Quantifying Intermittent Streamflow with Low-cost Sensors and Fully Integrated Hydrological Modelling of Upstream Runoff

Abstract

Water policy, planning and design problems can be challenging. They are often nested across multiple scales such that downstream issues cannot be addressed without detailed understanding of upstream processes. Complexity of catchment processes, their interdependence and their feedback mechanisms represent significant catchment heterogeneity that is not easily represented in hydrological models. As a result, many problems are bounded or idealized to focus on a subset of processes restricted to a specific scale. For instance, spatially lumped hydrological models—commonly used for policy, planning and design decisions—are often calibrated to gauging data at a catchment outlet, with limited regard for the hillslope processes leading to runoff generation. Traditional approaches to streamflow monitoring are limited by relatively high costs of gathering observations leading to sparse geographical coverage. This can be problematic when attempting to understand the non-linear and complex runoff behaviour present on hillslopes and reaches throughout an entire river network. The representation of runoff in hydrological models can be critical for supporting a range of spatially distributed problems. Many practical water management problems focus on the aggregation of processes to an outlet which can be inadequate for some questions that are nested across scales, such as management decisions relating to upstream land use, small farm dams and environmental flows. In the absence of streamflow monitoring at local scales, traditional approaches are less likely to be scalable and pose challenges in effectively representing surface and subsurface runoff generation at finer scales. Increasingly sophisticated low-cost and low-maintenance sensing technologies are an accessible means of addressing data gaps in the spatial coverage of streamflow. Environmental sensor technologies are continually being developed with miniaturization, wireless communication and reduced costs, enabling automated electronic data loggers with reliable high frequency measurements. New initiatives in hydrological and environmental monitoring provide opportunities for high-density and widespread environmental data collection providing opportunities to better understand water management questions which span multiple scales. This study combined process modelling with field data collected from low-cost distributed sensors to improve the representation of local scale flow processes. While process modelling can, in part, address some deficiencies, they are limited by known scaling issues of theoretical understanding and data availability of parameters that cannot be adequately measured. To compensate for these limitations, an alternative hydrological modelling approach for calibration was presented. The approach was supported with inexpensive data on the presence of water in the streambed, providing additional information on hillslope intermittency. A small 10km2 South Australian catchment, located in the Mount Lofty Ranges, was selected and instrumented across twelve sites with paired in-stream and on-bank temperature data loggers and pressure transducers required to evaluate results. A two-state hidden Markov model was applied to temperature data to classify whether the stream was ‘wet’ (flowing) or ‘dry’ (not flowing) for a given day. The accuracy of classifications was between 89% to 99% during calibration and 82% to 97% during evaluation of algorithm performance. The binary ‘wet’-‘dry’ classifications were used to calculate a number of intermittency signatures for each site (e.g. number of zero flow days, number of zero flow periods and average duration of zero flow periods) with the results demonstrating a high degree of heterogeneity of flow permanence within the small catchment area. A physically based hydrological model, HydroGeoSphere, was calibrated exclusively to discharge at the outlet to represent four conceptual models of runoff generation. The four competing conceptualisations were: (1) saturation excess dominated, (2) saturation excess and groundwater dominated, (3) groundwater dominated and (4) groundwater dominated but containing 17% infiltration excess. The conceptual models dominated by groundwater discharge showed a 20% increase in low flow days directly below at point of interception compared to upstream. This highlighted that conceptual assumptions about runoff generation mechanisms in intermittent river systems have significant implications on locally dependent water management problems. The four conceptual models were evaluated with classified ‘wet’-‘dry’ binary data showing that no single candidate calibration performed consistently well at all upstream sites. This result demonstrating that the high heterogeneity in headwaters means that catchment-average process simulations were not able to capture localized variability. Distributed data of flow intermittency in headwater catchments was able to improve process-based hydrological model calibration with eight out of nine sites showed significant improvement in performance, with only a small deterioration to the outlet. Headwaters, which are typically considered to include first to third order streams, are important for understanding intra-catchment fluxes and how these fluxes accumulate out of the catchment. With improved model performance there can be many benefits of getting the internal processes right, such as: simulating conditions that are outside a calibration period or infilling and completing data sets. As sensing and communications technologies continue to improve, there will be increasing opportunities to use information sources such as local-scale intermittency to supplement reliable streamflow records for representing hydrological processes across scales for more accurate water accounting and improve planning of water allocations for consumptive and environmental water needs.