Implications of 30–70-Day Boundary Effe
Transcription
Implications of 30–70-Day Boundary Effe
1338 JOURNAL OF CLIMATE VOLUME 17 MM5 Modeling of the Madden–Julian Oscillation in the Indian and West Pacific Oceans: Implications of 30–70-Day Boundary Effects on MJO Development WILLIAM I. GUSTAFSON JR.* AND BRYAN C. WEARE Atmospheric Science Program, Department of Land, Air, and Water Resources, University of California, Davis, Davis, California (Manuscript received 25 March 2003, in final form 7 October 2003) ABSTRACT The results of an experiment designed to isolate the initiation phase of the Madden–Julian oscillation (MJO) from 30–70-day boundary effects is presented. The technique used to accomplish this involves employing the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model (MM5), as first presented in the companion paper to this paper. Two runs, each 2 yr long, are integrated forward from 1 June 1990. The first run, called the control, uses the unmodified National Centers for Environmental Prediction (NCEP)–NCAR reanalysis (NRA) dataset for boundary conditions. The second run, called the notched, uses the same NRA dataset for the boundary conditions, with the exception that all signals with periodicities in the 30–70-day range have been removed. Any signals in the 30–70-day range subsequently generated by the notched run are then solely due to signals generated from within the model domain or from signals entering through the domain boundaries with frequencies outside of the MJO band. Comparisons between 2-yr means from each run indicate that filtering the boundaries does not significantly modify the model climatology. The mean wind structure, thermodynamic state, and outgoing longwave radiation (OLR) are almost identical in the control and notched runs. A 30–70-day bandpass filter is used to isolate MJO-like signals in the runs. Comparisons of 30–70-day bandpassed zonal wind, moist static energy (MSE), and OLR reveal that the notched run develops many of the expected characteristics of MJO episodes, but with a weaker signal. Largescale, organized structures develop that possess seasonal shifts in amplitude, mirroring observed MJO activity, have opposite wind directions in the upper and lower troposphere, and propagate eastward during most strong episodes. The results suggest that neither remnants from previous MJO episodes nor extratropical feedbacks within the MJO time band are necessary for MJO initiation. However, the control run is more organized than the notched run, implying that 30–70 signals outside the model domain influence the MJO signal. There is also some evidence that the recharge–discharge mechanism plays a role in MJO formation. 1. Introduction The Madden–Julian oscillation (MJO) encompasses variations of many atmospheric parameters that propagate eastward along the equator and have periodicities around 30–60 days. The MJO has long been described in terms of convection and equatorial wave dynamics (Madden and Julian 1994). Hendon and Salby (1994) connect the Kelvin–Rossby coupled wave system of Gill (1980) with observations through analysis of satellite outgoing longwave radiation (OLR) and temperature measurements. However, there are multiple hypotheses regarding initial MJO organization, including local instability processes (Fasullo and Webster 2000; Hu and Randall 1994, 1995; Kemball-Cook and Weare 2001), * Current affiliation: Pacific Northwest National Laboratory, Richland, Washington. Corresponding author address: Dr. William I. Gustafson Jr., Pacific Northwest National Laboratory, 3200 Q Ave., MSIN K9-30, Richland, WA 99352. E-mail: william.gustafson@pnl.gov q 2004 American Meteorological Society extratropical waves (Matthews and Kiladis 1999), and circumnavigating equatorial waves (Lau and Peng 1987). This paper is a companion to Gustafson and Weare (2004) and presents preliminary results from a new approach to investigating the MJO initiation process. As opposed to past modeling studies of the MJO, this approach uses a regional model where signals entering and leaving the domain can be controlled. This is done by filtering the lateral and surface boundary conditions to remove waves of certain types. In this paper, influences within MJO time scales are removed to determine how feedbacks and wave structures in the MJO time band from outside the MJO initiation region alter MJO development. These filtered influences include both feedbacks between the tropical and extratropical regions on MJO time scales, remnants of previous MJO events that have circumnavigated the equator, and feedbacks from far-field wave structures that would otherwise develop outside of the model domain and alter the MJO structure. In this paper these filtered influences will cumulatively be referred to as ‘‘boundary effects’’ and specifically refer to these influences for signals in the 30–70-day time band. 15 MARCH 2004 GUSTAFSON AND WEARE Studies to date are not convincing regarding the role of previous MJO episodes. One proposed theory for what determines the periodicity of the MJO is the influence of MJO waves that circumnavigate the equator and organize the next MJO convective episode as the waves pass through the Indian Ocean. There is at least some evidence that this may occur for stronger MJO episodes (Knutson and Weickmann 1987; Matthews 2000; Rui and Wang 1990). Modeling studies also tend to show this type of behavior (e.g., Bladé and Hartmann 1993). However, Hendon and Salby (1994) showed that the autocorrelation time between successive MJO episodes is only one period for the convection, which they interpret as evidence that circumnavigating waves are not the physical mechanism determining the MJO period. Furthermore, Bladé and Hartmann’s (1993) model experiments suggest that circumnavigating waves are not a prerequisite for periodic MJO-like signals to develop. However, these results are not conclusive because their model does not completely prevent previous waves from circumnavigating the globe, allowing for very small remnants to travel around the globe and possibly phase lock with the next episode. Questions as to the importance of circumnavigating waves is one area in which the new regional modeling technique excels. By using a regional model, one can filter the boundary conditions to remove nearly the entire influence of 30–70-day boundary effects. This provides a truer test than was possible in Bladé and Hartmann’s model of what determines the periodicity of the MJO. In the case presented in this paper, seasonality is also maintained in the boundary conditions so that the seasonal fluctuations of the MJO can be observed. Because the MJO signals are filtered from the boundary forcings, this also means that intraseasonal oscillations that develop within the model will be forced to have a scale on the order of the model domain size or smaller. Thus, a possible disadvantage of this approach, as presented here, is that the oscillations will not resemble the MJO exactly, but they do maintain many of the important characteristics. 