Increases in greenhouse gas concentrations are expected to impact the
terrestrial hydrologic cycle through changes in radiative forcings and plant
physiological and structural responses. Here, we investigate the nature and
frequency of non-stationary hydrological response as evidenced through water
balance studies over 166 anthropogenically unaffected catchments in
Australia. Non-stationarity of hydrologic response is investigated through
analysis of long-term trend in annual runoff ratio (1984–2005). Results
indicate that a significant trend (
Increases in atmospheric CO
Understanding spatial and temporal variability of catchment-scale water yield in relation to precipitation variability and ecosystem productivity is challenging, as it requires long-term observational records from unimpaired catchments. A plethora of modeling studies have been performed to predict the climate change impacts on vegetation productivity (Kergoat et al., 2002; Leuzinger and Körner, 2010) and global runoff (Betts et al., 2007; Piao et al., 2007). However, projections depend on the underlying model assumptions and structure, process representation and scale of application (Medlyn et al., 2011). Similarly, assessing climate elasticity of streamflow has shown that the degree of sensitivity of streamflow to various factors depends on the model structure and calibration approach (Sankarasubramanian et al., 2001).
Here, we investigate the nature and frequency of non-stationary hydrologic
response as evidenced through water balance studies over 166
anthropogenically unaffected catchments in Australia. Our assessment assumes
the non-stationarity to manifest itself through the annual water balance,
and more specifically, through the annual runoff ratio (
Non-stationarity of runoff ratio is caused by complex interactions between
precipitation, climate variability, plant physiological and structural
responses to elevated CO
An assessment of the non-stationarity of the runoff ratio across 166 anthropogenically unaffected catchments in Australia is presented next using long-term ground- and satellite-based observational records.
Daily stream discharge data are obtained from the Australian network of
hydrologic reference stations (HRSs) that consists of 221 gauging stations
(
Distribution of hydrologic reference stations across
Australia. Colored circles represent catchments with a significant trend in
annual runoff ratio. Colors represent catchment grouping based on the
classification framework of Fig. 6. The catchments with non-stationary
hydrologic response span over three significant ecoregions of the continent.
Ecoregion boundaries are from the World Wildlife Fund (
We used ground- and satellite-based observations to detect and investigate causes of non-stationarity of runoff ratio across HRS catchments. Our methodology consists of (1) detecting trends in annual runoff ratio, fractional vegetation cover, annual precipitation and precipitation seasonality indices at a catchment scale; (2) assessing long-term (27 years) water balance patterns across all catchments with non-stationary hydrologic response using hydrologic indices such as the Horton index (Troch et al., 2009); (3) exploring annual runoff ratio's sensitivities to water balance components and fractional vegetation cover at an individual catchment scale; and (4) formulating an ecohydrologic catchment classification framework.
The modified Mann–Kendall non-parametric test (Hamed and Rao, 1998; Kendall, 1970; Mann, 1945) that accounts for serial autocorrelation in the time series is performed to detect significant trends in annual runoff ratio at a 0.01 significance level across the 166 HRS catchments. The first and last 5 years of data are removed from the record to reduce the impact of edge effect for trend analysis (1984–2005). Similar trend analysis is performed for annual precipitation and average fractional vegetation cover of each catchment.
