The 2019 Australian mega fires occurred mainly in the eastern and southeastern coastal areas of Australian continent (Yang et al., 2021). The southeastern parts, including the state of Victoria and the southeastern part of the state of New South Wales, belong to a marine climate, where obvious existence of SLB (OE-SLB) is not clearly verified because of the influence of strong westerlies and water vapour accompanied with westerlies from the ocean (Shen et al., 2021). Note that OE-SLB means that SLB is significant from a climatological perspective. In other words, the SLB can be found during most time of the year. Details of the definition of OE-SLB can be found in Shen et al. (2021) and are not repeated here. Meanwhile, the wild fire events there were the most severe with a great density according to numerous reports, which could possibly have caused fire-induced complex flows and circulation in the form of fire–atmosphere interactions in the vicinity of a fire (Sun et al., 2019). Based on previous observation during mega fire events, the concentrated fire spots changed the local air pressure field and added a regional temperature-pressure field, bringing uncertainties to local wind speed and wind direction (Jia et al., 1987; Li et al., 2016). On the one hand, this could further interrupt the formation of SLB since it might make the background wind field more complicated. On the other hand, the detected SLB might not be accurate since it is likely to contain other wind disturbances at a small regional scale.

As shown in Fig. 1, we selected an urban site in Brisbane along the eastern coast of Australia as the study site, which was due to several considerations. First, alongside the eastern coastal areas of Australia which belong to monsoon climate, including Brisbane and areas to its south but to the north of the fire centre, the Australian monsoon system is not strong, so the OE-SLB can be verified from a climatological perspective, which also means integrated SLB circulation can be found during all seasons. Second, compared to rural sites, there are longer periods of high-time-resolution observation data at urban sites, which is necessary for the extraction of SLB signals. Third, the urban area of Brisbane is relatively small and is not very far from vast areas of forests which provide stable combustion environment, ensuring the persistent effect of wild fires. Fourth, the UHI effect, which could possibly interrupt SLB and bring errors when calculating SLB magnitude, should be small for the study region considering the small scale of urban areas. Also, the wild fires near suburban areas could further eliminate the UHI effect, even if it could exist through their heating impact on these areas. In contrast, the forest site is surrounded by or within great amounts of flora where the majority of solar radiation is absorbed and scattered by leaves, prohibiting the surface heating by solar radiation and then the formation and detection of SLB. Actually, due to the existence of photosynthesis, the endothermic process of leaves from solar radiation and the temperature rise of the “leaf surface” are different from those of Earth's surface. As a result, the traditional mechanism of SLB formation is not necessarily applicable when the site is in the forest or quite close to clusters of flora. Coastal sites to the north of Brisbane are too far from the fire centre, and they are mostly rural sites covered with flora as well. Considering all of this, we chose the site of Brisbane Archerfield located on the eastern coast of the state of Queensland (Fig. 1) as the study site.

Several types of data have been used in this study, including land-cover-type data, Modern-Era Retrospective analysis for Research and Applications version 2 (MERRA-2) data, Moderate Resolution Imaging Spectroradiometer (MODIS) data, ground site observation data, the Fifth Version of European Centre for Medium-Range Weather Forecasts (ECMWF) ReAnalysis (ERA5) data, fire spot and FRP data and Global Data Assimilation System (GDAS) data. The detailed data information is described below one by one.

Land-cover-type data. The land-cover-type data of Australia are from the Dynamic Land Cover Dataset (DLCD) with Version 2.1 provided by Geoscience Australia. In this study, the DLCD land-cover-type data were used to reveal the surrounding landscape of Brisbane Archerfield. The spatial resolution of the data is 0.002∘ × 0.002∘, which is based on the annual mean of satellite observations from 2014 to 2015.

