Atmospheric Reanalysis for Multi Centuries using ... · • Our activity (Yoshimura et al.) • Online or Offline? • Online DA: Back to DATUN (von Storch, 2000), Forcing Singular

Atmospheric Reanalysis for Multi Centuries using Historical Weather Archives and Isotopic Proxies

Kei Yoshimura (UT/SDSU), A. Okazaki, K. Toride, P. Neluwala, S. Shoji, T. Miyoshi, et al.

Contents based onToride et al., 2017, MWR

Okazaki and Yoshimura, 2017, CPYoshimura, 2015, JMSJ

H18OH

HD16O

(Atmospheric) Reanalysis"Reanalysis" means producing historical global objective analysis dataset for over last decades in a consistent manner with a fixed state-of-the-art numerical weather prediction data assimilation system. (from JRA25 HP)

Our Research Q: How can we make such high-temporal resolution global dataset much longer?

Analyses of long-term historical weather-climate variations

Reconstruction from climate proxies

Analyses from direct measurement

Long-term variations (>years)

Remote signals aredumped

No data before 19thC

Isotopic Proxy Data Assimilation

Old Weather Data Assimilation

Short-term variations (hr~day)

Limit Limit

Breakthrough!

Multi Century Atmospheric ReanalysisApplication: Validation of CGCMs, Causality of social changes

Data AssimilationIsotopic Proxies

Old weather diaries

Info-Measure Fusion Info-Measure Fusion

Our Strategy for “Multi-Cent. Atmos. Reanalysis”

Breakthrough!

Toride et al., 2017Okazaki and Yoshimura, 2017

dividesfading?

Integration

Long term (>1yr) variation:(Isotopic) Proxy Data Assimilation• It’s a recent Hot Topic!!

• Last Millennium Reanalysis (Hakim et al.) now V3• Franke data (Franke et al.) (Non-isotopic so far)• Our activity (Yoshimura et al.)

• Online or Offline?• Online DA: Back to DATUN (von Storch, 2000), Forcing

Singular Vector (van der Schrier and Barkmeijer 2005; 2007), Particle Filters (Crespin et al. 2009), Mathiot et al. 2013)

• Offline DA: Initiated by Goosse et al. (2006), Bhend et al. (2012), 1-member ensemble DA (Steiger et al. 2014)

• Reconstructed or Measured?• Very few examples of DA with measured values so far.

• Variances by time and ensemble space are equivalent each other: Ergodic theory “1-member ensemble”

• No prediction cycle is done. Only analysis step. “Offline” data assimilation

k=1

k=2k=3k=4

k=5

k=6

Period 1 Period 2 …

Time resolution of proxy data

☆：y--- : xb× : xa

● : Ensemble spread of xb

( )( ) 1−

+=

−+=

RHHPHPK

xHyKxxT

bT

b

bba

xa Analysis （n x 1）

xb First Guess （n x 1）

y Observation （m x 1）

K Kalman Gain（n x m）

H Observation Operator （m x n）

Pb Model Error Covariance （n x n）

R Observation Error Covariance （m x m）

n: (model grid number) x (vertical levels) x (number of variables)m: number of observation points

Y=1850

Y=1851Y=1852Y=1853

Y=1854

Y=1855

Result of Idealized Experiment

Correlation between the truth and the analysis The assimilation skill is high for temperature

and precipitation as well as for the assimilated variable

∵ δ is affected by temperature and precipitation

気温

δ18O

2m temp. Precip.

Rbar=0.32

Rbar=0.58 Rbar=0.35Observation siteNot significant（p>0.05）

Real proxy data assimilation

Correlation b/w analysis and HadCRUT (1970-1999)

Correlation is high for the similar areas as in the perfect model exp. ENSO is expected to be well reconstructed even with current model and data

“Simulation” is constructed from the simulation forced by modeled SST “Observation” is from real observed proxy

Shade: analysisBlack line: observation

Observation siteNot significant（p>0.05）

Annual Variation of Global Ave SATfrom AD850

850 2010

288.0

288.2

288.4

288.6

ice coreall (coral, tree ring, ice core)

Global

Preliminary result. (Shoji et al. in prep)

Short-term (hr-day) variations:Old Weather Data Assimilation• Historical Weather Data Base (HWDB)

https://tk2-202-10627.vs.sakura.ne.jp/

1809 Dec. 19: Cloudy. Snow from 8am. 27cm accumulated.1809 Dec. 20: Cloudy. Small rain at 10am. Changed to snow for all day

Decoding and Compilation

Digitization

Mapping with GUI

DOI: 10.1175/MWR-D-16-0288.1

Reconstructed DA

0

5

10

15

20

25

30

35

1600 1650 1700 1750 1800 1850 1900 1950

Year

Number of records

Official meteorological network started

Assume 18 Observation stations in Japan

1740 1870

Method

(Authentic) Online DA

Cloud Cover RMSE(noobs) – RMSE(assim)

Red: ImprovedBlue: Worsened

Cloud AssimNo obsReanalysis

Cloud Assim No obs

>>>> Cloud is reproduced well!!

