Comparison of Three Interpolation Schemes to Generate Daily and Monthly Gridded Precipitation Analyses by the Global Precipitation Climatology Centre (GPCC)

Präsentation zum Thema: "Comparison of Three Interpolation Schemes to Generate Daily and Monthly Gridded Precipitation Analyses by the Global Precipitation Climatology Centre (GPCC)"—  Präsentation transkript:

1 Comparison of Three Interpolation Schemes to Generate Daily and Monthly Gridded Precipitation Analyses by the Global Precipitation Climatology Centre (GPCC) and its Application to Produce Global Analyses Deutscher Wetterdienst, Global Precipitation Climatology Centre, Offenbach, Germany Ziese, M.; Schneider, U.; Meyer-Christoffer, A.; Rustemeier, E.; Finger, P.; Schamm, K.; Becker, A.

2 Background of GPCC  Collection, quality control and storage of observations, generation of gridded analyses  Established at the beginning of 1989 at Deutscher Wetterdienst (DWD) on invitation by WMO → more than 25 years of experience with precipitation  Analysis of precipitation on the basis of in-situ data for the land-surface  Contributing to GEWEX (Global Energy and Water Exchanges Project), GCOS (Global Climate Observing System) and land-analysis for GPCP  Many users world wide, analyses used in IPCC-AR5  Data sources:  SYNOP, CLIMAT, SYNOP from CPC  national meteorological services  CRU, FAO, GHCN  ECA&D, regional data collections 2

3 GPCC data basis 3 Monthly totals, collection since 1989 Daily totals, collection since 2012

4 Problems detected  Stations are sometimes located in the ocean or outside of the boundaries of the country  Unusual annual cycle or extreme outliers of monthly precipitation  Temporal shifts in the data  Factor*10 errors  Typing or coding errors  Errors in the conversion of inch, mm etc. (mostly with historical data)  Incorrect flagging of missing precipitation observations (might be misinterpreted as „0“) 4

5 Example of errors  Name of one station with different spellings:  Huddur  Huduur  Hudur  Hodur  Oddur  Xuddur  Xudur 5

6 Example of errors  earlier data set from Great Britain:  most stations had an incorrect longitude (factor- 10) – corrected in the precontrol step 6

7 Example of errors  Data shifted by one day 7

8 Example of errors  Wrong coding of missing values 8  Wrong zeros from CRU and GHCN corrected by FAO, indicated by surrounding stations

9 Example of errors  Shift in time 9 19821980 1953 19521951 1979

10 Comparison of Interpolation Schemes  used interpolation schemes:  modified SPHEREMAP  ordinary Kriging  arithmetic mean  two test months: July 1986 and January 1987  population divided into collectives with 300 stations  using 4800 stations as reference  50 runs with arbitrary selected reference stations to calculate skill scores  stepwise reduction of station density (input stations, not reference stations)  4 to 10 input stations  runs with interpolation of anomalies and absolute values  reduced station density in Germany (219 instead of more than 4000) 10

11 Kriging (Krige 1966)  statistical interpolation scheme  calculates correlations on basis of variograms  used for daily data sets at GPCC  uses at least 4, at most 10 stations  search radius depends on station density  applies only one variogram for global interpolations 11

12 SPHEREMAP (Willmott et al. 1985)  Application for monthly data sets at GPCC  Combination of distance and angular weighting  Distance weighting similar to IDW with given empirical weighting functions  Angular weighting to reduce influence of clustered stations  Compute gradients to preserve non-observed extremes 12 Grid point

13 SPHEREMAP – modifications at GPCC 13  defined other inner search radius  :  1 and  2   1 = 10% of grid size,  2 = 50% of grid size  if stations were found within  1 but not between  1 and  2 → arithmetic mean of stations within  1  if stations were found within  2 interpolation runs as usual  using more stations in case of high station density  4 to 10 input stations  interpolation runs on 0.25°/0.5° subgrid  using area weighting and land-portion weighting to calculate on final grid  runs operationally since 1995, as anomaly interpolation on basis of our Climatology since 2008 grid point  1  2 search radius

