The meteorology of the 2019 North Queensland floods

In January–February 2019, a monsoon low developed over northeastern Australia and brought extreme flooding to much of tropical Queensland. The European Centre for Medium Range Weather Forecasts reanalyses are used to explain the formation and severity of the event. Strong anticyclonic Rossby wave breaking to the southeast of Australia produced weak steering winds over northern Australia, and consequently the low remained nearly stationary for over a week, contributing to the extreme rainfall. Furthermore, the tropical moist margin was deformed by a series of disturbances along the monsoon trough, which drew moist maritime air over land. The event examined has much in common with the composite mean of slow‐moving potential vorticity anomalies in North Queensland, including weak background winds and heavy precipitation. An ensemble subseasonal forecast with the Australian Bureau of Meteorology's Australian Community Climate and Earth System Simulator Seasonal prediction version system 1 shows the same relationship between the background flow, the steering, and the subsequent rainfall. Hence, accurately forecasting the background winds is a prerequisite to forecasting such extreme rainfall.


INTRODUCTION
Synoptic-scale disturbances frequently occur within the summer monsoon season (December-March) in tropical Australia.These systems are a vital part of the hydrological cycle, bringing around 50% of total wet-season rainfall (Berry et al., 2012;Hurley & Boos, 2015).However, they also bring the risk of hazardous weather, such as strong winds, wind chill, and extreme rainfall leading to floods (Cowan et al., 2019;Godbole, 1977).
One class of tropical synoptic-scale disturbances is the "monsoon low" (also known as "monsoon depression").
Monsoon lows have a similar structure to tropical cyclones, particularly in the outer regions, but they are weaker and lack a characteristic eye (Davidson & Holland, 1986;Godbole, 1977).In the vertical, they have a warm-over-cold core, a potential vorticity (PV) structure with two maxima around 500 and 700 hPa, and cyclonic winds extending into the upper troposphere (Hunt et al., 2016;Murthy & Boos, 2019).
Monsoon lows typically form along the monsoon trough, a zonal band of low pressure associated with convergence and, in the Southern Hemisphere, westerly flow to the north and easterly flow to the south.Although the physical mechanisms responsible for monsoon low genesis have been debated in the literature, it is generally thought that they result from a combination of convective and barotropic instabilities along the monsoon shear line (Boos et al., 2015;Cohen & Boos, 2016;Diaz & Boos, 2019a, 2019b;Kilroy et al., 2016;Smith et al., 2015).
It is clear that the development of monsoon lows depends on the availability of moisture (Ditchek et al., 2016).In the Tropics, moisture is typically largest over the oceans and near the Equator.It is therefore not surprising that monsoon lows form mostly over the ocean, where they may intensify into tropical cyclones if the synoptic environment is favourable (Foster & Lyons, 1984).However, monsoon lows occasionally develop or intensify over inland Australia, and do so more frequently than other monsoon regions (May et al., 2008;McBride & Keenan, 1982;Zhu & Smith, 2020).Continental development of this kind is most common in the peak austral monsoon season (January-February), where the monsoon trough can be displaced southwards over the Australian continent.
Although monsoon lows are tropical weather systems, they may be affected by disturbances from the midlatitudes (Berry et al., 2012).For example, midlatitude Rossby wave breaking results in PV streamers that can either feed into and reintensify an existing tropical low (Leroux et al., 2020) or initiate new ones (Hoang et al., 2017).Similarly, monsoon bursts (sudden periods of heavy rain) are associated with an increased cyclonic circulation resulting primarily from absolute vorticity fluxes from the midlatitudes (Narsey et al., 2017).
This study examines a monsoon low that occurred over January-February 2019, causing major and widespread floods in North Queensland.The floods and associated wind chill led to five fatalities, major livestock losses, and over AUD 1 billion of property damage, with total losses of around AUD 5.68 billion including indirect costs (Deloitte, 2019).Two interesting aspects of this monsoon low were that (a) it intensified inland and (b) it remained near-stationary for over a week.
This event is of particular interest to the forecasting community since it was poorly predicted, particularly at lead times of around 1 week.Some studies have evaluated the skill with which the event was forecast with numerical weather prediction Hawcroft et al. (2021) and subseasonal Cowan et al. (2019Cowan et al. ( , 2022)); Tsai et al. (2021) models.On the numerical weather prediction time-scale, Hawcroft et al. (2021) find that forecast errors mostly originate from atmosphere-ocean coupling and the convection scheme, particularly near the west Cape York Peninsula coast.The same study also investigates the importance of large-scale flow to the accuracy of the forecast.Another study, by Callaghan (2021), focuses on the role of warm air advection in the floods and attributes the slow movement to the low being embedded in the monsoon trough between easterlies to the south and westerlies to the north.
Subseasonal models also failed to forecast the severity of the rainfall accumulations and wind chill with lead times longer than around a week, as discussed by Cowan et al. (2019Cowan et al. ( , 2022)); Tsai et al. (2021).The studies partially attribute this failure to the weak influence of large-scale climate drivers.However, synoptic-scale processes rather than seasonal-scale processes govern the severity of weather systems such as monsoon lows (Callaghan, 2021).Therefore, the present study analyses the dynamics of this event in greater detail.In particular, we demonstrate a link between midlatitude Rossby wave breaking and weak environmental steering, which meant the low lingered in the region.
This article is structured as follows.Figure 1 shows a map of location names referenced in the text.Section 2 describes the methods and data used in the study.Section 3 presents a climatology of synoptic disturbances in northeast Australia and highlights the fundamental differences between slow-and fast-moving systems.In Section 4 we examine the tracks, evolution, steering, and synoptic structure of the 2019 North Queensland event with European Centre for Medium-Range Weather Forecasts (ECMWF) reanalyses.Finally, Section 5 investigates the performance of the Australian Bureau of Meteorology's subseasonal model Australian Community Climate and Earth System Simulator Seasonal prediction version system 1 (ACCESS-S1) in predicting the monsoon low tracks, rainfall, structure, and the steering.Further discussion and conclusions are found in Section 6.

