Temperature Interrannual Variations

Summary

Linear regression analysis was applied to the 13.5-year-long (January 1994-June 2007) MLO deseasonalized monthly mean lidar temperature time series for each 1-km altitude bin between 15 and 85 km.
The regression analysis included components representing the Quasi-Biennial Oscillation (QBO), El Niño-Southern Oscillation (ENSO), and the 11-year solar cycle. Where overlapping was possible, the results were compared to those obtained from the twice-daily National Weather Service (NWS) radiosonde profiles at Hilo (5-30 km) located 60 km east-north-east of the lidar site, and the four-times-daily temperature analysis of the European Centre for Medium Range Weather Forecast (ECMWF). The analysis revealed the dominance of the QBO (1-3 K) in the stratosphere and mesosphere, and a strong winter signature of ENSO in the troposphere and lowermost stratosphere (~1.5K/MEI). Additionally, and for the first time, a statistically significant signature of ENSO was observed in the mesosphere, consistent with the findings of recent model simulations. The annual mean response to the solar cycle shows two statistically significant maxima of ~1.3 K/100 F10.7 units at 35 and 55 km. The temperature responses to QBO, ENSO, and solar cycle are all maximized in winter. Comparisons with the global ECMWF temperature analysis clearly showed that the middle atmosphere above MLO is under a sub-tropical/extra-tropical regime, i.e., generally out-of-phase with that in the equatorial regions, and synchronized to the northern hemisphere winter/spring.

Details

Between January 1994 and June 2007, more than 2000 temperature profiles were measured. The monthly mean profiles were calculated by averaging all the profiles within one month (weighted at each altitude by their precision). Figure 1 shows a histogram of the number of profiles obtained each month since January 1994. There are usually 10-20 profiles per month.The monthly means were deseasonalized, i.e., for each altitude, the climatological mean, and annual and semi-annual cycles were removed by subtracting the overall mean and a two-component harmonic fit (12-month and 6-month periods) to the monthly mean time series.

Figure 1
Figure 1. Histogram plot of the number of routine profiles measured each month. The dotted line marks the minimum number of profiles required for considering a valid monthly average from observations

A linear regression analysis was applied to the time series at each altitude, which can be expressed as:

$$ T(t) = \alpha \times trend + \beta \times solar + \gamma_{1} \times QBO_{1} + \gamma_{2} \times QBO_{2} + \epsilon \times ENSO + residual$$
Figure 2
Figure 2. Reference time series of solar cycle (top), ENSO (middle), and two normalized QBO EOF time series QBO1 and QBO2 (bottom).

Figure 3
Figure 3. Time series of deseasonalized monthly mean temperature (black) shown at 5 km altitude intervals between 15 and 85 km (lidar, top), and 5 and 30 km (radiosonde, bottom), and their corresponding linear regression fits (red) reconstructed using the QBO, ENSO and 11-year solar cycle components. The black vertical bars denote the total uncertainty of the deseasonalized monthly mean temperature. The temperature perturbation scale is 2 K/km. The black vertical bars denote the total uncertainty of the deseasonalized monthly mean temperature.

where a, b, g1, g2, and e are coefficients with the form A1 + A2coswt + A3sinwt + A4cos2wt + 5sin2wt, w = 2p/(12 months), to account for seasonal dependence. The reference time series “ solar”, “QBO1”, “QBO2”, and “ENSO” in the equation above are shown in figure 2a-2c. Figure 2a shows the time series of the F10.7 cm solar radio flux, and figure 2b shows the Multivariate ENSO Index (MEI) characterizing the ENSO signal. The two QBO time series shown in figure 2c correspond to two empirical orthogonal functions (EOF) deduced from the equatorial stratospheric zonal winds at 7 different levels (70, 50, 40, 30, 20, 15, and 10 hPa) measured by radiosonde over Singapore.

