Land-Locked Convection as a Barrier to MJO Propagation Across the Maritime Continent

.


10.1029/2022MS003503
2 of 20 morning, the land breeze, which is associated with weaker precipitation, propagates offshore and suppresses precipitation on the coast (Chen & Houze, 1997).
Numerical models often underestimate the precipitation in this region, in part due to a poor representation of the DC of precipitation around the islands (Neale & Slingo, 2003) (Baranowski, Waliser, et al., 2016).Increasing the model resolution reduces the dry bias over the MC and has been linked to better-resolved surface conditions and land-sea contrast (Schiemann et al., 2014), but models are still too quick to trigger precipitation over land and exaggerate the amplitude of the DC over land compared to what is simulated over water (Lee & Wang, 2021;Li et al., 2017;Love et al., 2011).
The MC acts as an obstacle to the eastward propagation of the MJO from the Indian Ocean toward the western Pacific -its barrier effect is responsible for weakening most MJO events that cross the region, and completely dissipating 45%-50% of them (Kerns & Chen, 2016, 2020;C. Zhang & Ling, 2017).Some studies focus on the physical effect related to the blocking of flow by topography (e.g., Wu & Hsu, 2009), and its direct consequences such as reduced air-sea fluxes over islands compared to the surrounding ocean (Birch et al., 2019;Sobel et al., 2010).More studies focus on dynamical barriers to MJO propagation, such as the westward propagation of dry air that meets the MJO over the MC (DeMott et al., 2018;Feng et al., 2015), the Warm Pool Dipole (L.Zhang & Han, 2020), and recently, the DC has been identified as an important contributor (e.g., Hagos et al., 2016;Ling, Zhang, et al., 2019).The MC barrier effect is exaggerated in most general circulation models (Ling et al., 2019), leading to a prediction barrier to the MJO.As one of the largest sources of tropical intraseasonal predictability, the MJO's downstream influences cannot be accurately resolved without capturing its propagation (or dissipation) over the MC.
The influence of the MJO on the DC over the MC is clear and can be explained by an influx of surface westerly winds which increase boundary layer convergence with the background trade winds, and a large supply of moisture that both accompany the active MJO (Lu et al., 2019).Rauniyar and Walsh (2011) and Oh et al. (2012) found that during the active phase of the MJO precipitation over water is increased, but precipitation over land is reduced, and the timing of peak precipitation is delayed.The DC of deep convective clouds was found to be amplified during active MJO over both land and water (B.Tian et al., 2006), but Peatman et al. (2014) show that over the islands of the MC, outgoing longwave radiation is no longer a good proxy for precipitation.Peatman et al. (2014) and Sakaeda et al. (2017) also note that the strongest amplitudes of the DC are seen in the convectively suppressed conditions before the arrival of active precipitation associated with the MJO, when the skies are most cloud-free.All these results show that the MJO is carried through the MC over water (C.Zhang & Ling, 2017).
The influence of the DC on MJO propagation is more difficult to infer, but land convection is frequently identified as the main culprit for the MC barrier effect related to the DC.Ling, Zhang, et al. (2019) find that one factor that separates crossing MJO events from those that dissipate is a strong increase in the DC ahead of precipitation (as described by Peatman et al., 2014).This increases soil moisture ahead of the MJO and dampens the land DC during active MJO-more so for crossing MJO events than the ones that dissipate.C. Zhang and Ling (2017) come to a similar conclusion in a different manner -they suggest that the inhibition of convective development over water could be the reason behind the barrier effect.The MC barrier effect seems to be strengthened either when precipitation over land is strong, or when precipitation over water is weak -or both.
Most other studies focusing on the MC barrier effect rely on modeling, where parameters are changed, and their effects examined.The observations of an enhanced DC of precipitation ahead of the active MJO are reproduced in cloud-resolving simulations, while topography plays a role in where precipitation develops and varies among islands (Wei et al., 2020).Inness and Slingo (2006) find that at low resolution, topography as a physical barrier is more important than the presence of islands themselves.But at higher resolution, many studies that modify the DC in one way or another find that weakening the DC over land leads to a weaker barrier to MJO propagation (e.g., Hagos et al., 2016;Oh et al., 2013;H. Tan et al., 2022;Zhou et al., 2021).
Though some studies have already performed similar terrain modifications as what we show here (H.Tan et al., 2022;Zhou et al., 2021), we go a step further and separate the effects of MC topography from the effects of its DC and land-sea contrast and identify physical processes through which those impact MJO propagation.We focus our analysis on high-resolution model simulations of a single well-observed MJO event that occurred in November-December 2011.The modeling configuration, MJO tracking, and our unique way of analyzing the DC 10.1029/2022MS003503 3 of 20 of precipitation are described in Section 2. Section 3 focuses on applying the methods to 20 years of precipitation data to establish baseline differences in the DC of precipitation between MJO and non-MJO environments.Section 4 addresses the MJO characteristics, while Section 5 describes the DC differences between the model simulations and observations.In Section 6, we establish enhanced land-locked convection as a physical mechanism that strongly contributes to the weakening of the MJO over the MC.The results are summarized and discussed in Section 7.

