Duration (IFD) - Innovyze Insider Blog
Transcription
Duration (IFD) - Innovyze Insider Blog
New Intensity-FrequencyDuration (IFD) Design Rainfalls Estimates Janice Green Bureau of Meteorology 17 April 2013 Current IFDs – AR&R87 Current IFDs – AR&R87 Current IFDs - AR&R87 • Options for estimating IFDs – Read the six master charts to obtain intensities for basic durations and frequencies; read regional coefficient of skewness from master chart; read short duration factors from appropriate curves; estimate intensities for LPIII distribution for basic ARIs and durations using Diagram 2.2; estimate LPIII intensities for other durations using Diagram 2.1 – Software developed to derive IFD relationships eg AusIFD www.ens.gu.edu.au/eve/research/AusIfd/AusIfdVer2 – Engage HAS at Bureau to provide estimates @ $100 – Computerised Design IFD Rainfall System (CDIRS) on-line www.bom.gov.au/hydro/has/ifd.shtml Current IFDs – CDIRS On-line Current IFDs – CDIRS On-line Current IFDs – CDIRS On-line Current IFDs – CDIRS On-line Current IFDs • Developed by Bureau of Meteorology nearly 30 years ago • Used a database comprising information primarily from the Bureau’s network of daily read and pluviograph stations • Adopted statistical techniques considered appropriate at the time • Focus of the IFDs was the design of structures on relatively large rural catchments and therefore durations of less than five minutes were not considered necessary. Basic approach Rainfall data Establishment of database Quality controlling of data Series of extreme values Frequency Analysis Regionalisation Gridding Annual Maximum Series Partial Duration Series Extraction of L-moments Choice of probability distribution Sub-daily and sub-hourly statistics Index rainfall approach Regions of Influence Explanatory variables Gridding technique Dissemination Outputs Medium AR&R87 IFDs – Adopted Approach Aspect ARR87 Data BoM stations only Record length ~ up to 1983; 7500 daily read > 30 years; 600 pluviographs > 6 years Frequency analysis Annual maximum series; method of moments; Log-Pearson Type III Daily to sub-daily Principal Component Analysis Mapping Subjective (meteorological analysis); ANUSPLIN used to translate hand drawn contours into grid format (CDIRS) Frequency ARIs 1 year to 100 year Duration 5 minute to 72 hour (3 day) Dissemination Maps; HAS; CDIRS on-line Climate change Stationary climate assumed; climatic trends negligible effect on IFDs IFD2015 – Adopted Approach Aspect IFD2013 Data Bureau stations plus stations from other data collecting agencies Record length up to 2011; 8074 daily read > 30 years (+2 >20 years); 2280 pluviographs > 8 years (754 Bureau; 1526 Regs) Frequency analysis Annual maximum series; L-moment; GEV Daily to sub-daily Bayesian Generalised Least Squares Regression Mapping ANUSPLIN Frequency 1 EY to 1% AEP Duration 1 minute to 168 hour (7 day) Dissemination New webpage on Bureau’s website Climate change Stationarity of long term stations explicitly assessed and no significant trends identified; therefore using complete period of record Data base • Bureau of Meteorology Australian Data Archive for Meteorology (ADAM ) – Contains 19711 daily read rainfall stations (both open & closed) for period from 1800 to 2011 – Contain 1467 continuous stations – both pluviograph & TBRG • Water Act 2007 identified the Bureau’s new responsibilities including • Water Regulations 2008 provided ready access to rainfall data collected by other organisations • In particular, data from dense continuous rainfall networks operated by urban water utilities and councils: – ~350 daily read rainfall stations – ~2175 continuous rainfall stations collecting and publishing water information Spatial coverage of daily read rainfall stations Spatial coverage of continuous rainfall stations • Data base • Finalisation of data base – Initially December 2010 Data base • Finalisation of data base – Initially December 2010 Data base • Finalisation of data base – Initially December 2010 – Updated to December 2011 Data base • Finalisation of data base – Initially December 2010 – Updated to December 2011 Data base • Finalisation of data base – Initially December 