Abstract

It is commonly understood that the observed decline in precipitation in southwestern Australia during the twentieth century is caused by anthropogenic factors. Candidates therefore are changes to large-scale atmospheric circulations due to global warming, extensive deforestation, and anthropogenic aerosol emissions—all of which are effective on different spatial and temporal scales. This contribution focuses on the role of rapidly rising aerosol emissions from anthropogenic sources in southwestern Australia around 1970. An analysis of historical long-term rainfall data of the Bureau of Meteorology shows that southwestern Australia as a whole experienced a gradual decline in precipitation over the twentieth century. However, on smaller scales and for the particular example of the Perth catchment area, a sudden drop in precipitation around 1970 is apparent. Modeling experiments at a convection-resolving resolution of 3.3 km using the Weather Research and Forecasting (WRF) Model version 3.6.1 with the aerosol-aware Thompson–Eidhammer microphysics scheme are conducted for the period 1970–74. A comparison of four runs with different prescribed aerosol emissions and without aerosol effects demonstrates that tripling the pre-1960s atmospheric CCN and IN concentrations can suppress precipitation by 2%–9%, depending on the area and the season. This suggests that a combination of all three processes is required to account for the gradual decline in rainfall seen for greater southwestern Australia and for the sudden drop observed in areas along the west coast in the 1970s: changing atmospheric circulations, deforestation, and anthropogenic aerosol emissions.

1. Introduction

Over the last century, southwestern Australia experienced a substantial decrease in precipitation. This poses a great challenge for the isolated region around Perth and its hinterland, which relies on reliable water resources for living, for industry, and for agriculture. Naturally, the question arises to what extent this decline in rainfall is human-induced and how much global and local environmental changes contribute to it. This topic motivated numerous studies and measurement campaigns already in the 1970s and has been debated widely since then (e.g., Bigg and Turvey 1978; Bates et al. 2008; Delworth et al. 2012; Delworth and Zeng 2014; Karoly 2014; Bigg et al. 2015).

Delworth et al. (2012) used the General Fluid Dynamics Laboratory (GFDL) global climate model GFDL CM2.5 to analyze the causes of the decline in precipitation. In Delworth and Zeng (2014), they concluded that many aspects of the observed reduction in rainfall can be attributed to anthropogenic changes in levels of greenhouse gases and ozone in the atmosphere, whereas anthropogenic aerosols do not contribute significantly. This stands in contrast to numerous studies of the impact of aerosols on the build-up of clouds and precipitation through the formation of cloud particles and by exerting persistent radiative forcing on the climate system that disturbs dynamics [see Tao et al. (2012) for a review]. Lee and Feingold (2013) investigated aerosol effects on cloud field properties of convective clouds and concluded that aerosols do have substantial influence on the spatiotemporal distribution of convection and precipitation. The coarse resolution (50 km) of the global model used by Delworth et al. (2012) and the simplified treatment of aerosols therein may explain this discrepancy.

Bates et al. (2008) reported from rainfall observations that the decrease in precipitation occurred in two distinctive steps around 1975 and 2000 rather than continuously. Likewise, they demonstrated a clear stepwise decrease in streamflow measurements at those times. Karoly (2014) also stressed that the simulations of Delworth et al. (2012) underestimate the decline in rainfall and that their identified drivers for this decline are usually associated with changes to the Southern Hemisphere climate in summer, whereas the bulk of the precipitation in this region occurs in austral winter.

Changes in atmospheric circulation due to a constant rise in greenhouse gases and a constant depletion of ozone are large-scale features and usually induce a continuous change in precipitation on longer time scales than observed in parts of southwestern Australia. Local changes to the environment, on the other hand, may have the potential to alter the local climate on very short time scales. With respect to the sudden drop in precipitation in the 1970s, several human-induced factors occurred just before or at this time:

  1. The conversion of natural forest to agricultural land after World War II led to an almost complete deforestation of a 130 000 km2 area by 1968, previously a biodiversity hotspot and now known as the “wheatbelt” (Saunders 1989; Bradshaw 2012). The deforestation had a strong impact on the aerosol concentration in this region through direct effects (surface deposition velocities; Gallagher et al. 2002) and indirect effects (rising groundwater levels in the absence of deep roots, and release of aerosols once the groundwater levels reach the surface; Ruprecht and Schofield 1991; Junkermann et al. 2009) with a time lag of about 15 years. Andrich and Imberger (2013) compared coastal and inland rainfall and showed empirically that land clearing alone can account for 55%–62% of the observed decline in precipitation for the wheatbelt area southeast of the vermin fence. This is also supported by modeling experiments by Kala et al. (2011), who showed that deforestation has been causing rainfall declines in this area, although their study was limited to two single events.

  2. In 1966, the Muja Power Station was commissioned 22 km east of Collie (see Fig. 1). The coal power plant had a total output of 974 MW and as such was the largest source of aerosols in this region. Further power stations burning coal from the Collie mine were added in 1973 (Kwinana, varying several times between coal, gas, and oil), in 1999 (Collie B), and in 2009 (Bluewaters). Additionally, the Kwinana refinery was continuously enlarged and eventually became the largest refinery in Australia. Airborne measurements taken during several flight campaigns in the 1970s led to an estimated total flux of 4 × 1019 particles per second from the Perth/Freemantle area, including the Collie region 150–200 km to the southeast, equivalent to a cloud condensation nuclei (CCN) production rate of 1 × 1019 particles per second (Andreae 2009). This value is close to the total natural CCN production of all of Australia at that time (Bigg and Turvey 1978).

Global circulation models suffer from a simplified treatment of aerosols and a relatively coarse resolution. Key components for the generation of rainfall such as convection and the interaction of CCN and ice nuclei (IN) cannot be resolved and therefore are parameterized. For instance, in GFDL CM2.5, only direct aerosol effects are included implicitly in the model (Delworth et al. 2012). Regional climate models, on the other hand, have been taken to higher and higher resolution over recent years. Newer, more sophisticated physics schemes have been added to explicitly treat the transport, growth, and interaction of CCN and IN.

The Weather Research and Forecasting (WRF) tool (Skamarock et al. 2008) is widely used in numerical weather prediction and regional climate simulations. Since version 3.6, released in April 2014, the ARW (Advanced Research WRF) core of WRF contains an aerosol-aware microphysics option, the Thompson and Eidhammer (2014) scheme. Its main features are a fundamental, first-order aerosol treatment and a direct coupling with radiation for aerosol indirect effects, which allows it to simulate the impact of aerosols on local weather and climate at a moderate increase in computational costs. In a first test of their new aerosol-aware scheme, Thompson and Eidhammer (2014) confirmed that increased aerosol number concentrations result in larger numbers of cloud droplets of overall smaller size, leading to an increase in cloud albedo (first indirect effect) and delays in the development of precipitation (second indirect effect). However, as pointed out by Thompson and Eidhammer (2014), recent large-scale, high-resolution studies have shown that aerosol impacts on cloud systems interplay with the dynamics in a “naturally buffered” system (Seifert et al. 2012; van den Heever et al. 2011; Grabowski and Morrison 2011), which means that even large changes in aerosols resulted in surface precipitation differences of only a few percent overall, but also that stronger effects may occur locally.

In this study, we investigate the effect of the addition of aerosol physics on the simulated weather and climate in southwestern Australia, and in particular address the question of whether the sudden increase in aerosols due to the emission from Muja Power Station can explain or at least contribute to the observed decrease in precipitation in the Perth/Freemantle area in the 1970s. We first revisit the historical rainfall observations of the Bureau of Meteorology to obtain a clear picture on the regional and temporal differences in rainfall decline over continental southwestern Australia. In section 2, we describe the observational datasets used and summarize the model configuration. In section 3, we present and discuss the results obtained from our modeling experiments, while section 4 is devoted to a summary and conclusions.

2. Methods

a. Observational rainfall, temperature, and pressure data

We refer to the Bureau of Meteorology (BOM) daily rainfall climate data (BOM 2015) for the reanalysis of the decrease in precipitation over an extended period from 1920 to 2015. The starting date is chosen to guarantee a sufficiently large number of recording stations with high availability of data (90% or more) for the entire period. Figure 1 displays all available stations in the area of study. A simple quality control is applied to the data to filter stations with zero or excessive annual precipitation.

Fig. 1.

(left) Model topography (terrain height in m) and (right) land-use classification for a subset of the 3.3-km domain, labeled as West Australia (WA). Indicated are the three regions West Coast (WC), Perth/Freemantle (PF), and Back Country (BC) used in the analysis, as well as the location of Perth (P) and Muja Power Station (M). The black dots represent meteorological stations of the BOM (BOM 2015) with a data availability of 90% or more between 1920 and 2015. The dominant land-use categories are evergreen broadleaf forest (red), woody savannas (ochre), and croplands (cyan), which clearly mark the northeastern border of the wheatbelt.

