Feedbacks in the climate system have the potential to exacerbate or alleviate extremes such as droughts. Over the land surface, feedbacks to precipitation are often mediated through changes in the soil moisture. These feedbacks can involve a number of processes and can be measured in a variety of ways (see Seneviratne et al., 2010). The multiple mechanistic pathways and the non-linear nature of the connection between the smoothly varying soil moisture field and highly episodic precipitation makes the feedback strength difficult to quantify with confidence. While some studies have used observations to quantify the soil moisture–precipitation feedback (Findell and Eltahir, 1999; Taylor et al., 2011, 2012; Catalano et al., 2016), more common is the use of model experiments to isolate and allow quantification of this behaviour (Schar et al., 1999; Koster et al., 2006; Seneviratne et al., 2013; Hirsch et al., 2014). Other slowly varying surface variables that have been found to provide feedbacks to precipitation include albedo (Charney et al., 1975; Lofgren, 1995; Zaitchik et al., 2007; Teuling and Seneviratne, 2008; Meng et al., 2014b) and vegetation (Pielke et al., 1998; Zeng and Neelin, 2000; Wang et al., 2006; Meng et al., 2014a). These feedbacks act on different timescales and can subdue or reinforce the feedback from the soil moisture field. This emphasises the difficulty in identifying feedback mechanisms when changes to all these surface fields are occurring simultaneously.

The influence of land–atmosphere feedbacks may be particularly important in the development of extreme events such as droughts. Such a connection has been recognised since Charney et al. (1975), who, using a global climate model (GCM), found that a change in surface albedo caused by a decrease in vegetation would cause a decrease in rainfall over the Sahara. This provided a positive feedback that enhanced drought conditions. Charney et al. (1977) extended this work, finding evidence for similar positive feedbacks in other semi-arid regions. Since these pioneering investigations, climate models have continued to improve, and many studies into the sensitivities of climate models to land surface conditions have been performed.

A number of these studies have focused on how the surface feedbacks affect the development in particular locations or drought events. For instance, Oglesby and Erickson (1989) used a GCM to examine the influence of soil moisture on drought in North America, also finding a positive feedback that enhances drought conditions. Hong and Kalnay (2000) used an RCM to investigate the role of local feedbacks in the development of the Texas, USA, drought in 1998. They found that the surface feedbacks were responsible for up to 30 % of the precipitation deficit during the drought. Schubert et al. (2004) investigated causes of the North American Dust Bowl drought in the 1930s. They attributed 50 % of the precipitation deficit to soil moisture–precipitation feedbacks. Zaitchik et al. (2007) examined the surface influence on a drought that occurred in the Middle East in 1999. Using a regional climate model (RCM), they found that vegetation and albedo changes had clear effects on the surface fluxes and planetary boundary layer (PBL) growth, but limited impact on the precipitation decrease (up to 4 %) compared to a normal year. Wu and Zhang (2013) performed an RCM investigation of soil moisture feedback on the 1999 drought in northern China, finding that the feedback accounted for up to 50 % of the precipitation decline in some places. Zaitchik et al. (2012) investigated the surface feedback on the southern Great Plains, USA, drought of 2006. They found that the precipitation decline during drought development increased by ∼ 10 % due to the feedback. Finally, Meng et al. (2014a, b) examined the role of changes in surface albedo and surface vegetation on the development of the 2002 drought in south-east Australia. They found that the precipitation reduction was enhanced by up to 20 and 10 % due to surface albedo and vegetation changes, respectively. Importantly, they identified differences in timescales over which changes in vegetation occur compared to changes in soil moisture or albedo, with the relatively slow vegetation changes tending to dampen the positive soil moisture–precipitation feedback. All of these studies showed that the land surface–precipitation feedbacks play an important role in drought development. However, the strength of this role is both space and time dependent.

In general terms, mechanisms that produce soil moisture–precipitation feedback involve a change in energy partitioning at the surface that subsequently changes the evolution of the planetary boundary layer (PBL) and the likelihood of triggering precipitation. That is, a decrease in soil moisture leads to more energy being used for sensible heating, and an increase in the PBL height with a related decrease in the moist static energy density, resulting in a decreased likelihood of triggering precipitation and a further reduction in soil moisture. Changes in other surface characteristics, such as albedo and vegetation cover, can also change the surface energy partitioning and produce a similar chain of effects that result in a feedback on the soil moisture conditions. Unlike soil moisture–precipitation feedbacks, however, the relationship between changes in these other surface characteristics and the surface energy partitioning can be quite complex. For example, an increase in albedo will reduce the net radiation at the surface but how will this reduction in available energy be partitioned between the surface fluxes? Similarly, a reduction in vegetation cover will mean more exposed soil from which water can evaporate quickly following a rainstorm, but a reduction in the vegetated area that can continue transpiring through a dry spell (up to a point). So, vegetation changes have a time-varying impact on surface energy partitioning but what is the cumulative impact on the surface energy fluxes? In reality, soil moisture, albedo and vegetation all change simultaneously. Here, we explore the impact of these changes on the development of drought in south-east Australia.