Data
For Sections 3 and 4 we use the ECMWF Reanalysis v5 (ERA5) (Hersbach et al., 2020).ERA5 has hourly data at 37 pressure levels from 1 to 1,000 hPa with a horizontal resolution of 0.25  Jones et al., 2009), which is based on surface gauges and has a resolution of 0.05 • .We also use the 1 • daily Global Precipitation Climatology Project (GPCP) dataset as it extends over the oceans (Adler et al., 2018).Section 5 examines the subseasonal forecast of the event by ACCESS-S1 Hudson et al. (2017).ACCESS-S1 produces daily forecasts at 85 pressure levels and a horizontal resolution of approximately 60 km.The forecasts comprise 33 ensemble members based on perturbed initial conditions.We examine the forecast initialised on January 24, 2019.Note that the low had already formed at this stage.The same model was evaluated by Cowan et al. (2019Cowan et al. ( , 2022)).

Tracking algorithms
Coherent PV anomalies are tracked on the 315 K surface for the extended summer season (November-March) over the years 1980-2020 using a similar methodology to Barnes et al. (2023); Berry et al. (2012); Hoang et al. (2017), which is briefly described here.In this study, we multiply the relative and PV by −1 so that positive values are cyclonic in the Southern Hemisphere.
To eliminate irrelevant small-scale features, the 315 K isentropic PV field is smoothed using two passes of a 13-point averaging smoother.Local maxima less than 0.01 PVU in magnitude are removed.Maxima that are not separated by at least 3 • are assumed to be part of the same maxima, with the less cyclonic of these discarded.PV maxima are then joined in coherent tracks every 6 h.A first estimate of the location of the PV maximum at a subsequent time step is made assuming simple advection using isentropic wind fields on the same level.PV maxima located within 5 • of the estimated location are joined to form a coherent PV anomaly track.Only coherent PV tracks that exist for at least 24 h are retained.Finally, since the focus of this study is on monsoon lows, tracks are only retained if the coherent PV anomaly is at least 0.5 PVU for at least one time step.A total of 2,311 tracks are found over the 41-year period.
The tracks of the coherent PV anomalies that move over the study region (135-145 • E, 15-25 • S) are then decomposed into slow-and fast-moving categories.The speed of the coherent PV anomalies is calculated using forward differences in time, and the mean speed is taken as the average over the time the track is inside the study region.Mean track speeds below the 20th percentile are defined to be slow, whereas those above the 80th percentile are defined to be fast.This gives 463 coherent PV anomalies in each category.Since the coherent cyclonic PV anomaly may accelerate or decelerate as it propagates through the study region, composites are taken at the slowest point in each slow track and the fastest point in each fast track.Anomalies of fields are taken with respect to the daily means from 1980 to 2020.Rainfall anomalies are calculated as departures of the AGCD daily rainfall from the climatological daily mean , and the composites are taken over the period in which the anomaly is in the study region.
In Section 4, we use an alternative tracking method based on the geopotential at 850 hPa.Because isentropic PV data are unavailable from ACCESS-S1, using geopotential allows for a more direct comparison of ERA5 tracks with the forecasts.The algorithm simply finds the absolute minimum of the daily mean 850 hPa geopotential height within the study domain 135-150 • E, 10-25 • S.
The ACCESS-S1 tracks in Section 5 are constrained further.First, the minimum 850 hPa geopotential height must be less than 1,500 m.Second, the low must move less than 7.5 • latitude and longitude per day.Third, the low must remain within the region 130-160 • E, 0-30 • S and not on the boundary, since this could imply the low lies outside the study domain.If any of the aforementioned criteria are not fulfilled, the track is said to terminate on the previous day.These constraints ensure the correct system is detected and prevent issues such as tracks remaining on the boundary or moving to other systems in the domain.

