1Trend included in the GAMM model.2Backtransformed trend which is interpreted as a
percentage of the flow. 3no trend in the model,
Mann-Kendall trend analysis on the residuals. The Mann-Kendall tau is
similar to a correlation coefficient and varies between -1 to +1
Figure Legends
Figure 1 Location of hydrologic reference stations used in the study
(http://www.bom.gov.au/water/hrs/)
Figure 2 Flowchart of the different approaches and comparisons of trends
and significance used in the methodology
Figure 3 Annual rainfall runoff relationships for all the catchments
based on the available 41 years of streamflow data and associated
gridded rainfall data.
Figure 4 Rainfall elasticities calculated from the original streamflow
and gridded rainfall data using a non-parametric method (Chiew, 2006)
(red triangles). Also plotted are the rainfall elasticities calculated
from the streamflow simulations of the two different rainfall runoff
models (SimHyd and GR4J) using the gridded rainfall data. As the model
calibrations were replicated 10 times to estimate parameter
equifinality, the elasticities from the rainfall-runoff models are
actually distributions. This equifinality is in most cases very small
and the boxplots show up as single horizontal gray lines. Outliers from
the boxplots are shown as light gray dots, median values of the boxplots
are shown as blue points.
Figure 5 Boxplots and median values (black dots) for goodness of fit of
the GR4J and SimHyd calibrations on the 1970 - 2010 daily data using the
gridded rainfall data. The goodness of fit is indicated by the Nash
Sutcliffe Efficiency, which is equal to 1 for a perfect fit, and 0 for a
fit equal to the mean daily streamflow. Because most of the catchments,
and particularly the GR4J model, had very small equifinality
(variability between repeated calibrations), the boxplots are only
visible as horizontal lines.
Figure 6 Mann Kendall trends (tau) for deseasonalised streamflow for the
different catchments, against the distribution of slope results of a
bootstrap analysis. Black points that are located outside the bootstrap
distribution can be confidently seen as highly significant and not
related to possible trends in random data.
Figure 7 Mann-Kendall tau values (Dot points) and tau results of the
bootstrapped residual trend analysis (distribution histograms) for the
residuals of the GAMM analysis.
Figure 8 Analysis of residual trends between modelled and observed data
for the models based on the 1970 – 2010 data. Red distribution boxes
are significant (p value < 0.05), while blue boxes are not
significant. The actual p-values can be found in the supplementary
material (document 6A). Boxplots represent the 10 replications of the
calibration. As in Figure 5, the variation between the calibration
replications is very small, especially for GR4J, in some of the
catchments resulting in the boxplots plotting as straight lines.