Increase in turbulent heat flux during the 1990s over the Kuroshio/Oyashio extension region



[1] Various turbulent heat flux data indicate remarkable increase in heat loss by latent and sensible heat fluxes over Kuroshio and Kuroshio/Oyashio Extension regions during the 1990's. This increase is found in net heat flux, and the heat flux reached its maximum during last 50 years. The slope is about 5.8 W m−2 year−1 on average and is found over the most part of Kuroshio, Kuroshio/Oyashio Extension regions and Japan Sea. The increase in latent heat flux dominantly contributes to the increase in turbulent heat flux mainly due to the increase in SST. Although the increase in wind speed also contributes to the increase in turbulent heat flux, the contribution is smaller than that of SST. In contrast, the increase of specific humidity which contributes the decrease in latent heat flux is also found.

1. Introduction

[2] It is well known that the Kuroshio and Kuroshio/Oyashio Extension regions are characterized by large ocean heat loss related to turbulent heat flux and one of the largest air-sea flux regions in the global ocean. This is caused by the warm Kuroshio water from the tropics and the cold and dry winds from Asian Continent in particular in boreal winter. The exchange of heat flux over the Kuroshio and Kuroshio/Oyashio Extension between the ocean and the atmosphere is quite important for many oceanic phenomena such as mixed layer variation and formation of mode water [Hanawa and Talley, 2001].

[3] Atmospheric signals in wind speed, air temperature and humidity have an effect on turbulent heat flux, while oceanic signals in SST also influence turbulent heat flux. In particular, in the Kuroshio/Oyashio Extension region, horizontal advection by ocean flow is also important to the SST or sub-surface temperature changes on interannual time scale [Vivier et al., 2002; Kelly, 2004].

[4] The decadal variability including a climatic regime shift in the Pacific Ocean has been studied by many investigators [e.g., Mantua et al., 1997; Minobe, 1997]. Some recent studies show that the abrupt changes occurred in 1998/99 over the North Pacific [Minobe, 2000; Hare and Mantua, 2000; Schwing and Moore, 2000]. For example, Minobe [2000] showed that air temperatures in winter and winter-spring over the Alaska were the coldest since 1977, as opposed to the warm regime (1977–1998). Minobe [2002] showed that in the 1998/99 change, SSTs and heat storage warmed abruptly both in the Kuroshio/Oyashio Extension region and central North Pacific, accompanied by cooling in the eastern North Pacific. This dramatic change in 1998/99 might be related with a possible major regime shift and affect variation of surface heat flux.

[5] In this study we demonstrate the remarkable increase in the 1990's in turbulent heat flux over the Kuroshio and Kuroshio/Oyashio Extension regions using various global turbulent heat flux data. Moreover, the relative significance of each surface meteorological parameter for this increase in turbulent heat flux is investigated.

2. Data

[6] We used two kinds of turbulent heat flux products which have different data source, i.e. reanalysis and satellite products. The reanalysis products are NCEP Reanalysis version 1 and 2 (NRA-1/2 [Kalnay et al., 1996; Kanamitsu et al., 2000]) and ECMWF 40-years Reanalysis (ERA40) data. The reanalysis products provide us longer time series compared with satellite products. The data during 1948–2003, 1979–2003 and 1979–2001 are used for NRA-1, NRA-2 and ERA40, respectively. In order to investigate the impacts of the turbulent heat flux change on net heat flux, we used net heat flux data from NRA1. The satellite products used in this study are HOAPS2 (, GSSTF2 [Chou et al., 2003] and J-OFURO [Kubota et al., 2002]. The comparison between all products are based on yearly mean value calculated from monthly mean data of turbulent heat flux (i.e., latent and sensible heat fluxes) and the related meteorological parameters (i.e., surface winds, specific humidity, air temperature and sea surface temperature) during 1992–2000, which all products are available.

