Assessing the sensitivity of hydro-climatological change detection methods to model uncertainty and bias. (December 2019)

• Record Type: Journal Article
• Authors: Jiang, Ze; Sharma, Ashish; Johnson, Fiona
• Is Part Of: Advances in water resources. Volume 134(2019)
• Publication Date: 2019-12
• Subjects: Detection & attribution -- Sensitivity -- Model uncertainty and bias -- Soil moisture; Hydrology -- Periodicals; Hydrodynamics -- Periodicals; Hydraulic engineering -- Periodicals; 551.48
• DOI: 10.1016/j.advwatres.2019.103430
• Languages: English
• ISSNs: 0309-1708

Abstract

Highlights: This study assesses the sensitivity of a routinely adopted detection methodology to model uncertainty and bias within a hydro-climatological context. The extent of uncertainty (as measured by the variance) and the consistency of trend play a critical role in changing the detection outcome. When trends are consistent, variance modulation through decadal averaging helps improve the D&A outcome. With inconsistent trends between simulations and observations, modulating variance leads to high uncertainty in D&A analysis. Abstract: Detection of systematic changes in the climate system resulting from anthropogenic forcing is a critical area of research. Detection and attribution of hydro-climatological change has been limited by model uncertainty and bias as well as the poor spatial-temporal coverage of observational data. This study assesses a routinely adopted detection methodology and its sensitivity to model uncertainty and bias within a hydro-climatological context. Using a synthetic case study, we establish the sensitivity of detection approaches to the magnitude and consistency of trend and variance along with the length of data available. It is found that the extent of uncertainty (as measured by the variance) plays a critical role in changing the detection outcome. Another important factor is the consistency of trend between simulations and observations. A case study of soil moisture in select locations within Australia shows that averaging over multiple years (e.g., five years to a decade) improves the detection of the climate change signal as long as consistency in the trends exists. Our results also demonstrate that there are substantial differences in simulated trends across climate models. Therefore, even though ensemble averaging is effective in modulating variance, it has the risk of canceling out the signal over models with markedly different responses.