Changes in precipitation seasonality across catchments with non-stationary
hydrologic response are explored by assessing the trends in two measures of
precipitation seasonality: the seasonality index (SI) (Walsh and Lawler,
1981) and days of a year at which the 10th, 25th, 50th,
75th and 90th percentiles of annual precipitation are reached
(Pryor and Schoof, 2008). The SI is calculated based on monthly
precipitation values (Walsh and Lawler, 1981):
Similarity of ecohydrologic response across all catchments with
non-stationary response is explored by examining the overall relationships
of long-term mean (1984–2010) annual fractional vegetation cover with annual
runoff ratio, precipitation and the Horton index (Troch et al.,
2009). The Horton index, the ratio of evapotranspiration to catchment wetting,
presents efficiency of catchments in using plant available water and is
reflective of water and energy availability in the catchment. The Horton index
ranges between 0 and 1, and incorporates the role of soil and topography in
the catchment wetting (Brooks et al., 2011; Troch et al.,
2009; Voepel et al., 2011). To estimate catchment-averaged ET and
wetting, the water balance equation (d
Sensitivities of annual runoff ratio to inter-annual variability of precipitation, ET and fractional vegetation cover in the 1984–2010 period are computed to identify factors that exert the largest sensitivity on annual runoff ratio. Normalized sensitivity of annual runoff ratio to precipitation is computed by estimating the slope of a linear regression between runoff ratio and precipitation, and multiplying it by the ratio of mean precipitation to runoff ratio (Fatichi and Ivanov, 2014; Hsu et al., 2012). Similarly, normalized sensitivity of runoff ratio to water balance ET and annual fractional vegetation cover is also computed. Normalized sensitivity of annual runoff ratio is equivalent to the streamflow elasticity approach of Zheng et al. (2009) that defined streamflow elasticity as the linear regression coefficient between the proportional changes in streamflow and a climatic variable (precipitation or potential evapotranspiration). Results of these analyses are used as the basis for formulating an ecohydrologic catchment classification.
Results of the modified Mann–Kendall trend test across 166 catchments
indicate that 20 catchments (with areas that range between 18.7 and
5158.3 km
Mean annual precipitation (
Based on the mean seasonality index using data from 1984 to 2010,
only two catchments exhibit a seasonal climate (0.6 < SI < 0.8)
(Table S2). However, all catchments have some degree of rainfall
seasonality (SI > 0.39) (Walsh and Lawler, 1981). Using the
modified Mann–Kendall trend tests, no significant trends in the 1984–2005 SI
values are observed in the catchments with non-stationary hydrologic
response (
Long-term annual average dryness index (PET
Mean and standard deviation of catchment-averaged
To explore differences between catchments with non-stationary or stationary
behavior, the cumulative absolute differences between consecutive annual
values of precipitation, fractional vegetation cover and runoff ratio for
each catchment are calculated and normalized by the total absolute
difference. In Fig. 3, the differences between catchments with
non-stationary and stationary hydrologic response are illustrated by
presenting the mean and standard deviations of normalized cumulative
differences for each group. As can be seen in Fig. 3, normalized
cumulative differences in annual precipitation and fractional vegetation
cover between the catchments with non-stationary and stationary hydrologic
response are very similar. However, large differences in the normalized
cumulative differences of annual runoff ratio exist between these
catchments. The catchment area ranges from 18.7 to 5158.3 km
Mean normalized cumulative absolute differences in annual precipitation, fractional vegetation cover and runoff ratio between catchments with non-stationary (20 catchments) and stationary (146 catchments) hydrologic response. The shaded areas represent standard deviations.
Although consistent patterns are observed in catchments' ecohydrologic response due to differences in mean annual precipitation in catchments with non-stationary hydrologic response, characterizing catchment-scale terrestrial ecosystem response to inter-annual precipitation variability is important for hydrologic predictions.
Normalized sensitivity of annual runoff ratio to annual fractional vegetation cover, ET and precipitation indicates greater sensitivity of runoff ratio to fractional vegetation cover than precipitation in most of the catchments with non-stationary hydrologic response (Fig. 4a). While runoff ratio sensitivities to precipitation are positive across all catchments with non-stationary hydrologic response, these sensitivities become negative in some catchments with increases in fractional vegetation cover. These results indicate the importance of incorporating vegetation dynamics in examining non-stationary hydrologic response.
