Madden Julian Oscillation

Content:


Surface Convergence and the Madden Julian Oscillation: Preliminary results from Probability Distributions for Surface Windfields

Ralph F. Milliff, Jan Morzel, Colorado Research Associates Division, NWRA
Roland A. Madden, National Center for Atmospheric Research

Collaborators: Mark Berliner, Tim Hoar, Doug Nychka, and Chris Wikle
Sponsors: NASA Ocean Vector Winds Science Team, NCAR Geophysical Statistics Project


Presented at IUGG, Sapporo, Japan, 9 July, 2003

Cover Page

Despite the global scale of the MJO in say perturbations in the 200 hPa geopotential height, propagation in the active convection regime of the MJO; i.e. the Indian and Western Pacific Ocean, should be evident in the smaller scale variability of convective processes at the surface.

The global coverage satellite data have reached scales sufficient to resolve structural details within convective complexes in the tropics. This talk will demonstrate that capability in the surface wind dataset from the SeaWinds scatterometer aboard the QuikSCAT satellite. We plan to use these data in conjunction with other satellite data; e.g. GMS brightness temperatures, to examine MJO propagation on shorter spatial scales than has been feasible in the past. It is perhaps timely to develop scatterometer wind analyses of the MJO. In December the Japanese Space Agency launched a second SeaWinds scatterometer on a spacecraft now called Midori-II. Once the Midori-II data become available, SeaWinds observations from the tandem missions will provide 90% global ocean surface wind field coverage in 6 hours.

QuikSCAT surface wind retrievals occur in 1600 km-wide swaths at 25 km resolution, in polar orbit patterns that sample a given tropical Indian Ocean or Western Pacific location twice a day; near 6am and 6pm local time. We will see a map depicting the daily coverage in a minute. The data within the swath have been shown to be reliable estimates of the synoptic winds, and they reveal many short-scale details of, for example, tropical storm systems (this is evident in papers presented by Dr. Tetsuo Nakazawa). Because limitations imposed by the polar-orbit and repeat time constraints, scatterometer winds do not uniformly cover the tropics and some blending with weather-center analyses must be done to provide synoptic maps, at 4-times daily resolution. This has been an activity of Milliff and co-workers, and we will use a new blended QuikSCAT and NCEP-FNL surface wind product in this paper.


Surface Wind BHM

The fields of blended winds in the tropics in this case occur as realizations from a posterior probability distribution that is the output of a so-called Bayesian Hierarchical Model (BHM). This is an emerging technique in atmosphere and ocean modelling in the presence of accurate data, and we will briefly review it in the context of the wind dataset here. For more of the details, there is a reference to a paper by Chris Wikle et al. (2001).

BHM makes explicit use of the fact that both our data and our process models for tropical dynamics are associated with different measures of uncertainty. For example in the case of satellite winds, the uncertainty is quantifiable from a very well-known measurement error model. In BHM, the uncertainties are used to define probability distributions for the data and dynamics. These distributions are combined in Monte Carlo procedures to produce a posterior probability distribution that is reflective of the total information content and uncertainty in both data and model. This is not traditional data assimilation, but it is related.

In the tropical wind BHM, we used relevant dynamics at two different scales to build the process model distribution. For large scales, we model the propagating modes of the equatorial &beta -plane, where the amplitudes for each mode are given distributions to reflect their uncertainty. For the smaller spatial scales that are sampled by the scatterometer, but absent from the analyses, we use a power-law relation observed in spectral transforms of the surface wind. This is imposed in a sequence of nested wavelet bases.

From the posterior distribution we extract 50 realizations of the wind field at 0.5° resolution, over a domain that spans the tropical Indian and Western Pacific Ocean from the horn of Africa to the date line. The fifty wind field realizations are provided 4-times daily for most of the QuikSCAT mission spanning the period December 1999 through March 2003. Each realization of the wind field, at each time, is physically consistent with the included dynamics including uncertainty in amplitude coefficients etc., and observationally consistent with the QuikSCAT winds where they are available and the analysis winds in the gaps. The process model includes propagation so the information from one timestep has an effect on subsequent timesteps.

To emphasize the probability distribution of surface winds, this graphic shows a 7 day time series of the histograms of zonal velocity at a single point in the domain (starting 5 January 2002, near 150° E on the equator). Each day, red distributions correspond to 0600 UTC, gold to 1200 UTC, green to 1800 UTC, and blue to midnight UTC. The local satellite overpasses are nearest the gold and blue distributions. The uncertainties are lowest then, so the histograms are most peaked around single values then (see the range and means projected on the "roof" of the diagram). Data for the intervening times are taken from the NCEP analyses and they are more uncertain on the small spatial scales of our grid. This is reflected in broader distributions in the zonal wind histograms that are red and green. Realize now that there are histogram time series like these for each grid point, and for the meridional wind component as well.