2.3
The tropical moist margin One way of measuring large-scale atmospheric moisture is with the TCWV.Mapes et al. (2018) found that the TCWV in the Tropics exhibits a bimodal distribution with a minimum value of 48 kg⋅m −2 .This value separates dry and moist regimes, with the vast majority of rainfall occurring in the moist regime.This moist margin can be perturbed by the synoptic flow, including monsoon lows.Following Mapes et al. (2018), this study defines the tropical moist margin by the 48 kg⋅m −2 contour.

CLIMATOLOGY OF SYNOPTIC DISTURBANCES IN NORTHEAST AUSTRALIA
To provide context for the 2019 North Queensland floods, this section first outlines the structural differences between slow-and fast-moving PV anomalies in northeast Australia.The composite structures for slow-(Figure 2a,c,e,g) and fast-moving (Figure 2b,d,f,h) PV anomalies are very different.For the fast-moving systems, prominent cyclonic anomalies on the 350 K surface lie to the south and southeast of the study region.Anticyclonic anomalies lie further to the west and east, with the sequence of cyclonic and anticyclonic anomalies resembling a typical Rossby wave train.At 315 K (where the PV anomalies are defined), cyclonic PV anomalies are prominent within the study region, and also to the west for the slow anomalies (not shown).The 250 hPa zonal winds are westerly in the study region and strongly easterly to the south.This is consistent with basic PV inversion principles, with the cyclonic PV anomaly inducing a clockwise circulation in the Southern Hemisphere.The mean wind speeds in the 850-500 hPa layer are mostly above 6 m⋅s −1 in the study region.The moist margin lies mostly over the ocean equatorward of around 10-15 • S. The rainfall is close to average over northern Australia, including the study region, except for a wet area near the eastern Queensland coastline.
In contrast, slow-moving PV anomalies have a much weaker PV pattern, with the cyclonic PV anomaly in particular almost absent.As a result, the 250 hPa zonal wind anomalies are much weaker and almost vanish within the study region.The 700 hPa geopotential height shows more prominent ridging that extends over eastern Australia.Slow-moving PV anomalies are also associated with a poleward shift of the moist margin over land.This shift is not just in the study region, but across the entirety of northern Australia.Similarly, the rainfall is enhanced across widespread regions of northern Australia, particularly along the west Cape York Peninsula.This pattern, where the monsoon trough strengthens and shifts southwards over the continent, is typical of an active monsoon (Hoang et al., 2017).Such a progression is also reminiscent of monsoon onset in northern Australia (Davidson et al., 1983).Following the arguments of Callaghan ( 2021), a monsoon low that is embedded in the monsoon trough over land is also likely to be slow moving.This is because the zonal winds vanish along the centre of the trough, where easterly winds to the south transition to westerly winds to the north.
As the choice of 20th and 80th percentiles to define the slow and fast categories is of course arbitrary, we have evaluated the composite structure instead using the 10th and 90th percentiles.The results are shown in Figure 3 and are largely similar.The main differences are (a) the slow anomalies have a more prominent upper level cyclonic PV anomaly to the southwest, and the PV structure to the south of the study domain resembles a Rossby wave shifted a quarter-wavelength zonally relative to the fast case, and (b) the fast anomalies have stronger upper level westerlies within the study domain.
A box plot showing the distribution of track durations within the study region is shown in Figure 4.As expected, the fast-moving anomalies exit the study region more quickly than the slow-moving ones, with a median time of 12 h for the fast-moving anomalies and 24 h for the slow-moving anomalies.The slow category also has a longer tail, with some systems persisting for more than 9 days.
The 2019 North Queensland event is marked red in the slow category of Figure 4 and persists for 8.25 days 1 .While not record-breaking, this event is clearly an outlier.

ANALYSIS OF THE 2019 MONSOON LOW
We now focus on the 2019 North Queensland monsoon low.In this section, we show that the synoptic structure of the 2019 event resembles the typical composite structure of the slow-moving PV anomalies described in the previous section.

