On the recent warming in the Murray-Darling Basin: Land surface interactions misunderstood



[1] Previous studies of the recent drought in the Murray-Darling Basin (MDB) have noted that low rainfall totals have been accompanied by anomalously high air temperatures. Subsequent studies have interpreted an identified trend in the residual timeseries of non-rainfall related temperature variability as a signal of anthropogenic change, further speculating that increased air temperature has exacerbated the drought through increasing evapotranspiration rates. In this study, we explore an alternative explanation of the recent increases in air temperature. This study demonstrates that significant misunderstanding of known processes of land surface – atmosphere interactions has led to the incorrect attribution of the causes of the anomalous temperatures, as well as significant misunderstanding of their impact on evaporation within the Murray-Darling Basin.

1. Introduction

[2] Previous studies have noted that during the recent drought in the Murray-Darling Basin (MDB) low rainfall totals have been accompanied by anomalously high air temperatures. In particular, Karoly et al. [2003] noted that whilst monthly rainfall totals were at extreme lows during the 2002 drought, the monthly average maximum temperatures were much higher than in previous droughts. This led the authors to state that (p.1) “…the higher temperatures caused a marked increase in evaporation rates, which sped up the loss of soil moisture and the drying of vegetation and watercourses. This is the first drought in Australia where the impact of human-induced global warming can be clearly observed…”

[3] Similarly, Nicholls [2004] investigated the anomalously high air temperatures that occurred during the 2002 cool season (May–October) in the MDB. This was achieved by comparison to an identified negative correlation between average monthly temperature and rainfall, between 1952 and 2002. Nicholls [2004] then examined the residual timeseries of the correlation and found a statistically significant increase toward higher air temperatures over the period of the regression data. It was then speculated that this was due to the increasing trend in atmospheric CO2 and other greenhouse gases, and that “the warming has meant that the severity and impacts of the most recent drought have been exacerbated by enhanced evaporation and evapotranspiration” [Nicholls, 2004, p. 334]. Such claims have previously been questioned given that drought is characterised by reduced rather than increased evapotranspiration [Roderick and Farquhar, 2004].

[4] This paper explores an alternative explanation of recent inter-annual variability and temporal trends in seasonal maximum air temperatures. First we present observations of sunshine hours (SSH) duration as an alternative explanatory variable for average maximum daily air temperature in direct comparison to average monthly rainfall. We then explore the role of elevated temperature in influencing land surface evapotranspiration. The relative role of the different alternative drivers of evapotranspiration is then evaluated using a coupled Land Surface - Planetary Boundary Layer (PBL) model.

2. Data

[5] Rainfall, temperature and sunshine hours duration (SSH) data were obtained from the Bureau of Meteorology. The temperature data are mean daily maximum temperatures, whilst the rainfall are mean monthly totals, both averaged over the MDB for the ‘cool’ season of May–October as utilised by Nicholls [2004].

[6] The SSH data consists of average monthly SSH observed at each station within the MDB (Figure 1). The MDB average for each year was computed by arithmetically averaging the SSH duration across the stations with data for each year. The SSH duration data was measured using a Campbell-Stokes sunshine recorder. Bright sunshine is recorded when the sky is clear and the direct solar irradiance exceeds a minimum threshold of 120 Wm−2 [World Meteorological Organization (WMO), 2003]. This threshold is the standard defined by the WMO. The World Meteorological Organization gives the SSH measurements an uncertainty of ±0.1hr [WMO, 2008].

Figure 1.

Location of Murray-Darling Basin in Australia and individual stations in the Murray-Darling Basin used for SSH data. The basin is outlined by the thick continuous line.

[7] The timeseries and fitted trends of maximum temperature, rainfall and SSH are shown in Figures 2 (top), 2 (middle), and 2 (top) respectively. Figure 2 (bottom) suggests that the mean maximum temperature has increased since 1952, whilst Figure 2 (middle) shows an accompanying decreasing trend in rainfall. Figure 2 (bottom) shows the SSH duration time series, and indicates an increasing trend, suggesting that sunshine hours may have utility in explaining the variance of observed temperatures.

Figure 2.

Time series of May–October (middle) rainfall totals, (top) mean maximum temperature and (bottom) SSH averaged across the Murray-Darling Basin from 1952 to 2008. The linear trend for each time series is also shown.