2. Data and methodology a. Model description The model used for this study is the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model (MM5) version 3.4 (Grell et al. 1995). The specific details of the model configuration can be found in the companion paper of this issue by Gustafson and Weare (2004). The model domain extends across the tropical Indian and west Pacific Oceans, approximately from 248S to 248N and from 448 to 1818E. This allows for waves to develop within the model that have sizes equivalent to what would be wavenumber 3 for a global model. The specific region has been chosen based upon the 1339 life span of the MJO with the convective activity developing in the western Indian Ocean, propagating east, and decaying near the date line. The time period for the runs presented in this paper covers a 24-month period beginning 1 June 1990. The model lateral and sea surface temperature (SST) boundaries are forced using variations of the National Centers for Environmental Prediction (NCEP)–NCAR reanalysis (NRA) dataset, which uses a copy of the Reynolds SST reanalysis (Kalnay et al. 1996). As shown in Gustafson and Weare (2004), the model is able to reasonably reproduce the mean equatorial climate as well as intraseasonal oscillations that resemble the MJO, based on comparison to the NRA and satellite OLR observations. The primary differences between the model and NRA means are that at 850 hPa the zonal wind has an overall easterly bias in the model as well as difficulty reproducing the wind patterns in the Indian monsoon region. Also, as compared to satellite OLR, the mean OLR peaks along the equator are shifted toward the edges of the domain leading to a reversal in the equatorial OLR gradient. The model-generated 30– 70-day intraseasonal oscillations reproduce many of the observed characteristics, with the noted exception of weaker and less organized OLR anomalies than in observations. b. Experimental procedure The goal of this experiment is to determine the role of 30–70-day boundary effects on successive MJO formation. To do this, two runs were made with MM5. The first, called the control, is a standard model run designed to determine the model climatology and the structure of MJOs within the model domain. The second, called the notched, differs from the control in that the lateral and SST boundary forcings are filtered to remove MJO signals. This filtering is done using an inverted 1201-point (equivalent to 301 day) Lanczos bandpass filter (Duchon 1979) to remove signals in the 30–70-day time band from the 6-hourly NRA data. The response function for this ‘‘notch’’ filter is shown in Fig. 1. By removing the 30–70-day signals, the MJO is effectively removed from the forcing data, allowing, as will be seen, the notched run to develop intraseasonal signals independently of any MJO signals on the boundaries while still experiencing the same seasonal shifts and higher-frequency forcings that may influence MJO formation. The portion of the signal removed with the notched filter has zero mean and does not change the overall signal outside of the filtered band. However, removing the 30–70-day signals reduces the average variance of the 200-hPa wind speed along the model boundaries by a factor of 0.89. With this filtering, the notched and control runs are compared in a manner similar to what is done in general circulation model perturbation/control experiments (Roeckner et al. 1999). 1340 JOURNAL OF CLIMATE FIG. 1. Response functions for the Lanczos digital filters used in this paper. The solid line is the 30–70-day notch filter constructed with weights over 301 days used to filter the model boundary conditions. The short dashed line is the 30–70-day bandpass filter constructed with 201 days of weights used to isolate the MJO in the model output. The long dashed line represents a perfect 30–70-day bandpass filter for comparison. c. Data postprocessing To identify MJO-like signals in the model runs, a series of diagnostic and statistical analyses are performed. Unlike most MJO investigations, the model runs in this experiment only cover a 2-yr time period and only one-third of the longitudinal extent of the globe. This poses strong limitations on the type of analyses performed, as well as on the establishment of significance for the statistical tests. For example, spatial filtering to isolate wavelengths close to wavenumber 1 or 2 is not possible. Also, over the 2-yr period, only sixteen 45-day periods can occur, limiting the number of degrees of freedom. Because of these and other limitations, the presented analyses have been chosen to provide a range of approaches, focusing on several wellobserved features of the MJO: eastward propagation, periodicity, organized vertical and horizontal wind structure, and organized convection. In order to simplify interpretation, data from the completed model runs are interpolated horizontally and temporally to a daily 2.58 grid using area-weighted averages of the corresponding grid cells. The original data, which is on sigma levels, is interpolated vertically to pressure levels. The resulting dataset is used when doing the bandpass and other calculations. To isolate the MJO signal a 201-point (equivalent to 201 days) 30–70-day Lanczos bandpass filter (Duchon 1979) is applied to the regridded, daily model output. The response function for this filter is shown in Fig. 1. The choice of 201 points is a compromise between maintaining enough data for analysis and the effectiveness of the filter. VOLUME 17 As discussed above, a suite of diagnostic and statistical techniques are used in this study to analyze the 30–70-day signals in the model output with each technique being chosen to identify one or more salient features of the MJO with some overlap between techniques. Using multiple techniques helps to identify the features based upon different basic assumptions and, therefore, gives more confidence in the results even though the time series are short. Brief descriptions of each technique follow with more detailed descriptions available in the accompanying paper by Gustafson and Weare (2004). Spectral analysis is used to identify the presence of activity in the 30–70-day time band for a series of eight points along the equator. These points (see Fig. 3a) are the same points used in Gustafson and Weare (2004), and the bandwidth is twice the Rayleigh frequency