The Murray–Darling Basin (MDB) is Australia's largest river system (Fig. 1). It covers a catchment area of ∼ 1 000 000 km2 or ∼ 14 % of Australia. Unlike most other major basins of the world, the MDB is predominantly semi-arid with a very low ratio of discharge to precipitation and exhibits high interannual hydrologic variability. The MDB is Australia's “food basket” (Nicholls, 2004), accounting for 40 % of Australia's agricultural production. The 1 500 000 ha under irrigation for crops and pastures represent 70 % of the total area under irrigation in Australia, with more than 80 % of the divertible surface water resources being consumed locally. From 2001 to 2008, the worst drought in recorded history was experienced in the MDB, with the 7-year averaged rainfall being the lowest since 1900 (Potter and Chiew, 2011). An investigation into the causes and impacts of the drought, referred to as the “millennium drought”, was performed by van Dijk et al. (2013). They found that large-scale climate modes, particularly in the Pacific Ocean, could explain about two-thirds of the precipitation deficit, leaving about 30 % of the deficit unexplained and potentially related to local feedbacks (Evans et al., 2011). This study extends the work of Meng et al. (2014a, b) who examined the effects of albedo changes and vegetation changes in isolation from one another. In reality, factors such as soil moisture, albedo and vegetation change simultaneously but over different timescales during drought development and intensification. This study adds value to previous work by examining the combined evolution of these factors and the resulting feedback on precipitation during the evolution of the millennium drought.

The RCM used here was built within the Weather Research and Forecasting (WRF) model framework. The WRF model is a widely used atmospheric model maintained at the National Center for Atmospheric Research (NCAR) in the US. The Advanced Research WRF is a nonhydrostatic, terrain-following, dry hydrostatic-pressure coordinate model designed to simulate or predict regional-scale atmospheric circulation. The WRF has been comprehensively evaluated across numerous investigations over south-east Australia and has been found to perform well (Cortés-Hernández et al., 2015; Evans and McCabe, 2010; Evans and Westra, 2012).

Version 3.1.1 (Skamarock et al., 2008) was applied in this study using the following physics schemes: the Kain–Fritsch cumulus physics scheme, the WRF single-moment five-class microphysics scheme, the Dudhia short-wave radiation scheme, the rapid radiative transfer model (RRTM) long-wave radiation scheme, the Yonsei University boundary layer scheme, Monin–Obukhov surface layer similarity and the Noah land surface scheme. The model simulation uses 6-hourly boundary conditions from the National Centers for Environmental Prediction (NCEP)–NCAR reanalysis project (NNRP – Kalnay et al., 1996) with an outer 50 km resolution nest and an inner 10 km resolution nest that covers south-east Australia (Fig. 1). Both nests used 30 vertical levels (see Evans and McCabe, 2010 for further details of the model setup).

The RCM simulations in this study began at the start of the year 2000 using the climate state produced by running the model for 15 years (1985–1999) to spin up the soil moisture states in a coupled environment. Four simulations were performed: one using the WRF default albedo and default vegetation fraction (WRF_CTL); one using the default vegetation fraction and observed albedo (WRF_ALB); one using the default albedo and observed vegetation fraction (WRF_VEG); and one using observed albedo and observed vegetation fraction (WRF_BOTH). The Noah land surface scheme is described in Chen and Dudhia (2001). In this implementation, the green vegetation fraction is used to determine the fraction of a grid cell that is covered by vegetation versus bare soil. It has a direct impact on the partitioning of evaporation between soil evaporation, canopy evaporation and transpiration. The albedo changes the amount of upward short-wave radiation and hence the energy available for use in other surface energy fluxes.

The Murray and Darling river basins are the focus regions of this study (Fig. 1). The Great Dividing Range to the east of both basins is a temperate zone that captures most of the precipitation that supplies the rivers. The Darling Basin has a subtropical region in the north but generally transitions through semi-arid grasslands towards desert in the west. The Murray Basin is dominated by the temperate region in the east and south, and contains grasslands in the north-west. In terms of rainfall, the Murray Basin is more consistently wet with winter-dominant precipitation, while the Darling Basin has large dry areas with summer-dominant precipitation.

The response of albedo and vegetation to the development of each of the droughts differs. The 2002 drought occurs after 2 years of declining precipitation from a normal/wet year in 2000, while the 2006 drought occurs after 1 year of declining precipitation that follows a normal year at the end of a dry period. In the Darling Basin, the 2002 drought is accompanied by a steadily increasing albedo and a decrease in vegetation fraction that is delayed by a year (Fig. 3). Neither the albedo nor the vegetation fraction recovers to pre-2002 drought levels before the 2006 drought arrives. The 2006 drought is accompanied by a smaller increase in albedo and a small decline in vegetation compared to the 2002 drought. This delayed vegetation response was explored in Meng et al. (2014a) who showed that while albedo responses and feedbacks occur on relatively short timescales (∼ 1 month), large-scale changes in vegetation occur over longer timescales (∼ 1 year) and hence become more important in multi-year droughts.