Correlation Coefficient

11

Time series @ observation station Blue: cloud assimRed: No obsBlack: Truth

Results improved by assimilating cloud!!

Specific humidity

12

Precipitation

Wind (Surface)

Wind (500hPa)

DistributionPrecipitation in Kyushu improved

13

Most of other variables clearly improved!

Surface Pressure

Precipitable water

Comparison with daily TCC obs.

Toride et al., 2017

Blue: WorsenedRed: Improved

Comparison with 6-hourly SAT obs.

Point (Okayama)( 35.07, 134.01 )1830 Jan –March

SST & Sea Ice concertation:Jörg Franke et.al (2017)

Initial Condition:1 year spin-up (single GSM) (1829-1830) starting with 1995/01/01.

Ensemble selection:20 ensembles using different dates from a GSM only run from 1830-1331 (Nature run)

Sola

r rad

iatio

n (w

/m2 )

TC C

loud

(%)

Wea

ther

Cat

egor

y W

eath

er C

ateg

ory

Application to 1830’s (Preliminary results)

Neluwala and Yoshimura,

in prep.

Summary

• Two key technologies for multi-century atmospheric reanalysis have been developed and tested.

• Isotopic Proxy Data Assimilation used offline & measured data assimilation. Good quality for interannual SAT/P variations for 19c-20c. Lack of proxy data further back to past is crucial.

• Old Weather Data Assimilation used online & reconstructed (TCC) data assimilation. Positive skill for TCC/SAT/q2m etc. Footprint may be extended remotely. Data availability is also crucial.

21Solar radiation (W/m2) vs Categories Cloud content vs Categories

Relationship between Jan-March JMA Downward shortwave Solar radiation, Total Cloud content vs weather categories derived from weather descriptions (1995 , 18 stations)

Relationship between cloud and weather categoriesJanuary to March (two months-1995)

Locations of diaries used

22

Application to 1830’s (Preliminary results)

-- No assimilation-- Assimilated

-- No assimilation-- Assimilated-- Category (1-3)

Japan Domain (18 grids average) 1830 Jan –March (daily mean)

Point (Okayama 35.07, 134.01 ) 1830 Jan -March

Cloud % Solar radiation (w/m2)

Solar radiation (w/m2)%

Neluwala and Yoshimura, in prep.

Data assimilation using Observed Cloud Data

from Japan Meteorological Agency

0

20

40

60

80

100

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31

Chosi

ncep jma

0

20

40

60

80

100

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31

Nemuro

ncep jma

0

50

100

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31

Hamada

ncep jma

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

R (JMA data & NCEP data)

Based on visual observation or equipment on ground

Correspondences with NECP data vary by observation point 17

25

Location of available personally diaries in 1830-1840-- Model grids (T62 grids)

Locations of diaries

Latit

ude

-->

Longitude -->

26

Experimental setup • Here we assimilate weather diary information.• Weather diary data

• The analog weather diaries has been converted to three categories (1,2,3) considering the clearness in the sky (i.e sunny, moderately cloudy, cloudy) (Mika Ichino, 1997)

• In this experiments 8 diary data are assimilated daily.• Diary data are converted to solar radiation using the emphatical

relation ship considering weather description and solar radiation data from 1979-199 (MikaIchino, 1997)

• Boundary condition • SST and sea-ice reconstruction data of Jörg Franke et.al (2017)

• Initial condition• Two year(1829-1831) GSM run was carried out with the initial

condition from a recent year(1995). • 1830 model run 20 ensembles were initiated from the above model

run 1830 January 1 to January 20 output.

27

Relationship between January clearness (KT) index vs weather categories(K) derived from weather descriptions) Japan metrological agency data from 1979-1999.

Ichino, M. (1995)

( )( )s

dT QInsolationTotal

QradiationsolardaccumulatedailyK −=

Total insolation can be calculated considering earth revolution and point location according to Kondo et. El (1995). Thus when weather category is known solar radiation can be calculated from about relationship

Relationship between Downward shortwave Solar radiation and weather

28

Next slides 1830 results

• I have plot the ensemble mean and ensemble spread (1 standard deviation ).

• Top two graphs shows the domain average Tc Cloud and solar radiation including the the ensemble spread .

• I would try to plot the bottom graphs as well

29

Other slides

• Next slides shows the analysis of the diurnal cycle which can be used for our next discussion

30

These are experiments using JMA data in 1995.JJMS data is converted to 5 categories and then solar radiation is assimilated using those 5 categories.

-- No assimilation – Assimilated -- Observation (JMA) -- Category (1-5)

31

-- No assimilation – Assimilated -- Observation (JMA) -- Category (1-5)After solar radiation assimilation we can observe some noise in temperature and precipitation, as these are important variables next task is to investigate this and improve these variables.