f R where f R 5 1/731 day 21 . The range within which the two spectra would be considered the same is determined based upon the 95% significance level, compared to a background red-noise spectrum, leading to a 5% indication that the two spectras would be different. Crossspectral analysis is also performed to identify how the activity in the 30–70-day band is related between these eight equatorial points. The resulting squared coherence between the points identifies the tendency for anomalies to fluctuate concurrently at each point and the phase difference between the points helps to identify the anomalous wave structure moving between the points. Propagation and, to an extent, spatial scale are seen using Hovmöller diagrams, although their interpretation tends to be quite dependent on the chosen contour levels. The structure of the anomalies is further elucidated through empirical orthogonal function (EOF) analyses of time series composed of vertical profiles and time series of variables on pressure levels. The EOFs reveal patterns of variance within the series, and the associated principal component (PC) time series identify the strength of the revealed pattern at any given time. The PCs are further evaluated using the Matthews EOF technique (Matthews 2000) where the PCs are used to generate amplitude and phase indices. The amplitude index serves to identify active MJO periods during the model runs, and the phase index identifies the approximate periodicity and propagation characteristics of the anomalies. In addition to the analysis techniques used in Gustafson and Weare (2004), singular value decomposition (SVD) analysis is used in this paper (Bretherton et al. 1992; Wallace et al. 1992). While similar to EOFs, in that SVDs isolate coherent patterns of variance in time series, the SVDs specifically isolate patterns of strong covariance between two different variables as opposed to patterns of variance for a single variable. This gives the advantage of being able to determine if strongly coherent covariance patterns develop between variables, such as OLR and 200-hPa zonal wind, and to see how this coupling changes over time. The spatial domain 15 MARCH 2004 GUSTAFSON AND WEARE 1341 FIG. 2. Two-year mean wind vectors and magnitudes for the notched run at (a) 200 and (c) 850 hPa. Increased shading indicates higher wind speeds, and the vectors are scaled based upon the 10 m s 21 reference vector. Also shown are the differences between the notched and control run wind vectors and magnitudes for (b) 200 and (d) 850 hPa. The reference vector for these plots is 1 m s 21 and the shading represents the difference in magnitude between the two runs. Vectors are plotted every 2.58 in latitude and every 58 in longitude. used for the SVD analysis extends across the full width of the domain but is limited to latitudes between 12.58S and 12.58N. Tests using the full latitudinal extent of the domain revealed that excess noise was introduced into the SVD results, particularly for the notched run along the north and south boundaries. Therefore, the region of transition between the model interior and the north and south lateral boundaries was excluded. Shading on the heterogeneous correlation plot is shown for correlations exceeding 0.5, which roughly corresponds to significant correlations at the 95% confidence level. 3. Model results and comparisons a. Reproduction of climatology To determine if filtering the boundaries has greatly modified the mean state of the model climatology a comparison is done for 2-yr means of the control and notched runs. By establishing that the mean state is similar between the two runs, it follows that the conditions under which the MJO forms are also similar. The first comparison is between the winds at 200 and 850 hPa. Figure 2 shows the 2-yr mean winds for the notched run and the difference between the two model runs. Overall, the two runs have very similar mean wind patterns. At 200 hPa the flow is characterized by easterly winds along the equator and westerly winds poleward of 158. Differences between the control and notched runs show little organized, large-scale patterns. The wind magnitudes at 200 and 850 hPa are not statistically different at the 95% level for more than 98% of the points based on Student’s t test and the magnitudes of the differences are almost entirely less than 1 m s 21 . At 850 hPa the flow is characterized by easterly winds south of the equator and for the entire eastern half of the domain. The differences at 850 hPa are almost all smaller than 0.5 m s 21 , and the direction of the differences is much more zonal than at 200 hPa, but again highly unorganized. It should be noted that the equatorial flow in the west Pacific differs from the equivalent region in the NRA. Along the equator in the west Pacific the model 850-hPa zonal wind is more easterly than in the NRA (Gustafson and Weare 2004). This could alter the feedback mechanisms associated with MJO propagation and development by providing an environment more susceptible to wind-induced sensible heat exchange feedbacks (Emanuel 1987). Figure 3 shows the mean OLR for the notched run and the difference between the notched and control runs. The convection, as represented by the OLR, occurs primarily along the equator and is strongest east of Papua, New Guinea. The control and notched OLR patterns are very similar with the differences smaller than 10 W m 22 . The differences show no large-scale organization and appear to be randomly distributed throughout the domain. Similar to the winds, the OLR means are not statistically different at the 95% level for more than 98% of the points, based on Student’s t test. However, Gustafson and Weare (2004) have noted that the peak model OLR in the Indian Ocean is west of the observed location leading to a reversal of the longitudinal OLR gradient in this region. The model also has a strong negative zonal gradient in the eastern West Pacific where the observations have a near-zero gradient. Comparisons between maps of mean temperature, mixing ratio, precipitable water, and moist static energy (MSE) (not shown) also reveal that the control and notched runs have very similar thermodynamic mean 1342 JOURNAL OF CLIMATE VOLUME 17 structures, the vertical wind structure, the eastward propagation, and the periodicity of the intraseasonal signals. 