MERRA-2 data. MERRA-2 belongs to the global atmospheric reanalysis product managed by the National Aeronautics and Space Administration (NASA). It is produced by the Global Modeling and Assimilation Office (GMAO), and the assimilation system of Goddard Earth Observing System (GEOS-5) is used to ensure the quality of this dataset. At major ground sites over Australia, Yang et al. (2021) compared the monthly aerosol optical depth (AOD) product with Aerosol Robotic Network (AERONET) observations and found their root mean square errors (RMSEs) were all smaller than 0.05. Thus, MERRA-2 should be reliable to be used for the analysis of the large-scale spatial distribution of AOD in Australia. Yang et al. (2021) also denoted that the 2019 Australian mega fires were the strongest in January 2020. Correspondingly, we used the monthly AOD in January at 550 nm from 2002 to 2020 to check the AOD difference between the mega fire year and years with no mega fires. The spatial resolution of MERRA-2 AOD data is 0.625∘ × 1∘.

MODIS data. The MODIS instrument is performed on Aqua and Terra platforms. In this study, we used the MODIS cloud product which belongs to the dataset of MCD06COSP_M3_MODIS. The cloud information includes cloud optical depth (COD) and cloud fraction for all January months during the period from 2003 to 2020 with a monthly time resolution. The Brisbane Archerfield site is located at 153.008∘ E, 27.57∘ S. So we used COD and cloud fraction data whose space range and resolution are 152.5–153.5∘ E × 28.5–26.5∘ S and 1∘ × 1∘ respectively. This spatial range covers the whole Brisbane area and the normal encroaching distance of SLB, which is about tens of kilometres (Rajib and Heekwa, 2010; Shen et al., 2019). In this study, their spatial averages were calculated to represent the local COD and cloud fraction every January from 2003 to 2020. Also, we used the MODIS monthly AOD product for comparison with that of MERRA-2, which belongs to the dataset of MOD08_M3. The spatial resolution of MODIS AOD data is 1∘ × 1∘, and the time range is the same as that of MERRA-2.

Ground site observation data. The wind and air temperature observation data are from National Oceanic and Atmospheric Administration (NOAA) global observation network at the site of Brisbane Archerfield (153.008∘ E, 27.57∘ S). We used data in January from 2001 to 2020 in this study. The time resolution is every 3 h at 02:00, 05:00, 08:00, 11:00, 14:00, 17:00, 20:00 and 23:00 UTC on most days. The continuity of the observation data is ensured; there are observations on each day in January throughout the whole study period, with only one missing observation data on each day of a small fraction time (approximately 3.5 %). The wind information includes wind speed and wind direction. The air temperature is measured in Fahrenheit, and we have converted it into Celsius. The observation data were the main data used in this study to show the variations of both SLB and air temperature during the fire.

ERA5 data. The monthly mean U wind (zonal) speed and V wind (meridional) speed in January 2020 from the ERA5 were used in this study to reveal the background meteorological field so as to assess its effect on aerosol transport. The spatial resolution is 0.250∘ × 0.250∘ at pressure levels of 1000, 975, 950, 925, 900, 875, 850, 825, 800, 775, 750 and 700 hPa.

Fire spot and FRP data. Fire spot and FRP data are from the MODIS product (MCD14). This product can catch and locate the active fire hotspots based on thermal anomalies of 1 km pixel resolution (Giglio et al., 2016). The time resolution is daily, and we used the monthly averages for January from 2002 to 2020 to look into the fire situations over the years in detail.

GDAS data. The GDAS data were used to perform the back-trajectory analysis from the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT). The spatial resolution of GDAS data is 1∘ × 1∘ with daily time resolution. The levels of GDAS data chosen in this study to help to perform HYSPLIT analysis were 500 m and 3 km respectively. The time range set in this study was the whole of January 2020.

