Project Director: Dr. A. Suryachandra Rao
The seasonal prediction of the Indian summer monsoon rainfall (ISMR) is very important for India, especially for planning strategies towards management of agricultural production and water resources. The seasonal prediction of the monsoon by dynamical models is based on the fact that the slowly varying boundary conditions like sea surface temperature (SST), soil moisture, snow cover etc. exert significant influence on atmospheric development on seasonal time-scales in the tropics. Although the seasonal mean monsoon seems to be potentially predictable, atmospheric GCM simulations have not shown enough skill in capturing the inter-annual variations in the monsoon rainfall. Indian Summer Monsoon has limited potential predictability. It has also been recognized that ocean-atmosphere coupling is crucial in determining the potential predictability of the monsoon. Therefore, a coupled ocean-atmosphere climate model is required for predicting the monsoon. It is essential to develop and improve a system of fully coupled ocean-atmosphere-land modelling system for dynamical prediction of the seasonal mean monsoon rainfall. IITM is developing such a system and it will be transferred to the India Meteorological Department (IMD). This model is useful for giving lot of spin-off in science, e.g., one should be able to study the role of air-sea interactions on monsoon variability and predictability in more details.
Recent studies have demonstrated the possibility of achieving improved skills in simulating the seasonal mean monsoon rainfall by using ocean-atmosphere coupled models. This improvement appears to result from more accurate representation of the coupled interactions between the Indian monsoon and the tropical oceans. During 11th and 12th Five Year Plan periods, IITM scientists have setup an ocean-atmosphere coupled model on its IBM P6 575 (Prithvi) & Aaditya High Performance Computing (HPC) system and made long period free runs as well as hindcast (retrospective) experiments to test the model with set of initial conditions (e.g., with ensembles of atmospheric and oceanic initial conditions). The model outputs have been analyzed and its performance for simulating Indian Summer Monsoon Rainfall(ISMR) was examined. Certain biases in model simulations have been identified and efforts are being made to reduce these model biases. In addition to the research efforts on the coupled model, IITM has provided, for the first time in India, reliable experimental coupled dynamical monsoon prediction to IMD for further dissemination to general public. Certain modifications (e.g., better physical parameterizations and better representation of air-sea interaction processes, higher resolution) has been incorporated for making this model better suitable for our region, leading to enhanced model skill of simulating ISMR.
These predictions were very close to the actually realised rainfall, i.e., 86% of long period average (LPA) over the country during the summer monsoon season (June-September) of 2015.
What needs to be done to further improve the ISMR prediction skill in CFS V2?
Recently, by using carefully designed coupled model experiments, the authors demonstrated that the prediction skill of the all India summer monsoon rainfall (AISMR) in Climate Forecast System version 2 (CFSv2) model basically comes from the El-Niño Southern Oscillation-Monsoon teleconnection. On the other hand, contrary to observations, the Indian Ocean coupled dynamics do not have a crucial role in controlling the prediction skill of the AISMR in CFSv2; however, these are important for the proper simulation of the inter-annual variance of AISMR. It is shown that inadequate representation of the Indian Ocean coupled dynamics in CFSv2 is responsible for this dichotomy. Hence, improvement in the Indian Ocean coupled dynamics is essential for further improvement of the AISMR prediction skill in CFSv2. It is also shown that the dry bias over the Indian landmass is primarily due to cold SST simulated in the tropical Indian Ocean. When SST bias is reduced in the Indian Slab run, the mean ISMR improved significantly over the Indian land mass (Fig. 2). [George G., Rao Nagarjuna D., Sabeerali C.T., Srivastava Ankur, Suryachandra A. Rao, Indian summer monsoon prediction and simulation in CFSv2 coupled model, Atmospheric Science Letters, 17, January 2016, DOI:10.1002/asl.599, 57-64]
Fig. 2: (a) Observed seasonal (JJAS) climatological SST, (b) observed seasonal (JJAS) climatological precipitation, (c) seasonal SST bias of CFSv2 control (CTL) run, (d) seasonal precipitation bias of CFSv2 control (CTL) run, (e) seasonal SST difference between ISLAB and CTL runs (ISLAB-CTL run), (f) seasonal precipitation difference between ISLAB and CTL runs (ISLAB-CTL run), (g) seasonal SST difference between PSLAB and CTL runs (PSLAB-CTL run), (h) seasonal precipitation difference between PSLAB and CTL runs (PSLAB-CTL run).
Large-scale teleconnection patterns of Indian summer monsoon as revealed by CFSv2 retrospective seasonal forecast runs
The seasonal prediction skill of Indian summer monsoon rainfall (ISMR) in the global ocean-atmospheric coupled model such as CFSv2 largely depends on the teleconnection of the model at advanced lead times. The current study analyses teleconnection relationships of large scale boundary forcing having predictive skill with advanced lead time e.g., El-Nino southern oscillation (ENSO)-monsoon teleconnection and Indian Ocean Monsoon teleconnections. Prediction skills in terms of correlations and teleconnection patterns are analysed based on hindcast runs with models initialised from February, March, April and May initial conditions. The model exhibits reasonable skill at a longer lead time (e.g., forecasts initialised with February initial conditions, February IC run, cc >0.5) that is reasonably better when compared with that with forecast initialised at shorter lead time [April/May IC runs, cc <0.5].
