Download presentation
Presentation is loading. Please wait.
1
June-Yi Lee and Bin Wang
How are seasonal prediction skills related to models’ systematic error? Good afternoon everyone (smile). Today, I’d like to talk about the impact of systematic errors in climate prediction models on seasonal anomaly prediction and ENSO-monsoon relationship. This work is one part of a series of assessment efforts of current climate models’ one-month lead seasonal prediction in APCC/CliPAS project which Profs. Bin Wang, In-Sik Kang, and J. Shukla lead supported by APEC Climate Center. I’d like to acknowledge contributions from all APCC/CliPAS investigators. June-Yi Lee and Bin Wang IPRC, University of Hawaii, USA In-Sik Kang, Seoul National University, Korea J. Shukla, George Mason University, USA C.-K. Park, APCC, Korea
2
CliPAS: Climate Prediction and Its Application to Society
The international project, the CliPAS, in support of APCC is aimed at establishing well-validated multi-model ensemble (MME) prediction systems for climate prediction and developing economic and societal applications. Acknowledge contributions from the following CliPAS/APCC Investigators BMRC: O. Alves CES/SNU: I.-S. Kang, J.-S. Kug COLA/GMU: J. Shukla, B. Kirtman, J. Kinter, K. Jin FSU: T. Krishnamurti, S. Cocke, FRCGC/JAMSTEC: J. Luo, T. Yamagata (UT) IAP/CAS: T. Zhou, B. Wang KMA: W.-T. Yun NASA/GSFC: M. Suarez, S. Schubert, W. Lau NOAA/GFDL: N.-C. Lau, T. Rosati, W. Stern NOAA/NCEP: J. Schemm, A. Kumar UH/IPRC/ICCS: B. Wang, J.-Y. Lee, P. Liu, L. X. Fu The Climate prediction and its application to society which the joint US-Korea project in support of APCC is aimed at establishing well-validated multi-model ensemble prediction systems for climate prediction and developing economic and societal application, taking advantage of existing advanced climate model predictions in Asia-Pacific sector. Since May 2005, the number of participating institutions and involved models have been expanding. Now more than 20 researchers in climate prediction field have been involved in this project from 11 different institutions in Korea, USA, Japan, China, and Australia. Especially I would like to acknowledge a big contribution from Jay and Arun in NCEP/CPC for providing the most comprehensive retrospective forecast among participating groups including one-tier as well as two-tier predictions.
3
Statistical-Dynamical SST prediction (SNU)
The Current Status of HFP Production Two-Tier systems One-Tier systems Statistical-Dynamical SST prediction (SNU) AGCM CGCM FSU 79-04, 2 times GFDL 79-04, 2 times NASA 80-04,2 times CFS (NCEP) 81-04,12 times SNU/KMA 79-02, 12 times CAM2 (UH) 79-03, 4 times SNU 80-02, 4 times SINTEX-F 82-04, 12 times Up to now, the CliPAS MME system consists of 7 two-tier and 7 one-tier retrospective forecasts for the period of approximately Most of model issue their forecast four times a year targeting seasonal climate prediction and 5 models issue every month. ECHAM(UH) 79-03, 2 times IAP 79-04, 4 times UH 82-03, 4 times GFDL 79-05,12 times *NCEP 81-04,4 times POAMA(BMRC) 80-02, 12 times * NCEP two-tier prediction was forced by CFS SST prediction
4
Multi-Model Ensemble Climate Prediction