Fig. 1.

(left) Model topography (terrain height in m) and (right) land-use classification for a subset of the 3.3-km domain, labeled as West Australia (WA). Indicated are the three regions West Coast (WC), Perth/Freemantle (PF), and Back Country (BC) used in the analysis, as well as the location of Perth (P) and Muja Power Station (M). The black dots represent meteorological stations of the BOM (BOM 2015) with a data availability of 90% or more between 1920 and 2015. The dominant land-use categories are evergreen broadleaf forest (red), woody savannas (ochre), and croplands (cyan), which clearly mark the northeastern border of the wheatbelt.

In addition, we use gridded rainfall and near-surface air temperature data from the University of Delaware (UDEL) long-term monthly means v3.01 (Willmott and Matsuura 2014) and from the Climate Research Unit (CRU) high-resolution time series dataset v3.23 (Harris et al. 2014) at 0.5° × 0.5° spatial resolution. Last, we include the HadSLP2 gridded global sea level pressure anomalies (Allan and Ansell 2006) at 5° × 5° spatial resolution in our analysis.

b. Aerosol-aware regional climate model

We employ version 3.6.1 of the regional climate model ARW, released August 2014, to study the effect of aerosols and changes in their concentration on local weather and climate at very high resolution for a 4-yr period from 1970 to 1974.1

Lateral boundary conditions are supplied by ERA-40 reanalysis data (Uppala et al. 2005) at 1.0° × 1.0° spatial resolution (110 km). For reasons explained below, a convection-resolving resolution of 3.3 km in a triple-nested approach is adopted (see Fig. 2). Because of computational constraints and the requirement to conduct multiple experiments with the innermost domain, we choose a one-way nesting approach (i.e., we switch off the feedback from the inner domains to the outer domains). Lateral boundary conditions from ERA-40 to the 30-km domain are supplied every 6 h, and every 3 h for the nested domains using the “ndown” utility of WRF. Spectral nudging of temperature, wind, and geopotential is applied above the planetary boundary layer to the outermost domain only to avoid a divergence of the regional model state from the forcing dataset (von Storch et al. 2000; Miguez-Macho et al. 2004). We use the MODIS 21-class land-use table to describe the land surface properties, which dates back to 2001 and thus matches the conditions after the creation of the wheatbelt in southwestern Australia in the 1960s. Several areas are selected for the analysis of the data, which are overlaid on the model topography and land-use map in Fig. 1: West (i.e., southwest) Australia (WA), the West Coast (WC), Perth/Freemantle (PF), and the Back Country (BC).

Fig. 2.

Triple-nested domain configuration with 30-, 10-, and 3.3-km resolution. Lateral boundary conditions are taken from ERA-40 reanalysis data at 1.0° × 1.0° spatial resolution (110 km).

Fig. 2.

Triple-nested domain configuration with 30-, 10-, and 3.3-km resolution. Lateral boundary conditions are taken from ERA-40 reanalysis data at 1.0° × 1.0° spatial resolution (110 km).

c. Aerosol-aware microphysics

The Thompson–Eidhammer aerosol-aware microphysics scheme was added in version 3.6 of WRF and allows the simulation of the effect of aerosols on local weather and precipitation for a moderate increase in computational costs by 16%, compared to the standard Thompson microphysics scheme. In this implementation, CCN are referred to as water-friendly aerosol particles and are created by summing sulfates, sea salts, and organic carbon, while IN are referred to as ice-friendly aerosol particles and are created by summing five size bins of dust. Aerosols are treated in a fundamental, first-order approach through activation of CCN and IN, depletion of aerosols (precipitation scavenging), and simplistic aerosol replenishment (surface emissions). The microphysics scheme is coupled directly to the RRTMG longwave (LW) and shortwave (SW) radiation schemes to account in principle for aerosol direct and indirect effects. It is important to note that in WRF version 3.6.1 this coupling is not complete: the calculation of the aerosol optical depth (AOD) is not informed by the new Thompson–Eidhammer scheme and thus assumes climatological aerosol concentrations (aerosol direct effect). However, the size of the aerosol particles, emitted by anthropogenic sources such as power plants and smelters in Australia and representing the bulk of the increase in aerosol concentration in the 1970s, has been measured for sizes between 5 and 100 nm (Junkermann and Hacker 2015). While their exact size depends on the distance from the source and the available time for coagulation and growth (Bigg 1973; Junkermann et al. 2009; Junkermann and Hacker 2015), they are well below the range in which direct effects through scattering and absorption are important (). Hence, these particles do not have a large interaction with solar radiation. Nonetheless, work is underway to implement a consistent calculation of the AOD using the microphysics number concentrations of CCN and IN in the radiation schemes (G. Thompson 2014, personal communication). Apart from the additional treatment of aerosols, physical consistency between the Thompson scheme and this new microphysics scheme is ensured. This allows us to assess the effect of the aerosol treatment—or, more precisely, the aerosol indirect effects of small, anthropogenic aerosol particles—on the model results.

The aerosol scheme is not coupled to any cumulus scheme, which means that there is no depletion of aerosols by convective precipitation and no subgrid-scale aerosol activation. It is thus required to use a very high, convection-resolving horizontal resolution. Further, a relatively high vertical resolution and the activation of the new namelist variable scalar_pblmix are required to ensure that aerosols get mixed in the vertical by subgrid turbulence. The specification of aerosols can be handled in two primary ways: 1) external datasets from climatology or other (chemistry) models, or 2) simplified vertical profiles, which are prescribed in the model.

To fulfil the above requirements, we use a horizontal resolution of 3.3 km for the innermost domain. At this resolution, it can be assumed that convection is resolved at grid scale (Weisman et al. 1997; Prein et al. 2013). To achieve a sufficiently high vertical resolution, in particular in the planetary boundary layer, we use 75 vertical levels with a lowermost level height of 25 m and 20 levels in the first 1000 m above surface. Such a small vertical grid spacing implies a reduction of the typical time step of 18 s (6 s km−1 horizontal resolution) to 4 s for model stability.

The initial aerosol concentrations are specified as simplified vertical profiles. The default vertical profile in WRFV3.6.1 depends on the terrain height and was designed to fit the continental United States, for which the near-surface value is found to exist within an idealized boundary layer of approximately 200 to 1000 m, depending on starting elevation. An exponential decay of aerosol number from the higher numerical value in the boundary layer to the lower free tropospheric number is used to complete the vertical profile (G. Thompson 2014, personal communication). This profile is adapted to describe different aerosol concentrations for southwestern Australia. First, a standard vertical profile is created based on the airborne measurements and analysis of Bigg (1973), Bigg and Turvey (1978), Junkermann et al. (2009), and Bigg et al. (2015). Bigg (1973) and Bigg and Turvey (1978) reported aerosol concentrations of up to 104 cm−3 (8 × 109 kg−1) up to 1.3 km above ground, measured in a “plume” stretching along the West Coast from 200 km south of Perth (i.e., the Collie area) to as far as 150 km north of the city. These numbers were updated in recent studies by Andreae (2009) and Bigg et al. (2015) to 30% and 10% of the original estimate, respectively, illustrating the uncertainty in the measurements. Outside of the plume, a typical CCN concentration of 700 cm−3 (5.7 × 108 kg−1) was reported. Since the actual CCN and IN concentrations in our models are a result of several processes adding and removing particles, we calibrate the standard CCN and IN profiles such that equilibrium values close to the observations are obtained (see also section 3). At the surface, a CCN number concentration of 2.1 × 108 m−3 is adopted at every grid point, which decreases only slightly to 2 × 108 m−3 within the boundary layer (i.e., approximately 1.5 km above ground). Between 1.5 and 3.5 km, an exponential decay to a CCN concentration of 9 × 107 m−3 is assumed, which decays further to 3 × 107 m−3 at 20 km above ground (i.e., at approximately the top of the model at 50 hPa). To approximate the measured IN number concentrations (Bigg 1973; Bigg and Turvey 1978), the CCN profile is scaled by 10−4.

The resulting profiles reflect the clean environmental conditions prior to the commissioning of the Muja, Kwinana, and Collie coal power plants. Figure 3 compares the so-obtained aerosol profile to the default profile in WRFV3.6.1 and to a climatological average, derived from multiyear (2001–07) global model simulations (Colarco et al. 2010), in which particles and their precursors are emitted by natural and anthropogenic sources and are explicitly modeled with multiple size bins for multiple species of aerosols by the Goddard Chemistry Aerosol Radiation and Transport (GOCART; Ginoux et al. 2001) model.

Fig. 3.