2010 – Updated to December 2011 – Updated to March 2012 – ????? Quality Control • Previous work had undertaken QCing on a largely manual basis • Enormous amount of data that needed to be quality controlled – – – – > 20 000 daily read stations - Bureau > 1500 pluviograph stations – Bureau ~350 daily read rainfall stations – Water Regulations ~2175 continuous rainfall stations – Water Regulations • Disparate amount and type of QCing undertaken by data providers • Necessitated automating as much of QCing as possible • However still required manual QCing using Bureau’s Quality Monitoring System (QMS) QCing daily read data • Quality Controlling of daily read data: – Infilling of missing data – Disaggregation of flagged accumulated daily rainfall totals – Detection of suspect data, identification and correction of: • Unflagged accumulated totals • Time shifts – Identification of gross errors - data inconsistent with neighbouring records but not either of the above two categories – Manual correction gross errors identified as having a high probability of being incorrect QCing daily read data using QMS QCing of continuous data • QCing of continuous rainfall data considerably more complicated: – significantly more data due to shorter time step – sparsity of continuous station network means fewer stations with which to compare – small rainfall depths extremely difficult to QC • Needed to reduce amount of data to be QC’d to a manageable amount • Developed approach QC’d the PDS – 5 x number of years of record • Durations of: – 5, 10, 15, 30 minutes – 1, 2, 3, 6, 12 hours – 1, 2, 3 days • Number of PDS values to be QC’d > 1,000,000 QCing continuous data • Issues with Continuous Rainfall Data – Time shifts of clock - DINES – Missed pulses – TBRG • QCing procedure compared values to: – AWAP (Australian Water Availability Product) gridded daily rainfall data – Co-located or nearby daily read stations – AWS (Automatic Weather Stations) – Synoptic stations Trialling of frequency distributions • Frequency analysis – Previously adopted Log-Pearson Type III fitted by method of moments – Used 58 Bureau long-term continuous rainfall stations to trial a range of distributions • From each of the 58 continuous rainfall stations extracted both the AMS and the PDS Trialling of frequency distributions • AMS & PDS extracted for durations of : – 6, 12, 18, and 30 minutes – 1, 2, 3, 6, and 12 hours • Calculated L-moments and fit five distributions: – – – – – • Generalised Logistic (GLO) Generalised Extreme Value (GEV) Generalised Normal (GNO) Pearson Type III (PE3) Generalised Pareto (GPA). Assessed goodness of fit using Hosking and Wallis (1997) goodness of fit measure ZDist Trialling of frequency distributions • AMS – the GEV gave the most acceptable fit for all durations except 3 and 12 hours – however, with the exception of the GPA, the other distributions also showed acceptable fits • PDS – the GPA gave the most acceptable fit for all durations Extraction of L-moments • L-moments used to summarise statistical properties of AMS and PDS – Index rainfall (mean) – L-skewness – L-CV • L-moments expected to be more robust against large outliers in the data, particularly for the higher order moments. • To reduce uncertainty in the parameter estimates, minimum station record lengths have been adopted – 30 years for daily rainfall stations and – 9 years for sub daily rainfall stations. Estimation of sub-daily rainfalls • Shift in focus to urban design on small catchments necessitating the provision of IFD estimates for durations as short as one minute • Far fewer continuous rainfall stations than daily read rainfall stations Spatial coverage of daily read rainfall stations Spatial coverage of continuous rainfall stations • Derivation of short duration IFDs • Need a method to improve spatial coverage of sub-daily data • Most commonly done using information from daily stations – Statistics of sub-daily data are inferred from those of daily data • Techniques adopted include: – Factoring of the 24 hour IFDs – Principal component analysis (PCA) – Partial least squares regression (PLSR) Derivation of short duration IFDs • However, major weakness of