Results
4.1 Regression Analysis in the Calibration
Periods
Rainfall-runoff relationships in the calibration and testing periods
were examined before developing the regression relationships. For
example, in severely burnt catchments such as Flowerdale, Taggerty and
Koornalla (with burnt percentages of 96.6%, 89.7% and 83.1%
respectively), there are obvious changes in the rainfall-runoff
relationship in the testing period compared with that in the calibration
period (see Figure 3). Similar results were obtained in the majority of
the selected catchments and the extent of the relationship change is
related to the burnt percentage as well as the burnt severity. The
results illustrate that there is more streamflow after the vegetation
loss caused by bushfire, which is consistent with our knowledge of how
bushfire affects streamflow in a catchment. This provides confidence in
our further analysis of bushfire impacts on streamflow.
Then, the regression relationships between rainfall and streamflow were
constructed before quantifying the bushfire and climate change impacts
on mean annual streamflow. The rainfall-streamflow relationship varies
largely from catchment to catchment in different time periods. In this
study, relationship between rainfall and streamflow in almost every
catchment impacted by bushfire can be fitted by linear regression model
in both calibration period and testing period (see Figure 3) and the
evaluation results of the regression model are listed in Table 2.
<Figure 3 Here Please>
Figure 3 shows the regression relationships obtained for each catchment
in the calibration and test periods and Table 2 summarizes the
statistical metrics of the linear regressions. There exist good linear
relationships between annual rainfall and annual streamflow in most
catchments in the calibration periods (Figure 3). The coefficient of
determination (\(R^{2}\)) result varying from 0.56 to 0.85 indicates
annual rainfall explains 56% - 85% variation of annual streamflow in
the calibration periods (Table 2). The modified coefficient of
efficiency (\(E_{1}\)) ranges from 0.26-0.65 and the modified index of
agreement (\(d_{1}\)) ranges from 0.62 to 0.82. The mean absolute error
(MAE) is in the range of 16.3-101.8. These results indicate
that the linear regressions are satisfactory in most of burnt
catchments, and can be used for estimating the streamflow under a given
precipitation during the testing period (i.e. the post-bushfire period).
<Table 2 Here Please>
4.2 Bushfire and Climate Variability Impacts on Mean Annual
Streamflow
To evaluate the total mean annual streamflow changes in the testing
period, the total annual streamflow change (ΔQtot)
were calculated based on Equation 10 and the results are summarized in
Table 3. The total mean annual streamflow of all the catchments
increases in the testing period (post-bushfire period), compared with
the calibration period (pre-bushfire period) except for Jamieson (with
decline of 22.3 mm). The increase in mean annual streamflow ranges from
5.9 to 141.4 mm. 12 out of the 15 selected catchments have increasing
average runoff ratio in the testing period, which varies from 0.002 to
0.106. In contrast, the average runoff ratio of Running Creek and
Tallarook decreases by 0.009 and 0.03 respectively while Frenchman Creek
Junction’s average runoff ratio remains stable.
To quantify the contribution of bushfire effects to the mean annual
streamflow change, streamflow changes due to vegetation change caused by
bushfire (ΔQveg) were determined from time trend
analysis method (see 3.1). As shown in Table 3, the calculated
streamflow increase due to vegetation change caused by bushfire
(ΔQveg) ranges from 1.6 (Frenchman Creek Junction) to
125.9 mm (Traralgon South) or -45% (Jamieson) to 98% (Murrindindi
above) of the observed total streamflow change
(ΔQtot). The results indicate that the bushfire caused
streamflow increases during the testing period in each burnt catchment.
Changes in rainfall and potential evaporation (PET) also contribute to
streamflow variations. Thus, the streamflow changes due to climate
variability (ΔQclim) were determined from
sensitivity-based method (see 3.3). The choice of w value in
sensitivity-based method was based on climate condition and agreement in
the calibration period between measured and estimated annual runoff
ratio (Zhang et al., 2011). The w value remained constant in
testing period to ensure equation 11 only represents the effect of
climate variability. Table 3 shows that climate variability in the
testing period increases annual streamflow in the majority of catchments
such as Running Creek and Rosewhite, but decreases annual streamflow in
the rest (e.g., Jamieson and Koornalla). The proportion of streamflow
change due to climate variability in total streamflow change ranges from
-7% to 157%.
<Table 3 Here Please>
In this study, percentage changes in mean annual streamflow caused by
bushfire and climate variability were calculated independently. When the
sum of the two percentage changes approaches 100%, it means that the
independent simulated streamflow changes are close to the actual
streamflow changes. Figure 4 summarizes the sum of the two proportions.
For most catchments except for Frenchman Creek Junction, the sum is
close to 100%, which means the sum of streamflow change due to bushfire
(ΔQveg) and climate variability
(ΔQclim) approaches the total streamflow change
(ΔQtot). This indicates that the two independent
methods are reliable for most of the bushfire impacted catchments. The
reason why Frenchman Creek Junction only has 28% total streamflow
change is because the absolute streamflow change before and after
bushfire is only 5.9 mm. Such a small change may amplify the error of
the actual total change of streamflow and the sum of estimated
streamflow change from bushfire and climate variability. As shown in
Figure 4, streamflow change between the two periods (pre- and post-
bushfire periods) is mainly contributed by bushfire impact in 11
catchments, but by climate variability in the other 3. It is worth
noting that among all the burnt catchments, only Jamieson has a decrease
in streamflow after bushfire. This occurs because the streamflow change
caused by bushfire increased but the climate variability induced a
greater decrease in streamflow, which is consistent with the fact that
Jamieson only has 36.3% burnt area with less burnt severity (see Figure
2) and the rainfall in the testing period is lower compared with the
calibration period (see Figure 1).
<Figure 4 Here Please>
To further investigate if the bushfire impacts on streamflow are related
to the burnt area, Figure 5 shows the relationship between burnt
percentage (fire scars area within a catchment as a proportion of the
total area of a catchment) and percentage of mean annual streamflow
increase due to bushfire. It is shown that the percentage of mean annual
streamflow increase due to bushfire is strongly related to the
percentage area burnt. The linear relationship between the two indicate
the burnt percentage can explain 68% of the percentage of mean annual
streamflow increase due to bushfire. This result can be used to estimate
or predict the annual streamflow changes due to bushfire if the burnt
percentage for a small to medium sized catchment is available. It is
worth noting that the relationship is suitable at annual time scale and
may be not appropriate at sub-annual time steps.
<Figure 5 Here Please>