3. The Remarkable Increase in Turbulent Heat Flux During the 1990's

[7] Figure 1 shows spatial distribution of the linear slopes calculated from NRA1 turbulent heat flux data during 1990–2000 over the Kuroshio and Kuroshio/Oyashio Extension regions. Figure 1 shows that there are positive linear slopes over the most part of the Kuroshio, Kuroshio/Oyashio Extension regions and Japan Sea. On the other hand, relatively small and negative values are found in the region around 37°N, 160°E. These features are also found in that for other products (not shown). In addition there are also large positive or negative slope values are found in global region (not shown).

Figure 1.

Spatial distribution of the linear trend calculated from yearly-mean turbulent heat flux derived from NRA1. Units in W m−2 year−1. The squared area is the maximal trend region (143–150E, 34–40N).

[8] In order to see the long time series of NRA1 turbulent heat flux in this region, the area-averaged time series of turbulent heat flux over 143–150°E, 34–40°N (squared in Figure 1) is shown in Figure 2. The significant increase of turbulent heat flux appeared during 1989–2001 and the change is the largest since 1948. After the increase, the rapid decrease in turbulent heat flux is appeared after 2001. The turbulent heat flux decrease from 270 W m−2 to 200 W m−2 during 2001–2003. Figure 2 shows that the increase is also observed in NRA1 net heat flux. This means that heat transfer from the ocean to the atmosphere is increasing during the 1990's in this region. It should be noted that the increase in annual mean heat flux is mainly caused by values in winter and autumn months, because the increase in spring and summer months are weak or not clear compared with in winter and autumn months (not shown).

Figure 2.

Time series of yearly averaged turbulent heat flux and net heat flux over the Kuroshio/Oyashio Extension region during 1948–2003. Turbulent and net heat flux were averaged in 143–150E, 34–40N (see Figure 1). Units in W m−2.

[9] In order to confirm this increase during the 1990's using other data set, Figure 3 shows the time series of the yearly mean turbulent heat flux, as well as Figure 2. There are large differences in mean values between each product in the whole period (10 ∼ 60 W m−2). Generally the reanalysis products show larger values compared with satellite products. On the other hand, the inter-annual variability is quite similar with each other except for J-OFURO. It was known that the J-OFURO latent heat flux data is a little bit problematic in continuity because of the change of the used satellite sensor and the procedure between 1995 and 1996.

Figure 3.

Time series of yearly-averaged turbulent heat flux over the Kuroshio/Oyashio Extension region during 1992 and 2000. Each turbulent heat flux (NRA1/2, ERA40, HOAPS2, GSSTF2 and J-OFURO) was averaged in 143–150E, 34N–40N (see Figure 1). Units in W m−2.

[10] An important and common feature given in Figure 3 is the linear increase in turbulent heat flux from 1992 to 2000. Obviously heat losses at the ocean surface increase year by year. Table 1 shows the linear slope values calculated from yearly averaged turbulent heat flux, latent heat flux and sensible heat flux. The significance of all linear slopes are evaluated using t-distribution [see Emery and Thomson, 1997], and all linear trends meet criteria of statistical significance of 95%. The significance were also assessed by the method described by Santer et al. [2000] which taken account the temporal autocorrelation, the results were same. In Table 1 the slope for all products for the turbulent heat flux are positive and its average is 5.8 W m−2 year−1 (5.1 W m−2 year−1 except for J-OFURO). The largest value for J-OFURO may be overestimated in associated with the aforementioned problem. In addition, from Table 1, it is clear that latent heat flux has large contribution to this increase, because the slope for sensible heat flux are small (20%) compared with the one for latent heat flux (80%).

Table 1. The Linear Slope Values for Each Producta
  • a

    Units in W m−2 year−1.

Turbulent heat flux5.526.573.494.535.459.19
Latent heat flux3.994.642.643.674.267.19
Sensible heat flux1.531.930.850.861.192.00

4. Cause of the Increase

[11] In order to investigate the cause of the large trend in turbulent heat flux during the 1990's, we investigate the surface meteorological parameters (here after refer as “bulk parameters”) which effect on the latent and sensible heat fluxes (i.e., surface winds, saturated specific humidity (Qs), surface specific humidity (Qa), air temperature and SST).