the previously adopted approaches is their inability to account for: – Variation in record lengths from site to site – Inter-station correlation • An approach that avoids these problems is Bayesian Generalised Least Squares Regression (BGLSR) Approach to be adopted • Statistics to be derived (predictands) are: – Index rainfall (mean) – L-skewness – L-CV • Predictors to be used are: – – – – – – – Location (latitude & longitude) Elevation Slope Aspect Distance from coast Mean annual rainfall Index rainfall, L-skewness & L-CV at 24, 48 & 72 hours Regionalisation • Regionalisation recognises for stations with short records – considerable uncertainty when estimating the parameters of probability distributions and – short records can bias estimates of rainfall statistics • Overcome by combining information from multiple rainfall stations – more accurate estimates of the probability distribution parameters can be made Regionalisation • Substitute space for time REGIONALISATION Regionalisation • Index rainfall approach adopted to do this (Hosking & Wallis) • Station point estimates have been regionalised using a Region of Influence Approach (ROI). • Trialled various approaches => ROIs defined as circle which is expanded until it includes 500 station years of record • Circular ROIs defined with distance defined in three dimensions – Latitude – Longitude – Elevation Region of influence Region of influence Region of influence Gridding • Regionalisation gave estimates of GEV parameters at all station locations – Combined with the mean of the AMS (index) at that site to estimate rainfall quantiles for any required exceedance probability. • However IFD estimates required across Australia, not just at station locations. • Results of the analyses needed to be extended in some way to ungauged locations. Gridding • Translation from point to gridded rainfall estimates carried out with thin plate smoothing splines implemented using ANUSPLIN. • ANUSPLIN (Hutchinson 2007) was chosen to grid the GEV parameters so that IFD estimates are available for any point in Australia. • GEV parameters are being gridded in ANUSPLIN, as: – earlier testing showed little difference in quantile estimates if point parameter or point rainfall depths gridded. – Gridding rainfall parameters gives more flexibility in the choice of exceedance probabilities that can be extracted and – requires fewer grids to be processed in ANUSPLIN. • Appropriate level of smoothing chosen through generalised cross validation by minimising the predictive error of the fitted surface Interpolation – point to grid SHAPE GRID SCALE GRID MEAN GRID Gridding Index Alpha Kappa Interpolation – point to grid SHAPE GRID SCALE GRID MEAN GRID Y = 1 in 100 AEP Y = 1 in 50 AEP Y = 1 in 20 AEP Y = 1 in 10 AEP Y = 1 in 5AEP Y = 1 in 2 AEP ANUSPLIN Output Example Outputs • Revised IFDs will be provided as depths in millimetres (not intensities) • Revised IFDs will be provided for standard durations of: – 1, 2, 3, 4, 5, 10, 15, 30 minutes – 1, 2, 3, 6, 12 hours – 1, 2, 3, 4, 5, 6, 7 days Outputs • Revised IFDs will be provided for standard EY and AEPs of: – 1EY (1 Exceedance Per Year) – 50%, 20%, 10%, 5%, 2%, 1% AEP – New AR&R probability terminology Revised IFDs • Revised IFDs will be disseminated – In electronic form – Via new web page accessed from Bureau of Meteorology website • Revised IFDs will be released 2 Phases Phase 1 • Phase 1 – Revised IFDs for a point – Functionality of Phase 1 web pages => same as current CDIRS web page • Except raw data from 6 master charts, skew & short duration maps – CDIRS and IFD 2013 run in parallel for ~ 18 months Phase 1 IFD Web page Phase 1 IFD Web page Phase 2 • Phase 2 - Enhancements – Multiple locations – Dynamic map – Areal reduction factors – Temporal patterns – Areal IFDs Phase 2 • Phase 2 - Enhancements – – – – – Seasonal IFDs Sub-annual IFDs eg 2EY, 12 EY Uncertainty estimates Advice provided for IFDs < 1 minute and > 7 days Rainfall frequency curve to 0.05% AEP (incorporating CRCFORGE estimates) – Climate change adjustments to IFDs More information…. Leave your business card Janice Green (02) 6232 3558 [email protected] or [email protected]