Track and evolution
Figure 5 shows the 850 hPa geopotential track, the 315 K PV track, accumulated rainfall from January 23 to February 10, and the evolution of geopotential height and mean-sea-level pressure.The geopotential track (Figure 5a) shows the low forming around Cape York Peninsula (143  completes the gap between January 28 and January 30, moving similarly to the first track.The separation of the PV track into multiple parts is perhaps not surprising, given that geopotential acts like a smoothed version of vorticity (through the inverse Laplacian operator).
The observed accumulated rainfall from January 23 to February 10 (Figure 5c) shows a maximum of well over 1,000 mm near Townsville.Another maximum of more than 500 mm lies inland where the low stalled, along with widespread values exceeding 250 mm over North Queensland.The southeastern part of the state is dry, except for accumulations around 20-100 mm along the coastline.

Steering of the low
The speed at which monsoon lows move is critically important, as slow-moving systems produce more accumulated rainfall for a given rainfall rate (Hall & Kossin, 2019).To first order, tropical weather systems move with the large-scale background wind, a process commonly known as "steering" (Chan, 2005).In northern Australia, synoptic disturbances typically propagate westward with the mean flow (Berry et al., 2012;Hoang et al., 2017).In this study, we use a steering layer of 850-500 hPa as it is more The 850-500 hPa average background winds (blue) and system speed (red) for the (a) zonal and (b) meridional directions.
The root-mean-square error (RMSE) between the two lines is also noted.
representative of the motion for systems below tropical cyclone strength (Velden & Leslie, 1991).
The background winds are calculated as the average over a 10 • × 10 • box centred on the daily 850 hPa geopotential track and in the vertical over 850-500 hPa.The system velocity is calculated using centred differences on the 850 hPa tracks.The position of the low is calculated at 0000 UTC and the background winds are calculated as a daily mean over 12 h each side of 0000 UTC, meaning the system velocity and background winds are aligned.
Figure 6 shows the zonal and meridional system motion and background winds.Overall, there is good agreement between the assumed steering flow and the observed motion, with a root-mean-square error (RMSE) of 1.8 m⋅s −1 for the wind speed.In the zonal direction, the period from around January 29-February 5 has both weak background winds and little system motion (less than 2 m⋅s −1 ).Errors are larger in the meridional direction, particularly from February 3 onwards (around 2 m⋅s −1 ).

Synoptic-dynamic structure
Four key dates have been chosen to represent the evolution of the event: January 23 (when the low first forms), January 26 (when the low moves inland), February 2 (peak intensity and the low is stationary), and February 8 (when the low drifts offshore).These dates are also labelled on the geopotential tracks in Figure 5. Figure 7 presents numerous daily-mean fields for these dates.
On January 23, a relative vorticity maximum lies just west of Cape York Peninsula (142 • E, 12 • S), indicating the monsoon low of interest (Figure 7b).with an upper level PV streamer extending east of New Caledonia (170 • E, 25 • S) into Queensland.The end of the streamer curves cyclonically and is detached, sitting about 10 • southwest of the low.Although it may be expected that this PV cut-off reinforces the lower level anomaly, by January 29 it is absorbed into a passing midlatitude trough (not shown).There is high rainfall to the east of the PV anomaly, and the tropical band of westerlies now extends further east, feeding the circulation associated with the monsoon low.
February 2 is the peak of the event, as defined by the minimum 850 hPa geopotential height.At this time a strong and coherent relative vorticity maximum lies near 20 • S, 140 • E. The system is also larger, as indicated by the 850 hPa winds and dashed red vorticity lines surrounding the centre of the low.Tropical moisture is drawn inland towards the low, and the moisture margin is perturbed well from its usual state (around 25 • S at its southernmost point).Rainfall is widespread across Queensland, with multiple maxima around the low.The tropical band of westerlies (5-15 • S) almost entirely feeds into the low, and strong easterlies are present inland to the south of the low centre.Widespread upward vertical motion can be seen surrounding the low with a similar structure to the rainfall.There is cyclonic PV (below 2 PVU) near the low and further to the south.
By February 8 the low is weaker and less coherent as it moves into the Coral Sea (148 • E, 16 • S).The TCWV inland is also considerably lower than on February 2. The easterlies located south of the low are also weaker, and the tropical band of westerlies now lies over the Pacific Ocean.