(Click on images for enlarged figures.)

We will use these realizations to reduce noise and amplify signal in the surface wind fields and convergence fields for individual MJO events. Before samples from a probability distribution were available, signal to noise issues had to be treated by compositing many MJO events from different times and background states. This can blur the short-scale information that might be critical to deducing propagation mechanisms for the MJO in the convection regime .


Data Stage Distributions:

Process Model Stage Distributions: Posterior Distribution Surface Winds:


Standard Deviations

This sequence of maps shows the spatial distributions of uncertainty in the surface wind ensemble from the BHM at four times during a single day. The contour interval is 0.2 m/s, from 0 to 2. The color bar has been created to emphasize low uncertainty (blue) in regions of QuikSCAT coverage. Note that in a daily average of these fields, most of the study region is covered by the satellite.


OLR anomaly time vs. longitude (Nov 2000)

We identify a single MJO event of interest here in the usual way. This is the time vs. longitude diagram of OLR anomaly averaged from 7.5° N to 7.5° S for the month of November 2000, across the Indian and W. Pacific ocean basins (52° E to the date line). The contour interval is 10 W/m2, from -110 to 70. The negative anomalies (blue and green colors) correspond to low OLR fluxes, and indicate coherent convective cloud complexes that are the familiar markers of the MJO. Not surprisingly, we see propagation at typical intraseasonal rates from 9-12 Nov near 60° E, to past 150° E by about the end of the month.


OLR maps Nov 10,15,20,25

This figure shows the map views of OLR (not anomaly) on 4 days separated by 5 days each during the November 2000 MJO. The contour interval is 10 W/m2, from 120 to 300. The propagation is harder to see in the map views of OLR because of the many sources of deep convection within the region of interest (e.g. off-equatorial convection, stationary convection at land-sea boundaries, etc.). Nonetheless the bulk of the minima in OLR anomaly (red) are seen to propagate eastward over period spanned by the 4 snapshots in the figure.


Zonal velocity maps Nov 10,15,20,25

These are the zonal velocity maps for the same 4 days in the November 2000 MJO. The contour interval is 2 m/s, from -12 to 12. Note that these are not anomaly zonal winds; these are 1 day averages of the ensemble of zonal wind realizations. The daily average zonal velocity field is an average of 200 realizations; 50 each from 4 times during the day. High spatial resolution admits small scale structures in the zonal winds, even in this average picture.

The boundary between westerly (red) and easterly (blue) zonal winds is sharply defined. The four panels depict eastward propagation in this boundary (westerly changing to easterly) over the 15 days in the period. The maritime continent complicates interpretation of propagation speed, as it did in the OLR maps in the previous slide.

Daily average fields of the meridional velocity for the period are also available, but we will jump instead to surface wind divergence.


Divergence maps Nov 10,15,20,25

Divergence of course, is the sum of velocity gradients and is therefore a very noisy field. The advantage of an ensemble of physically realistic surface wind fields for a single MJO event becomes evident. The divergence field for each realization from the ensemble is very noisy and the signals in the daily average to be discussed below are well hidden within the noise in each realization. But, because we are averaging over 200 fields to obtain the daily average maps shown here, the noise from the derivative operations has been driven down, and a clear physical signal emerges.

Again, we depict the same four days for the November 2000 MJO. The contour interval is 1x10-5 s-1, from -6x10-5 to 6x10-5. Features of deep blue(red) color correspond to strong surface convergence(divergence). The amplitude scale factor for each panel is 10-5 s-1. Realistic patches and bands of strong convergence are evident in the tropical Indian Ocean and along the region of the South Pacific Convergence Zone in the western Pacific. These features correspond with the OLR maps from a few figures ago.

The availability of a robust surface divergence field provides motivation for our future work. We are interested to look at average fields on 6-hourly timescales (i.e. averaging over 50 realizations) to see if we can deconvolve MJO propagation and stationary diurnal convection in the region. Our first looks at sub-daily maps of divergence are encouraging.


Nakazawa Super Cluster

A guiding hypothesis for our continued investigation of the high-resolution surface winds (and derivatives) in the convection regime of the MJO propagation is the now classic idea due to Tetsuo Nakazawa. Tetsuo elucidated a nested hierarchy of cloud variability within the large-scale eastward propagation of the MJO. His famous cartoon is reproduced here. Within the cloud super cluster bands that are the components of the eastward propagating envelope, Tetsuo identified westward propagation, and a convective cloud cluster life-cycle that occurs on diurnal to several day timescales. Since this insight, many related papers have emerged concerning the smaller-scale westward propagating features sometimes associated with the MJO. Many of the authors of these papers are in the room today (e.g. Takayabu, Wheeler, Kiladis, Liebmann, Hendon).

