Recent variations in surface specific humidity in the warm season over Japan

Abstract

Recent variations in the warm seasonal mean of surface specific humidity (q _sfc) in Japan were studied using 34 years of data (1976–2009) from 150 meteorological observation stations of the Japan Meteorological Agency. The regional trend in the warm seasonal mean q _sfc at 3 local standard time (LST) (UTC + 9 h), averaged over Japan, increased significantly with small changes in the relative humidity: the rate of increase was +1.93 g/kg per 100 years. The warm seasonal mean q _sfc over Japan, calculated using four-times-daily data, increased slightly: the rate of increase was +0.95 g/kg per 100 years. About 16 % of the stations exhibited a significant increasing trend in warm seasonal mean q _sfc calculated using four-times-daily data; these stations were concentrated in eastern Japan. The increasing trends at 40 % of the stations were more evident for the mean q _sfc at 3 LST than at other local times. The local time at which the diurnal cycle of the mean q _sfc reached a maximum was 15 LST in period 1 (1976–1992) and tended to be delayed until up to 21 LST in the interior in period 2 (1993–2009). In addition, the rate of increase in mean q _sfc at nighttime was larger around the coastal and mountainous areas on days with well-developed thermally induced local circulation.

Appendix

A large discontinuity in t _sfc and/or q _sfc may lower the quality of analysis. Time series of t _sfc and q _sfc for each station were statistically tested separately using “two-phase regression models.” This method was proposed to detect undocumented discontinuities in the climatological time series by Easterling and Peterson (1995) and revised by Lund and Reeves (2002) and Wang (2003). The revised method has been applied previously in several studies (e.g., Vincent and Coauthors 2005; Isaac and Wijngaarden, 2012). The procedure consisted of the application of two regression models. Model 1 was a simple linear regression and was fitted to the time series of t _sfc and q _sfc:

$$ {y}_i={a}_1+{b}_1\cdot {t}_i+{e}_i $$

(1)

Here, y _i is the warm seasonal mean value of t _sfc or q _sfc for year t _i , and e _i is the residual. Model 2 was also a linear regression but was fitted to the time series divided by a potential step at step year t _s. It is given as

$$ {y}_i={a}_2+{b}_2\cdot {t}_i+{c}_2\cdot I+{e}_i $$

(2)

Here, c ₂ is the step magnitude, and the variable I is the value 0 (1) before (after) the step year t _s. These two models were compared using an F test to determine if the step substantially improved the fit of the model 2:

$$ F=\frac{\left({\mathrm{SSE}}_1-{\mathrm{SSE}}_2\right)/1}{{\mathrm{SSE}}_2/\left(n-3\right)} $$

(3)

Here, SSE₁ and SSE₂ are the sum of squared error for models 1 and 2, respectively, and n is the number of data points. When the F statistic was greater than the 95th percentile (Wang 2003), model 2 was accepted, and it was concluded that there was a significant step in the time series at year t _s.

Figure 13a, b shows the time series of the number of stations with a significant step in the time series of t _sfc and q _sfc, respectively. Of the 150 stations used in the analysis, only three stations were found to exhibit a significant step with negative step magnitude in t _sfc data (temperature increased after the step year), and no stations exhibited a positive step magnitude. However, the nine stations with positive step magnitude in the q _sfc data were concentrated from 2000 to 2002, when averaged q _sfc over Japan had decreased as shown in Fig. 2a. These significant steps seem to correspond to real large-scale atmospheric variation, not to artificial changes. Eight stations were found to exhibit a negative step magnitude; however, no systematic distribution is apparent in Fig. 13b. Thus, it is concluded that, although the sensors (thermometers and psychrometers) were replaced several times for the JMA meteorological stations, these replacements did not produce significant discontinuities in t _sfc and q _sfc. Therefore, this study does not take into account discontinuities in t _sfc and q _sfc.

Fig. 13

Time series for the number of stations with a significant step in a t _sfc, b q _sfc, and c, d u and v components of surface wind. A bar indicates a positive (negative) step magnitude (c ₂) in Eq. 2

Figure 13c, d shows the time series for the number of stations with a significant step in the evolution of the u and v wind speed components at AMeDAS stations, respectively. Almost half the stations (46.3 %) have a significant step in these time series throughout the analysis period, although discontinuities in wind data were only reported for 15 stations (8.6 %, JMA 2014c). Environmental changes around the stations must be cause of these significant steps.

Figure 14 shows a histogram of step magnitudes for u and v components. In spite of a large number of significant steps, discontinuities were generally not large and step magnitudes of less than 0.2 m/s account for 50 % of the AMeDAS stations. Nevertheless, this study took the conservative approach of not including trends when time series showed a significant step.

Fig. 14

Histogram of step magnitude for u and v components of surface wind