Climate Prediction Models Multi-Model Ensemble Climate Prediction 13 coupled model retrospective forecasts for targeting seasonal climate prediction with 4 initial conditions starting from February 1st, May 1st, August 1st, and November 1st APCC/CliPAS One Tier APCC/CliPAS Two Tier DEMETER NCEP/CFS FSU CERFACS Meteo-France FRCGC/ SINTEX-F GFDL ECMWF Met Office SNU SNU Under the European Union, DEMETER project produced a series of 6-month multimodel ensemble hindcasts using 7 different coupled models until 2001. In this talk, I mainly utilize the 13 coupled retrospective forecasts which include 7 DEMETER and 6 APCC/CliPAS models for the common period of targeting one-month lead seasonal climate prediction with 4 initial conditions starting from Feb 1st, May 1st, Aug 1st, and Nov 1st. In addition, there will be comparison result between 7 one-tier and 7 two-tier MME predictions for summer and winter season. INVG GFDL Comparison NCEP GFS MPI LODYC POAMA/ BMRC IAP UH UH 1 NASA UH 2
5
Topics Objective: To identify the strengths and weaknesses of the seasonal prediction models, especially coupled models, in predicting seasonal monsoon climate. (1) The impact of the models’ systematic errors in mean state on its performance on seasonal precipitation prediction The fidelity of a model simulation of interannual variability has a close link to its ability in simulation of climatology (Shukla 1984; Fennessy et al. 1994, Sperber and Palmer 1996; Kang et al. 2002; Wang et al. 2004) and seasonal migration of rain belt (Gadgil and Sajani 1998). The objective of this study is to identify the strengths and weaknesses of the seasonal prediction models, especially coupled models, in simulating monsoon climate. Particular attention is paid to the impact of the model biases in the annual modes on its performance on seasonal prediction. I divide my talk into two parts. At the first part, particular attention is paid to the impact of the models’ systematic errors in mean state on its performance on seasonal precipitation prediction. Previous studies have found that the fidelity of a model simulation of interannual variability has a close link to its ability in simulation of climatology and seasonal migration of rain belt in the AMIP-type simulation. At the part, I will discuss the impact of the models’ systematic errors on ENSO-monsoon relationship. Recent studies have fount that improvements in a coupled model’s mean climatology generally lead to a more realistic simulation of ENSO-monsoon teleconnection in free run. The strength of the present model lies in objective and automatic selection of predictor grids. (2) The impact of the systematic errors on ENSO-monsoon relationship Improvements in a coupled model’s mean climatology generally lead to a more realistic simulation of ENSO-monsoon teleconnection (Lau and Nath 2000; Annamalai and Liu 2005; Turner et al. 2005; Annamalai et al. 2007)
6
13 Coupled Climate Models
Institute AGCM Resolution OGCM Ensemble Member Reference BMRC POAMA1.5 BAM 3.0d T47 L17 ACOM3 olat x 2olon L32 10 Zhong et al. (2005) FRCGC ECHAM4 T106 L19 OPA 8.2 2ocos(lat) x 2o lon L31 9 Luo et al. (2005) GFDL AM2.1 2olat x 2.5olon L24 MOM4 1/3olat x 1olon L50 Delworth et al. (2006) NCEP GFS T62 L64 MOM3 1/3olat x 5/8olon L27 15 Vintzileos et al. (2005) Saha et al. (2006) SNU T42 L21 MOM2.2 1/3olat x 1olon L40 6 Kug et al. (2005) UH T31 L19 UH Ocean 1olat x 2olon L2 CERFACS ARPEGE T63 L31 OPA8.2 2.0o x 2.0o L31 Deque (2001) Delecluse and Madec (1999) ECMWF IFS T95 L 40 HOPE-E 1.4x L29 Gregory et al. (2000) Wolff et al. (1997) INGV ECHAM-4 T42 L19 OPA 8.1 2.0x L29 Roeckner (1996) Madec et al. (1998) LODYC T95 L40 2.0x2.0 Meteo-France OPA 8.0 182GPx152GP Madec et al. (1997) MPI ECHAM-5 MPI-OM1 2.5x L23 Pope et al. (2000) Gordon et al. (2000) UK Met Office HadAM3 2.5x3.75 L19 GloSea OGCM 1.25x L40 Marsland et al. (2003) This is the description of the 13 coupled climate models from CliPAS and DEMETER projects.