Model Configuration and Simulations
The atmosphere-ocean coupled model used in this study is the Unified Wave Interface -Coupled Model (UWIN-CM) (Chen & Curcic, 2016;Chen et al., 2013).All simulations use the configuration that was described in Savarin and Chen (2022b) which includes convection-permitting resolution, atmosphere-ocean coupling, and a modification to air-sea flux parameterization to yield a good simulation of the observed MJO event.The modification to air-sea latent and sensible heat fluxes was implemented within the parameterization of the model surface layer by reducing the near-surface buoyancy contribution to wind speed within the flux calculation over water (see Savarin & Chen, 2022b).Simulations are initialized at 00Z on 22 November 2011 and integrated in time for 15 days (360 hr), ending on 7 December 2011.
Briefly, the simulations in this study use the Weather Research and Forecasting (WRF) model v3.6.1 with the Advanced Research (ARW) dynamical core (Skamarock et al., 2008) for the atmosphere component, and the Hybrid Coordinate Ocean Model (HYCOM) v2.2.99 for the ocean (Metzger et al., 2014).The simulated region encompasses the IO and MC with three nested domains of 36-, 12-, and 4 km resolution (Figure 1a); the outer domains use the Tiedtke convective parameterization (C.Zhang et al., 2011) and the 4-km domain does not use a convective parameterization.HYCOM grid spacing is a uniform 0.08°.Initial and boundary conditions for the simulations come from the European Centre for Medium-Range Weather Forecasts (ECMWF) operational forecast fields for the atmosphere (from CR37R2) and daily mean HYCOM global analysis for the ocean (Cummings, 2005;Cummings & Smedstad, 2013).Similar coupled model configurations have been successfully used to model the MJO (e.g., Wang et al., 2021).
The control simulation (CTRL) has real topography over the MC and is configured identically to AO4-FLX in Savarin and Chen (2022b).We then use the same initial and boundary conditions for two idealized simulations in which we modify topography and bathymetry over the MC to different degrees.In the FLAT experiment, MC topography is flattened to a uniform 10 m elevation, and the land-use category for the flattened terrain is changed to evergreen broadleaf forest (Figure 1b).Using the metgrid program provided by the WRF preprocessing system (WPS), atmosphere initial conditions are extended to the surface where topography has been modified, and the ocean initial conditions remain unchanged.In the WATER experiment, MC land is converted to 50 m deep ocean (Figure 1c).The atmosphere initial conditions are the same as in the FLAT simulation, but the newly created ocean has no currents, while temperature and salinity fields are interpolated from the nearby ocean and smoothed (a method similar to Chen et al., 2001)thus the ocean temperatures, salinities, and SSTs near the MC are smoother than in CTRL and FLAT simulations.

Data
Several observational data sets are used to evaluate the model simulations' performance and to explore the physical processes associated with the MJO and the MC.For precipitation, we use the Integrated Multi-satellitE  Retrievals for GPM (IMERG) satellite precipitation estimates (V06B; Huffman, Bolvin, et al., 2019), which are available in half-hourly intervals and at a spatial resolution of 0.1°.Twenty years of data (June 2000-June 2020) are used for MJO tracking and climatology, but when comparing with model simulations, only the 15 days from November 22-December 7 are considered.In addition to precipitation, we also use the Cross-Calibrated Multi-Platform (CCMP) gridded surface vector winds (V2.0;Atlas et al., 2011), which are available 6-hourly and at 0.25° spatial resolution.To create a distance-from-coastline reference framework for our analysis of the DC over the MC, we use the ETOPO1 data set, a global relief data set at a spatial resolution of 1 arcminute (NOAA National Geophysical Data Center, 2009).

Large-Scale Precipitation Tracking of the MJO
The large-scale precipitation tracking (LPT) algorithm (LPT; Kerns & Chen, 2016, 2020) is used to track the MJO-associated precipitation in the IMERG data set.The algorithm tracks a spatially smoothed 3-day precipitation accumulation that exceeds a chosen threshold over an area larger than 3 × 10 5 km 2 .Kerns and Chen (2020) use a 12 mm precipitation accumulation threshold on 20 years (1998-2018) TRMM 3B42 data and identify 215 MJO events.Before the application of the algorithm to the IMERG data set, precipitation data is conservatively re-gridded to 0.25° spatial and 3-hourly temporal resolution to match that of TRMM 3B42.After precipitation is tracked, additional constraints are used to separate MJO events from other large-scale systems, such as a minimum duration of 7 days, and consistent eastward propagation.In the IMERG data set, the November-December 2011 MJO event remains cohesive and propagates through the MC up to a precipitation threshold of 22 mm.When model simulations are compared to observations, a threshold of 17 mm is used instead of 12 mm to highlight differences between simulations, as the model tends to overproduce precipitation (see Savarin & Chen, 2022b).At lower thresholds, MJO propagation over the MC can be seen in LPT tracking, but it tends to present as a series of discrete longitude jumps, as the tracking algorithm attempts to connect distinct areas of precipitation with little overlap.

Diurnal Cycle Analysis
We analyze the DC relative to its distance from coastline, which can clearly show us the cycling between land and sea breezes in the MC.The method used is illustrated in Figure 2. We use the 1-arc-minute global relief model data set, ETOPO1 (Amante & Eakins, 2009) to define a land mask (where global relief is above sea level) and remove islands and bodies of water smaller than 400 km 2 -the modified land mask is shown in Figure 2a.Then the Haversine formula is used to calculate great-circle distances from each point to every other point on the globe, and for each point, the distance to its nearest coastline is chosen (Figure 2a).Negative distances denote inland areas, and positive distances denote areas offshore.In this study, we focus our attention to the western MC (90-120°E, 10°S-10°N), and only data from this region are considered whenever the DC is analyzed.The number of points in each 25-km distance bin within is MC region is shown in Figure 2b using a spatial resolution of 0.1° to match the GPM IMERG precipitation data set.
To construct DC composites, precipitation data is first converted to local solar time (LST), which only depends on longitude and is rounded to the nearest hour.The LST offsets inside the MC box range from UTC + 6 at 90°E to UTC + 8 at 120°E.Then precipitation data is binned into 25-km bins for the entire data record and averaged for every LST hour.The resultant distance-from-coastline Hovmöller diagram is shown in Figure 2c for 20 years (June 2000-June 2020) of IMERG data, and the DC is repeated twice for completeness.Analysis can then be simplified into a more quantitative line diagram in Figure 2d, where color represents the amount of precipitation at different LST.Displaying the DC in this manner clearly shows the cycling of precipitation between land and ocean (the alternation between land-and sea-breeze) and adds a spatial component to our analysis.
The method described here can be applied to any field, scalar or vector, regardless of whether it is gridded or not.With the additional calculation of bearing based on the Haversine formula, we can obtain the direction from any point to its nearest coastline.This allows us to project vector fields such as surface winds to their across-and along-coastline components with trigonometric functions.