Normalized sensitivity of annual fractional vegetation cover to
precipitation (Fatichi and Ivanov, 2014; Hsu et al., 2012) is
plotted against mean aridity index (PET
As vegetation productivity is controlled by plant available water (Brooks et al., 2011), fractional vegetation cover sensitivity to the Horton index is explored. Both positive and negative sensitivities between the fractional vegetation cover and the Horton index are observed in catchments with non-stationary hydrologic response (Fig. 4c). Positive sensitivities indicate increases in fractional vegetation cover as the Horton index increases. As a higher Horton index is indicative of a drier condition, removal of limiting factors like nutrient limitation is the likely cause of fractional vegetation cover increase in these catchments (Brooks et al., 2011). In a few of these catchments, light limitation may decrease fractional vegetation cover in wet years (positive correlations of sunshine hours with fractional vegetation cover; Table S3). In catchments with negative Horton index–fractional vegetation cover sensitivities, in which drier conditions decrease productivity, water availability is the primary factor in controlling vegetation growth. The annual runoff ratio's sensitivity to fractional vegetation cover was similar to the Horton index but with the opposite sign (Fig. 4d). This means that, in catchments with negative fractional vegetation cover–Horton index sensitivity, sensitivity of runoff ratio to fractional vegetation cover is positive, and vice versa. Across water-limited catchments (positive runoff ratio–fractional vegetation cover relationship), the runoff ratio sensitivities are smallest in catchments with the highest vegetation cover. As periods of higher productivity coincide with higher precipitation (positive precipitation–fractional vegetation cover relationship) in these catchments, runoff ratio increases in years with higher precipitation. It should be noted that the percentage of tree cover in these drier catchments are more than 60 %, with a few exceptions (Table S1). Negative runoff ratio–fractional vegetation cover sensitivities become more negative in catchments with higher fractional vegetation cover. Overall, mean annual runoff ratio and its variability (standard deviation) are smaller in drier catchments with smaller mean fractional vegetation cover (Fig. 2).
We used baseflow as a measure of catchment storage response to inter-annual precipitation variability. Baseflow sensitivities to mean annual aridity index are highest in drier catchments with non-stationary hydrologic response (Fig. 5a). Normalized fractional vegetation cover sensitivities to the baseflow decrease in catchments with higher annual baseflow index and even become negative at higher baseflow indices (Fig. 5b). This result suggests that, in catchments where groundwater constitutes significant component of streamflow, fractional vegetation cover exhibits smaller variability to changes in baseflow as vegetation roots have access to deeper water storage for transpiration and have less sensitivity to changes in baseflow.
Consistent patterns of fractional vegetation cover sensitivities to precipitation, baseflow and the Horton index across catchments with non-stationary hydrologic response present two distinct catchment response behaviors. We hypothesize plausible mechanisms to describe the likely causes of fractional vegetation cover sensitivity to inter-annual precipitation variability in order to distinguish between alternate catchment ecohydrologic responses.
At the global scale, precipitation is the main driver of vegetation productivity particularly in arid and semi-arid environments (Huxman et al., 2004). However, mean annual vegetation productivity becomes less sensitive to mean annual precipitation in humid environments (Schuur, 2003) as biogeochemical factors (nutrients, light, soil oxygen availability) or biotic factors (Yang et al., 2008) limit productivity (Fig. 6a). This is consistent with observed precipitation–fractional vegetation cover patterns across all the catchments with non-stationary hydrologic response (Fig. 2a). At a catchment scale, catchments can be classified into two main groups based on the annual precipitation and vegetation productivity relationship. We hypothesize four plausible mechanisms to explain catchment-scale ecohydrologic response to inter-annual climate variability in water- and energy-limited environments (Fig. 6b).
