Royal Observatory of Belgium

Transcription

1 R. Van Malderen 1, E. Poiaux 2, A. Klos 3, O. Bock 4, J. Bogusz 3, B. Chimani 5, M. Elias 6, M. Gruszczynska 3, J. Guijarro 7, S. Zengin Kazancı 8 and T. Ning

2 1. Moivaion and daase generaion 4. Assessmen of he performance of he homogenizaion ools on he synheic daase 5. 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 2

3 COST acion GNSS4SWEC Advanced Global Navigaion Saellie Sysems ropospheric producs for monioring severe weaher evens and climae WG3: Use of GNSS ropospheric producs for climae monioring (e.g. Inegraed Waer Vapour (IWV) = Precipiable Waer (PW)) From differen presenaions a differen GNSS4SWEC workshops, i urned ou ha differen groups were showing resuls from ime series analyses, someimes based on he same. They were dealing/sruggling wih he homogenizaion of heir. common aciviy, on a reference IWV daase. 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 3

4 he reference IWV daase From A. Klos 120 sies worldwide, wih homogeneous reprocessing from IGS repro 1: Inernaional GNSS Service, firs reprocessing screened, oulier removed, and ZTD convered o IWV by O. Bock 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 4

5 use of a reference IWV daase (ERA-inerim) good correlaion beween IGS repro 1 and ERA-inerim model oupu bu: no fully independen From O. Bock ERA-inerim is used o screen he GPS daa. ERA-inerim is used o conver he Zenih Toal Delay from GPS o IWV (surface pressure, weighed mean emperaure). ERA-inerim no homogenous eiher! 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 5

6 use of a reference IWV daase (ERA-inerim) daily monhly From O. Bock We will look for break poins/change poins in he GPS ERA-inerim IWV differences series. 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 6

7 based on he comparison of he epochs idenified by differen ools decided o build synheic wih known insered offses for he assessmen of he performance of he differen ools a sensiiviy analysis of his performance on he properies of he, we synheic of he IWV differences, 1. based on he characerisics of he real GPS ERA-inerim IWV differences: analysis of significan frequencies noise model linear rends gaps 2. based on he characerisics of he offses in he real IWV differences: number ypical ampliudes compuaions were performed wih Maximum Likelihood Esimaion (MLE) in he Hecor sofware (Bos e al., 2013). around 1000 epochs of insrumenal changes, repored in he saions meadaa, were manually checked in he IWV differences (164 confirmed + 57 new ones) also used for 1 s order correcion of he daa 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 7

8 combinaion of auoregressive process of firs order plus whie noise (AR(1)+WN) characerized by: coefficien, fracion and ampliude (AR), ampliude (WN). 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 8

9 combinaion of auoregressive process of firs order plus whie noise (AR(1)+WN) characerized by: coefficien, fracion and ampliude (AR), ampliude (WN). 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 9

10 of IWV differences For every saion (120 saions!), 3 varians of synheic ime series were creaed: 1. easy daase: seasonal signals (annual, semi-annual, er- and quaer-annual, if presen for a paricular saion) + offses + whie noise (WN) 2. less-complicaed daase: same as 1. + auoregressive process of he firs order (noise model = AR(1)+WN) 3. fully-complicaed daase: same as 2. + rend + gaps (up o 20% of missing daa = overshoo!) 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 10

11 of IWV differences For every saion (120 saions!), 3 varians of synheic ime series were creaed: 1. easy daase: seasonal signals (annual, semi-annual, er- and quaer-annual, if presen for a paricular saion) + offses + whie noise (WN) 2. less-complicaed daase: same as 1. + auoregressive process of he firs order (noise model = AR(1)+WN) 3. fully-complicaed daase: same as 2. + rend + gaps (up o 20% of missing daa) 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 11

12 Climaol J. Guijarro HOMOP B. Chimani PMTred T. Ning Nonparameric ess R. Van Malderen 2-sample - es M. Elias Pei es S. Zengin Kazancı Mehod 1 Mehod 2 Mehod 3 Mehod 4 Mehod 5 Mehod 6 Mehod 7 Operaor M. Elias R. Van Malderen Mehod / SW 2-sample - es R. Van Malderen J. Guijarro T. Ning S. Zengin B. Chimani 2 of 3 PMW CLIMATOL PMTred Pei HOMOP Daily/Monhly D+M D+M D+M D+M D+M D X Easy/Less/Full E+L+F E+L+F E+L+F L+F E+L+F E+L+F E+F 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 12

13 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 13

14 1. he idenificaion of he epochs of he insered breakpoins (+ sensiiviy analysis) in he synheic. 2. he esimaion of he rends ha were or were no imposed o he 3 ses of synheic IWV differences. Venema e al. 2012, Benchmarking homogenizaion algorihms for monhly daa, Climae of he Pas, 8, , doi: /cp Gazeaux e al. 2013, Deecing offses in GPS ime series: firs resuls from he deecion of offses in GPS experimen, J. Geophys. Res. Solid Earh, 118, doi: /jgrb h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 14

