Quantifying spatial and temporal patterns of flow intermittency using spatially contiguous runoff data

Abstract

River channel drying caused by intermittent stream flow is a widely-recognized factor shaping stream ecosystems. There is a strong need to quantify the distribution of intermittent streams across catchments to inform management. However, observational gauge networks provide only point estimates of streamflow variation. Increasingly, this limitation is being overcome through the use of spatially contiguous estimates of the terrestrial water-balance, which can also assist in estimating runoff and streamflow at large-spatial scales. Here we proposed an approach to quantifying spatial and temporal variation in monthly flow intermittency throughout river networks in eastern Australia. We aggregated gridded (5x5 km) monthly water-balance data with a hierarchically nested catchment dataset to simulate catchment runoff accumulation throughout river networks from 1900-2016. We also predicted zero flow duration for the entire river network by developing a robust predictive model relating measured zero flow duration (% months) to environmental predictor variables (based 2 on 43 stream gauges). We then combined these datasets by using the predicted zero flow duration from the regression model to determine appropriate 'zero' flow thresholds for the modelled discharge data, which varied spatially across the catchments examined. Finally, based on modelled discharge data and identified actual zero flow thresholds, we derived summary metrics describing flow intermittency across the catchment (mean flow duration and coefficient-of-variation in flow permanence from 1900-2016). We also classified the relative degree of flow intermittency annually to characterise temporal variation in flow intermittency. Results showed that the degree of flow intermittency varied substantially across streams in eastern Australia, ranging from perennial streams flowing permanently (11-12 months) to strongly intermittent streams flowing 4 months or less of year. Results also showed that the temporal extent of flow intermittency varied dramatically inter-annually from 1900-2016, with the proportion of intermittent (weakly and strongly intermittent) streams ranging in length from 3 % to nearly 100% of the river network, but there was no evidence of an increasing trend towards flow intermittency over this period. Our approach to generating spatially explicit and catchment-wide estimates of streamflow intermittency can facilitate improved ecological understanding and management of intermittent streams in Australia and around the world.

Evaluating a landscape-scale daily water balance model to support spatially continuous representation of flow intermittency throughout stream networks

“…Due to the fact that water balance models often over-predict the magnitude of very low flows (Ye et al, 1997), we adopted the same method used in Yu et al (2018) to estimate a threshold of zero flow from the model that correlated with measured cease to flow duration at each gauge. This involved three steps.…”

“…1) We used linear regression to model the cease to flow duration at each gauge as a function of catchment environment variables. The environmental variables were the same as those in Yu et al (2018), and included variables related to climate (annual daily maximum temperature), catchment geology topography (catchment area, catchment average slope, and catchment average elevation), and catchment soil properties (catchment average saturated hydraulic conductivity).…”

“…Spatially contiguous runoff data derived from water balance models provide another potential alternative to quantify spatiotemporal variations in flow intermittency. For example, Yu et al (2018) used runoff simulations obtained from a water balance model (Raupach et al, 2009) to generate spatially explicit and catchment-wide estimates of streamflow intermittency, but only at a relatively coarse monthly time step. Depending on the application, flow simulations at a finer temporal scale (e.g.…”

“…Lastly, we evaluated the effect of time step (daily vs. monthly) on the relative performance of the model in replicating observed patterns of cease to flow periods at reference gauges. A previous study conducted at the monthly time step (Yu et al, 2018) was used to benchmark flow intermittency estimated from the daily model. https://doi.org/10.5194/hess-2020-10 Preprint.…”

“…For the monthlyaggregated AWRA-L outputs, all days in a month had to have zero flow for the flows for that month to be zero. The AWAP model was developed at the Australian continental scale by CSIRO and BoM with a similar model structure to AWRA-L (Raupach et al, 2018), and has been used to quantify flow intermittency in SEQ in our previous research (Yu et al, 2018).…”