In group (A) catchments, a positive relationship between vegetation
productivity and precipitation increases exists and can be either caused by
(1) direct CO
Relationships between catchment-averaged annual
fractional vegetation cover and annual precipitation (left), water-balance-derived
annual ET (middle) and runoff ratio (right) against mean annual
fractional vegetation cover (1984–2010) across three catchments
representative of each class in Fig. 6. The Spearman rank correlation (
In group (B) catchments, vegetation productivity decreases in response to
annual precipitation increases. This negative feedback is most likely due to
biogeochemical constraints such as light, nutrients, temperature and soil
characteristics (Bai et al., 2008) despite changes in the stomatal
conductance due to CO
The prevalence of the four classes identified above is presented using time series of annual precipitation, water-balance-derived ET and runoff ratio as well as catchment-averaged fractional vegetation cover for the 1984–2010 period. Three constitutive relationships are established for every catchment at an annual scale between (1) precipitation, (2) runoff ratio and (3) ET versus catchment-averaged fractional vegetation cover. Catchment-scale transpiration data are not available for this classification. According to these relationships and Spearman rank correlations, catchments with non-stationary hydrologic response are grouped in three classes (A1, B1 and B2; Fig. 1). None of the catchments with non-stationary hydrologic response presented a relationship proposed for class A2 catchments. Figure 7 shows Spearman rank correlation values for one example catchment in each class.
As presented in Fig. 1 and Table 2, 12 catchments are classified as class
A1. The Spearman rank correlations between annual precipitation and
fractional vegetation cover in class A1 catchments are positive and
typically larger than class B1 and B2 catchments, and in 8 out of 12
catchments the correlation is significant (
Catchment properties and Spearman rank correlations (
While data on catchment-scale nutrient availability are not available, general ET–fractional vegetation cover relationships in group (B) catchments can be further explained by annual precipitation–temperature relationships. In wetter years, despite lower vegetation cover, ET will likely increase due to higher water availability in warmer years in B1 catchments (positive precipitation–temperature correlations) (Table S3). In the B2 class with negative precipitation–temperature relationships, cooler temperatures and light limitation decline ET.
Groupings of all A1, B1 and B2 catchments illustrate significant correlations
for all three constitutive relationships of Fig. 7 (
Box plots of Nash–Sutcliffe efficiency (NSE) values calculated between the regime curves of pre-drought and drought periods in catchments with non-stationary (20 catchments) and stationary (146 catchments) hydrologic response, respectively. Changes in daily precipitation and runoff and monthly fractional vegetation cover were larger in catchments with non-stationary hydrologic response.
According to our analysis, catchments with non-stationary hydrologic response present three distinct behaviors as a result of inter-annual variability in catchment water balance and vegetation fractional cover. In the following, we discuss whether the proposed catchment classification is consistent once other measures or data are used.
To explore whether HRS catchments have undergone similar changes during the period of analysis, regime curves based on daily runoff, precipitation and monthly fractional vegetation cover data for each catchment are developed using data from pre-drought (1984–1996) and drought (1997–2009) periods (Coopersmith et al., 2014). Regime curves are obtained by averaging daily values of precipitation or runoff for a given day over the length of the data. As daily fractional vegetation cover data are not available, monthly values are used to develop the regime curves. To summarize the differences between the regime curves for the pre-drought and drought periods, the Nash–Sutcliffe efficiency (NSE) criterion is calculated. As can be seen in Fig. 8, differences in daily precipitation and runoff and monthly fractional vegetation cover regime curves are much higher (indicated by negative NSE) in catchments with non-stationary hydrologic response than the catchments that do not exhibit non-stationary behavior.