Figure 3 shows the SLB day number in January from 2001 to 2020. As shown, the SLB day number in January was normally larger than 10. Among these 20 years, there were 25 % of the years whose SLB days in January accounted for more than half of the month. Note that it does not necessarily mean that there is no SLB on days that are not SLB days. It is obvious that there was a slump in the number of SLB days in 2020. The total SLB day number dropped to only 4 during mega fires, accounting for only 33.33 % of the average SLB day number during the past 20 years. Also, the year 2012 witnessed a low SLB day number (6 d) in January. There are a lot of potential influencing factors for SLB frequency, such as the background wind field (Miller et al., 2013) and the interruption of other small-scale circulations (Kusaka et al., 2000). Among all the influencing factors, cloud is one of the most important because it has a significant effect on in situ solar radiation, which is the direct cause of TDLS. We will discuss this in the following sections.

The monthly mean SW and LW speeds in January from 2001 to 2020 are shown in Fig. 4a. As can be seen, there were fluctuations in the trends of both SW and LW speeds. The SW speed was higher than LW speed, which conformed to many previous findings (Miller et al., 2013; Zhu et al., 2017). The averages were calculated as 3.70 m/s for SW speed and 2.86 m/s for LW speed, respectively. Figure 4b and c show the anomalies of both SW and LW speeds. In general, LW speed fluctuated more significantly than SW speed did. This is due to its lower level of kinetic energy which can make it more sensitive to any potential interruptions from the background meteorological field (Shen and Zhao, 2020). The negative anomalies of LW speed happened in 2001, 2004, 2008, 2010, 2011, 2015, 2016, 2017, 2018 and 2020. Different from other years, it is obvious that the negative anomaly in 2020 was higher than 0.6 m/s, which was beyond the multi-years' oscillation range. The anomaly accounted for 22.3 % of the multi-years' average LW speed. The negative anomalies of SW speed happened in 2004, 2008, 2009, 2010, 2011, 2013, 2014, 2015, 2017 and 2020 (Fig. 4c). For SW speed, the negative anomaly in 2020 was also obvious, but its value was still within the multi-year oscillation range. It was higher than 0.5 m/s, accounting for 14.8 % of the multi-years' average. It is interesting to find that there were obvious positive anomalies of both SW and LW speeds in 2003, whereas their absolute values were not the highest. Also, the SLB day number in 2003 was near the average. We will discuss this further, along with the aerosol emissions during that year, in the following sections.

It can be seen in Fig. 4b that there were also significant fluctuations in night-time land temperature over the years. There was a soar in land temperature during night-time in 2020 which approached nearly 24 ∘C. It was nearly 3 ∘C higher than the multi-years' average, exceeding the range of multi-years' oscillation. The fluctuation in land temperature during daytime was less significant than that during night-time. There was an obvious positive anomaly in 2020, indicating that the daytime land temperature was higher than that in normal years. Meanwhile, it was still within the range of multi-years' oscillation, though the positive anomaly was obvious. Fire spots have a heating effect on the nearby environment through either shortwave radiation of light from fires or heat conduction caused by a temperature gradient. It can be inferred that the mega wild fires in January 2020 contributed to the positive temperature anomalies during PTS (PTL) through the heating effect of fires, though they might not be the only cause. The heating effect during mega fires was more significant during night-time than during daytime, which is probably due to a colder background temperature field during night-time.

Basically, the decreased SW (LW) speed revealed that the TDLS during PTS (PTL) decreased. To be more specific, the temperature difference between the small regions where the upward stream and downward stream of SLB circulation lie became smaller during January 2020. Based on Fig. 4b and c, temperature during PTL seems to be generally negatively related to LW speed anomaly, while it is obvious that temperature during PTS does not show any corresponding relationship with SW anomaly.