Diurnal Cycle of Precipitation in MJO and Non-MJO Environments
We start by examining the impact of the MJO on the DC of precipitation over the MC in a climatological sense.Large-scale precipitation tracking tracking is used to separate the MC area (outlined in 2a) into two categories: 10.1029/2022MS003503 5 of 20 active MJO regions directly inside the convective envelope, and the non-MJO regions outside the convective envelope and its 5° filtering radius.The areas inside the 5° filtering radius between the MJO convective envelope and non-MJO regions are not considered for this analysis.The DC is then composited for each category and shown in Figure 3, which shows the Hovmöller diagrams of the DC in active MJO and non-MJO environments and their difference (a-c), as well as the more quantitative line diagrams of the DC (d-e), and the average precipitation and the DC amplitude between the two environments (f), all as a function of distance from coastline.The DC amplitude is defined as the difference between maximum and minimum rainrates that occur over the course of a day for each distance from the coastline.
Compared to non-MJO environments, the amount of precipitation is strongly increased during MJO events, which clearly increases the amount of precipitation within the DC.However, the precipitation increase is not uniform throughout the course of the day, nor is it uniform in its position relative to the coastline.The amount of precipitation over water is more than doubled during most of the day, with the greatest increase in the early morning.Precipitation over land shows a generally lesser increase than over water, and it is most prominent in the evening -in the early morning, there are even times when the DC far inland is unaffected by the MJO (Figure 3c).These results show that the active MJO increases precipitation over the MC in accordance with the DC but shows a clear preference for amplifying precipitation over water; it rains more where it would already be raining, over land in the afternoon, and over water in the morning.Over land, the morning precipitation is not much different between MJO and non-MJO composites, while afternoon precipitation is strongly amplified, and the mean precipitation is shifted upward by 2-4 mm day −1 .Over water, the MJO more than doubles the amplitude of the DC, but the precipitation throughout the entire day is also strongly increased, with the mean precipitation increasing by 10-20 mm day −1 (Figure 3f).These results confirm that the MJO (as defined by precipitation tracking) is largely carried through the MC over water, and that the DC persists even under large-scale MJO conditions.

MJO Characteristics in Model Simulations
To evaluate the relative effects of flattening topography and removing the land and its associated DC over the MC, we consider how the MJO is represented in observations and our simulations.The large-scale precipitation and surface wind fields for the simulations are shown in Figures 4 and 5 shows time series of precipitation over the MC, Figure 6 shows the LPT-tracked MJO events.Figure 7 summarizes some statistics that help characterize the differences in the MJO events among model simulations.
The observed November-December 2011 MJO event is the second MJO event observed during the intense observation period of the DYNAMO field campaign (Chen et al., 2016;Yoneyama et al., 2013).The observed MJO convection is organized in two strong eastward-propagating convectively-coupled Kelvin waves that initiate over the IO.The first Kelvin wave initiates on 22 November in the eastern IO (near 65°E) and dissipates over the western MC (near 100°) on 27 November.The second Kelvin wave initiates to the east of the first one, propagates through the MC and persists until the end of the model simulation period (as described by Baranowski, Flatau, et al., 2016;Zhu & Li, 2017).Within the eastward-propagating Kelvin waves, we can see the more frequent and shorter-lived westward features, which highligts the multiscale nature of MJO convection.
The CTRL simulation does a good job at representing the large-scale environment of the observed MJO event, as can be seen in the Hovmöller diagrams of rain rate and surface zonal winds in Figure 4 and was described in more detail in Savarin and Chen (2022b), their AO4-FLX experiment).The CTRL's precipitation signal is noisier than  in observations due to its higher resolution and a high bias in precipitation Savarin and Chen (2022b); the two distinct eastward-propagating Kelvin waves are apparent, though not as prominent as in observations.The post-MJO suppression in CTRL is not as strong as was observed, but it does propagate through the MC, even though the signal is not as smooth as over the IO or in observations (Figure 4b).The surface westerlies associated with the MJO are well reproduced, and they persist over the IO after the MJO has propagated east.
Flattening MC terrain results in small changes in the large-scale environment compared to the CTRL.The precipitation within the first Kelvin wave over the IO becomes more prominent (though it still dissipates over the western MC), while the second Kelvin wave is weakened.Over the MC, precipitation seems more scattered and the MJO-associated eastward-propagating precipitation is more difficult to distinguish until 3 December, where a heavy rainfall event forms near 130°E, to the east of our defined MC region.Surface westerly winds over the MC are stronger than in CTRL, which can be attributed to the removal of topographical barriers.When MC land is removed in the WATER simulation, we see a lot more precipitation, a much clearer eastward propagation associated with MJO convection and an amplification of the second Kelvin wave over the MC and to its east.After MJO passage, precipitation suppression over the IO is stronger than in previous simulations, as are surface westerly winds -this is a result of reduced friction over the entire MC.
Figure 5 shows time series of MC precipitation averaged over the MC box outlined in Figure 2 for IMERG observations and model simulations.All model simulations reproduce an increase in precipitation associated with the MJO that begins after November 25, ahead of the MJO centroid entering the region though the MJO's leading edge is already over the MC (Figure 7a).The WATER simulation produces the largest amounts of precipitation, and the most precipitation increase associated with the MJO, while simulations containing land produce less of both.This indicates that during MJO passage (the time range during which the MJO centroid is located over the MC is outlined with colored horizontal bars on the bottom of Figure 5), the presence of MC land is disruptive to MJO-associated precipitation enhancement.
These large-scale differences are reflected in LPT-tracked MJO events shown in Figure 6.At the 17 mm precipitation threshold, both the CTRL and FLAT simulations dissipate over the MC before the end of the simulation, while the observed and WATER MJO events propagate smoothly.The average 24-hr propagation speed in observations is 5.8 m s −1 , which is closely matched by 5.6 m s −1 in WATER.The propagation speeds of CTRL and FLAT MJO events are 3.4 and 3.5 m s −1 , respectively, so in addition to dissipating over the MC, they are also slower.This could be related to the strength of the second Kelvin wave that carries the MJO through the MC -it is strongest in WATER, but weakens when MC islands are present, with or without topography.
Soon after the bulk of CTRL and FLAT MJO events extends into the MC region (after 28 November, when the MJO centroids enter the MC, Figure 7a), the MJO areain both simulations is reduced by about 30% over the course of 2 days (Figure 7b).The MJO area then remains relatively steady for 4 days as the MJO propagates eastward, before beginning to dissipate on 3 December (Figure 7b).The FLAT MJO dissipates faster than the CTRL 9 of 20 MJO.The initial reduction in MJO size when first entering the MC is also present in observations -but after the initial weakening, the observed MJO's size remains relatively steady until the end of the simulation period.As there is no land present in the WATER simulation, the MJO area remains relatively steady throughout the simulation, with some size fluctuations as the tracking algorithm expands to capture convection over the western Pacific.Additionally, the average rain rate within the MJO convective envelope (Figure 7c) is lower in FLAT than in CTRL, indicating a weaker MJO event.
These results show that, as expected, when all obstacles are removed from the MJO's path (such as in WATER), its propagation is smooth, and its precipitation does not weaken.Removing mountains alone but keeping islands where they are (as in FLAT) has a much smaller impact on MJO propagation (compared to CTRL), and, surprisingly, that impact acts to weaken the MJO and impede its propagation even further.In the next section, we take a closer look at the diurnal precipitation patterns over the MC and how they can disrupt MJO propagation to explain this unexpected result.