[12] Figure 4 shows time series of yearly mean NRA1 latent heat flux, surface winds, specific humidity differences (DQ), Qs and Qa which are bulk parameters of latent heat flux averaged over the western Kuroshio/Oyashio Extension region (shown in Figure 1). This area is located in the region where the slope shows the highest value. From Figure 4, it is clear that all bulk parameters increase during the period from 1992 to 2000. The linear slope for each bulk parameter is given in Table 2. The slope is positive for all parameters. The positive slope means the increase in the latent heat flux except Qa. For Qa, the positive slope means the decrease of the latent heat flux. Therefore, it is concluded that the increase in the latent heat flux results from increasing of wind speed and Qs. Since Qs is a function of SST, the increase in the Qs is associated with the increase of SST.

Figure 4.

Time series of yearly-mean bulk parameters over the Kuroshio/Oyashio Extension region. Each parameter was averaged in 143–150E, 34–40N (see Figure 1).

Table 2. The Linear Slope Values for Turbulent Heat Flux and Each Bulk Parameter in NRA1a
  • a

    THF, turbulent heat flux (W m−2 year−1); LHF, latent heat flux (W m−2 year−1); SHF, sensible heat flux (W m−2 year−1); WND, wind speed (m/s/year); Qa, air specific humidity (g/kg/year); Qs, saturated specific humidity (g/kg/year); DQ, QS-QA (g/kg/year); SST, sea surface temperature (degree C/year).

Linear slope5.523.991.530.

[13] It is not clear which bulk parameter i.e. wind speed and SST (Qs) has larger contribution to the increase in the latent heat flux, because of its non-linearity of the bulk formula. Therefore, we carried out a similar analysis used by H. Tomita and M. Kubota (An analysis of the accuracy of J-OFURO satellite derived latent heat flux using moored buoy data and reanalysis products, submitted to Journal of Geophysical Research, 2005, hereinafter referred to as Tomita and Kubota, submitted manuscript, 2005) to quantitatively evaluate the contribution to the increase in latent heat flux. Firstly we prepare two types of data for all bulk parameters. One is raw data including the positive trend and the other is data after removing the trend by the linear slope regression. We calculate latent heat fluxes using the combination of two types of bulk parameters shown in Table 3. These latent heat fluxes are referred as “Fake” data. The bulk formula of Kondo [1975] and annual mean bulk parameters are used in this calculation. Since annual mean bulk parameters are used, the calculated Fake 0 latent heat flux is not identical to the values calculated from NRA1 latent heat flux exactly. The annual mean and the trend are about 70% and 40% below compared with values calculated from monthly mean, respectively. However, since the significant increase was found in results from the annual mean latent heat flux, we assume that the ratio of contribution of each bulk parameter to the increase in latent heat flux is not so different.

Table 3. The Combination of Bulk Parameters for Each Fake Analysis
Fake 0rawrawraw
Fake 1rawremovedremoved
Fake 2removedrawremoved
Fake 3removedremovedraw

[14] Table 4 shows the linear slope value for each Fake turbulent flux data using NRA1 data and its ratio to Fake 0. From Table 4 Fake 3 gives the largest slope value and its ratio to Fake 0. This means that the Qs (SST) has the largest contribution to the increase in latent heat flux. Fake 1 is also positive value but about 80% smaller than Fake 3. This means that wind speed's contribution is fairly small compared to the Qs (SST), although wind speed has a contribution to the increase in latent heat flux. Fake 2 shows a negative value and this means that the Qa contributes to the decrease in latent heat flux. However, its contribution is small and roughly canceled with that of wind speed.

Table 4. The Linear Slope Values for Each Fake Turbulent Flux Data Using NRA1 Dataa
 Fake 0Fake 1Fake 2Fake 3
  • a

    Units in W m−2 year−1.