In order to be more accurate, we carried out linear regression between temperature during PTL and LW anomaly and found that they had a negative linear relationship (p<0.02) with each other (Fig. 5). As the temperature increased by 10 ∘C, the LW speed anomaly decreased by 1.52 m/s. The correlation coefficient R was 0.52, which was at the medium level. However, considering the significance level as well as low level of sample number, it can be concluded that the LW speed is generally negatively correlated with night-time land temperature. Moreover, their R and significance level could be 0.69 and 0.0012 respectively if we excluded the only one abnormal point in 2019, which might be caused by some potential disturbances on coastal SST where the vertical stream of SLB lies. Considering all these, it can be concluded that the LW speed anomaly is generally negatively correlated with night-time land temperature. During night-time, the land is colder than the sea. As the land temperature increases, the TDLS decreases if the SST of the area where the upward stream of SLB lies remains relatively stable; the LW speed does too. Briefly, the good linear relationship reveals that the variation of temperature during PTL (night-time land temperature) could generally represent the variation of TDLS during PTL, while the daytime land temperature variation could not represent the TDLS variation during PTS. In our previous study, we also found through observation that the daily lowest temperature (DLT) was clearly negatively related to LW speed, while the SW speed was more related to in situ downwelling solar radiation rather than merely land temperature (Shen et al., 2021), which was similar to the findings here. It could be inferred that although the land temperature during daytime increased during mega fire events, TDLS was still narrowed during fire events. If we only consider the land temperature, the SW speed should have increased during fire events because SW circulation is formed due to warmer land and colder sea. Consequently, there should be other factors which could cause decreased TDLS during PTS, which is the direct cause of decreased SW speed. We would investigate this in the following sections.

Since the heating effect depends largely on the distance between the area heated and the heat centre, it is necessary to examine the distribution of fire spots in January over the years, which is shown in Fig. 6. It can be seen that fire spots are scattered all over the eastern part of Australia in January over the years. January is the middle of Australian summer, which is the season when wild fires happen most frequently (Yang et al., 2021). Apart from 2020, other years also witnessed considerable scattered fire spots all over the coastal and inland regions. It is obvious that there was an extreme fire centre in the southeastern corner of Australia with a great density of fire spots in January 2020. This was exactly the region where the 2019 Australian mega fires mainly happened. To be specific, it was the eastern corner of the state of Victoria and the southeastern corner of the state of New South Wales, which is in agreement with many reports in the media. There was also a great fire centre in the southeastern corner in 2003, although the scale was smaller than that in 2020. Considering the distribution of fire spots near the site, the density of fire spots nearby was not higher than in other years. Instead, there seems to be more fire spots nearby the site in 2003, 2005, 2006, 2010 and 2013 in the figure. If we restrained the nearby region to areas of smaller scales, the year 2003 and 2013 rather than 2020 would have the most nearby fire spots.

There is another possibility that although the fire spots nearby the site were not more concentrated with great density in 2020 than in other years, the FRP of fire spots in 2020 was higher. This means that the fire was greater regardless of the ordinary density of spots, which could also result in more fire-induced aerosol emissions. So we further examined the FRP of fire spots in 2020 and those in other years. In order to make it comparable and verifiable, the time period of data chosen here was the same as that in Fig. 6. As shown in Fig. 7a, both the nearby and local fire spots in 2020 were mostly within the lowest FRP range, which was less than 235 MW. There were some sparse fire spots with greater FRP (235–863 MW) scattered all over the eastern part of Australia. The FRP of the fire centre was higher than the FRP of other fire spots; there were many fire spots with greater FRP which belonged to the range of 235–863 MW or 863–2194 MW. Figure 7b shows the FRP of all fire spots from 2002–2019. The FRP of nearby or local fire spots also had the lowest values. As the number of years increased, the density of fire spots with higher FRP (235–863 MW) increased significantly, most of which were located in inland areas of the Australian continent. This indicates that scattered wild fires with low or medium FRP are common in Australia, but concentrated mega fires are not so common. There were also some fire spots which belonged to the range of 235–863 MW or 863–2194 MW in 2003, yet the number was less, and the distribution areas were smaller. Based on Fig. 7, one important point we found is that there was no discrepancy between FRP of nearby or local fire spots in 2020 and that of nearby or local fire spots in other years. So the possibility mentioned above was discarded.