7
Reconstruction of Annual Cycle in Climate Prediction
Annual cycle of prediction is reconstructed using retrospective forecasts for 4 initial conditions starting from 1 February, 1 May, 1 August, and 1 November. Thus, each month has different forecast lead time. 2-month forecast is used for March, June, September, and December, 3-month forecast for April, July, October, and January, and 4-month forecast for May, August, November, and February. Reconstruction of annual cycle using different forecast lead time for each month Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb 1mon 2mon 3mon 4mon Because we use four times a year prediction targeting one-month lead seasonal climate prediction, the annual cycle is reconstructed using different forecast lead time for each month. Spring forecast Integrating from February 1st Forecast lead time 1mon 2mon 3mon 4mon Summer forecast Integrating from May 1st 1mon 2mon 3mon 4mon Fall forecast Integrating from August 1st 1mon 2mon 3mon 4mon Winter forecast Integrating from November 1st
8
Current Status of Prediction of Seasonal Precipitation
: Temporal Correlation Skill for 13 Coupled Model MME (81-01) Before going to the mean stats performance, I would like to show you the current status of seasonal precipitation prediction using 13 coupled model MME in terms of temporal correlation skills for the period of The prediction skills for precipitation vary with space and season. The variations in the spatial patterns and the seasonality of the correlation skills suggest that ENSO variability is the primary source of the global seasonal prediction skill. Prediction in DJF, SON and MAM is evidently better than JJA due to the model’s capacity in capturing the ENSO teleconnections around the mature phases of ENSO. Precipitation predictions over land and local summer monsoon region have little skills.
9
One-Tier vs Two-Tier MME Prediction of JJA Precp
One-Tier vs Two-Tier MME Prediction of JJA Precp. /Anomaly Pattern Correlation & Normalized RMSE A-AM Region ENSO Region It is documented that the prediction skill of one-tier systems is better than the two-tier seasonal prediction system in boreal summer over both A-AM [40-160E, 30S-30N] and ENSO [ E, 30S-30N] regions in terms of anomaly pattern correlation skill and normalized RMS error.
10
Performance on Annual Mean
mmday-1 MME prediction reproduces the observed features which include (1) the major oceanic convergence zones over the Tropics, (2) the Major precipitation zones in the extratropical Pacific and Atlantic and (3) remarkable longitudinal and latitudinal asymmetries mmday-1 Underestimation over ocean convergence zone (2) Overestimation over Maritime continents and high elevated terrains where the wind-terrain interaction influences annual rainfall. Let’s look at the performance on annual mean first. This shows climatological annual mean precipitation in observation and one-month lead seasonal MME prediction. Overall, the MME prediction reproduces the observed features realistically, which to a large extent include (1) the major oceanic convergence zones over the Tropics: the intertropical convergence zone (ITCZ), the South Pacific convergence zone (SPCZ), and the equatorial-South Indian ocean convergence zone, (2) the major precipitation zones in the extra-tropical Pacific and Atlantic, which are associated with the oceanic storm tracks, and (3) remarkable longitudinal and latitudinal asymmetries, which many previous studies have discussed in detail (e.g., Wang and Ding 2007, Goswami et al. 2006, Chang et al. 2005, Annamalai et al. 1999). However, these models have common biases over the oceanic convergence zones where SST bias exists and the regions where the wind-terrain interaction is likely to produce annual rainfall such as Maritime continents and high-elevated terrains. The performance of the individual coupled models and their MME in predicting annual mean precipitation is evaluated by using a pattern correlation coefficient and a root mean square error normalized by the observed standard deviation over global tropics and subtropics between 30S-30N. The MME prediction is in good agreement with observation with a PCC of 0.93 and a NRMSE of In this case, two skill measurement show strong linear relationship implying that pattern related errors are dominant in this field rather than random errors.
11
Equinox asymmetric mode (13%) (AM minus ON mean precipitation)
Performance on Annual Cycle mm/day Equinox asymmetric mode (13%) (AM minus ON mean precipitation) Solstice global monsoon mode (71%) (JJAS minus DJFM mean precipitation) Forecast Skill The spring-fall asymmetry is exaggerated over the entire Indian Ocean, East Asia and South China Sea-Western North Pacific regions. The MME predicted a weaker-than-observed Asian summer monsoon. The first annual cycle mode represents a solstice global monsoon mode which account for 71 % of the total annual variance. Its spatial pattern can be represented extremely well by the difference between JJAS minus DJFM mean precipitation. The one-month lead seasonal prediction is reasonably close to the observed counterparts. The major deficiencies with the predicted first mode are found over the Bay of Bengal, South China Sea, Western North Pacific, and East Asian monsoon front implying that the MME predicted a weaker-than-observed Asian summer monsoon. The second EOF mode has also an annual period with a maximum in April-May and minimum in October-November and accounts for 13% of the total annual variance. The second mode represents an equinox asymmetric mode and its spatial pattern can be very well represented by the difference between AM and ON. The 2nd mode is captured realistically by the MME prediction but with less fidelity than the 1st mode with due regard to their respective amplitude. This spring-fall asymmetry is exaggerated over the entire Indian Ocean, East Asia and South China Sea-Western North Pacific regions. It is shown that the current coupled models can reproduce the observed 1st annual cycle mode realistically with a similar degree of skill as that for annual mean. On the other hand, the models tend to have difficulty in simulating the 2nd annual cycle mode.