Our questions then are:

As we have noted we are only beginning to address these questions, but the preliminary descriptive results to date are encouraging.


Meridional wind time vs. longitude Nov 2000

Finally, we show this time vs. longitude chart of the 4-times daily averaged meridional velocity to suggest that the surface wind ensemble dataset does contain westward-propagating, smaller-scale structures, possibly associated with the MJO. We have looked at 5 Northern Hemisphere wintertime MJO event over the period of the QuikSCAT record and similar behaviors in the meridional winds are evident in all of them. This chart depicts the November 2000 MJO that we have been focusing on today; where day 0 corresponds to 10 Nov 2000. The contour interval is 1 m/s, from -6 to 6.

There is evidence of westward propagation outside the region of the MJO envelope of convection. In addition, the MJO is interrupting and restarting westward propagation on 2 to 3 day timescales in the meridional winds at the surface. The strongest meridional winds are associated in several instances with the time-space location of the 0 contour in the OLR anomaly as it propagates eastward.


Summary


References

Wikle, C.K., R.F. Milliff, D. Nychka and L.M. Berliner, 2001: "Spatiotemporal hierarchical Bayesian modeling: Tropical ocean surface winds", JASA , 96, 382-397.

Hoar, T.J., R.F. Milliff, D. Nychka, C.K. Wikle, and L.M. Berliner, 2003: "Winds from a Bayesian Hierarchical Model: Computation for Atmosphere-Ocean Research", submitted.


Preliminary Results from Ocean Model with Forcing from BHM Surface Winds

February, 2005

Preliminary results are presented from an ocean model, forced with surface winds from the Bayesian Hierarchical Model (BHM) described above. The modeling was performed at NCAR, by Bill Large and Gokhan Danabasoglu. The BHM winds are used in the tropical regions of the Indian and Western Pacific Oceans (up to 180° E in the Pacifc). The time period is November 1, 2004 - February 8, 2001. The ocean model was forced in 50 runs with the 50-member realizations of BHM winds.

The following figures show daily averages at intervals of 5 days, starting on November 5, 2000. The top panel shows zonal windstress along the equator, indicating the statistical distribution of the 50-member wind ensemble: solid dots are averages, the vertical boxes indicate +/-st.dev, and the vertical lines are drawn from min to max. The central panel shows the average zonal velocity (in cm/s) of all 50-members (contour interval of 20cm/s), with contour lines of st.dev. superimposed (contour interval of 1cm/s, thick lines starting at 3cm/s). The bottom panel shows the average potential temperature of all 50-members (contour interval of 2° C), with contour lines of st.dev. superimposed (contour interval of 0.05° C, thick lines starting at 0.10° C).

(Click on images for enlarged figures.)

Fig. 1.1 November 5, 2000, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.2 November 10, 2000, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.3 November 15, 2000, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.4 November 20, 2000, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.5 November 25, 2000, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.6 November 30, 2000, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.7 December 5, 2000, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.8 December 10, 2000, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.9 December 15, 2000, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.10 December 20, 2000, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.11 December 25, 2000, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.12 December 30, 2000, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.13 January 4, 2001, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.14 January 9, 2001, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.15 January 14, 2001, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.16 January 19, 2001, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.17 January 24, 2001, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.18 January 29, 2001, along the equator, for (a) 50-members, and (b) NCEP control run.

Fig. 1.19 February 3, 2001, along the equator, for (a) 50-members, and (b) NCEP control run.

The following figures represent depth profiles of zonal velocity (cm/s) at (0° N, 147° E), from 0-300m, with data from TAO observations, and the 50-member ensemble average and NCEP control run.

Fig. 1.20 Zonal Velocity depth profiles, for (a) Nov 10, and (b) Nov 20, 2000.

Fig. 1.21 Zonal Velocity depth profiles, for (a) Nov 25, and (b) Nov 30, 2000.

Fig. 1.22 Zonal Velocity depth profiles, for (a) Dec 5, and (b) Dec 10, 2000.

Fig. 1.23 Zonal Velocity depth profiles, for (a) Dec 15, and (b) Dec 20, 2000.

Fig. 1.24 Zonal Velocity depth profiles, for (a) Dec 30, and (b) Jan 10, 2001.

The changes of the zonal velocity depth profile at (0° N, 147° E) over time (Nov 1, 2000 - Jan 15,2001) are presented in the next figure. The zero U contour line is indicated with a dotted line, and the standard deviation of the 50-members of U are shown with 1cm/s contour lines in the central panel.

Fig. 1.25 Zonal Velocity vs. time at (0° N, 147° E).

last modified on March 11, 2005
Go Back