Building Knowledge for a Changing Climate Specialist rainfall scenarios and software package Chris Kilsby Ahmad Moaven-Hashemi Hayley Fowler Andrew Smith Aidan Burton Michael Murray University of Newcastle School of Civil Engineering and Geosciences Paul Cowpertwait University of Massey
Overview Requirements for rainfall scenarios The RainClim approach - NSRP model - fitting to observed data and extremes - disaggregator - fitting to future climate data The RainClim package 2
Requirements for rainfall series Time resolution hours and minutes, not days Space resolution Specific to sites, not grid boxes Properties realistic amounts, intensities, extremes seasonality long time series multiple events 3
Approach Stochastic rainfall modelling To achieve downscaling in time and space To generate long series To interface with CRU weather generator Building on: UKCIP2 scenarios and UKMO climatology Consistency with the FEH extremes 15 years development at Newcastle The StormPac approach 4
Rainfall modelling - NSRP The NSRP model A stochastic rainfall modelling system Neyman-Scott Rectangular Pulses Can generate arbitrarily long series (e.g. 1 years) of rainfall Applied to historic, control and future climates Reproduces key statistical properties of rainfall series, e.g. mean, variance, dry hours, 2, 5, 1, 25 year annual maxima); Time resolution of 1-day or 1-hour 5
Rainfall modelling - NSRP Storm origins arrive in a Poisson process Each origin generates a random number of rain cells A rectangular pulse is associated with each rain cell intensity time time time The total rainfall at any time is the sum of all active rain cells total intensity time 6
Fitting NSRP to observed data Historic case Set up the GNSRP model for representative sites in UK Parameterise to match observed 1961-199 rainfall statistics, different parameter set for each calendar month: (a) Fit using mean, variance, proportion dry hours, skew etc (b) Validate using return period rainfalls - e.g. 2,5,1 or 25 year events using observed data and FEH DDF model 7
Observed data Hourly rainfall data 17 sites in 9 regions 8
Observed data Daily rainfall data UK Met Office 5 km grid of daily data 1958-22 Interpolated from gauges + 24 daily sites records 9
Fitting to observed data Some trade-off needed between matching extremes and mean:pd:var Good matching of mean, PD, variance, skewness (3 rd moment) Good reproduction of extreme values (annual maxima) using skewness to fit Good fits at both daily and hourly levels 1
Fitting to observed data f(x) f(x) x x mean variance Definition sketches for moments f(x) f(x) - mean, variance and skew x skew extremes 11
Fitting to observed data Mean Variance 3 Obs Sim Mean Depth (mm/day). 2 1 Jan Feb Mar Apr May Jun Jul Aug.2 Sep Oct Nov Dec Variance (mm 2 /hour).6 Obs.5 Sim.4.3 Variance Ringway.1 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 12
Fitting to observed data Skewness Extremes 7 6 Obs Sim Skew Skewness (daily) 5 4 3 2 1 2 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Ringway. Depth (mm/hour) 6 4 Obs Sim FEH 2 5 1-2 2 4 6 8 5 1 Standardised Gumbel Variate 1, T 13
Fitting to observed data.6.5 Variance - hourly Obs Variance Sim Skewness hourly Variance (mm 2 /hour).4.3.2.1 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Skewness (hourly) 3 25 2 15 1 Obs Sim Skew Elmdon 5 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 14
Fitting to observed data. Depth (mm/hour) 6 4 2 Obs Sim FEH Extremes - hourly 2 5 1-2 2 4 6 8 4 Standardised Gumbel Variate 5 1 Elmdon. 1, Depth (mm/day) 1 T 8 6 2 Obs Sim FEH Extremes daily 2 5 1-2 2 4 6 8 5 1 Standardised Gumbel Variate 1, T 15
Sub-hourly rainfall Requirement for 5, 1, 15-minute rainfall for urban drainage and roof drainage Approach is to disaggregate 1-hour data Use a separate stochastic Poisson cluster model Conserves hourly amounts (microcanonical) unlike other cascade or fractal models Previous models (e.g. Ormsbee method used in StormPac) under-estimate intensities Calibrated on observed data; validated on 3 year Farnborough data set. 16
Sub-hourly rainfall Depth (mm/h) 6 5 4 3 2 1 Observed Synthetic 6 12 18 24 Time (hour) Comparison between the observed hourly rainfall and the 5-minute disaggregated data. 17
Sub-hourly rainfall Farnborough 5min rainfall Frequency (in 3 years) 1 1 1 1 1 1 1 1.1 Observed Disaggregated 5 1 15 Rainfall depth (mm) Validation Farnborough 5min data aggregated to hourly Then disaggregated to 5min. Comparison with observed 3 year record shows good agreement. 18
Future climate Method perturb the observed rainfall statistics (m, pd, var, skew) change factors derived from climate model Issues HadRM3 rainfall statistics not always realistic extremes are sensitive to changes in skew hourly statistics must be derived from daily for future cases, and are not well conditioned 19
Future climate change factors 2
Future climate Fitting to factored mean, var etc 35 3 Obs Sim Skew Skewness (hourly) 25 2 15 1 5 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Skewness (hourly) 45 4 35 3 25 2 15 Fu_28 Sim Skew 1 5 Elmdon Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 21
Future climate Fitting to factored mean, var etc. 1 8 Present Fu_28 Fu_25 Daily Depth (mm/day) 6 4 2. 6 Present Fu_28 Fu_25 Hourly 2 5 1 5 1 1, -2 2 4 6 8 Standardised Gumbel Variate T Depth (mm/hour) 4 2 Elmdon 2 5 1 5 1 1, -2 2 4 6 8 T Standardised Gumbel Variate 22
RainClim Package developed for generating rainfall series, combining NSRP and disaggregator in a user interface RainClim v1. available for download: 17 sites around UK (+scale factors) Hourly and 5-minute series Graphing of time series and means 4 UKCIP2 future scenarios and 3 time-slices ( Present climate only in v1. ) Later version: Whole UK (5km grid) 23
End More tomorrow on: how to use RainClim future climate information (skew) how reliable is it, and how can we use it in RainClim That s all for now! 24