14 Interpolation of anomalies or totals  Interpolation of anomalies:  Deviation from long term means interpolated, finally added to gridded long term means  Also known as ‚Climatological Aided Interpolation‘ (CAI)  Need for many and long data series to compute long term means  Advantage: better results, also in data sparse regions  Interpolation of totals:  Observations are interpolated  Worse in data sparse regions 14

15 Possible anomalies  ‚absolute‘ anomalies:  Observation minus long term mean  Suitable for monthly data  ‚relative‘ anomalies:  Observation divided by monthly total or long term mean  Qualified for daily data 15

16 Used skill scores  mean squared error (MSE) [MSE] = mm²/month²  sensitive to outliers  mean absolute error (MAE) [MAE] = mm/month  measure of average error o – observed value at station y – interpolated value at station n - number of stations 16

17 Comparison July 1986 17  anomaly interpolation better than absolute interpolation  modified SPHEREMAP best for absolute interpolation (Climatology)

18 Comparison January 1987 18  anomaly interpolation better than absolute interpolation  modified SPHEREMAP best for absolute interpolation (Climatology)

19 Comparison SPHEREMAP and Kriging July 1986 Krigingmodified SPHEREMAP Kriging - SPHEREMAP  overall patterns look similar  Kriging produces smoother patterns  most differences due to different gradients of precipitation and in data sparse areas 19

20 Comparison of interpolation schemes, daily precipitation  Comparison of different interpolation schemes with cross-validation  Separation according to Köppen-Geiger climate zones  Best results for ordinary kriging with anomalies

21 Gridded GPCC products  Different products to tailor diverse user needs  Provide gridded analyses, no station data (except ITD) 21

22 PrecipitationStandard deviation Stations per gridKriging error Full Data Daily, 1997/07/06 22

23 Undercatch correction factor Fraction solid precipitation Stations per grid Precipitation Monitoring Product, 2015/05 23

24 Climatology, July 0.25° 1.0° 0.5° 2.5° 24

25 Suggestions for the interpolation of:  Daily minimum temperature (0.6 K/100 m)  Daily maximum temperature (0.6 K/100 m)  Wind speed  Vapor pressure (0.025 hPa/100 m)  Solar radiation  Lapse rate  Elevation correction 25

26 26

27 Zusätzliche Folien 27

28 First Guess Daily, First Guess Monthly First Guess Daily, 2015/07/02First Guess Monthly, 2015/07 28

29 GPCC quality control 29 SYNOP SYNOP + CLIMAT Historical data Automatic quality control using region depended fixed thresholds, consistence checks of overlapping intervals and weather observations; delete questionable observations; fill gaps Automatic quality control using region depended fixed thresholds; mark questionable observations for manuel checks, result: confirm, correct or delete values Test against station and grid based statistical thresholds; mark questionalbe observations for manuel checks, result: confirm, correct or delete values; spatial consistence of extreme values

30 non-used Stations in modified SPHEREMAP grid point  1  2 used station non-used station 0.25°/0.5° 30

31 Data base GPCC – spatial coverage 31  All stations in data bank  More than 100,000 stations

32 AOPC OOPC TOPC Data base GPCC – spatial coverage  Locations of 75 631 stations and lengths of their precipitation records  Only stations with records longer than 10 years, beginning no earlier than 1814 32 Figure 18 from GCOS report 195: Status of the Global Observing System for Climate, Full Report October 2015

33 Data base GPCC – spatial coverage 33 Start yearEnd year

34 Creation of long time series - causes of inhomogeneity  Example: Braemar in Scotland  A trend using unhomogenized data would lead to false conclusions  Error factor 10 in the beginning of the series  By now error in the data base is corrected

35 Creation of long time series - causes of inhomogeneity  Example: Braemar in Scotland  Correction as suggested by homogenization scheme  Factor 10 error leads to reduction of totals in later years

36 Anwendungsbeispiel: Vergleich von Bezugszeiträumen  Für jeden Bezugszeitraum selben Stationen verwendet  10 Datenjahre dürfen pro Bezugszeitraum fehlen  Daten nicht homogenisiert  Großräumige Muster gleich 36 1951-19801961-1990 1971-20001981-2010 Stationsbasis

