Research papersDisaggregated monthly hydrological models can outperform daily models in providing daily flow statistics and extrapolate well to a drying climate
Introduction
Obtaining daily timestep hydrological information is key for many water resource management purposes. This can include understanding hydrologic regimes, flood frequency analysis, providing the input for ecological, social, or economic models, and predicting the consequences of change – such as infrastructure development or a drying climate. A key challenge in understanding and managing hydrologic changes to river systems is how best to model impacts given limitations of existing modelling capability (Saft et al., 2016) and the significant uncertainties involved in predicting climatic changes (Hawkins and Sutton, 2011) and hydrologic responses (Andréassian et al., 2001, Knoben et al., 2020, McMillan et al., 2012)
The current approach to meet this challenge is to model impacts at a daily time-step to explicitly represent rainfall-runoff processes, river operational decisions and responses (such as ecological outcomes that respond to daily flow triggers). This is particularly the case for rainfall-runoff modelling which is dominated by daily timestep models (Knoben et al., 2019), specifically conceptual rainfall-runoff models calibrated to historic records. This dominance is understandable given that certain hydrometeorological data (e.g. precipitation) are commonly available at a daily timestep, and rarely finer. Thus, the daily timestep is usually the first choice for certain modelling purposes, such as exploration of dominant hydrological processes (Khatami et al., 2019) or to replicate historical river operations in models that support water allocation (Perera et al., 2005). However, the costs and benefits of daily conceptual models must be carefully weighed giving due consideration to the following challenges.
Firstly, it is well known that daily rainfall-runoff models are subject to significant trade-offs in performance (Euser et al., 2013, Gharari et al., 2013). For example, models calibrated to replicate floods simulate low-flow periods poorly, and vice versa (Pushpalatha et al., 2012). Similarly, models calibrated in wet periods typically do not perform well in dry periods (Coron et al., 2012), and also are particularly vulnerable to overestimation of flows during extended droughts (Saft et al., 2016). This can be problematic for ecological purposes, where multiple aspects of flow regimes such as flood timing and peak flow combine to support species needs (Poff et al., 1997). Often a lack of favorable trade-off solutions (Fowler et al., 2016) forces modelers to choose which aspect of performance to emphasize at the expense of the others (Li et al., 2012). Because of these trade-offs from daily calibration, we hypothesize that modelling with a monthly timestep paired with a disaggregation that directly uses observed flow patterns, may provide more favorable outcomes across a range of hydrologic metrics related to flow volumes, timing and sequencing. In this way, monthly calibration can focus more on volumetric accuracy whilst extra information is provided by incorporating observed daily flow patterns.
Second, from a practical perspective, more computational effort and model complexity is required to apply daily hydrological models especially when they are combined in integrated water resource models. This is particularly apparent for larger rivers or regional assessments where model inputs cannot be easily represented by area-averaged information and additional complexity is required to accommodate spatially correlated behavior at current and lagged time steps. Such integrated models may also have several sub-components that are heavily parameterized such as water demand or ecological response models, and their structure or parameterization may contribute significantly to output uncertainty (Arnold et al., 2009). Climate uncertainty in future conditions has contributed to the development of scenario-neutral or decision scaling approaches (Brown et al., 2012, Brown and Wilby, 2012), although these methods are not solely suited to climate change analysis. Unfortunately, their computational requirements can quickly become demanding, particularly when assessing complex systems with multiple objectives (Whateley et al., 2016). To overcome this, many modelers use simple statistical meta-models that are parameterized based on a few simulations of the more detailed system model (Poff et al., 2016, Turner et al., 2014, Whateley et al., 2016). However, these simple methods may obscure system dynamics and introduce additional uncertainty into modelling outputs, again, especially for ecological purposes where rates of species decline can depend on initial conditions (Bond et al., 2018, Wang et al., 2018). In modelling the hydrological impacts of climate change, many studies use simplified scaling approaches that provide daily outputs but do not simulate changes to the frequency distribution of daily rainfall (Anandhi et al., 2011). Although there are tradeoffs in computational effort and model ‘realism,’ there is a place for simpler simulation approaches as they more easily allow consideration of epistemic uncertainties (Helgeson et al., 2020).