1) 30–70-DAY FIG. 3. Two-year mean of (a) notched run OLR and (b) the difference between the mean notched and control run OLR. The contour interval is (a) 15 and (b) 3 W m 22 . Letters along the equator indicate longitudes used for spectral and other analyses, as discussed in the text. The lettered longitude locations are as follows: A, 458E; B, 608E; C, 82.58E; D, 102.58E; E, 1258E; F, 1458E; G, 167.58E; H, 182.58E. states. At 850 hPa the temperatures agree almost entirely within 0.25 K, the mixing ratio difference is almost entirely below 0.5 g kg 21 , and the MSE difference is mostly smaller than 0.6 3 10 3 J kg 21 , or about 0.2% of the mean value. The mean 1000–700-hPa precipitable water values differ by only 1.1 kg m 22 . In the upper troposphere the differences are again small with the temperature at 200 hPa generally within 0.2 K and the mixing ratio at 300 hPa within 0.03 g kg 21 for the two runs. Vertical profiles of mean MSE (not shown), similar to Fig. 5 in Gustafson and Weare (2004), also show very small differences. These differences are two to three orders of magnitude less than the mean values. b. MJO comparisons Even though 30–70-day intraseasonal time signals have been removed from the model boundary forcings, 30–70-day signals still develop within the model domain. This is shown clearly through spectral analysis of the 200-hPa zonal winds for eight points along the equator (Fig. 4). On the western and eastern domain edges, points A and H, the spectral power for the notched run shows gaps in the 30–70-day range due to the boundary filtering. For the interior points, B through G, the model develops clear signals in the 30–70-day time band for both the control and notched runs. Although the MJO signal is generally weaker in the notched run, the notched run spectral power is within a magnitude of the control run power. Given that intraseasonal time signals develop in the model independently of the boundaries, the next step is to determine what form these signals take and whether or not they are similar to the control run MJO. The specific MJO aspects to be identified are the large-scale, organized ZONAL WIND Hovmöller diagrams of 30–70-day 200-hPa zonal wind (Fig. 5) give a qualitative idea of the intraseasonal signals in the two runs, specifically identifying eastward propagation and relative oscillation strength. Periods of propagating westerly winds exist during boreal fall 1990, summer 1991, and to a lesser extent for the notched run, during fall/winter 1991. Anomalies propagate to the east during many of these periods with the exception of some weaker anomalies. Comparing the notched run 30–70-day zonal winds with the control run winds reveals several differences. First, the notched oscillations often develop farther east than in the control, roughly around 708–808E, whereas oscillations in the control run typically enter from the western domain boundary. Second, the notched oscillations are strongest in the western half of the domain and are less likely to propagate as far across the domain as in the control run. In particular the summer oscillations seem to weaken over the Maritime Continent in the notched run. Third, the 30–70-day 200-hPa zonal wind anomalies are weaker in the notched run, with a maximum value of 5.3 m s 21 , as opposed to 9.4 m s 21 in the control run. At 850 hPa (not shown), the difference between maximum values is less; the maximum values are 4.4 and 6.9 m s 21 for the notched and control run, respectively. The Matthews EOF technique is used to quantitatively identify active 30–70-day wind oscillations and the corresponding periodicity and propagation. This technique is more quantitative and less subjective than the Hovmöller plots. Figure 6 shows the two EOF maps used to construct the 30–70-day 200-hPa zonal wind Matthews indices. These maps have complimentary patterns with one map having mostly the same sign (EOF1 for the control, and EOF2 for the notched), and the other map having a dipolelike pattern (EOF2 for the control, and EOF1 for the notched). When the two EOF maps, along with the corresponding PCs, are used to reconstruct the zonal wind time series, one can see the eastward propagation of 30–70-day anomalies moving across the domain. In the notched run the first two EOFs are much noisier due, in part, to the smaller region over which propagation occurs. Animations of the reconstructed notched run time series shows most of the dominant, spatially organized patterns forming in the middle of the Indian Ocean and then propagating both west and east, but generally not farther east than the Maritime Continent. At 850 hPa the reconstructed notched run zonal wind time series also has coherent patterns that form in the mid-Indian Ocean, as well as occasional signals that propagate eastward from near the western boundary. Figure 7 presents the Matthews amplitude indices for 15 MARCH 2004 GUSTAFSON AND WEARE 1343 FIG. 4. Spectral power of 200-hPa zonal wind averaged from 58S to 58N for points A through H as indicated in Fig. 3a. Solid lines represent the notched run and dashed lines the control run. The shaded region represents periodicities between 30 and 70 days. The cross in the legend indicates the bandwidth (-) and the range ( | ) beyond which the two spectra are considered not to be the same at the 5% level. the 200- and 850-hPa zonal wind. The most notable feature of the notched run amplitude indices is the strong seasonality of the MJO signal, particularly at 200 hPa. During the two boreal winters and intervening summer the amplitude is much stronger than during the other portions of the year. At 850 hPa the winter periods have somewhat increased amplitude, but the summer peak is almost twice as big as the winter peaks. The 850-hPa summer peak occurs about 2 months prior to the corresponding 200-hPa peak, indicating some degree of separation between the activity at each level. This does not happen in the control run where the upper and lower peaks occur simultaneously. Another difference is that the notched run amplitude index magnitude ranges between about 20% and 60% of the control run magnitude during the peak times. Figure 7 also presents the phase indices for the 200and 850-hPa zonal winds. In the control run the phase advances smoothly from 08 through 1808 and from 21808 back to 08 during strong amplitude index peaks. This is indicative of the eastward propagation seen in the Hovmöller diagrams (Fig. 5). In the notched run the phase also advances during the summer peaks and for the 200-hPa winter 1990 peak. However, the phase decreases with time during winter 1992 and at 850 hPa during winter 1990. This decreasing phase is due to the strongest signal in the EOF maps being on the western half of the domain. During this first winter the strong MJO signal develops in the Indian Ocean and then propagates both east and west. Because the 850-hPa phase index identifies the strongest motion as in the western portion of the domain, the index identifies the westward propagation. The eastward propagating westerly wind signal