37 Anwendungsbeispiel: Vergleich von Bezugszeiträumen  Nordeuropa wird feuchter  Südeuropa wird trockener  Ausnahmen: Sizilien wird feuchter und Ostdeutschland und Polen trockener  Große Gradienten der Trends an den Alpen 37 61/90 minus 51/80 71/00 minus 51/8081/10 minus 51/80 71/00 minus 61/9081/10 minus 61/90 81/10 minus 71/00 Blau: Jahresniederschlag nimmt zu Rot: Jahresniederschlag nimmt ab

38 Beispiel: Full Data Monthly, 1997/07 Niederschlag Stationsanzahl pro Raster 38

39 Beispiel: Interpolation Test Dataset (ITD), Juli Niederschlag Stationsanzahl pro Raster 39

40 WZN-Dürreindex  GPCC-DI: gerasterter Dürreindex mit beinahe globaler Abdeckung  Kombination von SPI-DWD und SPEI  Niederschlagsdaten vom WZN; First Guess Monthly  Monatsmitteltemperatur vom CPC  Verwendet Mittelwert von SPI-DWD und SPEI, falls beide berechnet werden können, sonst nur den berechenbaren Index  Parameter basieren auf Full Data Monthly V.6, Referenzperiode 1961-1990  Mehrere Aggregationszeiträume: 1, 3, 6, 9, 12, 24 und 48 Monate  Verwendet nur gerasterte Felder, keine Interpolationen  Zurück bis Januar 2013 verfügbar  Abgabe im netCDF-Format  Wird am 10. bis 13. Tag des Folgemonats aktualisiert 40

41 Beispiel WZN-Dürreindex, Juli 2015, 1 & 3 Monate normal feuchter alstrockener als 1 Monat3 Monate normal feuchter alstrockener als 41

42 Example of errors  Wrong metadata (longitude) 42

43 Example of errors  Repeating data 43

44 Example of errors  Interchange between stations 44

45 Example of errors  Filled gaps with data from climatology 45

46 Anwendungsbeispiel: Einfluss der ENSO  El Niño/Southern Oscilation (ENSO) führt zu veränderten Niederschlagsmustern  Korrelation des Southern Oscilation Index (SOI) mit Full Data Monthly (V.7)  Blau: mehr Niederschlag als üblich bei El Niño  Orange: mehr Niederschlag als üblich bei La Niña 46

47 Anwendungsbeispiel: Einfluss der ENSO (Frühjahr) 47

48 Anwendungsbeispiel: Einfluss der ENSO (Sommer) 48

49 Anwendungsbeispiel: Einfluss der ENSO (Sommer) 49

50 Anwendungsbeispiel: Einfluss der ENSO (Herbst) 50

51 Anwendungsbeispiel: Einfluss der ENSO (Herbst) 51

52 Anwendungsbeispiel: Einfluss der ENSO (Winter) 52

53 Anwendungsbeispiel: Einfluss der ENSO (Winter) 53

54 Anwendungsbeispiel: Dürre in Kalifornien 54 GPCC-DI, Niederschlag über 12 Monate aggregiert 201120122013 20142015

55 Anwendungsbeispiel: Dürre in Sao Paulo  Vorzeitiges Ende der Regenzeit 2013/2014  Später Anfang und trockenere Regenzeit 2014/2015  Wassermangel für Bevölkerung und Industrie  Einfluss von Klimawandel und ENSO werden noch diskutiert 55

56 Anwendungsbeispiel: Trends in der Dürrehäufigkeit  Änderung der Anzahl der Dürren in 62 Jahren, nicht der Intensität  3 Monate aggregiert, Dürre über 5 Monate wurde dreimal gezählt  Blau: Dürren werden seltener  Braun: Dürren werden häufiger  Trends in Europa passen zu Trends im Niederschlag 56

57 Nachbetrachtung von Extremereignissen  Analogie der Jahrhundertfluten am Amur (August 2013) und in Deutschland (Mai/Juni 2013) bezüglich hoher Vorsättigung der Böden 57 Bodenfeuchteindex (H-SAF) vor und nach 6-Tage Starkregen Niederschlagsmenge (links) und relative Anomalie für 6 Tage (rechts)