Based on the analysis above, the nearby fire spot density and FRP in 2020 were both at the same level as in other years for local regions near the site. This implies that the heating effect of nearby fire spots did exist in 2020, contributing to the increase of land temperature to some extent (especially night-time land temperature), but it was not likely the major cause of the land temperature anomaly. Fluctuation in land temperature might be caused by combined mechanisms, including some other potential factors. In other words, the heating effect of fire spots does not necessarily correspond to the observed air temperature increase. For example, Fig. 4b and c show that there were negative land temperature anomalies in 2003, but actually this year witnessed a greater density of nearby or local fire spots. In a real situation, the scale of SLB is quite small. The fire spots might be quite a long distance away from the area where the vertical stream of SLB lies, as a result of which the heating effect is weak.

Large fires have great aerosol emissions which affect the in situ solar radiation and then the radiation budget. Based on the basic physical mechanism of SLB formation, the observed decreased SW and LW speeds demonstrated the decreased TDLS. As mentioned above, the heating effect of nearby fire spots was weak and did not become more significant in 2020. So the more important factors bringing about the decrease of SW and LW speeds should be more closely related to TDLS rather than the land temperature only. The TDLS during SLB formation is highly related to the in situ downwelling solar radiation. As the shortwave radiation increases, the TDLS becomes larger due to the different heat capacities between land and sea. SW forms and prevails when TDLS is enough to drive this thermodynamic circulation. During night-time, the land–sea system is a “heater” for the upper atmosphere as the land and sea both give out heat and undergo energy loss in the form of longwave radiation. As the outgoing longwave radiation increases, the TDLS also becomes larger due to the different heat capacities between the land and sea. Then the LW forms in a similar way to SW.

Based on discussions above, in situ downwelling solar radiation is a crucial influencing factor of SW speed. Considering that aerosol is an important factor affecting in situ downwelling solar radiation, it is necessary for us to check the temporal and spatial variations of aerosols over the years. Figures 8 and 9 show the spatial distribution of AOD of total aerosols (TA-AOD) over the years using MERRA-2 and MODIS aerosol products, respectively. It shows that except for a little overestimation of AOD in the fire centre in 2020, the overall distribution and value of AOD revealed by MERRA-2 agreed well with those revealed by MODIS. Both MERRA-2 and MODIS show that there was a burst of aerosols in the fire centre during January 2003 and 2020, and the latter was much more severe. Especially for the site learned in this study, the difference of AODs between MERRA-2 (approximately 0.26) and MODIS (approximately 0.29) was very small. Thus, MERRA-2 agreed well with both MODIS and AERONET in terms of AOD during mega fires, and it has higher spatial resolution than MODIS. Considering all these aspects and the focus of the study, we used the MERRA-2 product in the analysis on local aerosol variations in the following sections. Figure 8 shows that the background level of TA-AOD was generally low in Australia over the years, implying that Australia was less polluted as a result of human activities. The TA-AOD in 2020 increased significantly compared with the average level. It can be seen that there was a maximum value centre in the southeast corner, which overlapped the region of the fire spots' centre (Fig. 6). The peripheral area of the maximum value centre was covered with isopleths, showing the characteristics of free diffusion of aerosols in the air. There was also a maximum value centre in 2003 whose scale was smaller, overlapping the smaller region of the fire centre in 2003. Based on findings from these three aspects, it can be concluded that the mega fire centre was the main source of the large amounts of aerosols around the site location. In general, the TA-AOD was about 240 % of the multi-years' average level at the site, while the TA-AOD in the fire centre was at a more astonishing level, accounting for more than 420 % of that at the local site of Brisbane. Aerosol could significantly affect the in situ downwelling solar radiation through direct radiative forcing. Turnock et al. (2015) calculated the relationship between AOD and surface solar radiation (SSR) and found that when the background value is low over the years, the SSR increases by 10 % as AOD varies from 0.32 to 0.16. In this study, the TA-AOD increased even more significantly (240 %), considering the low background value. Normally, when we talk about the radiative forcing of aerosols in the form of SSR difference, it means the instantaneous radiative forcing. However, the formation of SLB is the result of different levels of radiation accumulations between the land and sea. So the effect of aerosols on the total in situ downwelling solar radiation can further accumulate in the process of SLB formation and results in even more significant impacts on the change of surface temperature.