seen in the corresponding Hovmöller is missed. During the second winter, Hovmöller diagrams of the notched zonal wind indicate that the strongest, organized activity is east of 1108E so the phase index again misses the eastward propagation. Further evidence of notched run propagation can be seen using cross-spectral analysis. Table 1 lists the phase 1344 JOURNAL OF CLIMATE VOLUME 17 FIG. 5. Normalized Hovmöller plots of 30–70-day bandpassed 200-hPa zonal wind averaged from 88S to 88N for the (a) notched and (b) control runs. Increased shading density reflects greater magnitudes and positive values are also contoured. Each plot is scaled by the maximum value in the series: 5.3 m s 21 for the notched run and 9.4 m s 21 for the control run. FIG. 6. First two EOF maps of 30–70-day bandpassed 200-hPa zonal wind for the (a), (b) control and (c), (d) notched runs. Positive values are shaded with solid contours, negative values are white with dashed contours, and the contour interval is 0.03 (units are incorporated into the PCs). 15 MARCH 2004 1345 GUSTAFSON AND WEARE FIG. 7. (a), (b) Matthews amplitude indices for the 30–70-day bandpassed zonal wind at 200 and 850 hPa, respectively. The amplitude index includes a 45-day running mean for smoothing. (c), (d) Matthews phase indices for the 30–70-day bandpassed zonal wind at 200 and 850 hPa, respectively. The phase index is only plotted for times when the smoothed amplitude index exceeds 35% of its maximum value. Black lines are for the notched run and gray lines are for the control run. angle and squared coherencies for base points C and E compared to the remaining points A through H. Overall, the two model runs have very similar phase angles for the 32 possible combinations of points; twenty-four points match within the 95% confidence intervals. Base point C shows the best propagating characteristics with the phase angles advancing from point A through G in the control run, and for the 58.2-day periodicity in the notched run. The 38.8-day periodicity has advancing phase angles up to point F. At base point E, both runs have more localized propagation with the phase angles advancing from point D through G for both periodicities. The number of points having a significant squared coherency at the 95% level is similar between the two runs. In addition to propagation, the Matthews phase indices can be used to roughly determine the periodicity of the intraseasonal signal in the models. This is done by measuring the time between successive zero crossings of the phase index. At 850 hPa the resulting periodicities for times with advancing phase angles range from 36 to 44 days in the control run and from 41 to TABLE 1. Squared coherency (Coh. 2 ) and phase from cross-spectral analysis of 30–70-day bandpassed 200-hPa zonal wind (U200bp) averaged from 58S to 58N for the two base points C and E and the two periodicities 38.8 and 58.2 days. Regions A through H are as indicated in Fig. 3a. Bold values indicate that the squared coherency differs from zero at 95% confidence. Ninety-five-percent confidence intervals for the phase are shown when they can be calculated. Base point 5 C Notched U200-bp Region A B C D E F G H 38.8 days Phase 255 17 0 28 50 6 52 6 43 60 65 6 16 51 2169 61 6 35 Base point 5 E 58.2 days Coh. 2 Phase 0.44 0.47 1.00 0.88 0.69 0.34 0.10 0.52 265 26 0 26 51 6 6 6 6 6 59 176 6 72 6 0.44 0.78 1.00 0.97 0.96 0.68 0.04 0.81 258 3 0 10 38 38.8 days Coh. 7 37 0 5 11 73 20 58.2 days Phase Coh. 0.84 0.50 1.00 0.87 0.75 0.30 0.39 0.65 2120 28 6 29 250 6 16 222 6 2 060 9 6 13 111 7.5 6 44 0.16 0.56 0.69 0.94 1.00 0.74 0.17 0.47 2123 212 251 223 0 7 69 6 15 6 11 62 60 6 10 114 4 6 13 0.79 0.70 0.75 0.96 1.00 0.78 0.25 0.72 0.85 0.89 1.00 0.82 0.47 0.38 0.21 0.63 276 222 239 6 2 226 6 1 060 19 6 11 87 25 6 10 0.26 0.76 0.96 0.98 1.00 0.76 0.13 0.79 2111 6 15 215 238 6 44 223 6 6 060 21 6 9 134 6 47 23 6 8 0.70 0.32 0.47 0.85 1.00 0.79 0.46 0.81 2 2 Phase Coh. 2 Control U200-bp A B C D E F G H 243 11 0 14 39 55 6 51 6 10 60 61 62 6 17 130 60 6 8 6 6 6 6 6 56 118 63 6 6 4 0 8 44 21 1346 JOURNAL OF CLIMATE FIG. 8. First EOF of vertical profiles of 30–70-day bandpassed zonal wind for the (a) notched and (b) control runs averaged from 58S to 58N for points B through G as indicated in Fig. 3a. 45 days in the notched run. At 200 hPa, the control run periodicity is 44 days, but the notched run is harder to calculate due to increased fluctuations of the phase index. Here, the periodicity ranges from 30 days for the one time when the phase index advances consistently to around 42 days when the phase index is decreasing in time. VOLUME 17 The vertical zonal wind structure of the model MJO is isolated by taking EOFs of vertical profiles of 30– 70-day zonal wind at points B through G (Fig. 8). The resulting profiles reveal the strong first baroclinic nature of the MJO signal with the wind direction changing once within the troposphere. The average percent of variance explained by these profiles is 50% and 67% for the notched and control runs, respectively. Overall, comparison of the sample profiles reveals that the notched profiles are similar to those in the control run. Some of the profiles, for example, point C, match almost exactly whereas points E and F near the Maritime Continent have larger differences. For these latter two points the level of wind directional change is much lower in the notched run. Also, the notched run profiles have smaller magnitudes above the tropopause. To concurrently identify the large-scale, organized wind patterns that form in the upper and lower troposphere, an SVD analysis is performed (Bretherton et al. 1992; Wallace et al. 1992). Figure 9 presents the two most important left and right heterogeneous correlation maps, which represent the 200- and 850-hPa zonal winds, respectively. A similar analysis for the NRA (not shown) reveals patterns akin to the control run. Parallel to the results of the EOF analysis, the first two SVD maps for each level in the control run form a coupled pair where the first SVD has uniform sign and the second has a dipolelike pattern with the zero crossing between FIG. 9. SVD heterogeneous correlation maps for 30–70-day bandpassed zonal wind at 200 and 850 hPa. For the control run: (a), (b) SVDs 1 and 2 at 200 hPa, and (c), (d) SVDs 1 and 2 at u850. For the notched run: (e), (f ) are SVDs 1 and 2 at 200 hPa, and (g), (h) SVDs 1 and 2 at u850. The contour interval is 0.25 and regions significant at the 95% level are shaded. 