3. Linear Regression Models of Maximum Air Temperature

[8] Following the method of Nicholls [2004], Figure 3a shows the correlation between the mean maximum temperature and rainfall data, whilst Figure 3b shows the resultant residual timeseries. The fraction of temperature variability explained by the rainfall is quantified using Pearson correlation (r2), whilst the statistical significance of the temporal trend in the residual is assessed using a p-value. This analysis indicates a reasonably strong relationship between temperature and rainfall (r2 = 0.426), and a highly statistically significant temporal trend in the residual series (p < 0.001).

Figure 3.

(a) Scatter diagram of rainfall versus mean maximum temperature, (b) the associated temperature residual, (c) scatter diagram of Murray-Darling Basin May–October SSH versus mean maximum temperature and (d) the associated temperature residuals, Murray-Darling Basin May–October 1952–2008. The linear trend of the residuals is shown by a broken line.

[9] The worth of SSH in explaining variations in maximum temperature can be examined using the same methodology. Figure 3c shows the correlation between the mean maximum temperature and SSH, whilst Figure 3d shows the resultant residual timeseries.

[10] Figures 3a and 3c clearly show that the correlation between SSH and max temperature (r2 = 0.632) is markedly stronger than the correlation to rainfall (r2 = 0.426). Moreover, Figure 3d shows that the temperature residual after removing the SSH-associated variability does not exhibit a statistically significant increase with time (p = 0.258). These results indicate that maximum temperature is better explained by SSH than by rainfall. Importantly, there is no significant evidence for any increase in temperature beyond that explained by an increase in SSH.

[11] A potential criticism of this analysis is that the simple arithmetic averaging of the available SSH data over the MDB may have led to a spuriously improved correlation to maximum temperature in comparison to rainfall. To exclude this possibility, the analyses were repeated for each individual station using the available individual rainfall, temperature and SSH data, with results shown in Table 1. The results indicate that SSH is a better explanatory variable of maximum temperature than rainfall for 14 of the 15 stations used.

Table 1. Locations of Individual Stations Used for SSH Analysis and the r2 Values of the Rainfall-Temperature and the SSH-Temperature Relationship
NumberLocationBOM Station IDYearsRainfall R2Sunshine Hours R2
1Moree Comp530481978–19940.3580.67
2Moree Aero531151995–20080.5710.662
6Griffith CSIRO750281963–19780.3540.486
8Canberra Forest700151952–19790.4230.241
9Canberra Aero700141978–20080.2190.786
14Lake Eildon880231977–20070.2170.381

[12] Consequently, data both at local and basin scale across the MDB demonstrate that SSH provides a more robust explanation of maximum air temperature variability, and that there is little support for an increase in underlying warming.

4. Simulation of Air Temperature and Evaporation

[13] Karoly et al. [2003] and Nicholls [2004] both proposed that increased temperatures were evident in the residual series, and moreover, that this leads to increased evaporation. Here we test these claims using a Planetary Boundary Layer (PBL) model.

[14] We employ a simple one dimensional PBL model, first presented by Tennekes [1973] and Carson [1973], and as used by Quinn et al. [1995]. The boundary layer model is coupled to a land surface model utilising the Penman-Monteith equation for estimating evapotranspiration [Monteith, 1965, 1981]. The land surface – atmosphere model simulates both potential and actual evapotranspiration, as well as heating rates in the near-surface atmospheric layer. Importantly, the PBL provides for the evolution of the boundary layer incorporating the interactions of energy and moisture fluxes at the landsurface and can therefore be used to assess the claims by Karoly et al. [2003] and Nicholls [2004] that temperature drives evapotranspiration fluxes.

[15] The PBL model was used to simulate wet and dry land surface conditions by specifying the surface resistance parameter in the Penman-Monteith equation as 50s/m and 500 s/m, respectively. For each wet and dry moisture condition, the actual evapotranspiration at the land surface and the potential evaporation from an open body of water were calculated at each timestep. The potential evapotranspiration is calculated to represent the expected evaporative losses from exposed surface water under both wet and dry landscape conditions.

[16] To assess the potential impact of elevated air temperatures, three initial temperatures were specified based on the mean May–October minimum temperature for the analysis period of 6°C. The 2002 drought was approximately 2°C warmer than the 1952–2008 average [Karoly et al., 2003]. To represent the impact of elevated temperatures, model simulations were initialised with both wet and dry bulb temperatures set at 4°C, 6°C and 8°C. The wet and dry bulb temperatures are set equal to ensure that no initial vapour deficit exists, and that any consequent divergence of evapotranspiration fluxes is due to the initial temperatures, and the different wet/dry scenarios.