Initial vertical aerosol profiles for (left) southwestern Australia and (middle) the continental United States. (right) The average profile for southwestern Australia, derived from multiyear (2001–07) global model simulations (Colarco et al. 2010).

Fig. 3.

Initial vertical aerosol profiles for (left) southwestern Australia and (middle) the continental United States. (right) The average profile for southwestern Australia, derived from multiyear (2001–07) global model simulations (Colarco et al. 2010).

The initial profile is applied once at the starting time of the model integration and for every grid point, both over land and sea. During the model integration, the CCN and IN variables are advected and diffused exactly as other scalars (e.g., cloud ice number concentration). A simplified surface aerosol emission tendency is computed as a 2D field based on the horizontal grid spacing and starting aerosol number concentration for the CCN variable [for details, see Thompson and Eidhammer (2014)]. No surface emission tendency is applied for IN in this version of the code. The 2D tendency field is added each time step to the first model vertical level CCN value. The initial aerosol profile for southwestern Australia, displayed in Fig. 3, corresponds to a surface emission rate of 7031 particles per kilogram second per column with area . For comparison, the standard aerosol profile for the continental United States corresponds to 11 180 particles per kilogram second.

In this study, we address two questions. First, we investigate the changes in the simulated weather and climate when aerosols are considered in the microphysics, using the initial aerosol profile presented above. Second, we study the impact of changes in the aerosol concentration from the clean environmental conditions to a polluted environment through modifications of the initial aerosol profile and the surface emission rates. In total, we conduct four model runs on the innermost domain for the period 1970–74:

  1. Standard run (wrf-std): In this configuration, we use the default Thompson microphysics scheme, which is coupled to the RRTMG LW/SW schemes, but does not treat aerosols explicitly. Because of the physical consistency between the Thompson and the Thompson–Eidhammer schemes, this run can be compared directly to the following runs to assess the impact of adding aerosol physics to the WRF Model.

  2. Aerosol run (wrf-aero): Here, we use the aerosol-aware Thompson–Eidhammer microphysics scheme with the initial aerosol profile (Fig. 3) and surface emission rates for southwestern Australia. This run allows us to investigate the effect of aerosols (natural and anthropogenic) on the model without the contribution of the Muja Power Station or other large pollutants.

  3. Aerosol boost run (wrf-aerox3): This configuration is identical to the aerosol run, but uses an initial aerosol profile 3 times as large for both CCN and IN. Accordingly, the CCN surface emission rate is also tripled to 21 093 particles per kilogram second. This run describes in a simple way the increase in aerosol concentrations due to the commissioning of the Muja Power Station and other sources of anthropogenic aerosols. The increase by a factor of 3 is motivated by differences in measured aerosol concentrations in the vicinity and far distance of the larger pollutants in the Muja/Collie area (Bigg and Turvey 1978; Junkermann et al. 2009) and leads to average CCN and IN number concentrations of and (see section 3).

  4. Muja Power Station run (wrf-muja): This configuration is identical to the aerosol run, but contains an additional source of anthropogenic aerosols injected into the model at the location of the power plant and with an emission rate as derived from observations of similar-sized power plants in Australia (Junkermann and Hacker 2015). A total emission rate of 4.6 × 108 particles per kilogram second is added to the surface emissions in a circle sector with 20-km radius and 35° opening angle in direction northeast at the location of the Muja Power Station (33.34°S, 116.26°W). To account for the elevated emission from power plants at 250–400 m above ground, this additional source term is distributed evenly across the first 1500 m in height at every grid point in this sector.

We refer to previous work (Fersch and Kunstmann 2014; Thompson and Eidhammer 2014) for a recommended WRF Model configuration for this specific region and research question, which is summarized in Table 1. It is important to remember that the differences between the standard Thompson scheme and the aerosol-aware Thompson–Eidhammer scheme are in the consideration of the aerosol indirect effects only: While the aerosol-aware runs compute size and number concentration of aerosols and thus cloud droplet numbers consistently, the standard run uses prescribed values for the cloud number droplets in the microphysics scheme. For both schemes, the aerosol direct effect is included through the calculation of the aerosol optical depth using climatological aerosol concentrations, which are independent of the initial aerosol profiles used for the different high-resolution model runs.

Table 1.

WRF Model configuration for the different domains at 30-, 10-, and 3.33-km resolution and for the different types of high-resolution runs (LBC = lateral boundary conditions, RRTMG = rapid radiative transfer model for GCMs, BMJ = Betts–Miller–Janjić, MYJ = Mellor–Yamada–Janjić).

WRF Model configuration for the different domains at 30-, 10-, and 3.33-km resolution and for the different types of high-resolution runs (LBC = lateral boundary conditions, RRTMG = rapid radiative transfer model for GCMs, BMJ = Betts–Miller–Janjić, MYJ = Mellor–Yamada–Janjić).
WRF Model configuration for the different domains at 30-, 10-, and 3.33-km resolution and for the different types of high-resolution runs (LBC = lateral boundary conditions, RRTMG = rapid radiative transfer model for GCMs, BMJ = Betts–Miller–Janjić, MYJ = Mellor–Yamada–Janjić).

3. Results and discussion

a. Long-term rainfall trends in southwestern Australia

Figure 4 displays the annual precipitation compiled from the BOM (2015) daily rainfall climate data for the entire year, the wet season April to September, and the dry season October to March for the areas described in Fig. 1. Displayed are the mean of all stations with a data availability of 90% or more and the stations with maximum and minimum rainfall over the entire period 1920–2015. Stations with minimum rainfall show no decrease over the entire period, while stations with maximum rainfall exhibit a dramatic decline in rainfall predominantly in the wet season. Independent of the area, the decrease in mean annual precipitation is observed entirely in the rainy season, while there is no change in the low amount of rainfall during the dry season. The selection of stations used in the analysis matches that of Andrich and Imberger (2013) with the exception that the quality control flags of the station data and possible relocations of stations are not taken into account in our study. Instead, we apply a simplistic quality control filter. Stations are excluded if one or more of the following criteria apply: 1) commissioned after 1920, 2) shut down before 2015, 3) less than 90% data availability for the entire period, 4) 0 mm recorded precipitation for an entire year, and 5) more than 2000 mm recorded precipitation for a single month. Further, our classification into dry and wet season differs from that of Delworth and Zeng (2014), who defined the wet season from March through August. However, we find that on average the September precipitation is larger than the March precipitation for the daily rainfall data provided by the BOM and that the separation between the dry and the wet season is clearer using our definition.

Fig. 4.

Annual precipitation compiled from the BOM (2015) daily rainfall climate data for the areas described in Fig. 1 (top to bottom: WA, WC, PF, BC) and for (left) the entire year, (middle) the wet season April–September, and (right) the dry season October–March. Displayed are the mean of all stations with a data availability of 90% or more and the data of the station with the maximum/minimum rainfall over the entire period. Stepwise and linear fits to the mean annual rainfall are displayed with their coefficients.

Fig. 4.

Annual precipitation compiled from the BOM (2015) daily rainfall climate data for the areas described in Fig. 1 (top to bottom: WA, WC, PF, BC) and for (left) the entire year, (middle) the wet season April–September, and (right) the dry season October–March. Displayed are the mean of all stations with a data availability of 90% or more and the data of the station with the maximum/minimum rainfall over the entire period. Stepwise and linear fits to the mean annual rainfall are displayed with their coefficients.

Stepwise and linear fits to the mean annual rainfall are displayed with their corresponding coefficients of determination, defined as

 
formula

The data presented here and the fits to it at first glance contradict the current perception of a sudden, stepwise decrease in precipitation in the 1970s and at the beginning of the twenty-first century in southwestern Australia (Bates et al. 2008). A single stepwise fit to the data results in a step of −246 (PF) to −42 mm yr−1 (WA) around 1970 with a similar correlation coefficient as a continuous, linear decline by −430 (PF) to −71 mm yr−1 (WA) from 1920 to 2015. For the three regions WA, WC, and BC, the observations and our fits to them imply that a continuous decline in annual precipitation by 20% between 1920 and 2015 matches the observations as well as a sudden drop by 10% around 1970. (Table A1 in the  appendix summarizes these results and the list of BOM stations used for the analysis.)

