Skip to main content
ARS Home » Research » Publications at this Location » Publication #307754
Title: Validation of non-stationary precipitation series for site-specific impact assessment: Comparison of two statistical downscaling techniques

Author
item MULLAN, DONAL - Liverpool University
item BRISSETTE, FRANCOIS - University Of Quebec
item Zhang, Xunchang

Submitted to: Climate Dynamics
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/20/2015
Publication Date: 4/26/2015
Citation: Mullan, D., Brissette, F., Zhang, X.J. 2015. Validation of non-stationary precipitation series for site-specific impact assessment: Comparison of two statistical downscaling techniques. Climate Dynamics. doi:10.10074/s00382-015-2626-x.

Interpretive Summary: The synthetic generation of realistic future precipitation scenarios is crucial for assessing their impacts on a range of environmental and socio-economic impact sectors. A scale mismatch exists, however, between the coarse spatial resolution at which global climate models (GCMs) output future climate scenarios, and the finer spatial scale at which impact modellers require projections. Statistical downscaling (SD) methods have become a popular, low-cost and accessible means of addressing this scale mismatch, with various different SD techniques used for a wide range of applications across the world. The Generator for Point Climate Change (GPCC) model and the Statistical Downscaling Model (SDSM) were compared in terms of their ability to generate precipitation series under non-stationary conditions in a wide range of climatic zones. Ten stations were selected to reflect a broad range of global climates, with observed precipitation series split into a calibration and validation period. The mean, maximum and a selection of distribution statistics for daily, monthly and annual precipitation were compared between the models and the observed series for the validation period. Results indicate that both methods generally produce precipitation series that compare closely when precipitation amounts are lumped to monthly and annual resolutions, whilst mean daily and annual maximum daily precipitation is not as well simulated. The distribution of precipitation series is generally well simulated at the same temporal resolutions, but the GPCC method tends to overestimate higher precipitation amounts, whilst SDSM tends to underestimate these. The standard deviation and skewness statistics are found to have a key role to play in controlling these precipitation extremes. These results indicate that GPCC is more likely to overestimate the effects of precipitation on a given impact sector, whilst SDSM is likely to underestimate the effects. Despite this, the results reveal that two contrasting SD methods can generate daily precipitation series that generally closely mirror observed series for a wide range of non-stationary climates, but the precipitation extremes can be considerably different from the observed period for both means and distributions. This illustrates the importance of users performing a thorough validation in order to determine the influence of simulated precipitation on their impact sector. This work will provide useful information to climate and hydrological modellers who are interested in simulating the potential impacts of climate changes on soil and water conservation and crop production.

Technical Abstract: The generation of realistic future precipitation scenarios is crucial for assessing their impacts on a range of environmental and socio-economic impact sectors. A scale mismatch exists, however, between the coarse spatial resolution at which global climate models (GCMs) output future climate scenarios, and the finer spatial scale at which impact modellers require projections. Statistical downscaling (SD) methods have become a popular, low-cost and accessible means of addressing this scale mismatch, with various different SD techniques used for a wide range of applications across the world. The Generator for Point Climate Change (GPCC) model involves first spatially downscaling GCM/RCM (regional climate model) monthly precipitation projections from a grid box to a target station using transfer functions, and then temporally disaggregating monthly projections into daily series using the CLIGEN weather generator. The Statistical Downscaling Model (SDSM), in contrast, develops transfer functions between observed precipitation for a target station and large-scale atmospheric variables from a grid box at a daily resolution, and a subsequent forcing of these transfer functions using the same set of large-scale atmospheric variables as output by GCMs for the future. This paper compares these two contrasting SD methods in terms of their ability to generate precipitation series under non-stationary conditions in a wide range of climatic zones. Ten stations were selected to reflect a broad range of global climates, with observed precipitation series split into a calibration and validation period in a manner which maximised the difference of mean annual precipitation between the two records. The mean, maximum and a selection of distribution statistics for daily, monthly and annual precipitation were compared between the models and the observed series for the validation period. Results indicate that both methods generally produce precipitation series that compare closely when precipitation amounts are lumped to monthly and annual resolutions, whilst mean daily and annual maximum daily precipitation is not as well simulated. The distribution of precipitation series is generally well simulated at the same temporal resolutions, but the GPCC method tends to overestimate higher precipitation amounts, whilst SDSM tends to underestimate these. The standard deviation and skewness statistics are found to have a key role to play in controlling these precipitation extremes. These results indicate that GPCC is more likely to overestimate the effects of precipitation on a given impact sector, whilst SDSM is likely to underestimate the effects. Despite this, the results reveal that two contrasting SD methods can generate daily precipitation series that generally closely mirror observed series for a wide range of non-stationary climates, but the precipitation extremes can be considerably different from the observed period for both means and distributions. This illustrates the importance of users performing a thorough validation in order to determine the influence of simulated precipitation on their impact sector.