Vertical structure
Figure 8 shows the vertical structure of the low on January 26 and February 2. On January 26, the cyclonic circulation surrounding the low extends upwards to 100 hPa in the zonal cross-section, although there is a displacement of the winds to the west above 400 hPa.The upper level cyclonic winds are likely to be associated with the upper level PV cut-off in Figure 7d, which is also seen here as a deep intrusion of high-PV air extending from the stratosphere.Cyclonic winds similarly extend upwards to 400 hPa in the meridional cross-section, where there are westerlies associated with the jet stream south of the low.The upper level isentropes bow slightly downwards towards the centre of the low, and the isentrope nearest the surface bows upwards, exhibiting the classic warm-over-cold core structure of the low.The PV associated with the vortex is mostly smaller than 1 PVU, although a double-maximum structure is evident at around 800 hPa and 550 hPa, consistent with previous studies of monsoon lows (Hunt et al., 2016;Murthy & Boos, 2019).The upper level PV streamer can be seen in the meridional cross-section (Figure 8b) at around 25 • S. Note that as the low-level PV structure is disconnected from the upper level structure the low-level PV anomaly is therefore likely generated diabatically.The vertical structure is stronger and more coherent on February 2, extending upwards to 200-250 hPa in both the zonal and meridional directions.There are two strong PV peaks (larger than 2 PVU) at around 500 hPa and 700 hPa.
The upper level warm structure is clearer with the isentropes bowing downward more strongly.There is also very little tilting in any of the fields, suggesting the low is near-barotropic at this time.

Tracks and rainfall
Figure 9 shows the tracks, intensity, and rainfall for the ensemble mean.There is a large spread between the tracks of the individual members.As the tracks are relatively insensitive to the details of the numerical method used, the large spread is indicative of forecast uncertainty rather than the tracking method.The ensemble mean track is slightly too far to the west at first, and many ensemble members move westward or southward outside the plotted domain.The ensemble mean track recurves to the southeast from around January 31.Note that some tracks terminate or move outside the plotted domain at this point, and hence thereafter the ensemble mean comprises fewer members.Although the ensemble mean correctly forecasts that the low reintensifies (albeit slightly too early, on January 27 instead of January 29), it is too deep.Widespread rainfall exceeding 250 mm develops over areas of north Queensland (equatorward of 20 • S) in the ensemble mean, but the maximum values are substantially smaller than observed (see Figure 5c).Furthermore, the ensemble mean produces widespread regions of rainfall at higher latitudes (poleward of around 25 • S), which is not observed.Presumably, this poleward rainfall is due to those ensemble members with tracks that extend further to the south.To quantify these differences, we calculate the 90th percentile of rainfall within the domain 138-150 • E, 10-22 • S, which is where the heaviest rainfall occurred.The 90th percentile in the observations is 642 mm.In ACCESS-S1, ensemble members 2 and 24 have similar values of 627 mm and 665 mm respectively, and all other ensemble members produce values varying from 250 to 600 mm.For the ensemble mean, the 90th percentile is 427 mm, about 35% smaller than what was observed and similar to the mean of all ensemble members (460 mm).Therefore, the reduced rainfall in the ensemble mean is largely due to most individual ensemble members producing too little rainfall.
Figure 9 also shows ensemble member 2, which has the lowest RMSE to the reanalysis track.This member has an accurate track that captures the inland stall and subsequent eastward drift off the coast.However, the low reintensifies inland too late on February 1 and subsequently intensifies rapidly after moving into the Coral Sea.The rainfall patterns are also closer to reality than the ensemble mean, with a double maxima of more than 500 mm near Townsville (146 • E, 18 • S) and further inland.Nonetheless, the rainfall is still too large over large areas of southern Queensland.
Figure 10 shows the mean-sea-level pressure, 850 hPa winds, and midlevel ascent for the ensemble mean and member 2 on the key dates January 26 and February 2. On January 26, both plots show a coherent vortex in the east of the Gulf of Carpentaria (approximately 140 • E, 15 • S) and large-scale ascent over the Cape York Peninsula, as in the reanalysis (see Figure 7).By February 2, the structures are considerably different from reanalysis.The ensemble mean shows the vortex centred over the Northern Territory and Western Australia (130 • E, 20 • S), consistent with the tracks that move westward outside the domain of Figure 9.The vertical velocities are very small, which is perhaps expected when taking the ensemble mean of a noisy field with lower spatial resolution.Ensemble member 2 produces a relatively coherent vortex in approximately the correct location, although it is much weaker than reality.In particular, the strong inland easterlies to the south of the low are missing.The upward motion is also generally too far southeast compared with reanalysis.Additionally, there is another weak low-pressure system to the west at around 125 • E, 20 • S.

Steering
Figure 11 shows the relationship between background winds and system motion for the ensemble mean and ensemble member 2. There is a very strong relationship in the ensemble mean, with a combined zonal and meridional RMSE of 0.93 m⋅s −1 until January 30 (when some tracks terminate).This is smaller than the RMSE in the reanalysis steering (see Section 4.2).The steering relationship is also strong for ensemble member 2, with an RMSE of 1.37 m⋅s −1 .The ACCESS-S1 forecasts underscore the fundamental link between the patterns of rainfall, the track and the background winds that steer the vortex.Moreover, the large spread of ensemble tracks highlights the importance of correctly predicting the background winds.This is similar to the results of Hawcroft et al. (2021), who speculate that errors in the large-scale flow may result from the growth of smaller errors in model physics, ultimately resulting in a less accurate forecast.

