effect of low-level cloud, but the primary driver for frontal ΔSST might be oceanic forcing associated with the western boundary currents. It is also hard to assume that T_sub anomaly at deeper part was modulated by the radiative flux anomaly with changes in the low-level cloud properties. This represents further evidence of the active role of the SST anomaly in modulating low-level cloud.

Fig. 13

Horizontal map of a estimated sea surface temperature (SST) response modulated by the shortwave cloud radiative effect (SWCRE) response shown in Fig. 4.10c, using the mixed-layer temperature equation for the increased strength of the sea surface temperature, b the ratio of the estimated SST response to the actual response (shown in Fig. 4.5) (unit: %). Only positive ratios are shown. The area-mean ratio of the SST response in the northern (southern) flank of the Isoguchi Jet 2 SST front (165–175° E) is 51% (47%)

On inter-annual timescales, several previous studies have found a significant linkage between the westerly jet and SSTF in the mid-latitude (e.g. Nakamura et al. 2004, 2008; Ogawa et al. 2012; Wang et al. 2019; Chen et al. 2019; Parfitt and Kwon 2020) via baroclinicity modulation processes. The dynamical state associated with a fluctuation of the jet position is one of the important drivers of the mid-latitude cloud variability. Low-level cloud responses to inter-annual variations of the positions of the jet and Hadley cell edge have also been examined using the satellite observational records and CMIP experiments (Tselioudis et al. 2016; Grise and Medeiros 2016; Lipat et al. 2017; Zelinka et al. 2018). They showed that both dynamical and thermodynamical factors— i.e. anomalies of T_adv and EIS, respectively— can be the driver of low-level cloud responses to the jet fluctuations, although the primary factor is still under debate. While the present study focused on only the SSTF properties, we concluded that the local thermodynamical factors (i.e. SST and/or EIS) play key roles in modulating the low-level cloud variability around the Oyashio Extension in summertime. To fully understand the inter-annual variability of low-level cloud properties in the mid-latitude, we need to further investigate the relative contributions from the dynamical and thermodynamical components.

Finally, it is useful to comment on the spatial–temporal scales in the processes of the low-level cloud response to the SST front variations presented in this study. In terms of the spatial scales, our analysis was conducted using the cloud datasets with 1° horizontal resolution. This presents the difficulty in depicting the impact on low-level cloud properties on a SST frontal scale. Our recent work (Takahashi et al. 2020) investigated the sensitivity of low-level cloud properties to the increased S_SSTF using Weather Research and Forecasting numerical simulations with different SST boundary conditions. The results suggested that the SSTF impact on low-level cloud can be explained by not only one-dimensional vertical process, but also two-dimensional horizontal processes with air-mass advection across the front (Schneider and Qiu 2015; Kilpatrick et al., 2014, 2016; Bai et al. 2019). However, its impact is limited within a hundred kilometer from the front position. In other words, the impacts of SST anomalies on larger spatial scale than the frontal scale can be mostly explained by the vertical process with local SST anomalies. Thus, our results in the present study are limited to depict the impacts of only local SST anomalies along the western boundary currents. In terms of time scales, the dominant timescale of the low-level cloud variability is shorter than inter-annual (e.g. de Szoeke et al. 2016). Atmospheric synoptic disturbances in the mid-latitude NP strongly affect the short-term variability of low-level clouds (Xu et al. 2005; Norris and Iacobellis 2005; Takahashi and Hayasaka 2020). In addition to the atmospheric variability, oceanic eddies can also affect cloud formation in atmospheric boundary layer via vertical mixing and pressure adjustment mechanisms, as shown in recent studies based on research cruise observations (Kawai et al. 2015, 2019; Jiang et al. 2019). Such synoptic timescale air-sea interactions are difficult to investigate using the coarse temporal observational datasets, and they are also not represented well in general circulation models. Future research is needed to examine the detailed spatial and temporal characteristics of the interaction among the synoptic disturbances, oceanic state, and cloud using a new generation of geostationary satellites with higher spatial–temporal resolutions.