15 Mean epoch difference Mean epoch difference Moivaion & 1. defining a proper ime window daily monhly daily monhly In his presenaion, a ime window of 62d (2 monhs), will be assumed. 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 15

16 1. deecion scores good performance TP + TN > 40% FP < 40% FN < 40% EASY + A.R. Noise LESS COMPLICATED Performance Increase Adaped from Gazeaux e al good performance for he majoriy of he ools for he easy and less complicaed daase 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 16

17 1. deecion scores LESS COMPLICATED + gaps and rend FULLY COMPLICATED Adaped from Gazeaux e al performance decreases drasically for almos all he ools when adding gaps and a rend in he benchmark ime series 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 17

18 PSS PSS PSS Moivaion & 1. deecion scores EASY = POD - POFD LESS COMPLICATED FULLY COMPLICATED Venema e al T: Precip. : h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 18

19 POFD POD Moivaion & 1. deecion scores FULLY COMPLICATED 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 19

20 1. ampliudes of offses EASY DAILY EASY TRUTH 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 20

21 1. ampliudes of offses EASY DAILY EASY MONTHLY 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 21

22 2. EASY FULLY COMPLICATED Hecor = oupu from MLE wih simulaed offses large variey of rends (and uncerainies) for he differen correced ime series 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 22

23 2. FULLY COMPLICATED Hecor = oupu from MLE wih simulaed offses mos rends differ wihin ±0.05 mm/yr. 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 23

24 more deailed assessmen of he performance of he differen ools (and sensiiviy analysis): CRMSE, The epochs and ampliudes (and feedback) of he offses in he synheic is given o he paricipans who already provided heir soluions fine-uning of heir ools We highly welcome oher conribuions (ools, soluions) in our aciviy! Ineresed? Please conac me a roeland@meeo.be. A nex generaion of a fully complicaed synheic daase will be generaed: fully complicaed II? gaps decoupled from rend(s)? based on he difference of he synheic IGS repro 1 minus he real ERAinerim? A second round of blind homogenizaion on his nex generaion daase(s) will end in Sepember. 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 24

25 applicaion of he good performing ools on he IGS repro 1 define a common sraegy o correc he IGS repro 1 daase, based on crieria as (examples!): break poins should be deeced by a minimum of homogenizaion ools. break poin should be presen in he meadaa file of he saion. he ampliude of he offse should be above a cerain limi. break poins should be deeced in oher IWV differences ime series (e.g. IGS minus NCEPNCAR reanalysis) Thereafer, he focus will be on a GNSS IWV daase cenred above Europe, homogeneously reprocessed, wih abou 100 sies (ou of 280) wih daa from ( neighbour-based approach possible?) We are looking forward o geing feedback/inpu/paricipaion from your communiy! 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 25

26 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 26

27 Ning Elias Van Malderen e al. Bock e al. KTU (Tanır Kayı Klos e al. monhly monhly daily monhly daily daily daily manual albh 15/10/ /10/2002 2/05/ /05/2002 albh 15/03/2000 9/07/1998 albh 15/02/2006 2/07/2000 albh 12/03/2001 albh 18/01/2005 algo 7/02/ /05/ /10/ alic 20/04/ /04/ /04/ /08/ /10/ /07/ alic 15/08/ /08/ /04/ /06/ alic 6/05/ alic 11/10/ ankr 15/09/ /10/ /10/ /10/ /10/2001 3/01/ /05/2005 7/02/ ankr 15/08/ /09/ /05/ /07/ ankr 15/09/ /09/ /07/ ankr 16/09/ ankr 4/07/ ankr 24/11/ ankr 6/05/ ankr 4/06/ ankr 16/09/ ankr 26/11/ breakpoins deeced in meadaa & visual inspecion, bu no by any of he groups? breakpoins deeced by a number (all) ools, bu no meadaa informaion? ime window! When are breakpoins coinciden? 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 27

28 28 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology IWV i n j j i i i i i i i i i i H d j i h g f e d c b a IWV cos 8 sin 6 cos 6 sin 4 cos 4 sin 2 cos 2 sin If significan frequencies were no accouned for, i may lead o wrong rend esimaes. Compuaions were performed wih Maximum Likelihood Esimaion (MLE) in he Hecor sofware.

29 Characerizaion of real daa: Annual signal. The larges differences beween seasonal curves were found for KIT3, IISC and POL2 due o a shif in phase. 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 29

30 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 30

31 Why did we go for synheic differences (DIFF_syn)? We examined ERAI_syn-GPS_syn and we couldn have covered some par of power There is remaining unmodelled signals and noise in he ime series of he real differences. 4 h Conference on Spaial Inerpolaion Techniques in Climaology and Meeorology 31