Diurnal Cycle of Precipitation
In this section, we examine the DC of precipitation over the MC (90-120°E, 10°S-10°N) for the period from 22 November to 6 December 2011 to evaluate how well the DC is represented in CTRL, and how it changes among model simulations.Figure 8 shows the average rain rates over all MC land and ocean points (a, b), and the percentage of total rain that falls over them (c) relative to LST.To put our 15-day period into broader context, observations for the model period are shown in solid colors and bars, while the dashed lines and hatched bars show the 20-year IMERG climatology.Unsurprisingly, the amount of precipitation in the 15-day period is higher than the 20-year aver ages at all times of day.This is due to two factors: first, we are considering a shorter period, so extreme rain rates would contribute more strongly to the average, and second, the 15-day model period contains an MJO event, which increases the amount of precipitation over the MC-especially over water (Figure 3).The signature of the MJO can be inferred from the fact that at any time of day, the portion of precipitation that falls over water is greater in the 15-day composite than in the 20-year one (Figure 8c).Apart from the difference in magnitude, the 15-day, and the 20-year DC composites over land and water have the same characteristic timing, indicating that the method we use for analyzing the DC is appropriate even for such short time periods.
We noted previously that both the CTRL and FLAT simulations tend to overproduce precipitation (Figure 4), but when considering only precipitation over the MC ocean points (Figure 8b), the average rain rates in model simulations accurately reproduce IMERG observations both in intensity and timing of precipitation extrema.In the WATER simulation, the amount of precipitation over water is higher over the course of the day, and closer to the DC we would see over open ocean, with smaller amplitude and a precipitation maximum slightly earlier in the day (Nesbitt & Zipser, 2003).Over MC land, the timing of the diurnal precipitation extrema still matches that of observations, but the precipitation intensity is consistently exaggerated (Figure 8a), which results in proportionally more rain falling over land.The land-sea contrast present in observations and CTRL and FLAT simulations results in land-locked convection in the afternoon, with convective systems that are much more intense than what we see over water.
The separation of land and water points for the DC of precipitation in Figure 8 shows that in model simulations, convection over land is more intense than in observations, with a slightly lower-intensity and longer-lasting precipitation peak in FLAT.However, this way of looking at land and sea precipitation obscures the changes in precipitation patterns over land that arise from imposed terrain modifications.To investigate those, Figure 9 shows distance-from-coastline relative Hovmöller diagrams of rain rate for the 15-day period in observations and CTRL and FLAT simulations (top), and their 15-day composites (bottom).Seen in this manner, we can note that flattening MC terrain results in changes in the location of precipitation, as well as in precipitation frequency.
Compared to observations, the DC of precipitation in CTRL is stronger and more regular, with sea breeze precipitation propagating far inland on most afternoons while observations show more day-to-day variability.As inferred from Figure 8, the amount of precipitation over land is exaggerated by up to 80% while the amount of precipitation over water is simulated more accurately (Figure 9e); therefore, the precipitation in the model is more land-dominated than in observations.Flattening terrain results in precipitation pattern changes that can be separated into two regions over land: the near-coastal region (within 100 km inland), and the far-inland region (more than 200 km inland).In the near-coastal region, the FLAT DC is diminished but remains regular while in the far-inland region, it is strongly amplified in intensity but reduced in frequency.Near the coast, the reduction in peak precipitation is due to two effects-concurrent effects of sea-and valley-breezes that amplify onshore flow in the early afternoon, and mountains near the coast (along the west coast of Sumatra) that both act to amplify precipitation in CTRL but not in FLAT.But the more interesting changes are happening far inland, where sea breezes from different sides of islands (mainly Borneo) in FLAT converge and grow into organized mesoscale convective systems that are more intense, larger, and last longer than in CTRL.Outside of MJO conditions (before 29 November), these systems persist into the next morning and suppress precipitation for the rest of the day, creating a 2-day cycle.During active MJO, the increased moisture supply means that these large systems are formed every day.In a 15-day composite from the FLAT simulation (Figure 9f), inland precipitation peaks are significantly stronger than in CTRL even though they occur less frequently.
To summarize, these results show that model simulations with land-sea contrast simulate the DC over MC water accurately, but show more, and more intense land-locked convection in the afternoon.When terrain is flattened, we see the coastal sea-breeze precipitation is diminished in amplitude, but a convergence of sea-breezes from all around islands, which are no longer disrupted by terrain, results in an amplification of convection far inland.In the next section, we focus on the differences in convective systems that arise from modifying MC terrain.).The Maritime Continent area is defined from 90 to 120°E, 10°S-10°N as in Figure 2a, and it is bounded by vertical lines at 90°E and 120°E in (a).Thin lines in (c) show hourly rain rates while thick lines show its 24-hr running mean.Observations are from IMERG; CTRL simulation (orange) contains real topography, which is leveled between 90°E and 160°E in FLAT (green), and removed in WATER experiments (blue), as shown in Figure 1.