The model shows unrealistic teleconnection of ISMR with Indian Ocean Dipole (IOD) and is unable to represent the large scale rainfall pattern over the Indian land region. The Equatorial Indian Ocean Oscillation (EQUINOO) shows unrealistic teleconnection with ISMR as well as equatorial Pacific from all the initial condition runs. Unrealistic EQUINOO and the IOD teleconnection suggest that the air-sea interaction in the Indian Ocean requires to be improved in the model. The relationship between El Nino-Southern Oscillation (ENSO) and monsoon in CFSv2 is realistic in terms of spatial pattern though it is somewhat stronger than that in observations. The ENSO-monsoon teleconnection spatial pattern shifts westward with decreasing lead time, resulting in unrealistic patterns as compared with observations, and causing a loss of prediction skill at shorter lead times. (Fig. 3) [Chattopadhyay R., Suryachandra A. Rao, Sabeerali C.T., George G., Rao Nagarjuna D., Dhakate A., Salunke K., Large-scale teleconnection patterns of Indian summer monsoon as revealed by CFSv2 retrospective seasonal forecast runs, International Journal of Climatology, online, December 2015, DOI:10.1002/joc.4556]
Fig.3. (a) Climatological JJAS mean rainfall. (b)-(e) bias in mean rainfall from February, March, April and May initial conditions respectively. (Units mm/day). The black circles represent the shift in rainfall bias as lead time is reduced.
Predictability of global monsoon rainfall in NCEP CFSv2This study evaluates the actual and potential prediction skill of the global monsoon rainfall using hindcast simulations by NCEP CFSv2 at zero to three lead forecast months (L0-L3). It is shown that the model has moderate skill in global monsoon rainfall (GMR) prediction, where the boreal summer monsoon rainfall forecast is more skillful than that of the austral summer. In general, the prediction skill of the GMR (actual and potential) increases with the decrease in lead forecast time, which is true for all the major regional monsoons, except the Australian monsoon (Fig. 4). Over the Australian monsoon region, both actual and potential prediction skills in rainfall increase with increase in lead forecast. The forecast skill of tropical SST during austral summer is a maximum at 3-month lead forecast (i.e., July initial conditions) and that is associated with spring predictability barrier. Using partial least square (PLS) regression method, it is shown that the major predictor (first latent vector) of the boreal and austral summer monsoon rainfall variability is ENSO, and the influence of ENSO on rainfall variability is much stronger in the model as compared to the observation. The second PLS regression mode is associated with the non-ENSO variability like tropical Atlantic, Indian, subtropical northwest Pacific Ocean variability, mid-latitude interactions, etc. However, the model has very poor skill in reproducing the second mode, particularly during the boreal summer monsoon season. It is also shown that a significant part of the Indian summer monsoon rainfall variability is controlled by other than ENSO variability and the model has limited success in capturing that. [Saha Subodh K., Sujith K., Pokhrel S., Chaudhari H. S., Hazra A., Predictability of Global Monsoon Rainfall in NCEP CFSv2, Climate Dynamics, online, December 2015, DOI:10.1007/s00382-015-2928-z]
Fig. 4: Potential prediction skill (Rlimit, based on ANOVA) calculated using GMR from CFSv2 at all lead times.
Prediction of seasonal summer monsoon rainfall over homogenous regions of India using dynamical prediction system
Seasonal prediction of Indian summer monsoon rainfall is a challenging task for the modeling community and predicting seasonal mean rainfall at smaller regional scale is much more difficult than predicting all India averaged seasonal mean rainfall. The regional scale prediction of summer monsoon mean rainfall at longer lead time (e.g., predicting 3–4 months in advance) can play a vital role in planning of hydrological and agriculture aspects of the society. Previous attempts for predicting seasonal mean rainfall at regional level (over 5 Homogeneous regions) have resulted with limited success (anomaly correlation coefficient is low, ACC 0.1–0.4, even at a short lead time of one month). The high resolution Climate Forecast System, version 2 (CFSv2) model, with spectral resolution of T382 ( 38 km), can predict the Indian summer monsoon rainfall (ISMR) at lead time of 3–4 months, with a reasonably good prediction skill (ACC 0.55). In the present study, we have investigated whether the seasonal mean rainfall over different homogenous regions is predictable using the same model, at 3–4 months lead time? Out of five homogeneous regions of India three regions have shown moderate prediction skill, even at 3 months lead time. Compared to lower resolution model, high resolution model has good skill for all the regions except south peninsular India. High resolution model is able to capture the extreme events and also the teleconnections associated with large scale features at four months lead time and hence shows better skill (ACC 0.45) in predicting the seasonal mean rainfall over homogeneous regions. [Dandi A. Ramu, Suryachadra A. Rao, Prasanth A. Pillai, M. Pradhan, G. George, D. Nagarguna Rao, S. Mahapatra, D.S. Pai, M. Rajeevan, Prediction of seasonal summer monsoon rainfall over homogenous regions of India using dynamical prediction system, Journal of Hydrology, 546, January 2017, 103–112]
Fig 1: Seasonal rainfall anomaly composites for deficient, excess and normal years for observations (a–c), T126 (d–f) and T382 (g–i). Geographical locations of 5 homogeneous regions of India are marked in (e). Pattern correlation between model and observations is included in top of each model plots (d–i).
Sub-project: Seasonal Prediction
Associates : Dr. R. S. Maheshkumar, Sci-D