The extreme numbers for PF are a result of a very small number of three stations only that meet our criteria on data availability and for which two of the stations, 9010 Churchman Brook and 9031 Mundaring Weir, show a significant decrease in precipitation from 1500 to about 1000 mm yr−1 for Churchman Brook, and from 1100 to about 800 mm yr−1 for Mundaring Weir, respectively. For these particular stations, the long-term records indeed resemble a sudden decrease in annual rainfall precipitation around 1970, during which time the station Churchman Brook also has a few gaps in the recorded data. We consulted the station metadata provided by the BOM2 to verify that no change in equipment or relocation of the station took place during this time or within the entire period 1920–2015. It is interesting to note that both Churchman Brook and Mundaring Weir are located in the Perth catchment area at the Canning/Mundaring surface water storages3 and that the sudden drop in precipitation around 1970 for these stations is qualitatively consistent with the observed dam levels displayed in Karoly (2014, Fig. 1a therein): The measured streamflow into the Perth dams dropped from about 340 GL (average 1911–74) by 48% to 180 GL (average 1975–2000), while Churchman Brook (Mundaring Weir) recorded 33% (27%) less precipitation on average since 1970 compared to 1920–70. To illustrate the importance of these stations on the observed decline in precipitation, Fig. 5 displays the long-term annual precipitation trends for the regions WA and WC as in Fig. 4, but without the two stations 9010 and 9031. Despite the large number of stations included in the trend analysis for WA, the effect on the mean values is noticeable and results in a 13%–17% smaller decrease in precipitation of that of the entire set of stations (see also Table A1) and comparable coefficients. For the West Coast WC with only a quarter as many stations, the effect is more pronounced with a 20%–22% smaller decrease in precipitation of that of the entire set of stations and larger values.

Fig. 5.

As in Fig. 4, for regions WA and WC, but without the “outlier stations” Churchman Brook (9010) and Mundaring Weir (9031) located in the Perth/Freemantle area PF.

Fig. 5.

As in Fig. 4, for regions WA and WC, but without the “outlier stations” Churchman Brook (9010) and Mundaring Weir (9031) located in the Perth/Freemantle area PF.

Hence, our general findings of a continuous decline in precipitation for the WA, WC, and BC areas do not contradict the sudden drop in observed river discharges reported by Bates et al. (2008) and Karoly (2014) for the Perth catchment area, for which small changes in circulation may have led to shifts in precipitation bands on a regional scale. In addition, anthropogenic factors such as irrigation and deforestation (Andrich and Imberger 2013) may have influenced the dam water levels at that time.

b. Aerosol effects in high-resolution regional climate modeling

Our findings in the previous section suggest that observational rainfall data can be interpreted as a continuous decline in precipitation or as a sudden decrease around 1970, depending on the area. In the following, we address the question of whether a sudden increase in small anthropogenic particles can in principle cause such a drop in precipitation through first and second aerosol indirect effects. To derive a consistent picture of the impact of aerosols through indirect effects, a number of meteorological parameters are investigated.

1) CCN and IN number concentrations

Figures 6 and 7 display the number concentration of CCN and IN for the three different high-resolution model runs using aerosol-aware microphysics (wrf-aero, wrf-aerox3, wrf-muja) both as time series of the sum over the entire column, averaged over the area WA, and as contour-plot average for the wet season (April–September) and the dry season (October–March) 1970–74. Near-surface wind vectors are overlaid on the CCN contour plots, and wind vectors at PBL height on the IN contour plots. The CCN number concentration of the wrf-muja run, averaged over the area WA, exceeds that of the wrf-aerox3 run (i.e., 3 times the conditions prior to the commissioning of Muja Power Station and other large sources of anthropogenic aerosols). Despite being emitted by a tiny area around the location of Muja Power Station, the ultrafine aerosol particles are distributed widely and result in a higher CCN concentration along the West Coast and over the wheatbelt. While the annual average shows a symmetrical distribution around the emitting source (not displayed), the direction in which the ultrafine aerosol particles travel depends on the seasonality of the near-surface winds. In austral summer, the dominant near-surface wind direction is toward the northwest over land, whereas in austral winter the bulk of the CCN are pushed to the southeast of the area WA (Fig. 7). The emissions from Muja Power Station are not affected by the wind field at PBL height, which suggests that the bulk of the CCN particles are advected horizontally rather than mixed up into higher layers. The CCN number concentration varies with time around its initial values as a result of the continuous removal and replenishment through CCN activation, rain/snow/graupel collecting aerosols, cloud/rain evaporation, and surface emissions [Thompson and Eidhammer 2014, their Eq. (3)].

Fig. 6.

Mean value of CCN and IN number concentrations (kg−1) for region WA as function of time, summed up over the entire column at each grid point.

Fig. 6.

Mean value of CCN and IN number concentrations (kg−1) for region WA as function of time, summed up over the entire column at each grid point.

Fig. 7.

Contour plots of (top) CCN and (bottom) IN number concentrations (kg−1) for the wet season and the dry season averages 1970–74, summed up over the entire column at each grid point. Overlaid are 10-m surface winds at top, and winds at PBL height at the same scale at bottom. The black dots represent the positions of the BOM stations Churchman Brook (9010) and Mundaring Weir (9031); the white dot is the location of the Muja Power Station.

Fig. 7.

Contour plots of (top) CCN and (bottom) IN number concentrations (kg−1) for the wet season and the dry season averages 1970–74, summed up over the entire column at each grid point. Overlaid are 10-m surface winds at top, and winds at PBL height at the same scale at bottom. The black dots represent the positions of the BOM stations Churchman Brook (9010) and Mundaring Weir (9031); the white dot is the location of the Muja Power Station.

The initial IN number concentration drops rapidly for all three runs with a constant factor of 3 between wrf-aero/wrf-muja and wrf-aerox3. Without additional surface emissions, the equilibrium level of IN is maintained by a minimum value of 1% of the initial IN concentration (5 × 103 kg−1 for wrf-aero/wrf-muja and 1.5 × 104 kg−1 for wrf-aerox3), hardcoded in the aerosol-aware microphysics scheme, and additional terms stemming from IN activation, rain/snow/graupel collecting aerosols, and cloud ice sublimation [Thompson and Eidhammer 2014, their Eq. (4)]. These additional source and sink terms lead to an equilibrium value of 6 × 103 kg−1 for wrf-aero/wrf-muja and 1.8 × 104 kg−1 for wrf-aerox3. The spatial distribution of IN is positively (negatively) correlated with the wind field at PBL height in austral winter (summer), which suggests that IN are located at higher levels than CCN and that cloud ice sublimation (melting) occurs during austral winter (summer).

2) Precipitation

The main focus of our investigation is the change in model precipitation when including aerosol physics in the microphysics schemes (i.e., the difference between wrf-std and the aerosol-aware runs) and when changing the aerosol concentration (i.e., the difference between wrf-aero and wrf-aerox3/wrf-muja). Figures 8 and 9 display mean monthly rainfall and mean monthly rainfall anomaly of the different model runs and of observational data from UDEL and BOM for the four regions of interest over land only, averaged over the entire period 1970–74. The monthly precipitation is derived as the average of all stations in the region with a data availability of 90% or more between 1970 and 1974 (BOM) or all grid points in the corresponding region (all others). The anomaly is calculated as the root-mean-square error (RMSE) of each month’s precipitation with respect to the mean monthly precipitation. Averaged over the entire region WA and the Back Country region BC, the gridded observations from UDEL and the station data from BOM agree well. For WC and even more so for PF, the UDEL observations show consistently smaller values than the BOM station data, which is a result of the small number of stations contributing to the BOM average and the “outlier” stations Churchman Brook and Mundaring Weir. The  appendix (see Table A2) provides additional statistics on biases, coefficients of variation, and Pearson correlation coefficients for the seasonal cycle (mean monthly rainfall) and interannual variability (mean monthly rainfall anomaly) of the gridded datasets displayed in Figs. 8 and 9.

Fig. 8.

Mean monthly precipitation 1970–74 for the different model runs and observational datasets. For each region, the mean value is taken over all stations (BOM) or over all grid points (all others), respectively.

Fig. 8.

Mean monthly precipitation 1970–74 for the different model runs and observational datasets. For each region, the mean value is taken over all stations (BOM) or over all grid points (all others), respectively.

Fig. 9.

Mean monthly precipitation anomaly 1970–74 for the different model runs and observational datasets. For each region, the mean value is taken over all stations (BOM) or over all grid points (all others), respectively. The anomaly is calculated as the RMSE of each month’s precipitation with respect to the mean monthly precipitation 1970–74.

Fig. 9.

Mean monthly precipitation anomaly 1970–74 for the different model runs and observational datasets. For each region, the mean value is taken over all stations (BOM) or over all grid points (all others), respectively. The anomaly is calculated as the RMSE of each month’s precipitation with respect to the mean monthly precipitation 1970–74.