Land-Locked Convection and Suppression of MJO Precipitation Over Water
In this section, we take a closer look at the differences in land-locked convection between the CTRL and FLAT simulations.As noted previously, the FLAT simulation produces inland convective systems that are more intense, larger, and longer-lasting than in the CTRL simulation (Figure 9, where mountains disrupt the convergence of sea breezes from different sides of the island.In Figure 10, we show an example of the evolution of one such convective system in the FLAT simulation that developed overnight between 26 and 27 November.Figure 11 demonstrates that the case shown in Figure 10 is not unique, and the associated patterns of low-level convergence and moisture supply are illustrated in Figure 12. The evolution of sea-breeze fronts into a large mesoscale convective system (MCS) in the FLAT simulation in Figure 10 shows precipitation(color), surface winds (vectors), and negative 500-hPa vertical velocity regions over water (brown shading) on the left.On the right are vertical cross-sections of hydrometeor mixing ratio greater than 0.1 g kg −1 to indicate clouds (dotted in black), potential temperature anomaly (color) from the previous hour, and zonal and vertical wind components vectors.The vertical cross-sections are plotted with longitude and averaged between 1°S and the equator, and the vertical winds have been multiplied by a factor of 10 for better visibility.The three rows of figures correspond to three different times, one in the early stage of MCS development (21-22 LST on 26 November), when convection is beginning to converge inland, along the south-east coast of Borneo (top), one in the mature stage (00-01 LST on 27 November) when convection is organized on a very large scale (middle), and one in the dissipating stage (05-06 LST on 27 November), when convection is dissipating in this region, but intense precipitation has migrated to the north and west (bottom).
The evolution of the convective system in Figure 10 indicates that in the FLAT simulation, a large, organized, and robust MCS and its associated circulation develop and propagate over Borneo.During the early stage, the system is just beginning to propagate and develop inside the averaging box near 115°E (Figure 10a).There is some upward motion, and a weak warm anomaly is beginning to develop in the mid-troposphere, coincident with where clouds are present (Figure 10d).Three hours later, the system has matured and is more than 200 km across, with strong updrafts and strong mid-and upper-tropospheric heating, indicating that the MCS is beginning to develop its own circulation and a broad upper-level region of stratiform clouds (Figures 10b and 10e).The presence of a deep inflow layer we can see in Figure 10e is associated with mature MCSs, where large regions of stratiform clouds and precipitation are likely present, and has been numerically shown by Mechem et al. (2002Mechem et al. ( , 2006)).Five hours later, the system has grown to over 300 km across and precipitation is dissipating inside the averaging box as the system is propagating away from it.There is subsidence from the mid-troposphere, and warming has moved closer to the surface, while upper levels begin to cool (Figures 10c and 10f).
To show that the development of the MCS described in Figure 10 is not a singular occurrence but a systematic difference between CTRL and FLAT simulations, we take the five most intense convective events that occur far inland and compare the results between the simulations.The five most intense events are determined based on average rainfall rate more than 200 km inland and marked with stars next to the Hovmöller diagrams in Figure 9. Figure 11 shows the time series of far-inland rain rate (top), with the highlighted intense convective events and the rain rate thresholds that need to be exceeded for each simulation.The thresholds have been chosen so that  and 120°E).The stars on the left indicate the five most intense convective events that occurred more than 200 km inland (see Figure 11).CTRL simulation contains real topography, which is leveled between 90°E and 160°E in FLAT, and removed in WATER experiments, as shown in Figure 1.