12
Systematic Errors in JJA Monsoon Climate
Reduced precipitation over BoB, SCS, WNP, and East Asia Enhance precipitation over MC, WIO and TP Strong warm bias over land and cold bias over ocean enhancing the zonal and meridional land-sea thermal contrast Let’s look into the systematic errors in JJA monsoon climate. The most striking deficiency of the one-month lead MME coupled prediction is underestimated precipitation over major Asian monsoon regions including the Bay of Bengal, South China Sea, WNP, and East Asian regions. The most striking deficiency of the one-month lead coupled prediction from our work is underestimated precipitation over the South China Sea and WNP regions, which is one of the most important atmospheric heat sources that drive atmospheric circulation. This common bias in the coupled model is likely to link with the bias of seasonal anomalies over that region. Enhanced AC over IO and MC and northward shifted AC over NP Strong low level div. over India and weakening of V over BoB and SCS, Strong conv. over MC Weakening of divergence and anti cyclonic circulation in upper level monsoon flow (a) Precipitation ( mmday-1) (2) 2m air temperature (degree) (3) stream function (shading, 1x106m2s-1) and wind (vector,ms-1) at 850 hPa, (d) stream function (shading) and velocity potential (contour, 2x106m2s-1)
13
Performance on Mean States and its Linkage with Seasonal Prediction
Pattern Correlation over Global Tropics [30S – 30N] Combined annual cycle skill of the 1st and 2nd EOF modes by weighting their eigenvalues The seasonal prediction skills are positively correlated with their performances on both the annual mean and annual cycle in the coupled climate models. The MME prediction has much better skill than individual model predictions for all metrics
14
Metric: Anomaly pattern correlation skill over 0-360E, 30S-30N
Annual Mode vs Seasonal Precipitation Prediction / One-Tier vs Two-Tier MME (a) Climatology vs IAV (b) 1st Annual Cycle vs IAV NCEP CFS NCEP CFS NCEP T2 NCEP T2 Metric: Anomaly pattern correlation skill over 0-360E, 30S-30N
15
One Tier vs Two Tier / The 1st Annual Cycle Mode
Mean biases against CMAP precipitation Model spread against multi-model ensemble mean Let’s look into the systematic errors of the two systems in detail. Upper panels show the mean biases of the 1st AC mode in one-tier and two-tier MME prediction against CMAP precipitation, and lower panels show model spread against multi-model ensemble mean. Much larger biases and model spread are evident in the two-tier MME than in the one-tier MME prediction. It is obvious that the regions that have large mean biases from observation tend to have large model spread against MME prediction, especially in the two-tier case. Nonetheless, the spatial distribution of mean biases in one-tier MME is quite similar to that in two-tier MME except few regions, such as Maritime continents, even though the biases are much alleviated. The common biases in the two types of systems may arise from uncertain model physics and problematic land surface processes. The spatial distribution of mean biases in one-tier MME is quite similar to that in two-tier MME except few regions, although the biases are much alleviated. The common biases in the two types of systems may arise from uncertain model physics and problematic land surface processes.