The monthly precipitation of the models is given by nonconvective precipitation for the 3.3-km runs, where no cumulus scheme is employed, and by the sum of convective and nonconvective precipitation for the 10- and 30-km runs. The amount of rainfall is clearly related to the horizontal resolution, with lower values for larger grid spacing. All 3.3-km models overpredict rainfall for WA/BC in both seasons and for WC/PF in the dry season, but match the WC/PF winter precipitation closely with mean biases of less than 10 mm month−1. The coarser-resolution models provide better estimates of WA/BC precipitation for both seasons and for WC/PF in the dry season, but largely underestimate the WC/PF winter precipitation. Among the high-resolution runs, the non-aerosol-aware run wrf-std generates least precipitation, followed by wrf-aerox3, wrf-muja, and wrf-aero with largest precipitation amounts, independent of the region and the season. This will be discussed in detail below. The general overprediction of summer precipitation by the high-resolution models is reflected in the coefficients of variation of RMSE, with large errors for region BC due to the small amount of rainfall of less than 20 mm month−1.

The ability of the models to capture the interannual variability is demonstrated by the Pearson correlation coefficient of the time series data (see Table A2). In general, the high-resolution runs show large improvements in the correlation for all regions and both seasons. While a detailed analysis of the underlying reasons is beyond the scope of this study, we speculate that this is due to the improved representation of the topography, in particular the mountainous region in the northeast, in the 3.3-km models. For the coastal regions WC and PF, the correlation is higher during the dry season than during the wet season for all models. We attribute this to a smaller interannual variability for WC and PF during the dry season due to the dominant wind direction from the southeast (i.e., over dry continental planes). The lowest correlations overall are found for the BC region, where absolute precipitation amounts are smallest.

An appendix table (Table A3) also compares the total amount of modeled precipitation, accumulated over the entire 4-yr period 1970–74, to the UDEL observations for each of the four regions and split into dry season and wet season. Contrary to the statistics shown in the earlier table (Table A2), which address the temporal correlations of area-averaged rainfall, the statistics in this table represent spatial correlations of accumulated rainfall amounts after 4 years of simulation. Averaged over area WA, the coarse-resolution runs wrf-30km and wrf-10km have smaller RMSEs than the high-resolution runs. This is also true for area BC, where the RMSEs of wrf-30km and wrf-10km are 11%–18% for the wet season (24%–45% for the dry season), while they are considerable larger (worse) for the 3.3-km runs (38%–55% for the wet season, 62%–88% for the dry season). The Pearson correlation coefficients for WA and BC of wrf-30km and wrf-10km are similar to that of the high-resolution runs for the wet season, but smaller (worse) for the dry season. The coastal areas WC and PF are fit significantly better by the high-resolution runs, in particular for the wet season in austral winter, with relative RMSE values of 13%–16% and spatial correlations above 0.92. The coarser runs wrf-30km and wrf-10km show RMSE values of 37%–59% and spatial correlations between 0.78 and 0.90. Among the four high-resolution runs, the standard run wrf-std with least precipitation performs best and in particular matches the observed WC and PF winter precipitation closely.

The separation of the 30- and 10-km runs from the 3.3-km runs can be explained by the differences in model physics: The former require a cumulus scheme, the Betts–Miller–Janjić (BMJ) scheme in this case, to generate convective precipitation in the model, while the latter are able to resolve these processes at grid scale. Previous parameter studies of different WRF physics schemes over Australia and other regions of the world (e.g., West Africa) have shown that the BMJ cumulus scheme tends to produce drier conditions than other cumulus schemes, while the default Thompson microphysics scheme (without aerosol physics) tends to produce wetter conditions than other microphysics schemes (Fersch and Kunstmann 2014; Noble et al. 2014; Klein et al. 2015). Since the aerosol-aware Thompson–Eidhammer scheme is identical to the standard scheme except for the treatment of aerosols, this argument applies to all four high-resolution runs.

By comparing the accumulated precipitation values of the wrf-std and the wrf-aero runs, we find that the addition of aerosol physics with pre-1960s aerosol concentrations to the microphysics scheme leads to an increase in rainfall by 8.1% (WC), 9.5% (PF), 10.6% (WA), and 12.1% (BC). This effect is more pronounced for the dry season (between 11.2% and 14.6%) than for the wet season (between 7.2% and 10.6%). Increasing the aerosol concentration, however, reduces the amount of precipitation in all areas: Of the three aerosol runs, wrf-aero shows the largest values of accumulated precipitation and wrf-aerox3 (with 3 times larger CCN and IN concentrations) shows the smallest values: Compared to the pre-1960s run, precipitation is reduced by 3.1% for WC, 4.0% for PF, 5.4% for WA, and 6.5% for BC. Again, this effect is more pronounced for the dry season (between 5.5% and 8.7%) than for the wet season (between 2.4% and 4.4%). The wrf-muja run with additional CCN emissions from Muja Power Station, but otherwise standard aerosol concentrations from wrf-aero, lies in between. It is nevertheless interesting to correlate the wrf-muja seasonal differences in decrease in precipitation for the four regions with the surface wind fields displayed in Fig. 7. For regions WA and WC, the ratios between the decrease in summer (dry season) and winter (wet season) precipitation relative to wrf-aero are 3.1 (3.7%/1.2%) and 3.3 (3.6%/1.1%), respectively. For region BC, where the bulk of the additional surface emissions from Muja Power Station are transported to in austral winter, we find a ratio of 2.9 (4.7%/1.6%). The opposite holds for region PF, which is most affected by the aerosol inflow from Muja Power Station in austral summer and for which we find a ratio of 11.0 (6.7%/0.61%). These numbers demonstrate the potential impact of aerosol emissions from a single source on local rainfall amounts.

Last, we investigate the impact of aerosol emissions on extreme events. No statistically significant change in extreme rainfall events or the length of dry spells can be detected between the four model runs (not shown), which is in line with previous findings by Thompson and Eidhammer (2014) and the results of an observational and modeling study over eastern China by Qian et al. (2009), who conclude that changes in aerosol concentrations primarily impact light precipitation systems.

3) Investigating the effects of aerosol-aware physics

The addition of aerosol physics to the microphysics scheme first leads to an increase in modeled precipitation, but further increasing the amount of CCN and IN in particular decreases the amount of rainfall. In this section, we aim at exploring the reasons for this behavior. We concentrate on the wet season April–September, for which the bulk of the precipitation is generated. Figure 10, top panel, displays the accumulated rainfall for the wet season, averaged over the entire period 1970–74, for the four 3.3-km models as well as for gridded observational data from CRU and UDEL. In Fig. 10, bottom panel, the differences with respect to wrf-aero are displayed. Large discrepancies between the two observational datasets CRU and UDEL can be found. The close match of UDEL and BOM station data (see Fig. 8 and the discussion above) suggest that CRU is underestimating precipitation in this case.

Fig. 10.

(top) Mean accumulated precipitation for the wet seasons 1970–74, with 400-mm isohyates in black contours; (bottom) difference in mean accumulated precipitation with respect to wrf-aero for the wet seasons 1970–74.

Fig. 10.

(top) Mean accumulated precipitation for the wet seasons 1970–74, with 400-mm isohyates in black contours; (bottom) difference in mean accumulated precipitation with respect to wrf-aero for the wet seasons 1970–74.

The comparison of observed and modeled total rainfall shows that the standard run without aerosol physics reproduces the observed precipitation best and overestimates inland rainfall least among all four models. It should be noted here that the low-resolution models wrf-30km and wrf-10km fail to generate sufficient precipitation along the coast, but capture the drier inland conditions better (not displayed). Focusing on the effect of adding aerosol physics to the model, we find that coastal and inland rainfall are larger for the aerosol-aware model runs than for the standard run. However, increasing the aerosol concentration leads to lower precipitation amounts from the northeast to the northwest. The 400-mm isohyates displayed in Fig. 10 are pushed slightly to the southwest. Averaged over the entire domain and the wet seasons 1970–74, precipitation is reduced from 429 mm yr−1 (wrf-aero) by 10 mm yr−1 for wrf-muja and by 26 mm yr−1 for wrf-aerox3. The standard model without aerosol physics wrf-std produces 51 mm yr−1 less rain than wrf-aero.

Figure 11 displays mean near-surface temperature and mean sea level pressure for the wet seasons 1970–74. Gridded observational data for temperature are taken from CRU and UDEL. For mean sea level pressure, observations are obtained from HadSLP2 and reanalysis data from the forcing ERA-40 dataset. All model runs are showing lower temperatures of about 1°C toward the south of the domain, compared to the observations, and very little difference between them. The resulting mean sea level pressure (psl), however, is significantly different: While the gradient in psl in the high-resolution model runs follow that of the forcing ERA-40 data, the difference from the southwest corner to the northeast corner is larger and around 4 hPa, compared to about 2.5 hPa for ERA-40. We attribute this difference to the improved representation of the topography in the high-resolution model runs. While the difference in psl between the southwest and northeast corners is comparable for the high-resolution runs, the absolute values are lowest for wrf-std, followed by wrf-aerox3, wrf-muja, and wrf-aero.

Fig. 11.