DISCUSSION & CONCLUSION
The synoptic structure and evolution of the monsoon low responsible for the 2019 North Queensland floods has been analysed.Although the monsoon low was not particularly extreme, the rainfall and associated impacts were extreme because it remained relatively stationary.We showed that there is a close relationship between background winds and system motion (Figure 6) and suggested that anticyclonic wave breaking in the Tasman Sea created a weak steering wind that allowed the system to linger in the region.We have also placed the event in the context of a climatology of coherent PV anomalies in tropical northeastern Australia.The structure of the low in 2019 agrees well with the composite structure of slow-moving low-level PV anomalies, although it is still an outlier in terms of how long it persisted and the total precipitation accumulated.
We now discuss some of these findings in more detail.The results show that steering is the dominant contribution to the slowness of the low.A more complete explanation of the motion of the low would involve analysing the circulation associated with vortex asymmetries, which tend to advect tropical weather systems poleward and westward of the mean flow (Boos et al., 2015;Chan, 2005).However, this theory is generally applied to tropical cyclones, and may not be suitable for the weaker monsoon low in this case study.
Presumably, the weak steering is a result of Rossby wave breaking.When a strong midlatitude trough passes to the south, the westerly winds associated with cyclonic PV anomalies quickly advect the system to the east (as seen in Figures 2 and 3).If no such trough passes, the system persists for longer.Moreover, Rossby wave breaking stirs the PV, creating weak PV gradients, and therefore weak background winds.This idea can be illustrated as follows.Let  be the relative vorticity and  the streamfunction.Then  = ∇ 2  with v = ∕x and u = −∕y.Hence, ∕x = ∇ 2 v and ∕y = −∇ 2 u, which scale like −v and u respectively, implying that the horizontal winds  (c, d) "best" ensemble member 2. Dotted lines for the ensemble mean indicate the point at which tracks terminate in some ensemble members (from January 30).The root-mean-square error (RMSE) between the two lines is also noted (for solid lines only).
are proportional to the horizontal gradient of vorticity.This argument can be extended to PV, where the principles of PV inversion imply that horizontal winds are proportional to PV gradients (Hoskins et al., 1985).Therefore, the weak PV gradients generated by Rossby wave breaking and the associated PV stirring result in weak background winds and steering, which explains why the low remained relatively stationary.
Anticyclonic Rossby wave breaking also results in PV streamers that can extend into the Tropics and interact with or initiate tropical weather systems (Berry et al., 2012;Hoang et al., 2017;Leroux et al., 2020).Indeed, a PV streamer and later cut-off low is found near the monsoon low around January 26 (see Figure 7).One may have expected this PV cut-off to cyclonically reinforce the low-level system, or to assist in drawing moisture from the northwest.Instead, the anomaly is soon absorbed by a passing midlatitude trough on January 29.It is still unclear what role, if any, this PV cut-off played in the development of the monsoon low.A PV inversion could provide an estimate of the low-level winds induced by the PV anomaly, but it may not be appropriate at the low latitudes analysed here.
It is unlikely that the upper level PV anomaly was responsible for the initial development of the monsoon low, as the low had formed some days earlier (January 20-23, depending on the tracking method) at a much lower latitude (near 10 • S).The monsoon low was most likely born from a combination of barotropic and convective instability within the monsoon trough (e.g., Diaz & Boos, 2019a, 2019b).This inference is consistent with the series of near-equally spaced low-pressure systems forming around January 26 (see Figure 7).Further analysis would be necessary to verify the role of barotropic instability in this case study; for example, through linear instability theory.
The analysis presented here shows a clear effect of midlatitude Rossby wave breaking on rainfall in the Tropics.Although not the focus of this study, the period examined contained multiple days of extreme heat in southern and eastern Australia, including days with widespread maximum temperatures above 45 • C from January 22 to January 26.This configuration of an active monsoon in the north (including a tropical cyclone) with heatwaves in the south is common in Australia (Boschat et al., 2015;Engel et al., 2013;Parker et al., 2013).
Other factors contributing to this event, but not investigated in detail here, include the Madden-Julian oscillation (MJO; Madden & Julian, 1971), an equatorial Rossby wave (ERW), and a blocking high.The frequency of monsoon lows in the northern Australian region increases by 90% during the active phase of the MJO (Haertel & Boos, 2017), and previous work by Cowan et al. (2019Cowan et al. ( , 2022)); Tsai et al. (2021) suggests that the MJO played a part in this event.Indeed, there is a strong band of equatorial westerlies in the reanalysis (see Figure 7), characteristic of the MJO.These westerlies are part of the northern branch of the cyclonic circulation associated with the low at its peak.However, the MJO alone does not explain this event, particularly the inland reintensification and relative stationarity of the low.In their study, Tsai et al. (2021) identify an ERW that propagates westward from the tropical central Pacific and modulates the rainfall in North Queensland.Although the ERW is not clearly evident in our analysis, it is not the focus of this study.Finally, the blocking high is a factor emphasised by Cowan et al. (2019), responsible for directing persistent low-level easterly winds onto the Queensland coast.Blocking anticyclones are a result of Rossby wave breaking (Berrisford et al., 2007), which further emphasises the importance of wave breaking.
The event is also analysed through ACCESS-S1 model forecasts initialised on January 24.The large spread of tracks between ensemble members indicates the high uncertainty associated with this event (Figure 9).Only a small portion of members (4 out of 33, not shown) capture the observed inland stall.The model also produces rainfall that is too widespread, and it underestimates the observed maximum rainfall near Townsville.These findings are in line with Cowan et al. (2019Cowan et al. ( , 2022)), who also find that forecast skill with ACCESS-S1 reduces significantly at lead times longer than 1 week (January 24, the same date as used here).
The relationship between background winds and system motion is represented well in ACCESS-S1.Combined with the high ensemble spread of the tracks, this relationship emphasises that accurate rainfall forecasts rely on accurate track forecasts, which in turn rely on accurate background wind forecasts.This finding is supported by Hawcroft et al. (2021).These results also highlight the need to consider more than just the ensemble mean in operational forecasting.
One aspect that has not been discussed in this study is the instability within the monsoon trough that led to the initial formation of the low.In the case of the Indian monsoon, where the background state is characterised by strong easterly vertical wind shear, it has been argued that baroclinic instability is necessary for the growth of monsoon lows (Krishnakumar et al., 1992;Mishra & Salvekar, 1980;Saha & Chang, 1983).However, a study by Cohen and Boos (2016) finds that the upshear tilt of the PV necessary for baroclinic instability is not observed in intensifying monsoon lows in the region, implying that baroclinic instability is not the primary mechanism.More recent research suggests that monsoon lows form due to meridional gradients in the zonal wind (barotropic instability) and possibly the humidity ("moisture vortex instability"; Adames & Ming, 2018;Adames, 2021).In particular, barotropic instability has been found to explain the growth of monsoon lows based on reanalysis data (Diaz & Boos, 2019a, 2019b) and shallow-water models (Suhas & Boos, 2023).In the present study, there is both a meridional shear of zonal wind (Figure 7c) and a poleward-increasing moisture gradient (Figure 7b) in the development phase, which are necessary conditions for barotropic and moisture-vortex instability respectively.It is plausible that these are the mechanisms by which the monsoon low formed in North Queensland in 2019.A detailed analysis, however, is left to future research.
An alternate method for describing the development is through vorticity budget analysis (Raymond et al., 2014), following similar case studies by, for example, Kilroy et al. (2016); Smith et al. (2015); Zhu and Smith (2020).Robinson (2020) examined the circulation and moisture budgets for this event, but the work was inconclusive due to difficulties in closing the circulation budget.The analysis on steering could also be extended to include the effect of vortex asymmetries.Finally, a PV inversion (if successful) may elucidate the role of the upper level PV anomaly in this event.
Carpentaria (140 • E, 13 • S) southeastwards and inland on January 26, by which time it has strengthened to 1,430 m.For the next few days, the low drifts southward and westward and weakens slightly to around 1,450 m on January 29.The system reintensifies to a minimum of around 1,370 m on February 2. The low remains near stationary until February 7, when it drifts eastward, reaching the Coral Sea (147 • E, 17 • S) and weakening to above 1,410 m by February 8.The corresponding 315 K PV track detected by the objective algorithm in Section 3 (Figure 5b) shows a broadly similar evolution, although the track is divided into three parts.The first track (green line) follows the initial development near Cape York Peninsula on January 20 to its inland decay on January 28.The second track (blue line) forms on January 30 near Townsville (145 • E, 19 • S), moves westward and stalls north of Mt Isa (140 • E, 19 • S), before moving offshore to the east.This is the track highlighted in Figure 4.A third track (yellow line) F I G U R E 3 Same as Figure 2, but defining slow and fast potential vorticity anomalies as the 10th and 90th percentiles respectively.