In the Murray Basin, the 2002 drought is accompanied by a steady albedo increase that occurs in summer only, while the related vegetation fraction decrease continues into 2003. The 2006 drought sees a similar decrease in albedo but the vegetation fraction experiences a 1-year decrease with a similar magnitude to that experienced in 2002. It should be noted that, in the Murray Basin during the peak drought years, an entire phenological cycle is skipped and replaced by a steady decline in vegetation. This ability for vegetation to skip phenological cycles during drought years is an adaptation found commonly in Australian arid or semi-arid zones (Broich et al., 2014). Here, we see that even temperature zone species are able to do this to an extent in extreme drought years.

The spread of the experiments is much smaller than the drought-related precipitation declines (Fig. 5) indicating that the droughts are mostly driven by external processes, with albedo and vegetation changes producing a smaller effect on precipitation. This concurs with van Dijk et al. (2013) who attribute about two-thirds of the rainfall deficit to El Niño–Southern Oscillation (ENSO). Comparing Figs. 5 and 6 allows an estimate of the role played by changing albedo and vegetation on the total precipitation decline in each basin. In the Murray Basin, from 2000 to 2002, the precipitation declined by ∼ 18 mm month-1 and the WRF_BOTH simulation declined an extra 2 mm month-1 compared to the WRF_CTL simulation. From 2005 to 2006, the precipitation declined ∼ 10 mm month-1 and the WRF_BOTH simulation declined an extra 1 mm month-1 compared to the WRF_CTL simulation. In the Darling Basin, from 2000 to 2002, the precipitation declined by ∼ 20 mm month-1 and the WRF_BOTH simulation declined by an extra 2 mm month-1 compared to the WRF_CTL simulation. From 2005 to 2006, the precipitation declined 8 mm month-1 and the WRF_BOTH simulation declined by an extra 1 mm month-1 compared to the WRF_CTL simulation. In each case, the albedo and vegetation changes combined can account for ∼ 10 % of the precipitation decline. This is double the contribution found for albedo and vegetation changes to a drought in the Middle East (Zaitchik et al., 2007).

In all of the simulations in this study, the soil moisture–precipitation feedback is present. Previous studies that explicitly quantified the magnitude of the soil moisture–precipitation feedback on the development of drought found that the feedback accounted for anything from 10 to 50 % of the precipitation decline (Hong and Kalnay, 2000; Schubert et al., 2004; Wu and Zhang, 2013; Zaitchik et al., 2012). Here, we find that the addition of vegetation fraction and albedo changes add a further 10 % to the drought-related precipitation decline. While the wide range found for the soil moisture–precipitation feedback appears to be location and model dependent, the fact that the impact of albedo and vegetation changes falls within this range suggests that they should not be ignored. In reality, soil moisture, albedo and vegetation changes all occur simultaneously, albeit over different timescales. The challenge for the land surface modelling community is to predict changes in all these surface characteristics in a physically consistent way, over drought-relevant timescales, that will allow simulation of the total surface feedbacks and hence produce more realistic drought development within climate models.

In this study, we have examined the ability of a regional climate model (WRF) to simulate the extended drought that occurred from 2002 to 2007 in south-east Australia. In particular, the ability to reproduce the two drought peaks in 2002 and 2006 was investigated. Overall, the RCM was found to reproduce both the temporal and the spatial structure of the drought-related precipitation anomalies quite well, despite using climatological seasonal surface characteristics such as vegetation fraction and albedo. This concurs with previous studies that found that about two-thirds of the precipitation decline can be attributed to ENSO. Satellite-based observations show substantial inter-annual variability in albedo and vegetation fraction, particularly in drought periods. These variations were found to be highly anti-correlated, showing that, in reality, simultaneous changes in both quantities are occurring and need to be accounted for. Experiments that allow for the vegetation fraction and albedo to vary as observed show that the intensity of the drought is underestimated by about 10 % when using climatological surface characteristics. These results suggest that in terms of drought development, capturing the feedbacks related to vegetation and albedo changes may be as important as capturing the soil moisture–precipitation feedback. In order to improve our modelling of multi-year droughts, the challenge is to capture all these related surface changes simultaneously, and provide a comprehensive land surface–precipitation feedback during the drought's development.

The WRF is introduced and referenced in the text. The code can be downloaded from http://www2.mmm.ucar.edu/wrf/users/. The simulation experiment data are several terabytes in size. These data are available upon request to the lead author.