Mean near-surface temperature and mean sea level pressure for the wet seasons 1970–74.

Fig. 11.

Mean near-surface temperature and mean sea level pressure for the wet seasons 1970–74.

To explain these findings, Table 2 summarizes the most important diagnostics for the four model runs averaged over the entire area WA and the wet seasons 1970–74. The key quantities responsible for the differences between the model runs are the surface CCN emission rate and the (atmospheric) CCN and IN number concentrations. The latter two values are prescribed and constant for wrf-std and significantly higher for the IN number concentration, while the CCN number concentration is slightly lower than for the aerosol-aware runs. It is important to note that those numbers are fixed values that are not used for calculating aerosol indirect effects, since the number of cloud droplets is also prescribed in the microphysics scheme (non-aerosol-aware microphysics scheme). In that respect, the wrf-std run corresponds to an aerosol-aware run with significantly larger IN (and comparable CCN) number concentrations in the atmosphere, albeit being inconsistent in the formulation of the microphysics processes. Among the aerosol-aware runs, wrf-aero shows the lowest values of CCN/IN number concentrations and surface CCN emission rate, and wrf-aerox3 the highest values (except for the CCN number concentration, which is slightly lower than for wrf-muja).

Table 2.

Model diagnostics for the four high-resolution runs wrf-std, wrf-aero, wrf-aerox3, and wrf-muja for the entire area of West Australia (WA), averaged over the wet seasons 1970–74 and over land only. The CCN/IN number concentrations and the various mixing ratios are calculated as the mean value over the entire column at each grid point. For wrf-std, the CCN/IN number concentrations are prescribed, while for the three aerosol-aware runs these numbers are a result of the continuous advection, diffusion, and near-surface replenishment as described in detail in section 2c. Listed are also cloud cover and CCN/IN number concentrations for the three subregions WC, PF, and BC.

Model diagnostics for the four high-resolution runs wrf-std, wrf-aero, wrf-aerox3, and wrf-muja for the entire area of West Australia (WA), averaged over the wet seasons 1970–74 and over land only. The CCN/IN number concentrations and the various mixing ratios are calculated as the mean value over the entire column at each grid point. For wrf-std, the CCN/IN number concentrations are prescribed, while for the three aerosol-aware runs these numbers are a result of the continuous advection, diffusion, and near-surface replenishment as described in detail in section 2c. Listed are also cloud cover and CCN/IN number concentrations for the three subregions WC, PF, and BC.
Model diagnostics for the four high-resolution runs wrf-std, wrf-aero, wrf-aerox3, and wrf-muja for the entire area of West Australia (WA), averaged over the wet seasons 1970–74 and over land only. The CCN/IN number concentrations and the various mixing ratios are calculated as the mean value over the entire column at each grid point. For wrf-std, the CCN/IN number concentrations are prescribed, while for the three aerosol-aware runs these numbers are a result of the continuous advection, diffusion, and near-surface replenishment as described in detail in section 2c. Listed are also cloud cover and CCN/IN number concentrations for the three subregions WC, PF, and BC.

Model runs with larger aerosol number concentrations exhibit larger cloud water mixing ratios and smaller water vapor and rainwater mixing ratios through increased numbers of cloud droplets, which suppresses warm rain events (second aerosol indirect effect; Albrecht 1989; Thompson and Eidhammer 2014). Consequently, the near-surface relative humidity is reduced. Larger aerosol number concentrations also correspond to lower near-surface temperatures, which leads to shallower (i.e., more stable) boundary layers. The resulting near-surface or mean sea level pressure is determined by the density and the temperature in the column through the equation of state, . While the mean column air density is lower for the model runs with larger water vapor concentrations (i.e., smallest for wrf-aero), this effect is outweighed by the lower temperatures of the runs with smallest water vapor concentrations and highest aerosol number concentrations (i.e., lowest for wrf-std). Consequently, the resulting mean sea level pressure is lowest for wrf-std and highest for wrf-aero.

Increasing numbers of cloud droplets of overall smaller size for larger aerosol concentrations should in principle also lead to changes in cloud cover and radiation budgets (first aerosol indirect effect). However, we do not find a clear correlation between incoming and outgoing shortwave/longwave radiation and the CCN/IN number concentrations among the four model runs (not shown). Also, the effect on cloud cover is less clear: while we do find that larger CCN and IN number concentrations lead to slightly greater cloud cover, increasing the CCN number concentrations close to the surface only (wrf-muja) leads to slightly smaller cloud cover for the entire region WA and for BC, and to slightly greater cloud cover for WC and PF (cf. Table 2; differences of the order of 0.1%). While a detailed investigation is beyond the scope of this study, we speculate that the absence of a clear signal of the first aerosol indirect effect is due to the relatively small changes in aerosol loading between the runs [the sensitivity experiments in Thompson and Eidhammer (2014) use 10 times smaller/larger values of CCN than their control run] and the fact that the near-surface CCN emissions from Muja Power Station are only slowly mixed into higher layers, whereas IN are generally found higher up in the atmosphere. Further, as discussed in Thompson and Eidhammer (2014), while changes in aerosol concentrations can have strong effects on scales of single events and small areas due to shifts in location of clouds and rainbands, these effects might be smeared out to in the context of long-term simulations.

4. Summary and conclusions

In this study, we investigate the potential impact of changing aerosol concentrations on rainfall distribution and amount in southwestern Australia. Prior to this, we revisit historical long-term observations of rainfall to differentiate between the nature of the processes contributing to the observed decline in precipitation in the twentieth century. On larger spatial scales for continental southwestern Australia, we determine a continuous decline in precipitation rather than a sudden drop. This is in line with Karoly (2014) and Delworth and Zeng (2014), who concluded that many aspects of the observed reduction in rainfall can be attributed to anthropogenic changes in levels of greenhouse gases and ozone in the atmosphere and changes to the transport and advection patterns over the Indian Ocean.

On smaller spatial scales and for the particular case of the Perth/Freemantle area, the observed decline in precipitation is too strong to be explained by changes of the large-scale atmospheric motion only. Further, we detect large differences between the individual stations and we identify in particular two stations (9010 Churchman Brook, and 9031 Mundaring Weir) within the Perth water storage basin with a dramatic decline in annual precipitation by around 30% between 1920 and 2015. More importantly, a significant share of this decrease falls into the 1970s. This coincides well with the observations of rainfall and streamflow measurements reported by Bates et al. (2008) and Karoly (2014). We conclude that further processes of likely anthropogenic nature occurring on shorter time scales and smaller regional scales must have been involved. Candidates therefore are irrigation and deforestation (Andrich and Imberger 2013) with subsequent release of ultrafine particles from salt lakes (Junkermann et al. 2009; Kamilli et al. 2015), and anthropogenic aerosols emitted by coal power plants and smelters.

Here, we focus on the possible role of anthropogenic aerosols only and assume a constant land-use classification in our models, which dates back to 2001 and thus matches the conditions after the land clearing in southwestern Australia. We consider in particular the emissions from large pollutants such as the Muja Power Station, commissioned in 1966 approximately 200 km southeast of the Perth catchment area, and the impact of a consistent treatment of aerosols on precipitation through first and second aerosol indirect effects using four different regional climate modeling experiments with a convection-permitting grid spacing of 3.3 km. We create pre and postindustrial aerosol profiles of ultrafine and fine water-friendly aerosol particles (CCN) and ice-friendly aerosol particles (IN), based on airborne measurements (Bigg 1973; Bigg and Turvey 1978; Junkermann et al. 2009; Bigg et al. 2015).

First, we show that the emissions of ultrafine particles from Muja Power Station alone increase the amount of cloud condensation nuclei (CCN) by a factor of 3 over the entire area and that this leads to a reduction in precipitation in the model along the West Coast, including the Perth/Freemantle area, as well as farther inland. We further show that the emissions from Muja Power Station are only slowly mixed up into higher layers and follow the near-surface wind fields, which leads to a large inflow of aerosols into the Perth/Freemantle region in austral summer and into the Back Country region in austral winter. Accordingly, precipitation around Perth is suppressed to a greater extent in austral summer than in austral winter compared to the average over the entire region of southwestern Australia, whereas the opposite holds for the Back Country region.

The poor documentation of the history of the western Australian power stations in the National Pollution Inventory does not give insight into when flue gas cleaning filters, affecting the partitioning between fine and ultrafine particles, were installed and how much CCN and IN particles were emitted prior to it. Hence, in a second experiment we increase both CCN and IN by a factor of 3 compared to pre-1960s levels. By inspecting a number of diagnostics and meteorological parameters, we find that more aerosols lead to more cloud droplets, higher cloud water mixing ratios and lower water vapor and rainwater mixing ratios through coagulation processes, and accordingly to the suppression of predominantly light rain (second aerosol indirect effect). We also find that cloud coverage is increased slightly for larger CCN and IN concentrations, but not for larger CCN concentrations only. Contrary to Thompson and Eidhammer (2014), we do not observe the first aerosol indirect effect in the form of increasing outgoing shortwave radiation at the top of the atmosphere and downward longwave radiation at the surface. This is presumably due to the relatively small changes in aerosol loading and the long simulation times in our study, over which potentially strong effects on spatiotemporal scales of single events are smeared out.