In this paper, we tackle these challenges by investigating whether alternative approaches to daily modelling of hydrological inputs can provide equivalent information for multiple hydrological metrics and purposes for less computational effort. We developed a method that combines monthly hydrological modelling and a disaggregation approach to provide daily timestep flows as an alternative to direct daily hydrological modelling using conceptual rainfall-runoff models calibrated on historic records. Our emphasis on model performance is on characterizing a streamflow regime rather than the simulation of catchment response to a specific climatic sequence. In other words, “history matching” to a daily observed record is not a requirement, and this allows disaggregation approaches to be applied in a way that replicates daily statistics relevant to the full flow regime. While other studies have looked at the reproducibility of daily flow statistics using synthetic flow models, including disaggregation (You et al., 2014), this approach has not been compared to conceptual daily hydrological models and evaluated for a wide range of hydroclimatic conditions at large scale, nor tested when extrapolated to hydroclimatic periods outside of the range of calibration data.
Section snippets
Methods
We compare the more traditional approach of using daily conceptual rainfall-runoff modelling to generate daily streamflow with an alternative approach that uses a monthly rainfall-runoff model and then disaggregates to daily streamflow based on sampling informed by catchment wetness. The approach is firstly applied to two unregulated systems of differing catchment conditions in the Goulburn River in Victoria, Australia. It is then also tested using a range of unregulated catchments across
Case study for Goulburn River tributaries
Selected model calibration statistics for the daily and monthly rainfall runoff models are given in Table 2. Note that the statistics are calculated on monthly values for the WAPABA model and daily values for the Sacramento model, hence are not directly comparable for a given catchment. PBIAS (or the percentage of model bias) is calculated as proposed by Moriasi et al., (2007). Nonetheless, both the Sacramento and WAPABA models show good performance in replicating historic flows across wet and
Strengths of disaggregation method
The calibration strategy used in hydrological modelling will affect the ability of models to meet certain purposes, particularly for drying climates (Fowler et al., 2018). A key strength of the disaggregation approach is that it avoids some of the trade-offs that are normally required in calibrating daily rainfall-runoff models. For example, when high flows are important a calibration strategy might focus on Nash-Sutcliffe efficiency or mean-squared error as an objective function that weights
Conclusion
It is clear from the detailed case study and wider exploration on catchments across Australia that the disaggregation approach is not only a viable alternative to daily modelling, but that it outperforms the daily model for a large range of purposes. For example, water resource management is assisted by better replication of typical runoff yields and variability. Assessing flood risk is aided through improved simulation of high flows and annual maxima. Ecological modelling and management
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
This study was funded by the Australian Research Council (ARC Linkage Project LP170100598), Australian Commonwealth Government under a Research Training Program Scholarship, and several partner agencies including the Department of Environment, Land, Water and Planning; the Victorian Environmental Water Holder and the Bureau of Meteorology. Avril Horne was funded by Australian Research Council DECRA DE180100550.
References (88)
- et al.
Impact of imperfect rainfall knowledge on the efficiency and the parameters of watershed models
J. Hydrol.
(2001) - et al.
A comparison of multi-site daily rainfall downscaling techniques under Australian conditions
J. Hydrol.
(2011) - et al.
Development of representative indicators of hydrologic alteration
J. Hydrol.
(2009) - et al.
Using spatial interpolation to construct a comprehensive archive of Australian climate data
Environ. Model. Softw.
(2001) Operational estimates of areal evapotranspiration and their significance to the science and practice of hydrology
J. Hydrol.
(1983)- et al.
Computer software tool REALM for sustainable water allocation and management
J. Environ. Manage.
(2005) - et al.
Change-signal impacts in downscaled data and its influence on hydroclimate projections
J. Hydrol.
(2018) - et al.
A review of efficiency criteria suitable for evaluating low-flow simulations
J. Hydrol.
(2012) - et al.
Monthly versus daily water balance models in simulating monthly runoff
J. Hydrol.
(2011) - et al.
The examination of reproducibility in hydro-ecological characteristics by daily synthetic flow models
J. Hydrol.
(2014)