10.1029/2022MS003503
13 of 20 they result in the same number of hours within the simulation during which the threshold is exceeded.A rain rate threshold of 3.6 mm hr −1 in FLAT results in 20 hr separated between 5 convective events, while the same number of hours and convective events are identified with a threshold of 2.2 mm hr −1 in CTRL -the convective events are over 60% more intense in FLAT.The precipitation for the highlighted times is composited together for the CTRL and FLAT simulations (Figures 11b and 11c), and their difference is shown in Figure 11d, with red colors indicating where rain rates are higher in FLAT than in CTRL.
Large land-locked convective events in the FLAT simulation are spread over larger and more central areas of islands, and the precipitation that occurs in them is more intense when compared to the systems that develop in CTRL.These large MCSs suppress precipitation over the surrounding waters (brown shading in Figure 10) where the MJO is attempting to enhance precipitation around the same time of day.In FLAT, the larger and more intense MCSs can develop because there is no terrain disrupting the convergence of sea breezes from different sides of the islands, and there are no mountains forcing upward motion in specific locations.
The systematic differences in precipitation patterns between CTRL and FLAT are also evident in the accompanying patterns of low-level convergence and water vapor supply shown in Figure 12.The compositing is done in the same manner as for precipitation in Figure 11, and we can see large-scale convergence over most of Borneo in the FLAT simulation (Figure 12c).In CTRL, the convergence region is smaller and less contiguous, and we also see the signature pattern of elevated topography, with dipoles of convergence and divergence in the north of the island (Figure 12a).The convergence and water vapor mixing ratio shown are averages for a layer that spans between 1,000 and 700 hPa, indicating that the low-level convergence is not confined solely to the boundary layer, which implies a presence of mature MCSs and elevated mid-level moisture (Mechem et al., 2002).But the sharper difference is evident when looking at low-level moisture availability -due to elevated terrain in CTRL, large areas of Borneo show much lower water vapor content near the surface than in FLAT (Figures 12b and 12d).In fact, during the most intense convective events, low-level convergence in FLAT is 58% higher than in CTRL, and the low-level water vapor shows a 3% increase, indicating that convection mainly grows due to increased and widespread convergence.Figure 12 explains why convective systems can grow larger and stronger in FLAT in a physical sense -the collocation of low-level convergence and moisture supply can support precipitation.In CTRL, though we see large areas of convergence, the moisture supply is lower, so the systems can only grow in a limited capacity.
These effects can be seen in the differences between the composite DC in the CTRL and FLAT simulations separated into the MJO and non-MJO environments shown in Figure 13, following the same method as in Section 3. We can clearly see the amplified enhancement of far-inland convection in the early morning on the FLAT simulation, while far-inland convection is slightly suppressed by the MJO in CTRL.At the same time, the enhancement of precipitation over water in MJO environments is smaller in FLAT than in CTRL, indicating that the mountains in CTRL present a physical barrier to MJO flow, but they also disrupt the large-scale organization of convection due to convergence of multiple sea breeze fronts.The resulting MCSs that develop over land in CTRL occur on a smaller scale -which is still disruptive to the MJO, but to a lesser extent.So, in a way, mountains can help the MJO propagate across the MC by disrupting the large convective systems that would develop in their absence.