While the results of trend analysis are impacted by defining the significance
level, the above analysis indicates that catchments with non-stationary
behavior have undergone larger changes. To further assess the impact of
significance level on the results of the trend analysis, the approach of
Douglas et al. (2000) for computing the field significance of regional trend
tests is implemented. In this approach, time series of runoff ratio for every
catchment are resampled 10 000 times using the bootstrap approach. In the
next step, Kendall's
Positive precipitation–fractional vegetation cover relationships in class A1
catchments is consistent with positive normalized fractional vegetation
cover sensitivities of individual catchments to annual precipitation (Figs. 4b
and S2) and indicate that water availability primarily controls
fractional vegetation cover increase in A1 catchments. A positive Spearman
rank correlation between the coefficient of variation (CV) of annual
fractional vegetation cover and CV of annual precipitation (
In group (B) catchments, negative normalized sensitivities of fractional
vegetation cover to precipitation exist (Fig. 4b). This pattern is followed
by a negative correlation between the CVs of these two factors across all
group (B) catchments (
Our classification framework suggests that group (A) catchments are more
sensitive to increases in CO
To assess whether observed precipitation–productivity relationships are the
artefacts of remote sensing data, two independent remote sensing vegetation
products are used: vegetation optical depth (VOD) and enhanced vegetation
index (EVI) (Huete et al., 2002, 2006). A global long-term
(1988–2010) annual VOD dataset from passive
microwave satellites with 0.25
Here, we assumed that changes in catchment storage at the annual scale are zero to compute annual water balance ET. However, this assumption is likely not correct in all years. Using AWAP actual annual ET similar relationships between fractional vegetation cover and annual ET are obtained, except in three catchments in class B2 (Table 2). AWAP ET is based on daily transpiration and soil evaporation values obtained from the WaterDyn model that simulates terrestrial water balance across Australia at 5 km resolution (Raupach et al., 2009). In addition to inter-annual water storage carryover, inter-annual non-structural carbon storage across years (a wet year can result in greater biomass/leaf area in the following year) can impact precipitation–vegetation relationships.
The period of analysis is limited to 1984–2010 in this study due to availability of AVHRR fractional vegetation cover data for Australia. To assess sensitivity of precipitation–fractional vegetation cover relationships to data length and catchment condition, these relationships are developed for two time periods: 1984–1996 and 1997–2009. It should be noted that 1997–2009 corresponds to the millennium drought in Australia (Chiew et al., 2014). Results indicate similar precipitation–fractional vegetation cover relationships in 1984–2010 in class A1 as well as in group (B) with a few exceptions (Fig. S2). Despite these exceptions, the drier conditions of 1997–2009 resulted in higher mean fractional vegetation covers in group (B) compared to the 1984–1996 period consistent with the classification framework. Results suggest that the record length is important in catchments where productivity is limited by resources besides water availability.
We used precipitation–fractional vegetation cover relationships for first-order groupings of catchment-scale ecohydrologic response in 20 catchments with non-stationary hydrologic response, located in different hydroclimatic regions of Australia. Our results illustrate that fractional vegetation cover is more sensitive to increases in precipitation (stronger Spearman rank correlations) in class A1 catchments (12 catchments). This inference is consistent with the result of meta-analysis of productivity response to precipitation across the globe (Wu et al., 2011). The drawback of using precipitation as the main driver of vegetation productivity is that the impact of confounding variables that covary with precipitation is ignored (Wu et al., 2011). Fractional vegetation cover sensitivity to precipitation and Horton index provided consistent results with our catchment classification framework, except in two catchments. These catchments (408202, 410061) have smaller rank correlation between precipitation and fractional vegetation cover compared to the rest of class A1 catchments. A total of 8 out of 20 catchments with non-stationary hydrologic response present negative precipitation–fractional vegetation cover relationships impacted by nutrient or light availability.
While determining the exact causes of non-stationarity requires detailed
modeling experiments, non-stationarity of runoff ratios could be attributed
to changes in precipitation amount, intensity and seasonality, increases in
air temperature and CO
The datasets used in this research are publicly available from the following
websites. The daily streamflow
values are obtained from the hydrologic reference stations (HRS) website
(
Remotely sensed vegetation products are available for download from the
following websites: (1) Australian monthly fPAR
dataset version 5:
(
This research was funded by the Australian Research Council linkage grant (LP130100072), the Australian Bureau of Meteorology and WaterNSW. We acknowledge the Australian Bureau of Meteorology for providing the hydrologic reference station data supported by the Australian Government through the Water Information Program. We would like to acknowledge Yi Liu for providing the VOD data. Edited by: A. Wei Reviewed by: two anonymous referees