Figure 3 shows the time series (plotted every 5 kilometers) of the deseasonalized monthly mean temperature (black) observed by the lidar (top) between 15 and 85 km, and by the Hilo radiosonde (bottom) between 5 and 30 km. The red superimposed curves are the corresponding reconstructed linear regression fits obtained by summing the trend, solar cycle, ENSO, and QBO components included in equation (1) (i.e., without the residuals). The linear regression analysis generally captures well most of the longer timescale variability below 50 km. The effect of the strong 1997/1998 El Niño (warm ENSO) event is very clear below 20 km in both lidar and radiosonde temperature time series.

The vertical profiles of percentage variance is shown in figure 4. The ENSO signal is clearly dominant in the troposphere and explains 60-70% of the total variance (55-60% just above the tropopause). Not surprisingly, the QBO is the dominant signal in the stratosphere. In the lower stratosphere, the QBO accounts for ~70% and ~45% of total variance for the radiosonde and lidar time series respectively. Compared to the ENSO and QBO, the solar cycle signature is generally weaker, and accounts for less than 15% of the total variance at all altitudes. The total percentage of explained variance is small in the mesosphere, most likely due to the decreasing lidar signal-to-noise ratios at these altitudes causing the increase in the amplitude of regression analysis residuals. One exception is at 72 km, with 22% explained by ENSO thus revealing a statistically significant ENSO signature in the middle mesosphere. Although an ENSO signature in the mesosphere has been predicted by numerical models, this feature is the first of this type ever observed experimentally.

fig_4
Figure 4: Vertical profiles of percentage of total variance explained by the 11-year solar cycle, ENSO, and the QBO components in the regression analysis applied to the lidar time series (left) and the radiosonde time series (right), The 1994-2007 mean lidar temperature profile is over-plotted in black solid curve.

QBO

Figure 5 shows the 2D color contours of the temperature QBO. An approximate one-year phase lag in 2000-2001 is clearly observed, leading to the reversal of the QBO phase before and after 2001, i.e., the cold QBO phase at 35 km is synchronized to the winter/spring of even years before 2000, and odd years after 2001. In the stratosphere, the calculated QBO amplitude of ~2 K maximizes at ~25, 35, and 45 km. The maximum found at ~45 km extends up to at least 56 km whereas a downwardly propagating structure is observed below. Though it is not as clear as that in the stratosphere, a QBO signature is also observed by lidar in the mesosphere, with amplitude of ~1 K at 60 km and ~2 K in the upper mesosphere.


fig 5
Figure 5. 2D contours of the temperature QBO perturbations reconstructed from the regression analysis applied to the lidar (a), radiosonde (b), and ECMWF (c) time series. The superimposed white contour lines in (a) and (b) show the equatorial zonal winds measured by radiosonde above Singapore (thicker contour lines denote westerlies and thinner contour lines easterlies)
Figure 6

Figure 6. Contours of global QBO at 24km and 36km obtained from ECMWF temperature regression analysis. The white straight lines denote the MLO latitude of 19.5N

Figure 6 shows the contour plots of the QBO signatures calculated at 24 km (top) and 36 km (bottom) from the ECMWF zonal mean temperature time series. Typical in-phase / out-of-phase perturbations of several degrees (K) near the equator and at mid- and high latitudes are clearly seen.As a result of the vertical sampling, the QBO at 36 km (top) appears out of phase with that at 24 km (bottom), and has weaker amplitude near the Equator and stronger amplitude at midlatitudes than at 24 km. At midlatitudes, the QBO at 36 km is well synchronized to the winter of each hemisphere.

