EGU General Assembly 2022
© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

How does the performance of rainfall-runoff models degrade due to multi-annual drought? A large-sample, multi-model study.

Luca Trotter, Margarita Saft, Murray Peel, and Keirnan Fowler
Luca Trotter et al.
  • University of Melbourne, Faculty of Engineering and Information Technology, Infrastructure Engineering, Australia

We studied the effect of a 13-year long dry period on the performance of five conceptual rainfall-runoff models in 155 catchments in the Australian state of Victoria in order to identify (1) which aspects of the flow regime are harder for models to reproduce during and after the drought; and (2) how model performance during this persistent drought compares to that during similarly dry individual years in the historical record. Persistent dry conditions in recent decades have affected hydrological processes and water partitioning in several regions globally; given the increased risk of drought posed by climate change, studying these historical long-lasting droughts and their effects on model reliability can inform climate adaptation strategies in many drought-prone regions worldwide.

The Millennium drought (MD), which affected more than 1×106 km2 of south-eastern Australia between 1997-2009, is one of such events and arguably the most reported on in literature. Research to date identifies significant shifts in catchment-level annual rainfall-runoff relationships in most catchments affected by the MD, many of which have failed to recover several years after the end of the meteorological anomaly. These shifts affect the reliability of models’ projections; however, by focusing on a handful of performance metrics only, currently published research falls short on identifying which specific aspects of model performance are affected and how.

Here, we focused on a wider range of performance metrics to assess models’ ability to represent a variety of aspects of the hydrograph and the flow-duration curve during and after the MD. For objective (1), we used a statistical metric derived from Wilcoxon signed-rank test (known as matched-pairs rank-biserial correlation coefficient) to compare changes in model performance during and after the drought from a pre-drought benchmark across metrics and catchments. For objective (2), we analysed changes in the relationship between annual model performance and annual rainfall using a regression technique.

We observed extensive degradation of model performance during the drought across most of the metrics studied. Overestimation of flow volumes drives the decline, while representation of shape and variability of the hydrograph and the flow-duration curve are more resilient to prolonged drought. This means that volumes’ overestimation is not associated to specific flow regimes, but results from flow declining proportionally throughout the hydrograph, suggesting that multiple catchment processes interact to cause the observed changes across high and low flows as well as through faster and slower routes. We obtained very similar results in the decade after the drought, indicating that model performance, similarly to rainfall-runoff relationships, often does not recover after the dry spell ends. Additionally, regression analysis of annual performance and rainfall showed disproportional decline of model reliability during the multi-year event compared to single dry years before the drought, suggesting that the persistency of the drought is likely responsible for exacerbated performance decline due to accumulation and aggravation of errors over subsequent dry years.

How to cite: Trotter, L., Saft, M., Peel, M., and Fowler, K.: How does the performance of rainfall-runoff models degrade due to multi-annual drought? A large-sample, multi-model study., EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-6866,, 2022.


Display file

Comments on the display

to access the discussion