Linear slope1.631.01–1.123.33
Ratio to Fake 01.000.43–0.481.42

5. Summary and Discussions

[15] We demonstrated the remarkable increase in turbulent heat flux during the 1990's for both of reanalysis (i.e. NRA1/2 and ERA40) and satellite data (i.e., HOAPS2, GSSTF2 and J-OFURO) which are basically independent data. Both of latent and sensible heat fluxes increase during this period (in particular winter and autumn months). However, the latent heat flux has larger contribution to the increase in turbulent heat flux. The increase in the Qs (SST) has the largest contribution to the increase in latent heat flux. Although the wind speeds also contribute to the increase in turbulent heat flux, its impact is small compared with the Qs (SST). The Qa also has a positive slope, but it acts on decrease of turbulent heat flux and is quantitatively canceled with contribution of wind speed.

[16] In general, it is well known that large ocean turbulent heat loss by changing wind speed and Qa is dominant and control SST variation in this region [Alexander and Scott, 1997]. However, the features described in this study suggest that the ocean heat loss by changing SST is dominant in this region. This feature is quite similar to results by Tanimoto et al. [2003]. They show that SST anomalies on decadal time scale play the primary role in determining turbulent heat flux anomalies in the part of North Pacific. The present study also suggests that the increase in SST determine the trend in turbulent heat flux in this region.

[17] Kelly [2004] investigated the relationship between oceanic heat transport and surface heat fluxes in the western North Pacific using NRA1 heat flux data during 1970–2000. She also showed anomalous surface heat loss in the 1990's, in particular late the 1990's. This result is consistent with our results. She also revealed that these surface flux anomalies on inter-annual time scale are balanced with lateral flux anomalies. This feature consists with our results that oceanic signals determine the trend of turbulent heat flux. Her results suggest that the remarkable increase in turbulent heat flux during 1990's over the Kuroshio/Oyashio region is related to inter-annual variation of heat flux. Actually the variability seems to include two independent inter-annual variations (early 1990's and late 1990's). On the other hand, the present results suggest that the dramatic pan-Pacific change in 1998/99 [Minobe, 2000; Hare and Mantua, 2000; Schwing and Moore, 2000] might caused the anomalous ocean to atmosphere heat flux.

[18] There is rapid decrease in turbulent heat flux over the after the increase in 1990's. The turbulent heat flux decrease from 270 W m−2 to 200 W m−2 during 2001–2003. It is possible that the anomalous heat loss occurred in late 1990's causes the rapid decrease in SST during 2001–2003.

[19] The quantitative difference between each product for turbulent heat flux is large even for yearly averages. In particular NRA1/2 products may overestimate turbulent heat loss. A systematic bias in the reanalysis products is also reported by several studies [e.g., Smith et al., 2001; Moore and Renfrew, 2002; Sun et al., 2003]. For example, Sun et al. [2003] showed that NRA1/2 overestimate turbulent heat loss in the Atlantic Ocean compared with flux obtained from buoy data. They revealed that the systematic bias in reanalysis products mainly is caused by flux algorithm. Tomita and Kubota (submitted manuscript, 2005) also conducted evaluation of various global latent heat flux data using buoy data including buoys moored around Japan including Kuroshio region. They also showed large overestimation of reanalysis latent heat flux compared with JMA buoy located in the Kuroshio region (29°N, 135°E). Although there is a systematic bias in each product, the variation pattern is similar at least on the interannual time scale and the increasing of turbulent heat flux year by year. However, for more quantitative study the accurate flux data should be required. Therefore, the accuracy of each turbulent heat flux data should be evaluated with in situ observations in this region in the future.


[20] This research was partly supported by Japan Aerospace Exploration Agency and the Category 7 of MEXT RR2002 Project for Sustainable Coexistence of Human, Nature and the Earth. We would like to thank two anonymous reviewers for their helpful comments and suggestions.