16
SEOF Modes for Precipitation over Global Tropics
Source of Seasonal Predictability of Precipitation in Couple Model MME SEOF Modes for Precipitation over Global Tropics [0-360E, 30S-30N] How many modes are predictable? In the current MME prediction, ENSO variability is the primarily source of the global seasonal precipitation prediction. We found that the MME prediction skill of seasonal precipitation basically comes from the first three leading modes of SEOF which are strongly related with ENSO variability with different lead-lag relationship. I will discuss this problem in detail tomorrow afternoon 4pm. However, the predicted ENSO-monsoon relationship has errors associated with systematic bias in mean states. We found that the MME prediction skill of the seasonal tropical precipitation basically comes from the first four leading modes of SEOF based on the following criteria the first four leading modes of SEOF based on the following criteria: (1) the percentage variances of observed leading modes, (2) the pattern correlation between the observed and predicted eigenvector, and (3) the temporal correlation between the observed and the observed and predicted principal component (PC) time series (Fig. 2). The first two SEOF modes are very well predicted. The third and even the fourth modes are also reasonably well predicted. But all other higher modes are not predictable as shown by the insignificant correlation skill in the spatial structures (Fig. 2). Thus, we consider the first four major modes as predictable part of the interannual variations. The fractional variance is obtained from the ratio of the variance associated with a single SEOF mode to the total variance (Wang and An 2005). If we take these four predictable modes together, about 60% of the total variance can be captured by those observational modes.
17
Systematic and Anomaly Errors of JJA SST Forecast
The errors in El Nino amplitude, phase, and maximum location of variability in coupled models are related with mean state errors such as colder equatorial Pacific SST and stronger easterly wind over western equatorial Pacific.
18
ENSO Composite / Precipitation (Shaded) & SST (Contoured)
(Normalized anomaly field) The breaking relationship between ENSO and Indian monsoon is evident in observation, whilst the MME produce clear negative relationship. The anomalous precipitation and circulation are predicted better in the ENSO decaying JJA than ENSO developing JJA.
19
ENSO Composite /Velocity Potential at 850 (shaded) and 200 hPa (contoured)
(63,68,72) (82,91,97) Divergence (Dashed line) Convergence (Solid line) The shift of variability centers in onset summers and exaggerated variability in decay summers are evident in the atmospheric circulation field.
20
Summary The skills of one-month lead MME prediction of seasonal mean precipitation vary with space and season. The variations in the spatial patterns and the seasonality of the correlation skills suggest that ENSO variability is the primary source of the global seasonal prediction skill. Prediction in DJF, SON, and MAM is evidently better than JJA due to the model’s capacity in capturing the ENSO teleconnections around the mature phases of ENSO. 1 The state-of-the art coupled models can reproduce realistically the observed features of long-term annual mean precipitation. However, these models have common biases over the oceanic convergence zones where SST bias exists and the regions where the wind-terrain interaction is likely to produce annual rainfall. 2 The seasonal prediction skills are positively correlated with their performances on mean states in the coupled climate models. The MME prediction has much better skill than individual model predictions. 3 To summerize, The errors in amplitude, phase, and maximum location of El Nino variability in model are associated with mean state errors such as colder equatorial Pacific SST and stronger easterly wind over western equatorial Pacific, resulting in errors in ENSO-Monsoon teleconnection. The breaking relationship between ENSO and Indian monsoon is evident in observation, whilst the MME produce clear negative relationship. 4
21
Thank You !
22
Model Descriptions of CliPAS System
APCC/CliPAS Tier-1 Models Institute AGCM Resolution OGCM Ensemble Member Reference FRCGC ECHAM4 T106 L19 OPA 8.2 2o cos(lat)x2o lon L31 9 Luo et al. (2005) GFDL R30 R30L14 R30 L18 10 Delworth et al. (2002) NASA NSIPP1 2o lat x 2.5o lon L34 Poseidon V4 1/3o lat x 5/8o lon L27 3 Vintzileos et al. (2005) NCEP GFS T62 L64 MOM3 1/3o lat x 1o lon L40 15 Saha et al. (2005) SNU T42 L21 MOM2.2 1/3o lat x 1o lon L32 6 Kug et al. (2005) UH T31 L19 UH Ocean 1o lat x 2o lon L2 Fu and Wang (2001) APCC/CliPAS Tier-2 Models Institute AGCM Resolution Ensemble Member SST BC Reference FSU FSUGCM T63 L27 10 SNU SST forecast Cocke, S. and T.E. LaRow (2000) GFDL AM2 2o lat x 2.5o lon L24 Anderson et al. (2004) IAP LASG 2.8o lat x 2.8o lon L26 6 Wang et al. (2004) NCEP GFS T62 L64 15 CFS SST forecast Kanamitsu et al. (2002) SNU/KMA GCPS T63 L21 Kang et al. (2004) UH CAM2 T42 L26 Liu et al. (2005) ECHAM4 T31 L19 Roeckner et al. (1996)
23
Current Status of ENSO Prediction
/ Correlation Skill of Nino 3.4 SST
24
ENSO Composite (Velocity Potential)
(ECMWF model) Divergence (dashed line) Convergence (solid line)
25
MME predicts weaker-than-observed monsoon precipitation
Systematic Bias of Model in JJA Strong warm bias over land and cold bias over ocean enhance the zonal and meridional land-sea thermal contrast in the prediction models. Oceanic anticyclones are enhanced especially over Indian Ocean and maritime continent. North Pacific anticyclone is shifted northward. Associated with enhanced anticyclones, cross-equatorial meridional wind is weaken over east of maritime continent and South China Sea. Meridional wind over Bay of Bengal is also weaken. Precipitation is reduced over Bay of Bengal, SCS, WNP, and east Asian monsoon region and enhanced over maritime continent and western North Indian Ocean. Reduced precipitation over SCS-WNP region results in weakening of divergence over same region and anticyclone over Indian Ocean at 200 hPa.