Tripling the pre-1960s aerosol concentrations of both CCN and IN (CCN only) corresponds to a decrease in annual precipitation between 3.1% (1.7%) for the West Coast and 6.5% (2.8%) for the Back Country. We also compare the pre-1960s aerosol run wrf-aero with the standard run wrf-std, which uses prescribed and constant cloud droplet numbers and CCN/IN number concentrations in the radiation and microphysics schemes. These values are significantly larger in the case of cloud droplets and IN number concentrations, and comparable in the case of CCN number concentrations. Relative to wrf-aero, the decrease in annual precipitation ranges from 8.1% (WC) to 12.1% (BC) for wrf-std and is thus stronger than for wrf-aerox3 and wrf-muja.

In summary, our modeling results suggest that anthropogenic aerosol emissions can contribute to the observed sudden rainfall decline in the Perth/Freemantle area. While a decrease of around 10% could account for the majority of the observed drop in the 1970s, it remains to be answered whether the combined CCN and IN emissions from Muja Power Station and from other large emitters in the area, for example the Kwinana Oil Refinery (commissioned in 1955), are sufficient. Further, as shown by Andrich and Imberger (2013), the vast deforestation occurring in the same period can also lead to a decrease in precipitation.

This leads us to speculate that the gradual decline in rainfall for large parts of southwestern Australia over the last 100 years and the sudden drop in the 1970s on smaller spatial scales (and in particular for the Perth catchment area) are a combination of three distinct processes: 1) continuous changes of the large-scale atmospheric circulation through anthropogenic changes in levels of greenhouse gases and ozone in the atmosphere, 2) a nearly complete deforestation of natural indigenous vegetation to give room to agriculture in the 1950s to 1970s, and 3) a dramatic increase in anthropogenic aerosols as a result of the commissioning of oil refineries, smelters, and power plants in southwestern Australia within the same period.

Changes in greenhouse gases, ozone, and large-scale internal climate variability are occurring on longer time scales and are included implicitly through the forcing reanalysis dataset used in our study. Hence, future work should try to disentangle the effects of deforestation and anthropogenic aerosols through a series of experiments with different land-use classifications or different aerosol concentrations and a combination of the two, based on detailed land-use maps and precise CCN and IN emissions from all major pollutants. We also suggest repeating the experiments presented here for the East Coast of Australia, where no large-scale deforestation took place and the effect of rising anthropogenic aerosol emissions can be studied in isolation.

Acknowledgments

Funding of all authors from the Helmholtz Association within the programme “Atmosphere and Climate” (POF III, subtopic 12.02.03, “Regional Climate and Water Cycle Variability”) is gratefully acknowledged. The modelling experiments presented here required more than 2 Mio CPUh and were conducted on the Karlsruhe Institute of Technology Steinbruch Centre for Computing (KIT-SCC) ForHLR1 supercomputer. The authors acknowledge the European Centre for Medium-Range Weather Forecasts (ECMWF) for the dissemination of ERA-40, the NOAA/OAR/ESRL PSD, Boulder, CO, for providing the UDEL air temperature and precipitation data and the HadSLP2 sea level pressure data, the University of East Anglia, Climate Research Unit, for access to the CRU air temperature and precipitation data, and the Bureau of Meteorology, Australia, for the dissemination of the daily rainfall climate data. The authors are particularly grateful for the support of Greg Thompson (NCAR) in the design of the experiment and the setup of the WRF Model and for the constructive comments from three anonymous reviewers.

APPENDIX

Data Availability and Auxiliary Tables

The configuration files required to reproduce this modeling experiment and the model results discussed in this contribution are available at https://doi.org/10.1594/PANGAEA.859952. For further details, see the documentation of files in the archive.

Appendix Tables A1A3 appear below.

Table A1.

Stations and fit parameters used to investigate the type of long-term precipitation decline in southwestern Australia and in selected areas of investigation. Also listed are the stations with minimum/maximum precipitation in each area for the entire period 1920 to 2015.

Stations and fit parameters used to investigate the type of long-term precipitation decline in southwestern Australia and in selected areas of investigation. Also listed are the stations with minimum/maximum precipitation in each area for the entire period 1920 to 2015.
Stations and fit parameters used to investigate the type of long-term precipitation decline in southwestern Australia and in selected areas of investigation. Also listed are the stations with minimum/maximum precipitation in each area for the entire period 1920 to 2015.
Table A2.

Bias/coefficient of variation of RMSE of the seasonal cycle (mean monthly precipitation) and Pearson correlation coefficient of the interannual variability (mean monthly precipitation anomaly) for the 4-yr simulation period from 1 Jan 1970 to 1 Jan 1974 for the different model runs and observations from UDEL, for each of the regions over land only and split into wet season and dry season. Mean monthly precipitation values of UDEL gridded rainfall data (mm month−1) per region and season (wet/dry) are (WA), (WC), (PF), and (BC).

Bias/coefficient of variation of RMSE of the seasonal cycle (mean monthly precipitation) and Pearson correlation coefficient of the interannual variability (mean monthly precipitation anomaly) for the 4-yr simulation period from 1 Jan 1970 to 1 Jan 1974 for the different model runs and observations from UDEL, for each of the regions over land only and split into wet season and dry season. Mean monthly precipitation values of UDEL gridded rainfall data (mm month−1) per region and season (wet/dry) are  (WA),  (WC),  (PF), and  (BC).
Bias/coefficient of variation of RMSE of the seasonal cycle (mean monthly precipitation) and Pearson correlation coefficient of the interannual variability (mean monthly precipitation anomaly) for the 4-yr simulation period from 1 Jan 1970 to 1 Jan 1974 for the different model runs and observations from UDEL, for each of the regions over land only and split into wet season and dry season. Mean monthly precipitation values of UDEL gridded rainfall data (mm month−1) per region and season (wet/dry) are  (WA),  (WC),  (PF), and  (BC).
Table A3.

Accumulated precipitation (mm) and spatial correlation at the end of the 4-yr simulation period from 1 Jan 1970 to 1 Jan 1974 for the different model runs and observations from UDEL, for each of the regions over land only and split into wet season and dry season.

Accumulated precipitation (mm) and spatial correlation at the end of the 4-yr simulation period from 1 Jan 1970 to 1 Jan 1974 for the different model runs and observations from UDEL, for each of the regions over land only and split into wet season and dry season.
Accumulated precipitation (mm) and spatial correlation at the end of the 4-yr simulation period from 1 Jan 1970 to 1 Jan 1974 for the different model runs and observations from UDEL, for each of the regions over land only and split into wet season and dry season.