15 MARCH 2004 FIG. 10. First EOF of vertical profiles of 30–70-day bandpassed MSE for the (a) notched and (b) control runs averaged from 58S to 58N for points B through G as indicated in Fig. 3a. the positive and negative regions roughly splitting the domain in half. In the notched run, the sign reversal is also apparent between the left- and right-heterogeneous correlation maps for both SVDs 1 and 2. However, the single-sign/dipolelike appearance between the two SVD maps is more difficult to interpret due to smaller regions with significant values. The fact that organized regions result from this analysis indicate that the notched run is indeed developing organized systems within the domain that exhibit the required sign reversal observed for the MJO. 2) MOIST 1347 GUSTAFSON AND WEARE STATIC ENERGY Changes to the thermodynamic structure of the atmosphere during MJO episodes are examined using EOFs of 30–70-day MSE profiles for points B through G. The overall shape of the profiles is similar for the notched and control runs, as seen in Fig. 10. The value of the profiles is completely of one sign with the largest anomalies just above the top of the boundary layer, around 850 hPa. This leads to increased low-level gradients, and, thus, shifts in the low-level stability as the MJO develops. The main difference between the notched and control runs is that the increased low-level gradients seen in points D and G of the control run do not exist in the notched run. Instead, these profiles have shapes similar to the remaining profiles. The timing of these stability changes is not consistent enough in the notched run to statistically determine a relationship between the 30–70-day zonal wind and the MSE profiles. In the control run significant negative correlations exist between these profiles at the 95% level for the MSE, leading the zonal wind by about 10 days. However, in the notched run the only significant correlations occur for point F and then for the MSE lagging FIG. 11. (a) Matthews amplitude indices for the 30–70-day bandpassed OLR. The amplitude index includes a 45-day running mean for smoothing. (b) Matthews phase indices for the 30–70-day bandpassed OLR. The phase index is only plotted for times when the smoothed amplitude index exceeds 35% of its maximum value. Black lines are for the notched run and gray lines are for the control run. the zonal wind by about 2 days. For comparison, this relationship is also seen in the control run at point F. 3) OUTGOING LONGWAVE RADIATION Examining OLR provides a more stringent test of how the dynamic and thermodynamic variables in the model interact because the cloud field is very dependant upon the feedbacks between them. Also, OLR is not directly forced by the boundary conditions and, therefore, is entirely model dependant. Parallel to the analyses performed on the zonal wind, similar techniques are used on the OLR to identify the following: active MJO times; large, organized patterns in the time series; propagation; and periodicities. Consistent, organized structures identified by the Matthews EOF technique clearly show the seasonality of the MJO. Figure 11 presents the Matthews EOF amplitude and phase indices for the notched and control run 30–70-day OLR. The seasonality of the OLR amplitude indices is even clearer than seen in the zonal wind. The timing of the amplitude peaks matches well between the two runs. The largest difference occurs during the second winter where the notched run peaks about 40 days before the control run. The notched run summer peak coincides with the summer peak in the 850-hPa zonal wind. The summer peak for the notched run 200hPa zonal wind corresponds to a local minimum in the OLR. This may be due to the 850-hPa-level winds having a greater influence on convective development. Propagation of the 30–70-day OLR can be identified through multiple techniques, including the Matthews phase index, cross-spectral analysis, and, subjectively, with Hovmöller diagrams. Because of the noisiness of 1348 JOURNAL OF CLIMATE VOLUME 17 FIG. 12. First two EOF maps of 30–70-day bandpassed OLR for the (a), (b) control and (c), (d) notched runs. Positive values are shaded with solid contours, negative values are unshaded with dashed contours, and the contour interval is 0.03 (units are incorporated into the PCs). the OLR time series, the results differ between the techniques. First, the results from the Matthews phase indices will be examined. The 30–70-day OLR phase indices consistently advance during times of strong amplitudes for both runs, indicating consistent changes in the OLR patterns during strong intraseasonal oscillations. While advancing phase angles have indicated propagation for the zonal wind, this is not as clear for the OLR. Figure 12 presents the two EOF maps corresponding to the phase index. Because the OLR field is so noisy, these maps are not the expected ideal single valued and dipole patterns indicative of the eastward propagation typically associated with the MJO. Animations constructed from these OLR maps and the corresponding PCs identify more locally propagating patterns, particularly in the equatorial Indian Ocean, and also smaller north–south propagating patterns. In the control run, propagation in the west Pacific favors turning toward the South Pacific convergence zone, but in the notched run the signals maintain a more constant latitude. Figure 13 presents the 30–70-day OLR Hovmöller diagrams for both runs. Even though the Matthews EOF technique was able to identify coherent fluctuating patterns in the full OLR fields, identifying these patterns in the 88S–88N region averaged for the Hovmöller diagrams is difficult. During the winter periods rough, eastward propagation across the domain can be seen in the control run. Unfortunately, in the notched run convection appears to occur in ‘‘bursts’’ with little activity in between each burst. During the first winter, these bursts can be connected to approximate eastward propagation, but during the second winter no argument can be made for propagation. One feature that the Hovmöller diagrams identify is the tendency for what appears to be a stationary oscillation in the OLR near 708 and 1108E. Both runs have this pattern during the first winter and spring, but to a lesser extent later in the runs. This feature is corroborated by the animations made from the first two OLR EOF maps. In another attempt to quantify the propagation of the model OLR, cross-spectral analysis for base points C and E has been done and is shown in Table 2. Overall, few points have significant squared correlations at the 95% level. However, for 58.2-day periodicities both runs show propagation, albeit with large uncertainties: the notched run has advancing phase values from points C through G and the control run has advancing phase values from points B through E. The notched run reveals no noticeable propagation at the 38.8-day periodicity in comparison to the control, which shows no pattern for this periodicity at base point C. Even with the weakly propagating characteristics of the notched run 30–70-day OLR, the coherent OLR patterns still fluctuate on a relatively consistent time scale. Using the phase indices in Fig. 11 to gauge the periodicity of the notched run intraseasonal oscillations, the periodicity ranges from 41 to 47 days, depending on the episode. This compares to 40–44 days in the control run. In an attempt to identify the relationship between the zonal wind and OLR, SVD analyses were performed between OLR and 200-hPa zonal wind and also for OLR and 850-hPa zonal wind. Due in part to the short, 2-yr time series, as well as the noisiness of the OLR, significant spatial patterns were not identified by these analyses. 