26
Annual Cycle of NCEP Models
SST Precipitation
27
Source of Predictability and Error
Indian Monsoon MME system predicts realistic annual cycle of precipitation over the Indian monsoon region, while it has no skill in seasonal anomaly prediction of precipitation. Systematic Bias: Cold bias of SST over the entire North Indian Ocean Weak upper level easterly Major error source: Systematic bias in ENSO-Indian monsoon teleconnection SCS-WNP Monsoon MME system has large systematic bias in annual cycle of precipitation, it has moderate skill in seasonal anomaly prediction Systematic Bias: Cold bias of SST Enhance precipitation in cold seasons and reduced one in warm season Weak mean precipitation and its variance in JJA Weak upper level divergence Predictability source: ENSO (MME reproduce realistic ENSO-WNP relationship) Error source: unrealistic simulation of ISO in models is related to weak mean precipitation and its weak variance
28
The definition of monsoon domain
(red) The definition of monsoon domain The regions in which the annual range (summer mean minus winter mean) exceeds 2mm/day and the local summer monsoon precipitation exceeds 35% of annual rainfall. Here, summer means JJA in the NH and DJF in the SH (Wang and Ding 2006). Here we use the monsoon domain as one of the metrics to evaluate coupled model’s capability in simulating global precipitation distribution. According to Wang and Ding (2006), the “monsoon domain” is defined as the regions in which the annual range exceeds 2mm/day and the local summer monsoon precipitation exceeds 35% of annual rainfall. The annual range of precipitation is measured by the local summer-minus-winter precipitation. This criteria is very similar as Dr. Zhang’s which showed yesterday. Figure shows monsoon domain depicted by CMAP (black contour) and the one-month lead seasonal MME prediction (red contour). The prediction can realistically capture the major monsoon domains in South Asia, Indonesia-Australia, North and South Africa, and Central and South America. However, the MME prediction has difficulty in capturing the western North Pacific monsoon regions and the East Asian subtropical Maiyu-Baiu region. In those regions, the coupled models show a large discrepancy in terms of defining the monsoon domain. The model spread in these regions is depicted by the number of models which capture monsoon domain at each grid point. Gray color indicates that all 10 models capture monsoonal precipitation characteristics at the point. The deficiency arises from the fact that some models cannot predict correctly the seasonal distribution of precipitation in the East Asian subtropics and western North Pacific, nor in the southwest Indian Ocean monsoon regions.
29
Temporal Correlation and Normalized RMSE of Precipitation Prediction
Figure 4. Temporal correlation coefficients (upper panels) and normalized RMSE (lower panels) of precipitation between observation and one-month lead seasonal prediction obtained from APCC/CliPAS MME system in summer (left-hand panels) and winter (right-hand panels) seasons, respectively. In (a) and (b), dashed line is for 0.3 and solid line is for 0.5 correlation coefficient. Solid contour indicates 0.9 in (b) and (d).
30
Performance on Mean States and its Linkage with Seasonal Prediction
Pattern Correlation skill over the A-AM Region [40-160E, 30S-30N]
31
Performance on Mean States and its Linkage with Seasonal Prediction
Pattern Correlation skill over the global Tropics [30S-30N]
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.