Summary and Discussion
This study investigates the MC barrier effects to MJO propagation through a systematic analysis of the impacts of MC terrain and land-sea contrast.Three atmosphere-ocean coupled simulations at convection-permitting resolution are conducted to evaluate the responses in MJO evolution and eastward propagation to changes in MC topography.Including atmosphere-ocean coupling and convection-permitting resolution in our model precipitation in the afternoon (the sea breeze; Figure 8a).The resulting convective systems over land can be separated into two types.The near-coastal convection that is directly forced by sea-breezes (and the background flow) is present on each day of the model simulation.The far-inland convection that is forced by the convergence of multiple sea breezes (and the background flow) is present on each day during active MJO conditions, but only occurs every other day before MJO arrival (Figure 9c).The far-inland sea breeze convergence results in the formation of very large organized MCSs that develop their own circulation, produce heavy-precipitation, and last well into the morning hours (Figures 9c and 10).The long-lasting systems then suppress the far-inland convection on the following day (when there is no MJO) due to reductions in mid-upper-level moisture and insolation-induced surface heating.During active MJO conditions, the intense far-inland MCSs are triggered every day due to increased background moisture and upward vertical velocity.
The large daily MCSs that form far inland in the FLAT simulation last into the next morning, suppressing precipitation that is supposed to be initiating over coastal waters at the same time due to the land breeze.The early morning is also the time during which, climatologically, the MJO tends to most enhance precipitation over water, and this local suppression works against that.Therefore, the large land-locked MCSs that develop in the later afternoon and persist until morning reduce the precipitation enhancement over water that happens due to the MJO (Figure 13c) and result in a weakened MJO with a discontinuous propagation across the MC (Figures 6c and 7).In addition to being land-locked, these MCSs repeatedly form over central Borneo due to the island's size and shape, and remain stationary with respect to the eastward-propagating MJO.
The mountainous terrain added in CTRL provides a disruption to the FLAT DC that results in a change in diurnal precipitation patterns.Compared to FLAT, the amount of precipitation falling over land is increased in CTRL (Figure 8a), but it is distributed in smaller systems that are less disruptive to the MJO.The amplitude of the DC near the coastline, both of land and water) is increased (Figure 9e) while the systems that develop far inland are smaller in area and less intense (Figure 11).The low-level convergence associated with these systems is much weaker than in FLAT due to flow disruption by mountains, and as they cannot grow as large, they induce less suppression to precipitation developing over nearby waters (Figures 12,13a and 13b.Compared to FLAT, the MJO propagation in CTRL is smoother, contains more intense precipitation, and dissipates later (Figures 6b and 7c).This implies that considering that the land-sea contrast is disruptive to the MJO, mountains act to reduce the disruption to MJO propagation, because they disrupt the even-stronger MCSs that would develop in their absence.This disruption of large-scale convection within the MJO can also be seen in the strength of the second convectively coupled Kelvin wave embedded within the MJO convective envelope which carries the observed MJO  2a, and the MJO and non-MJO environments for the modeled period are defined as in Section 3 and Figure 3. CTRL contains real topography, while in FLAT, topography over the MC is leveled to 10 m above sea level, as shown in Figure 1.across the MC (Figure 4).This second Kelvin wave is strongly apparent in WATER, but appears weakened in CTRL, and further weakened in FLAT.It is possible that the Kelvin wave that is weakened by persistent and stationary MCSs that form over Borneo is then unable to organize the further eastward propagation of the smaller-scale convection within the MJO, which leads to its early dissipation in CTRL and FLAT simulations.
The presented results showing the disruptive role of the MC islands on the MJO through the forcing of land-locked convective systems (and the exaggeration when topography over the region is flattened) are consistent with the moisture mode models of the MJO (Sobel & Maloney, 2013).Based on the moisture mode theory and previous work with the UWIN-CM coupled modeling system (Savarin & Chen, 2022b), surface flux feedbacks are important for the eastward propagation of the MJO.In our WATER simulation, the surface enthalpy fluxes over the MC remain high, maintaining the low-level moisture supply as the MJO propagates through the region due a combination of high winds and warm SSTs.In CTRL and FLAT simulation, land surfaces, with their lower specific heat capacity, reduce the surface enthalpy fluxes and the available moisture in addition to forcing precipitation to shift between land and ocean over the course of the day.
Our results show that the active MJO in IMERG observations increases the amount of precipitation throughout the MC, and thus increases the amplitude of the DC over both land and water (Figure 3), though the increase over water is dominant.These results disagree with previous studies on the subject, which found that while the amplitude of the DC over water is increased by an active MJO, the amount of precipitation over land is reduced (Oh et al., 2012;Rauniyar & Walsh, 2011).We believe the reason for this discrepancy lies in the methodology of MJO and DC identification.Most other studies of the MJO identify events based on the Real-Time Multivariate MJO Index (RMM, Wheeler & Hendon, 2004), or similar indices based on global anomaly fields.Our MJO identification method relies on LPT, which directly tracks MJO precipitation, and only considers the points that lie inside the MJO convective envelope as active, so that at any one time, parts of the MC can be inside the MJO, while other parts are not.
Our results also disagree with the earlier study by Inness and Slingo (2006) which finds that it is the mountains, and not the presence of islands, that blocks MJO propagation through the MC.However, their model simulations were performed at very low resolution (2.5° × 3.75°), and many studies have shown that increasing resolution helps with the representation of the MJO (Love et al., 2011;Savarin & Chen, 2022a), so their findings could be attributed to something other than the barrier effect of the MC.
A similar set of convection-permitting simulations with real and flattened topography was performed by H. Tan et al. (2022) and by Zhou et al. (2021), both without dynamic atmosphere-ocean coupling and for two different MJO events.H. Tan et al. (2022) find similar high biases in the DC of land precipitation that are characteristic of our simulations, but also show a low bias in the amplitude of the DC over water, indicating that air-sea coupling could be an important contributor to the variability of precipitation over water.Their results generally agree with our study in that when topography is removed, the peak precipitation over land is reduced, but tapers off more slowly than when topography is present (e.g., Figure 8a).Though their analysis focuses on different aspects of the DC, the fact that they find similar differences in their simulations makes the results of our study more robust.
We recognize that the afternoon peak land-locked convection in our coupled model simulations is higher than indicated by IMERG observations (Figure 8a), though it is unclear whether the land precipitation bias is as large as it appears.Previous studies have found the resolution of IMERG to be high enough to accurately represent the DC of precipitation (e.g., J. Tan et al., 2019), and our results qualitatively compare well with precipitation radar studies in the region from the TRMM era (e.g., Biasutti et al., 2012).But the accuracy of hourly IMERG precipitation retrievals over the MC region's sharp land-sea contrast areas and dynamic terrain has not yet been thoroughly evaluated.Some evaluation studies indicate that IMERG tends to underestimate precipitation associated with tropical cyclone precipitation over the United States (e.g., Mazza & Chen, 2023;F. Tian et al., 2018), while a study by Hayden and Liu (2021) showed both regional under-and over-estimates in the tropics.In addition, many modeling studies performed at higher resolutions show a high bias in land convection over the MC; at lower resolutions, the timing of the DC as well as its amplitude are frequently misrepresented (e.g., Love et al., 2011;Watters et al., 2021).
Though this study only contains model simulations of a single (though well-observed) MJO event, our findings have large implications for numerical modeling of the MJO and its propagation over the MC.Specifically, we expose the role of mountainous and diverse terrain over the MC as important to disrupting the formation of very large MCSs over land that could act to obstruct MJO propagation.In models run with low-resolution terrain (such as in climate simulations), MC mountains would appear smoother and flatter, and their effects on the DC would be smaller.Based on the results of this study, they would provide a lesser disruption to the formation of large land-locked MCSs, and, consequently, they would provide a greater barrier to MJO propagation over the MC.UWIN-CM model simulations and help with the LPT tracking algorithm used for identifying MJO events.The manuscript was also improved based on comments from anonymous reviewers, which we would like to thank.This study was supported by research grants from NOAA CVP (NA15OAR4320063 and NA21OAR4310263) and the first author was supported in part by the NASA Earth and Space Science Fellowship (NESSF 80NSSC17K0748).

Figure 1 .
Figure 1.Domain configuration and relief in model simulations.(a) CTRL topography (m) and initial time SST (°C); (b) FLAT topography and bathymetry (m); and (c) WATER topography and bathymetry (m).Black rectangles in (a) show the boundaries of nested domains in the atmospheric component of the model.CTRL simulation contains real topography, which is leveled between 90°E and 160°E in FLAT, and removed in WATER experiments, as shown in this figure.

Figure 2 .
Figure 2. Illustration of diurnal cycle (DC) analysis.(a) Distance from coastline over the Maritime Continent (MC) (km, negative distances are over land), with the outlined MC area where the DC is analyzed; (b) number of points in each 25-km distance bin within the MC; (c) distance from coastline Hovmöller composite of 2000-2020 IMERG rain rate DC (mm day −1 ), repeated twice; (d) quantitative composite of the IMERG rain rate DC (mm day −1 ), with color representing local solar time.

Figure 3 .
Figure 3. 20-year IMERG diurnal cycle (DC) composites in (a, d) Madden-Julian Oscillation (MJO) and (b, e) non-MJO environments and (c, f) MJO-non-MJO DC composite differences.The color bar in (a and b) is the same as in 2c up to 20 mm day −1 for easy comparison, and new colors have been added for rain rates above 20 mm day −1 .In (f), the solid lines show the amplitude of the DC, and the dashed lines show the mean diurnal precipitation, red for areas inside the MJO convective envelope, and black for areas outside the envelope and its 5° filtering area.The percentages in the top right corner of (a and b) denote the percentage of time that the Maritime Continent (MC) is within MJO and non-MJO environments, respectively.For the remaining 6% of the 20-year time period, the MC is outside the MJO convective envelope, but inside its 5° filtering area.

