Introduction

The effects of climate change will influence water resource availability in the future (Huntington, 2006; Trenberth et al., 2014). This is particularly of importance in Australia where major droughts have drastically increased public concerns about water availability (Leblanc, Tweed, Van Dijk, & Timbal, 2012). As a result, there have been several studies focussing on climate change impacts on water resources (Betts et al., 2007; Chien, Yeh, & Knouft, 2013; Chiew & McMahon, 2002; Chiew et al., 2009).
Most of the earlier studies on future water availability have used rainfall-runoff models with climate data from downscaled global circulation models to highlight future scenarios (e.g. Chien et al., 2013; Chiew et al., 2009; Leblanc et al., 2012; Vaze, Davidson, Teng, & Podger, 2011). For Australia, most of these studies indicate a more variable future and a decreased water availability (van Dijk et al., 2013). The main explanation is decreased rainfall and increased evaporation and therefore decreased streamflow. A drawback of such studies is that they introduce at least two types of errors: the calibration uncertainty of the rainfall-runoff model and the associated uncertainty in the prediction; and uncertainty in future climate data, either from downscaling global models (Leblanc et al., 2012), or from scaling of historical data (Chiew, 2006; Fu, Charles, & Chiew, 2007).
Reductions in streamflow in Australia have also been attributed to increased CO2 fertilisation of the vegetation due to changes in the atmospheric CO2 concentrations (Trancoso, Larsen, McVicar, Phinn, & McAlpine, 2017; Ukkola et al., 2015). However, as Ajami et al. (2017) highlight, this effect varies along aridity gradients and depends on other factors such as nutrient availability. Biological work (Franks et al., 2013) suggests that tree stomata adapt to changes in climate, i.e. there is a certain level of bio-plasticity to reduce how water is transpired as a function of temperature. This highlights further other, possibly interacting and counteracting, factors that could influence future changes in streamflow, well beyond the traditional forcing variables.
Another issue with the earlier work, including some of the latest work on the changes in catchment responses with drought (Ajami et al., 2017; Booij, Schipper, & Marhaento, 2019; Saft, Western, Zhang, Peel, & Potter, 2015), is that very few of the studies have controlled for land cover and therefore might have stationarity issues (Montanari et al., 2013; Navas, Alonso, Gorgoglione, & Vervoort, 2019). Ajami et al. (2017) analysed vegetation cover and found that runoff sensitivity to precipitation varied with fractional vegetation cover. Moreover, they confirmed that increased vegetation cover decreases runoff sensitivity, confirming earlier work (Chiew, 2006). However, Ajami et al. (2017) also point out that fractional vegetation cover covaries with precipitation, therefore making the comparison difficult.
An alternative approach to biophysical modelling with future climate scenarios to understand changes in future water yield, would be to analyse historical data sets, as several older studies in Australia have done (Booij et al., 2019; Cai & Cowan, 2008; Chiew, Whetton, McMahon, & Pittock, 1995; Petrone, Hughes, Van Niel, & Silberstein, 2010; Potter, Petheram, & Zhang, 2011; Trancoso et al., 2017). However, there are several reasons why the issue of trends in streamflow might need revisiting. Many of the older studies were based on annual data, a limited geographical area (Petrone et al., 2010) and varied in terms data periods and data aggregation in space and time. Some studies focused on a few rivers (Petrone et al., 2010; Potter & Chiew, 2011), some focussing on the whole Murray Darling water balance (Cai & Cowan, 2008) and some focussing on a large data base of catchments (Booij et al., 2019; Chiew, 2006; Trancoso et al., 2017).
Finally, the models and types of analyses varied across the studies, including simple linear regression on un-transformed and transformed data (Cai & Cowan, 2008; Chiew, 2006; Chiew et al., 2014; Potter et al., 2011), non-parametric Mann-Kendall and step change methods (Petrone et al., 2010; Potter & Chiew, 2011), statistical analyses and using rainfall runoff models (Chiew, 2006; Potter, Chiew, & Frost, 2010) and applying an energy framework (Booij et al., 2019) and focussing on baseflow (Trancoso et al., 2017). As a result, despite a consistent view of a sharply declining streamflow in response to changing rainfall, it is difficult to compare across studies.
Rather than using conceptual or deterministic modelling approaches, data based statistical models allow quick identification of the main variables explaining variations in streamflow. If the variables in the statistical model represent real physical processes (Fu et al., 2007), the model removes all variance based on the physical processes and the remaining unexplained trend in the residuals can be analysed. Modern regression approaches also allow non-linearity and the identification of uncertainties (Wood, 2006), rather than being limited to linear relationships. The disadvantage of statistical models is that they are essentially a “black box” as they do not explicitly include any physical processes. This means that, while the model can show correlation, it does not necessarily prove causation, and a lack of understanding of processes within a system may lead to an incorrect interpretation of statistical relationships.
The above review highlights that it is therefore important to have consistent methodology that includes several different statistical analyses to cross check the trend results, as each analysis included methodological and data uncertainty.
Given the range of approaches, the aim of this paper is 1) to offer a combination of statistical approaches to analyse climate change trends in streamflow data in forested catchments; 2) to use the increased observed data length to revisit the work by Chiew (2006) and other past research to investigate climate change trends on a subset of the Australian hydrological reference stations across the continent; and 3) to highlight the spatial and temporal variation in these rainfall and temperature effects on streamflow across the Australian continent.
While the data in this paper focuses on Australia, the outcomes and methodology are relevant for other regions that are experiencing changes in rainfall and temperature and have predicted changes in streamflow.
In light of the complexity of the topic, and the number of existing papers with somewhat conflicting results, all the data, explanations and code for the analyses used in this paper have been stored as supplementary data in an on-line repository, which can be accessed via: https://doi.org/10.5281/zenodo.3757041