F
Frequency distribution of the duration of each track in the study region (as shown in Figure 2).The 2019 North Queensland event is marked red.F I G U R E 5 (a) Track of the monsoon low based on 850 hPa geopotential height and (b) 315 K potential vorticity.(c) Total Australian Gridded Climate Data accumulated rainfall from January 23 to February 10.(d) Time series of the minimum 850 hPa geopotential height and minimum mean-sea-level pressure (MSLP).
Another maximum lies north of the Western Australia coast (125 • E, 16 • S)-this soon develops into Tropical Cyclone (TC) Riley, although it never crosses the coastline.The tropical margin extends from TC Riley to the monsoon low, covering parts of northern Australia.A meridional band of moisture (though below 48 kg⋅m −2 ) extends from TC Riley to the southern coast of Western Australia (125 • E, 35 • S).The rainfall is generally low except near TC Riley, and there is an upper level cyclonic PV anomaly over central-eastern Australia.The overturning PV contours near New Zealand (150-180 • E) indicate Rossby wave breaking.At 850 hPa winds, a strong belt of westerlies lies over Indonesia and northernmost Australia (120 • E, 10 • S), indicating an active phase in the monsoon.There are weak, onshore easterlies along the Queensland coast due to a blocking high in the Tasman Sea (165 • E, 30 • S), and the vertical motion is largely coincident with the relative vorticity maxima.By January 26, the relative vorticity maxima for the monsoon low are larger and more coherent.TC Riley to the west is considerably stronger, and the moist margin is greatly perturbed by these two systems (as well as another low near the western border of the domain), suggestive of barotropic instability along the monsoon trough.A maximum in TCWV lies just east of the Queensland coast and is co-located with strong vertical motion.The anticyclonic Rossby wave breaking (150-180 • E) is more pronounced, F I G U R E 7 (a, d, g, j) Global Precipitation Climatology Project daily precipitation (shading) and 350 K cyclonic potential vorticity (red contours).Solid contours denote 1 PVU intervals starting from 2 PVU and dashed contours denote 0.5, 1, and 1.5 PVU.(b, e, h, k) Total column water vapour (shading) and 700 hPa cyclonic relative vorticity (orange contours).The 48 kg⋅m −2 threshold is indicated by the purple contour.Solid vorticity contours denote 1 × 10 −4 s −1 intervals and dashed contours denote 0.25 × 10 −4 , 0.5 × 10 −4 , and 0.75 × 10 −4 s −1 .(c, f, i, l) Mean-sea-level pressure (MSLP; grey contours), 850 hPa normalised wind vectors and speed (shading), and 400-600 hPa average vertical velocity (dark blue contours).Vertical velocity contours are drawn at 0.5 Pa⋅s −1 and 1.0 Pa⋅s −1 , and MSLP contours are drawn every 4 hPa.The four rows represent key dates January 23, January 26, February 2, and February 8 in that order.