Figure 4 .
Figure 4. 5°S-5°N Hovmöller diagrams of rain rate (left, mm hr −1 ) and surface zonal wind (right, m s −1 ) in observations and model simulations.The products are ordered from top to bottom as follows: observations (IMERG precipitation and Cross-Calibrated Multi-Platform surface winds), CTRL, FLAT, and WATER simulations.The dashed vertical lines at 90°E and 120°E denote the longitudinal bounds of the Maritime Continent analysis region as shown in Figure 2. CTRL simulation contains real topography, which is leveled between 90°E and 160°E in FLAT, and removed in WATER experiments, as shown in Figure 1.

Figure 5 .
Figure 5.Time series of average rain rates over the MC.Thin lines show hourly precipitation while thick lines show its 24-hr running mean.The horizontal bars indicate the time during which the Madden-Julian Oscillation centroid is over the Maritime Continent.Observations are from IMERG; CTRL simulation (orange) contains real topography, which is leveled between 90°E and 160°E in FLAT (green), and removed in WATER experiments (blue), as shown in Figure 1.

Figure 6 .
Figure 6.Large-scale precipitation tracking tracking of the Madden-Julian Oscillation (MJO) convective envelope in (a) IMERG observations, (b) CTRL, (c) FLAT, and (d) WATER simulations at the 17 mm precipitation accumulation threshold.The colors represent the MJO convective area at a given time.CTRL simulation contains real topography, which is leveled between 90°E and 160°E in FLAT, and removed in WATER experiments, as shown in Figure 1.

Figure 7 .
Figure 7. Summary of Madden-Julian Oscillation (MJO) tracking with time.(a) Longitudinal location of the MJO centroid (solid lines) and its trailing and leading edges (dashed lines), (b) MJO area (x10 6 km 2 ), and (c) the average rain rate within the MJO convective envelope (mm hr −1).The Maritime Continent area is defined from 90 to 120°E, 10°S-10°N as in Figure2a, and it is bounded by vertical lines at90°E and 120°E in (a).Thin lines in (c) show hourly rain rates while thick lines show its 24-hr running mean.Observations are from IMERG; CTRL simulation (orange) contains real topography, which is leveled between 90°E and 160°E in FLAT (green), and removed in WATER experiments (blue), as shown in Figure1.

Figure 8 .
Figure 8.Diurnal cycle (DC) of precipitation over (a) land, and (b) ocean areas of the Maritime Continent (MC) as defined in Figure 2a.(c) percentage of total precipitation over the MC that falls over land (left axis), or ocean (right axis).The dashed black lines show the 20-year composite DC, while the solid black lines are only for the period of the model simulation.2. Observations are from IMERG; CTRL simulation (orange) contains real topography, which, between 90°E and 160°E, is flattened in FLAT (green), and removed in WATER experiments (blue), as shown in Figure 1.

Figure 9 .
Figure 9. Distance-from-coastline diurnal cycle (DC) composites over the Maritime Continent (MC) region outlined in Figure 2. Top: Hovmöller diagrams of rain rate (mm day −1 ) with local solar time for (a) IMERG observations, (b) CTRL, and (c) FLAT simulations.Bottom: 15-day composite DC of rain rate (mm day −1 ) for (d) IMERG observations, (e) CTRL, and (f) FLAT simulations.The vertical bars on the right edge of Hovmöller diagrams denote the times during which the Madden-Julian Oscillation centroid is located over the MC (between 90and 120°E).The stars on the left indicate the five most intense convective events that occurred more than 200 km inland (see Figure11).CTRL simulation contains real topography, which is leveled between 90°E and 160°E in FLAT, and removed in WATER experiments, as shown in Figure1.

Figure 10 .
Figure 10.Evolution of a large, long-lasting mesoscale convective system in the FLAT simulation (flattened topography over the Maritime Continent).Left: 3-hourly averaged precipitation (mm hr −1 ), 10 m winds over land (vectors), and 500 hPa vertical velocity over water (brown shading where vertical velocity is negative) centered on 26 November at (a) 14Z (21-22 local solar time (LST)), (b) 17Z (00-01 LST on November 27), and (c) 22Z (05-06 LST on November 27).Right: 1°S-0° averaged vertical cross-sections of zonal and vertical winds (arrows), potential temperature change from the previous hour (K hr −1 ; red-blue shading), and cloud area approximated by hydrometeor content (black hatching where total hydrometeor content ≥0.1 g kg −1 ) for the corresponding times.Vertical velocity is multiplied by 10 to emphasize the pattern, and red and blue contours outline a temperature change of 0.25 K hr −1 .

Figure 11 .
Figure 11.Comparison of inland convective systems over Borneo in CTRL and FLAT simulations.(a) Time series (in local solar time) of rain rate (mm hr −1 ) averaged over land areas more than 200 km inland, highlighting the five most intense convective events for each simulation.(b, c) Rain rate (mm hr −1 ) composites of the five most intense convective events in CTRL and FLAT simulations, respectively.(d) The difference in precipitation (mm hr −1 ) associated with intense convective events between FLAT and CTRL simulations.CTRL contains real topography, while in FLAT, topography over the Maritime Continent is leveled to 10 m above sea level, as shown in Figure 1.

Figure 13 .
Figure 13.Comparison of diurnal cycle (DC) composites for the CTRL (real topography, top) and FLAT (flattened topography over the Maritime Continent (MC), bottom) simulations under Madden-Julian Oscillation (MJO) (left) and non-MJO environments (right).The DC is composited over the MC region shown in Figure 2a, and the MJO and non-MJO environments for the modeled period are defined as in Section 3 and Figure 3. CTRL contains real topography, while in FLAT, topography over the MC is leveled to 10 m above sea level, as shown in Figure 1.