[17] Figures 4a and 4b show the simulated actual evapotranspiration under wet and dry conditions, respectively. As expected, the evaporative heat fluxes are higher in the wet scenario due to the increased availability of soil moisture. Importantly, both Figures 4a and 4b demonstrate a very minor increase in evaporative fluxes associated with increased temperature.

Figure 4.

Evolution of actual and potential evapotranspiration during a sunny day for initial temperatures of 4, 6 and 8°C (a) the evolution of actual ET under wet conditions, (b) the evolution of actual ET under dry conditions, (c) the evolution of potential ET under wet conditions, (d) the evolution of potential ET under dry conditions. For an initial temperature of 6°C (e) the temperature evolution under dry and wet conditions and (f) the vapour pressure deficit under dry and wet conditions.

[18] Figures 4c and 4d compare the simulated potential evaporation as would occur from surface water under both wet and dry land surface conditions, respectively. Again, the influence of air temperature is minor. Importantly, the potential evapotranspiration is actually enhanced in the dry land surface scenario. This is due to the relatively high atmospheric moisture demand (vapour pressure deficit, shown in Figure 4f), which itself is due to the lack of actual evapotranspiration in the dry scenario.

[19] Figure 4e compares the simulated air temperature for the wet and dry scenarios. It clearly shows that higher air temperatures occur under the dry land surface scenario, due to enhanced heating rates resulting from the lack of available moisture for evapotranspiration.

[20] All presented results demonstrate that potential evaporation under dry conditions is elevated not as a result of the air temperature, but as a result of the lack of actual evaporation. This is accompanied by increased sensible heat fluxes which increases air temperatures. This is an entirely natural consequence of the dynamics of drought. Importantly, it is shown that antecedent temperature increases do not lead to significant increases in actual or potential evapotranspiration.

5. Relative Influence of Temperature and SSH on Actual Evapotranspiration

[21] The relative roles of air temperature and SSH on evapotranspiration rates can be approximated through simple calculations. The 2002 drought had an extra 1.5 hours of bright sunshine per day than the 1952–2008 average. Using the PBL model under wet land surface conditions and forced with typical sunny conditions, the average actual evapotranspiration was simulated. On average, an extra 1.5 hours of bright sunshine, instead of the alternative cloudy conditions, provides approximately 0.32mm of additional evapotranspiration. In contrast, an increase in air temperature of 2°C causes only an additional 0.076 mm of evapotranspiration over the entire day.

[22] It is therefore apparent that the increased occurrence of direct solar insolation by 1.5 hours per day is more hydrologically significant than the additional evapotranspiration loss given an increase in temperature of 2°C. It is therefore clear that increased air temperatures play a relatively minor role in driving evapotranspiration.

6. Conclusions

[23] The empirical correlations between temperature and rainfall identified by Nicholls [2003, 2004] and Karoly and Braganza [2005] are valid and statistically significant. However, to accept the correlation as the sole basis for the attribution of cause to human emissions is to implicitly assume that the correlation represents an entirely correct model of the sole driver of maximum air temperature. This is clearly not the case. The simple adoption of the residual timeseries as an anthropogenic signal is to assume that any unexplained variance is entirely due to CO2 and other greenhouse gases. As an attribution technique, this is clearly overly-simplistic given the known complexity of land surface hydrological processes and hence is susceptible to both statistical and interpretive bias.

[24] As an alternative explanation, the correlation between temperature and sunshine hours is more statistically significant than the rainfall relationships previously identified. However, despite the improved correlation offered by SSH, this model still represents a gross simplification of the known physical processes of land surface – atmosphere interactions. Consequently, to simply replace rainfall with SSH as a valid model of detection and attribution of anthropogenic influences would be as incorrect and inappropriate as the methodologies proposed by Nicholls [2003], Karoly and Braganza [2005] and the subsequent and related climate change impact assessment techniques of Cai and Cowan [2008] and Chiew et al. [2008].

[25] It is stressed that the results of the analyses presented here in no way negate genuine concerns over anthropogenic climate change. However, the science of assessing future hydroclimatic risk is not aided by premature claims of recent severe drought being incorrectly attributed to enhanced evaporation due to increased atmospheric carbon dioxide. Given the prominence of the cited techniques in providing analyses of seemingly observed climate change impacts to both the IPCC and the Australian Government [Hegerl et al., 2007; Trenberth et al., 2007; Cai and Cowan, 2008; Chiew et al., 2008], it is imperative that the assumptions, limitations and inferences of such techniques are reviewed as a matter of some urgency.


[26] NL is funded by a University of Newcastle Research Scholarship with additional support from the Faculty of Engineering and Built Environment.