ENSO

fig 7
Figure 7. Annual mean response to ENSO as derived from the regression analysis defined by equation (1) and applied to the lidar (black solid curve), radiosonde (blue dashed), and ECMWF (red dashed) time series. The horizontal bars indicate the 1s confidence level of the calculated fits
Figure 7 shows the annual mean of the temperature response to ENSO derived from the lidar (dark solid), radiosonde (blue dash), and ECMWF (red dot) time series. The annual mean mainly reflects the strong response calculated during northern hemisphere winter (see figure 8). Two positive maxima of 1.2 K/MEI and 1 K/MEI were observed near 10 and 70 km respectively, and a negative maximum of -1 K/MEI was found near 20 and 42 km. The temperature responses to ENSO obtained from lidar, radiosonde, and ECMWF below 30 kmshow excellent agreement. The agreement between ECMWF and radiosonde was expected however because the ECMWF analysis includes the radiosonde profiles in their assimilation scheme. The response to ENSO as a function of altitude and season is plotted in figure 8. The positive response in the troposphere and middle mesosphere, and negative response in-between, are statistically significant only in the northern hemisphere winter, with maxima  of 1.5 K/MEI, -1.5 K/MEI, 2.5 K/MEI, and 2.5 K/MEI near 10, 20, 42, and 73 km respectively. One last, but not least, result is the statistically significant response in the mesosphere. Though such a response has been predicted by recent modeling, it is reported from observations for the first time here. The National Center for Atmospheric Research (NCAR)/WACCM model simulations suggest that a middle atmosphere temperature response to ENSO can be produced due to the anomalous upward propagation of planetary and gravity waves. This is especially significant in the northern hemisphere winter when the westward planetary waves can propagate upward through the stratospheric westerly zonal mean zonal winds.

The dissipating or breaking of planetary waves in the upper stratosphere and lower mesosphere can decelerate the westerly mean zonal wind and facilitate the anomalous propagation of eastward gravity waves into the mesosphere. The resulting residual mean meridional circulation is enhanced in the stratosphere during El Niño events, causing anomalous warming in the polar regions and cooling in the tropics in the stratosphere, and vice-versa in the mesosphere. These modeling results are consistent with our observational results.

fig_8
Figure 8. Seasonally-dependent response to ENSO as calculated by the regression analysis defined by equation (1) for the lidar time series (left), and the radiosonde time series (right). The shaded regions indicate that the results are not significant at the 2s confidence level, the blue dash contour lines denote the negative values and the red solid contour lines denote positive values. The dark dashed horizontal lines mark the lower and upper boundaries for each time series.

The annual mean ENSO temperature signature at 20 and 10 km obtained from the ECMWF time series is shown in figure 9 as a function of longitude and latitude. At 20 km, a zonally symmetric negative response in the tropics, and positive response with a  zonal wave number 1 in the high latitudes of the southern hemisphere are clearly identified, in agreement with the results from satellite observations in the lower stratosphere. At 10 km, a wave like ENSO pattern is identified with typical massive positive perturbation in the tropical eastern pacific basin surrounded by negative perturbations in the subtropics and midlatitudes, and consistent with early satellite observations. At 10 km, a wave like ENSO pattern is identified with typical massive positive perturbation in the tropical eastern pacific basin surrounded by negative perturbations in the subtropics and midlatitudes, and consistent with early satellite observations. As pointed out by the black plus sign on figure 9, MLO is located at the northern periphery of the zonally symmetric negative temperature response at 20 km, and at the northwestern periphery of the geographical.

fig_9
Figure 9. 2D longitude-latitude contour plots of the annual mean response to ENSO at 20 (top) and 10 km (bottom) calculated from the ECMWF time series. The geographical location of MLO is marked by a black plus sign

11-year Solar Cycle

Figure 10
Figure 10. Same as figure 7, but for the 11-year solar cycle

Figure 10 shows the temperature response to the 11-year solar cycle derived from the regression analysis on the time series of lidar (black solid), radiosonde (blue dash), and ECMWF (red dash dot), as a function of altitude. The monthly mean solar cycle response (not shown) was found to reach 2 K/100 F10.7 unit near 35 and 55 km, and -1 K/100 F10.7 unit near 10 km in the winter, and 1.5 K/100 F10.7 unit near 45 km in the summer. The seasonal response at 55, 35 and 10 km yields a statistically significant annual mean response of 1.4 K/100 F10.7 unit, 1.3 K/100 F10.7 unit, and -0.7 K/100 F10.7 unit respectively. in the stratosphere, it is attributed to direct absorption of solar radiation by ozone

References

Li, T., T. Leblanc, and I. S. McDermid (2008), Interannual variations of middle atmospheric temperature as measured by the JPL lidar at Mauna Loa Observatory, Hawaii (19.5°N, 155.6°W), J. Geophys. Res., doi:10.1029/2007JD009764, in press.

Back to Top