4. Discussion and conclusions Based on the experiment presented in this paper external forcing in the form of incipient MJO episodes from outside the Indian Ocean is unnecessary for the formation of 30–70-day intraseasonal oscillations. However, with the current model setup, once initially developed, the intraseasonal oscillations do not fully conform to observed MJO characteristics. Two model runs have been made using MM5 with a domain extending from eastern Africa to the date line. The first run functions as the control and is forced by the NRA. 15 MARCH 2004 1349 GUSTAFSON AND WEARE FIG. 13. Hovmöller plots of 30–70-day bandpassed OLR averaged from 88S to 88N for the (a) notched and (b) control runs. Increased shading density reflects greater magnitudes, and positive values are also contoured. TABLE 2. Squared coherency (Coh. 2 ) and phase from cross-spectral analysis of 30–70-day bandpassed OLR (OLR-bp) averaged from 58S to 58N for the two base points C and E and the two periodicities 38.8 and 58.2 days. Regions B through G are as indicated in Fig. 3a. Bold values indicate the squared coherency differs from zero at 95% confidence. Ninety-five-percent confidence intervals for the phase are shown when they can be calculated. Notched OLR-bp Base point 5 C 38.8 days Base point 5 E 58.2 days Region Phase Coh. B C D E F G 175 060 168 6 36 162 153 56 0.02 1.00 0.51 0.13 0.28 0.25 267 060 2129 169 2161 88 0.11 1.00 0.04 0.11 0.22 0.06 2 Phase 38.8 days Coh. 2 58.2 days Phase Coh. 2 Phase Coh. 2 0 17 30 37 10 0.21 1.00 0.67 0.55 0.50 0.78 153 6 26 2162 92 060 78 6 42 2176 0.58 0.13 0.24 1.00 0.48 0.13 130 244 6 30 4 060 25 6 37 234 6 77 0.01 0.55 0.35 1.00 0.50 0.39 211 6 34 060 12 6 7 67 6 29 35 1 6 17 0.52 1.00 0.84 0.55 0.25 0.67 2169 2169 226 060 45 64 0.22 0.11 0.03 1.00 0.36 0.11 2131 267 234 060 14 230 0.15 0.55 0.32 1.00 0.52 0.32 0 19 44 36 6 5 6 6 6 6 6 Control OLR-bp B C D E F G 1350 JOURNAL OF CLIMATE The second run is forced by a filtered version of the NRA in which all signals with periodicities in the range of 30–70 days are removed. This effectively removes the MJO signal from the boundaries of the notched run without altering the overall flow outside these periodicities. Comparing the control and notched runs confirms that removal of the 30–70-day band has not altered the climate in the notched run. Only small differences have been found between the two runs when comparing 2-yr means. Even more importantly, with the 30–70day periodicities removed, the notched run still develops large-scale, propagating oscillations in the 30–70-day band. Based upon the spectral power of 200-hPa zonal wind, these periodicities develop across the entire domain, including close to the western boundary where 30–70-day boundary effects would have the most influence. The strength of the 30–70-day intraseasonal signals is less in the notched run, compared to the control. Even though the exact difference in strength varies between the particular event and the method used to define the amplitude, the general consensus among the techniques used in this paper is that the notched oscillations are roughly 50%–75% as strong as in the control run, with direct spectral measurements giving a larger difference. Depending on whether one compares the Matthews amplitude indices or the Hovmöller diagrams, and then depending on the variable, the first winter produces the strongest notched oscillations with the strength diminished during the second winter. The reason for the differences between winters is currently not understood. Many of the defining characteristics of the MJO are generated. However, due to the lack of a consistent relationship between the relatively noisy OLR and wind, it is not possible to demonstrate a coherent relationship between these variables, as seen in observations. Examination of notched run zonal wind, MSE, and OLR reveals systematic, coherent anomalies that develop within the 30–70-day band. The notched run develops anomalous patterns with periodicities near 40 days, with peak amplitudes during boreal winter and late summer. Propagation is also seen, most clearly in the zonal wind, originating in the Indian Ocean and moving away from this region. Propagation of the OLR anomalies occurs over smaller distances and is less evident due to the noisiness of the OLR signal. The vertical structure of the 30–70-day anomalies matches well with what would be expected for MJO-type signals in the model. Comparing the 30–70-day vertical zonal wind profiles of the notched and control runs reveals very similar patterns with one sign reversal within the troposphere and peak wind speeds near the tropopause. SVD analysis of the 850- and 200-hPa zonal wind reveals that the sign reversals form over large coherent regions of the model domain. Finally, vertical profiles of 30–70-day MSE are also very similar between the two runs, having uniform sign and the largest anomalies just above the top of the boundary layer. This variation near the boundary layer VOLUME 17 top gives credence to the buildup of energy prior to MJO passage, as identified by Kemball-Cook and Weare (2001), and further implies that the boundary layer recharge mechanism is important for MJO formation. The MJO event is associated with MSE building up in the lower troposphere and then being released during peak MJO convection. With the formation of 30–70-day intraseasonal oscillations in the notched run, some arguments can be made regarding the role of 30–70-day boundary effects on MJO development. The notched run was able to develop the intraseasonal oscillations without any influence from remnants of previous MJO episodes or extratropical influences in the 30–70-day time band. This leaves at least two other possibilities for triggering the episodes. The first is higher-frequency waves propagating into the Indian Ocean region. This possibility has been suggested in studies by Matthews