F
I G U R E 8 (a, c) West-to-east and (b, d) north-to-south cross-sections of the low on (a, b) January 28 and (c, d) February 2. Plotted are wind velocity perpendicular to the cross-section (red/blue shading), potential temperature (green contours), and potential vorticity (black contours).Solid potential vorticity contours denote intervals of 1 PVU, and dashed contours denote intervals of 0.2 PVU from 0.2 to 1 PVU.The vertical dashed line indicates the centre of the low.

F
The Australian Community Climate and Earth System Simulator Seasonal prediction version system 1 (a, d) 850 hPa track, (b, e) intensity, and (c, f) total accumulated precipitation from January 25 to February 10. (a)-(c) The ensemble mean, with individual ensemble members in grey lines where applicable.Red markers and lines indicate the point at which tracks terminate in some ensemble members (from January 30).(d)-(f) The "best" ensemble member 2. The European Centre for Medium-Range Weather Forecasts Reanalysis v5 850 hPa track (as in Figure 5) is shown as the dashed black line.The region of heavy rainfall 138-150 • E, 10-22 • S is marked by a black box in the rainfall panels.

F
I G U R E 10 The Australian Community Climate and Earth System Simulator Seasonal prediction version system 1 mean-sea-level pressure (MSLP; grey contours), 850 hPa normalised wind vectors and speed (shading), and 500 hPa vertical velocity (dark blue contours) for key dates (a, b) January 26 and (c, d) February 2. (a, c) The ensemble mean and (b, d) the "best" ensemble member 2. MSLP contours are drawn every 2 hPa, and vertical velocity contours are drawn at intervals of 0.05 m⋅s −1 .

F
I G U R E 11 The Australian Community Climate and Earth System Simulator Seasonal prediction version system 1 850-500 hPa average background winds (blue) and system motion (red) for (a, c) zonal and (b, d) meridional directions.(a, b) The ensemble mean and