Apart from aerosols, clouds could play an even more important role in the radiation budget. The COD and cloud fraction anomaly at this site are shown in Fig. 10. The time range was from 2003 to 2020 due to data availability. It can be seen that both the cloud fraction and COD in 2003 were at an obvious low level, while both the cloud fraction and COD in 2020 showed a tiny negative anomaly. Based on the spatial distribution of TA-AOD, both 2003 and 2020 witnessed a soar in TA-AOD at the site, while TA-AOD increased more significantly in 2020. Figure 3 shows that there was a slump in SLB number in 2020 but not in 2003, while Fig. 4 shows that there were positive anomalies of both SW and LW speeds in 2003. Many previous studies on SLB have pointed out that a high level of in situ downwelling solar radiation is favourable for SLB formation and SLB speed increase (Shen and Zhao, 2020; Shen et al., 2021b; Miller et al., 2013). Our previous study in a monsoon climate region also showed that there was a positive linear relationship between in situ downwelling solar radiation and SW speed (Shen and Zhao, 2020). As we know, the in situ downwelling solar radiation is determined by both cloud and aerosols through their combined “umbrella effect”. The finding shown in Figs. 3 and 4 could be explained by the radiative cooling effects of aerosols and clouds. Although there was a positive anomaly of TA-AOD in 2003, the COD and cloud fraction were less than the average, offsetting the aerosols' negative radiative forcing effect. In situ downwelling solar radiation of the regional sea–land system was still ensured so that the SLB happened with a normal frequency (Fig. 3) and with an even larger speed (Fig. 4). The in situ downwelling solar radiation in January 2020 should be lower than the average, considering the tiny negative anomaly in both COD and cloud fraction and the significant increase in TA-AOD. The increased radiative forcing effect of TA-AOD was accumulated during the formation of SW. In conclusion, during daytime, the negative radiative forcing effect of total aerosols was the determinant factor to weaken the in situ downwelling solar radiation, resulting in lower level of TDLS and then decreased SW speed.

Mega fire events are significant in the way that they emit large amounts of carbonaceous aerosols, which include OC and BC. The OC is a very good scatter to solar radiation. Thus, among all the aerosols, OC could be an important contributor to the weakened TDLS during SW formation. Figure 11 shows the spatial distribution of OC over the years. The spatial distribution of OC was also similar to the fire spot distribution, which further confirmed that the source of great aerosol emissions was the mega fire centre. There were extreme value centres in the fire centre in both 2003 and 2020. The same as what we found earlier, it can be seen that the large values spread further in 2020 than 2003, indicating that the fire events were more severe in 2020 than in 2003. Similarly, the background value of OC at the site was low on average. The specific value of organic carbon AOD (OC-AOD) at the Brisbane site in 2020 was about 630 % of the multi-years' average, which was even higher than that of total aerosol. This is easy to understand because the fire centre is also covered with plants and trees, and their combustion can result in significant amounts of carbonaceous aerosols. Zhang et al. (2017) estimated the radiative forcing of OC globally using the BCC_AGCM2.0_CUACE/Aero model, which showed that Brisbane was within the large value area, with high levels of negative radiative forcing at the top of atmosphere. They also attributed this to biomass combustion. Thus, both total aerosol and OC made great contributions to the SW speed decrease by decreasing in situ downwelling solar radiation in January 2020.