and Kiladis (1999) who noted a greater incidence of higher-frequency waves entering the Indian Ocean during the MJO formation phase. The second possibility is that the formation and periodicity is determined by local feedback mechanisms. Studies of radiosonde soundings by Kemball-Cook and Weare (2001) suggest a regional building up of instabilities that are then released as MJO convection develops. Future experiments using the method presented in this paper will be aimed at determining the role of frequencies less than 30 days on MJO formation. The difference between the control and notched runs, particularly the weaker MJO signals and propagation in both the west and east directions, suggests that the boundary filtering has altered important physics related to the MJO. The simplest explanation for the differences would be the limited domain size in which the notched run MJO must form. Without 30–70-day signals in the boundaries, the notched run MJOs are completely confined within the model domain. This is a severe constraint, given the global influence of the MJO. A second explanation is that direct feedbacks are required that are not contained in the model domain. For example, even though circumnavigating MJO signals do not appear to be necessary for MJO formation, they could be important for organizing nascent MJOs developing in the Indian Ocean. To improve the model MJO structure, work is being done to determine why the current runs do not form as well organized mesoscale convective complexes, as seen in satellite observations of OLR. The noisiness of the MM5 OLR field has proven disappointing in this analysis. This is most likely due to inadequacies in the Betts– Miller convective scheme used for the runs, or at least in the tuning of this scheme within MM5. A second strong contender is the boundary layer scheme and how it handles fluxes to and from the ocean. Because most MJO theories rely on feedbacks between convection and dynamics, the lack of large, organized convective regions proves to be a substantial shortcoming of the current model configuration. It is believed that if the model 15 MARCH 2004 GUSTAFSON AND WEARE can be improved to better capture this phenomenon, a truer MJO signal will be produced. A more robust analysis could also be done by extending the current methodology to include an ensemble of model integrations. With only one control and one notched run, the cause of the differences between the runs cannot be uniquely identified. While a portion of the differences is most likely due to the filtered boundary conditions, the exact percentage cannot be stated because some of the differences are also due to natural variability. Multiple model integrations could be used to quantify the natural variability of MJO activity with the current model configuration and then to determine more exactly the significance of the boundary forcing. Acknowledgments. We would like to thank Daniel Hodyss for his helpful discussions. NCEP–NCAR reanalysis data were obtained from the NCAR mass storage system through NCAR University Projects 36131021 and 36131022. This work was partially supported by NSF Grant ATM-9613779 and by the University of California Office of the President through the Campus Laboratory Collaboration Program. Further support was received from the Department of Energy (DOE), under the auspices of the Atmospheric Sciences Program of the Environmental Sciences Division of the Office of Biological and Environmental Research, under Contract DE-AC06-76RLO 1830 at the Pacific Northwest National Laboratory. Pacific Northwest National Laboratory is operated for the U.S. DOE by Battelle Memorial Institute. REFERENCES Bladé, I., and D. L. Hartmann, 1993: Tropical intraseasonal oscillations in a simple nonlinear model. J. Atmos. Sci., 50, 2922– 2939. Bretherton, C. S., C. Smith, and J. M. Wallace, 1992: An intercomparison of methods for finding coupled patterns in climate data. J. Climate, 5, 541–560. Duchon, C. E., 1979: Lanczos filtering in one and two dimensions. J. Appl. Meteor., 18, 1016–1022. 1351 Emanuel, K. A., 1987: An air–sea interaction model of intraseasonal oscillations in the tropics. J. Atmos. Sci., 44, 2324–2340. Fasullo, J., and P. J. Webster, 2000: Atmospheric and surface variations during westerly wind bursts in the tropical western Pacific. Quart. J. Roy. Meteor. Soc., 126, 899–924. Gill, A. E., 1980: Some simple solutions for heat-induced tropical circulation. Quart. J. Roy. Meteor. Soc., 106, 447–462. Grell, G. A., J. Dudhia, and D. R. Stauffer, 1995: A description of the fifth-generation Penn State/NCAR mesoscale model (MM5). NCAR Tech. Note NCAR/TN-3981STR, 122 pp. Gustafson, W. I., and B. C. Weare, 2004: MM5 modeling of the Madden–Julian oscillation in the Indian and west Pacific Oceans: Model description and control run results. J. Climate, 17, 1320– 1337. Hendon, H. H., and M. L. Salby, 1994: The life cycle of the Madden– Julian oscillation. J. Atmos. Sci., 51, 2225–2237. Hu, Q., and D. A. Randall, 1994: Low-frequency oscillations in radiative–convective systems. J. Atmos. Sci., 51, 1089–1099. ——, and ——, 1995: Low-frequency oscillations in radiative–convective systems. Part II: An idealized model. J. Atmos. Sci., 52, 478–490. Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471. Kemball-Cook, S. R., and B. C. Weare, 2001: The onset of convection in the Madden–Julian oscillation. J. Climate, 14, 780–793. Knutson, T. R., and K. M. Weickmann, 1987: 30–60 day atmospheric oscillations: Composite life cycles of convection and circulation anomalies. Mon. Wea. Rev., 115, 1407–1436. Lau, K. M., and L. Peng, 1987: Origin of low-frequency (intraseasonal) oscillations in the tropical atmosphere. Part I: Basic theory. J. Atmos. Sci., 44, 950–972. Madden, R. A., and P. R. Julian, 1994: Observations of the 40–50day tropical oscillation—A review. Mon. Wea. Rev., 122, 814– 837. Matthews, A. J., 2000: Propagation mechanisms for the Madden– Julian oscillation. Quart. J. Roy. Meteor. Soc., 126, 2637–2651. ——, and G. N. Kiladis, 1999: The tropical–extratropical interaction between high-frequency transients and the Madden–Julian oscillation. Mon. Wea. Rev., 127, 661–677. Roeckner, E., L. Benotsson, J. Feichter, J. Lelieveld, and H. Rodhe, 1999: Transient climate change simulations with a coupled atmosphere–ocean GCM including the tropospheric sulfur cycle. J. Climate, 12, 3004–3032. Rui, H., and B. Wang, 1990: Development characteristics and dynamic structure of tropical intraseasonal convection anomalies. J. Atmos. Sci., 47, 357–379. Wallace, J. M., C. Smith, and C. S. Bretherton, 1992: Singular value decomposition of wintertime sea surface temperature and 500mb height anomalies. J. Climate, 5, 561–576.