REFERENCES

REFERENCES
Albrecht
,
B. A.
,
1989
:
Aerosols, cloud microphysics, and fractional cloudiness
.
Science
,
245
,
1227
1230
, doi:.
Allan
,
R.
, and
T.
Ansell
,
2006
:
A new globally complete monthly historical gridded mean sea level pressure dataset (HadSLP2): 1850–2004
.
J. Climate
,
19
,
5816
5842
, doi:.
Andreae
,
M. O.
,
2009
:
Correlation between cloud condensation nuclei concentration and aerosol optical thickness in remote and polluted regions
.
Atmos. Chem. Phys.
,
9
,
543
556
, doi:.
Andrich
,
M. A.
, and
J.
Imberger
,
2013
:
The effect of land clearing on rainfall and fresh water resources in Western Australia: A multi-functional sustainability analysis
.
Int. J. Sustain. Dev. World Ecol.
,
20
,
549
563
, doi:.
Bates
,
B. C.
,
P.
Hope
,
B.
Ryan
,
I.
Smith
, and
S.
Charles
,
2008
:
Key findings from the Indian Ocean Climate Initiative and their impact on policy development in Australia
.
Climatic Change
,
89
,
339
354
, doi:.
Bigg
,
E. K.
,
1973
:
Ice nucleus concentrations in remote areas
.
J. Atmos. Sci.
,
30
,
1153
1157
, doi:.
Bigg
,
E. K.
, and
D.
Turvey
,
1978
:
Sources of atmospheric particles over Australia
.
Atmos. Environ.
,
12
,
1643
1655
, doi:.
Bigg
,
E. K.
,
S.
Soubeyrand
, and
C.
Morris
,
2015
:
Persistent after-effects of heavy rain on concentrations of ice nuclei and rainfall suggest a biological cause
.
Atmos. Chem. Phys.
,
15
,
2313
2326
, doi:.
BOM
,
2015
: About climate data: Daily rainfall. Bureau of Meteorology, data accessed 27 October 2015. [Available online at http://www.bom.gov.au/climate/cdo/about/about-rain-data.shtml.]
Bradshaw
,
C. J. A.
,
2012
:
Little left to lose: Deforestation and forest degradation in Australia since European colonization
.
J. Plant Ecol.
,
5
,
109
120
, doi:.
Colarco
,
P.
,
A.
Da Silva
,
M.
Chin
, and
T.
Diehl
,
2010
:
Online simulations of global aerosol distributions in the NASA GEOS-4 model and comparisons to satellite and ground-based aerosol optical depth
.
J. Geophys. Res.
,
115
,
D14207
, doi:.
Delworth
,
T. L.
, and
F.
Zeng
,
2014
:
Regional rainfall decline in Australia attributed to anthropogenic greenhouse gases and ozone levels
.
Nat. Geosci.
,
7
,
583
587
, doi:.
Delworth
,
T. L.
, and Coauthors
,
2012
:
Simulated climate and climate change in the GFDL CM2.5 high-resolution coupled climate model
.
J. Climate
,
25
,
2755
2781
, doi:.
Fersch
,
B.
, and
H.
Kunstmann
,
2014
:
Atmospheric and terrestrial water budgets: Sensitivity and performance of configurations and global driving data for long term continental scale WRF simulations
.
Climate Dyn.
,
42
,
2367
2396
, doi:.
Gallagher
,
M. W.
,
E.
Nemitz
,
J. R.
Dorsey
,
D.
Fowler
,
M. A.
Sutton
,
M.
Flynn
, and
J. H.
Duyzer
,
2002
:
Measurements and parameterizations of small aerosol deposition velocities to grassland, arable crops, and forest: Influence of surface roughness length on deposition
.
J. Geophys. Res.
,
107
,
4154
, doi:.
Ginoux
,
P.
,
M.
Chin
,
I.
Tegen
,
J. M.
Prospero
,
B.
Holben
,
O.
Dubovik
, and
S.-J.
Lin
,
2001
:
Sources and distributions of dust aerosols simulated with the GOCART model
.
J. Geophys. Res.
,
106
(
D17
),
20 255
20 273
, doi:.
Grabowski
,
W. W.
, and
H.
Morrison
,
2011
:
Indirect impact of atmospheric aerosols in idealized simulations of convective-radiative quasi equilibrium. Part II: Double-moment microphysics
.
J. Climate
,
24
,
1897
1912
, doi:.
Harris
,
I.
,
P. D.
Jones
,
T. J.
Osborn
, and
D. H.
Lister
,
2014
:
Updated high-resolution grids of monthly climatic observations—The CRU TS3.10 dataset
.
Int. J. Climatol.
,
34
,
623
642
, doi:.
Junkermann
,
W.
, and
J. M.
Hacker
,
2015
:
Ultrafine particles over eastern Australia: An airborne survey
.
Tellus
,
67B
,
25308
, doi:.
Junkermann
,
W.
,
J. M.
Hacker
,
T.
Lyons
, and
U.
Nair
,
2009
:
Land use change suppresses precipitation
.
Atmos. Chem. Phys.
,
9
,
6531
6539
, doi:.
Kala
,
J.
,
T.
Lyons
, and
U.
Nair
,
2011
:
Numerical simulations of the impacts of land-cover change on cold fronts in south-west Western Australia
.
Bound.-Layer Meteor.
,
138
,
121
138
, doi:.
Kamilli
,
K. A.
,
J.
Ofner
,
B.
Lendl
,
P.
Schmitt-Kopplin
, and
A.
Held
,
2015
:
New particle formation above a simulated salt lake in aerosol chamber experiments
.
Environ. Chem.
,
12
,
489
503
, doi:.
Karoly
,
D. J.
,
2014
:
Climate change: Human-induced rainfall changes
.
Nat. Geosci.
,
7
,
551
552
, doi:.
Klein
,
C.
,
D.
Heinzeller
,
J.
Bliefernicht
, and
H.
Kunstmann
,
2015
:
Variability of West African monsoon patterns generated by a WRF multi-physics ensemble
.
Climate Dyn.
,
45
,
2733
2755
, doi:.
Lee
,
S. S.
, and
G.
Feingold
,
2013
:
Aerosol effects on the cloud-field properties of tropical convective clouds
.
Atmos. Chem. Phys.
,
13
,
6713
6726
, doi:.
Miguez-Macho
,
G.
,
G. L.
Stenchikov
, and
A.
Robock
,
2004
:
Spectral nudging to eliminate the effects of domain position and geometry in regional climate model simulations
.
J. Geophys. Res.
,
109
,
D13104
, doi:.
Noble
,
E.
,
L.
Druyan
, and
M.
Fulakeza
,
2014
:
The sensitivity of WRF daily summertime simulations over West Africa to alternative parameterizations. Part I: African wave circulation
.
Mon. Wea. Rev.
,
142
,
1588
1608
, doi:.
Prein
,
A. F.
,
A.
Gobiet
,
M.
Suklitsch
,
H.
Truhetz
,
N. K.
Awan
,
K.
Keuler
, and
G.
Georgievski
,
2013
:
Added value of convection permitting seasonal simulations
.
Climate Dyn.
,
41
,
2655
2677
, doi:.
Qian
,
Y.
,
D.
Gong
,
J.
Fan
,
L. R.
Leung
,
R.
Bennartz
,
D.
Chen
, and
W.
Wang
,
2009
:
Heavy pollution suppresses light rain in China: Observations and modeling
.
J. Geophys. Res.
,
114
,
D00K02
, doi:.
Ruprecht
,
J.
, and
N.
Schofield
,
1991
:
Effects of partial deforestation on hydrology and salinity in high salt storage landscapes. II. Strip, soils and parkland clearing
.
J. Hydrol.
,
129
,
39
55
, doi:.
Saunders
,
D.
,
1989
:
Changes in the avifauna of a region, district and remnant as a result of fragmentation of native vegetation: The wheatbelt of western Australia. A case study
.
Biol. Conserv.
,
50
,
99
135
, doi:.
Seifert
,
A.
,
C.
Köhler
, and
K. D.
Beheng
,
2012
:
Aerosol-cloud-precipitation effects over Germany as simulated by a convective-scale numerical weather prediction model
.
Atmos. Chem. Phys.
,
12
,
709
725
, doi:.
Skamarock
,
W.
, and Coauthors
,
2008
: A description of the Advanced Research WRF version 3. NCAR Tech. Rep. NCAR/TN-475+STR, 113 pp., doi:.
Tao
,
W. K.
,
J. P.
Chen
,
Z.
Li
,
C.
Wang
, and
C.
Zhang
,
2012
:
Impact of aerosols on convective clouds and precipitation
.
Rev. Geophys.
,
50
,
RG2001
, doi:.
Thompson
,
G.
, and
T.
Eidhammer
,
2014
:
A study of aerosol impacts on clouds and precipitation development in a large winter cyclone
.
J. Atmos. Sci.
,
71
,
3636
3658
, doi:.
Uppala
,
S.
, and Coauthors
,
2005
:
The ERA-40 Re-Analysis
.
Quart. J. Roy. Meteor. Soc.
,
131
,
2961
3012
, doi:.
van den Heever
,
S. C.
,
G. L.
Stephens
, and
N. B.
Wood
,
2011
:
Aerosol indirect effects on tropical convection characteristics under conditions of radiative–convective equilibrium
.
J. Atmos. Sci.
,
68
,
699
718
, doi:.
von Storch
,
H.
,
H.
Langenberg
, and
F.
Feser
,
2000
:
A spectral nudging technique for dynamical downscaling purposes
.
Mon. Wea. Rev.
,
128
,
3664
3673
, doi:.
Weisman
,
M. L.
,
W. C.
Skamarock
, and
J. B.
Klemp
,
1997
:
The resolution dependence of explicitly modeled convective systems
.
Mon. Wea. Rev.
,
125
,
527
548
, doi:.
Willmott
,
C.
, and
K.
Matsuura
,
2014
: University of Delaware air temperature and precipitation long term monthly means V3.01. Accessed 2 November 2015. [Available online at http://www.esrl.noaa.gov/psd/data/gridded/data.UDel_AirT_Precip.html.]

Footnotes

1

A patch of the WRF Model is required for this version to ensure correct restart results when using the Thompson–Eidhammer aerosol-aware microphysics scheme; see http://www2.mmm.ucar.edu/wrf/users/wrfv3.6/known-prob-3.6.1.html for details.