This investigation will be done using JMA data

32

ReferencesFranke, J., Brönnimann, S., Bhend, J., Brugnara, Y., & Wanner, H. (2017). A

monthly global paleo-reanalysis of the atmosphere from 1600 to 2005 for studying past climatic variations. Scientific Data, 4, 170076. http://doi.org/10.1038/sdata.2017.76

Ichino, M and Takehiko Mikami (2003), Spatial and Temporal Differences of Global Solar Radiation: Applicability of Mean Daily clearness index, Geographical reports of Tokyo Metropolitan University, Vol.38, p.15-21.

Ichino, M. (2007), Climatological study on distribution of the frequencies of global solar radiation for each weather categories at Tokyo. Ochanomizu Geotechnical Societyお茶の水地理, Vol.47, p.15-26. Retrieved from http://hdl.handle.net/10083/12722

市野美夏・元尚美・増田耕ー・三上岳彦(2001)天候記録を用いた全天日射

量の推定. 法-歴史時代の気候復元に向けて,天気Vol.48 No.11別刷日本気

象学会

市野, 美夏 (2007). 東京における天気別全天日射量分布の気候学的考察, 紀要論文 ,お茶の水地理学会, Vol.47, p.15-26. Retrieved from http://hdl.handle.net/10083/12722

Goal: Millennium Reanalysis • Much longer records than man-made observation

– Oceanic sediment δ18O (millions yBP)– Icesheet cores δ18O・δD (~800 kyBP)– Icecap cores δ18O・δD (~20 kyBP)– Speleothem δ18O (~2000 yBP)– Treering δ18O (~1000 yBP)– Coral δ18O (~400 yBP)

• Man-made, but not direct observation– Old weather diaries (~4000 yBP?)

33

Thank you very much!

keiyoshi08@gmail.com

天明の飢饉（1782-1788）

• 気候値からのアノマリを図示• 気候値は1767-1803平均とする

Preliminary Results

天保の飢饉（1831-1837）

• 気候値からのアノマリを図示• 気候値は1816-1852平均とする

Preliminary Results

38N, 135E

天明 天保

天明 天保

復元値は左軸 再解析は右軸

25.00

26.00

27.00

28.00

29.00

30.00

31.00

32.00

33.00

1820 1840 1860 1880 1900 1920 1940 1960 1980 2000

Chart Title 山形の7月気温推定値（1830－1980年，平野ほか,2013 地理評）

Preliminary Results

The University of Tokyookazaki@rainbow.iis.u-tokyo.ac.jp

Reproducibility by the model

Cellulose, Corals are OK as Precip Bad for Ice cores!! (Due to severe location)

red…Good（p<0.05）black…NG

00.10.20.30.40.50.60.70.8

有意

に再

現され

る

サンプル

の割

合

Precip Cellulose Coral Icecore

△…d18Ocell◇…d18Ocoral○…d18Oicecore

/2538

Frac

tion

of g

ood

sites

Okazaki and Yoshimura, in prep.

World’s Reanalysis Products

• NCEP/NCAR (1948-)• ERA15 (1979-)• NCEP/DOE (1979-)• ERA40 (1957-)• JRA25 (1979-)• ERA-Interim (1979-)• MERRA (1979-)• CFSRR (1979-)• 20CR (1871-)• JRA55 (1958-)

Most data has >100km horizontal scale in >6-hourly

39

Data Usage (in case of JRA25)

Contents

• Background– Reanalysis Products– Climate Reconstruction

• What we have done– Vapor isotope assimilation– Weather diary assimilation– Isotope proxy assimilation

• By combining the those…• Final Remarks

PAGES (PAst Global changES) 2K Network(See workshop program appendix “The PAGES 2k Network”)

2K consortium paper

PAGES 2k — Temperature reconstructions30-year standardized means (SD) (wrt 1200-1965)

• Most coherent feature among regions = pre-20thC long-term cooling trend

• No globally synchronous warm or cold intervals that define a worldwide Medieval Climate Anomaly or Little Ice Age

• 1971-2000 = warmest 30-year period; some regions experienced 30-year periods that were warmer

Cook et al., 2013

Precipitation Reconstruction

• 196 data longer than 900yr and shorter than 50yr interval are used.• Published data with hydrological signal is qualified. • Standardized.

Ljungqvist et al., 2016

Proxy/Model comparison

• The area coverage depends on data availability. The model results are treated as similar as proxy (standardized manner).

Proxy Model (CMIP5)

ResultInterannual variability of δ18O in ice core

Because first guess is climatology, it is drawn as horizontal line

First guess(before assimilation)

Analysis(after assimilation)

TruthFirst guessEnsemble member

TruthAnalysisEnsemble member

δ18O

[‰]

δ18O

[‰]

Result

First guess(before assimilation)

Analysis(after assimilation)

TruthFirst guessEnsemble member

TruthAnalysisEnsemble member

Interannual variability of surface air temperature

T2 [K

]T2

[K]

DA of Reconstructed values vs Measured values

Okazaki and Yoshimura, 2017