The result above is analysed based on the impacts of aerosols on solar radiation. However there is almost no shortwave radiation during night-time. Then one question pops up: why was the slump of LW speed more significant? This indicated that the TDLS was significantly weakened at night in January 2020. While the heating effect of fire spots on night-time land temperature did exist, which was more significant than that during daytime, it was not likely the main cause of weakened TDLS based on FRP and fire spot distribution analysis. We next investigated the spatial distribution of BC over the years in Fig. 12. It shows that the black carbon AOD (BC-AOD) at the site was about 425 % of the multi-years' average level, with the extreme value centre overlapping the area of that of fire spots' density. Similar to the distribution of TA-AOD and OC-AOD, the peripheral areas of the maximum value centre are covered with isopleths, showing the characteristics of free diffusion. BC is well known as a kind of absorbing aerosol, which is reported to have a wider range of absorbing band than greenhouse gases, which can absorb broadband radiation from visible light to infrared wavelength (Zhang et al., 2017). During the daytime, it can absorb solar radiation, longwave radiation from the warmer land and shortwave radiation from local fires. During night-time, it has a warming effect on both the atmosphere and Earth's surface through longwave radiation. As a result, it has a warming effect on the Earth–atmosphere system, including the surface of the regional land–sea system, so there is a soaring of temperature, shown in Fig. 4b. The soaring BC during the mega fire heated the local atmosphere, which was like adding a heater in the air. The heater then gave out downward longwave radiation to the regional land–sea system. Just like the Sun during daytime, this could trigger a SW circulation anomaly, weakening LW circulation. Considering the BC burst during mega fires, there is nothing weird about its dominant role in the local land temperature increase during night-time. The mechanism proposed above can be summarized as follows. During night-time, the formation of LW originates from the process of heat release from both land and sea. As they both lose heat at different paces due to different heat capacities, the TDLS is enlarged. During the mega fires, the upper atmosphere of the regional land–sea system is heated, so that the vertical temperature gradient is weakened, which is unfavourable for heat release from both sea and land surfaces. As a result, the TDLS is significantly weakened.

Another potential contributing accelerator is CO2, which is also the product of fires due to the combustion of plants and trees. CO2 is a kind of greenhouse gas which is likely to be engaged in the same mechanism as BC to reduce TDLS during night-time, except that CO2 cannot affect the downwelling solar radiation. Details about this are not repeated again. However we should note that the effect of CO2 is based on theoretical analysis rather than observational verification due to the lack of accurate observation data. Both BC and CO2's warming effects increase TDLS during daytime, which partially offsets the strong negative radiative forcing effect of total aerosols, but their combined warming effect is more significant during night-time than during daytime. That is most likely the reason (at least partially) that SW speed had a negative anomaly but was less significant than LW speed.

What we discussed above are all factors whose influences were restrained to a small scale. Although SLB is a small-scale system, it can still be affected by the variations of signals on a large scale, since the local temperature is affected by both regional forcing and the variation of the large-scale background temperature field. In our previous study, we weighed their contributions qualitatively (Shen et al., 2019). In this study, we simply discuss the potential effect of the change in large-scale SST. Hirsch and Koren (2021) emphasized the effect of record-breaking aerosol emission from this mega fire on the cooling of the oceanic areas. On a large scale, its average radiative forcing on sea surface was -1.0 ± 0.6 W/m2. The temperature decrease of large-scale sea surface could have negative forcing on the SST at a regional scale, though the specific temperature variation of the sea surface where the SLB vertical stream lies might not be the same.

We summarized all the influencing factors of TDLS at both regional and large scales in Table 2. Among all these factors, aerosols, BC, OC and CO2 had direct forcing on TDLS by changing the solar radiation reaching the regional sea–land system. In contrast, the heating effect of fire spots and large-scale SST signal had forcing on land temperature and regional SST respectively, thus further having different forcing effects on TDLS during daytime and night-time. During the 2019 Australian mega fires, TDLS during daytime and night-time both decreased under their combined forcing effects, which could be inferred from the anomalies of SLB speed. Clearly, the directions of all forcing effects of different factors were the same during night-time. That was why LW speed decreased much more significantly than SW speed did. The negative radiative forcing effect of total aerosols was the determinant cause for TDLS decrease during daytime, which could only be partially offset by other factors.