Multi-Method Experimental Forecasts of Summer Monsoon 2003
The experimental forecast system at C-MMACS is aimed at developing and testing methodology for improved range, scope and reliability. The neural network forecast method, termed the cognitive network, has been used to generate all-India summer monsoon rainfall (ISMR) since 1995. There is, however, urgent need for-improving the scope for the forecast in terms of spatio-temporal scales. In view of this, and the failure of most methods to predict the unusual monsoon of 2002, we have adopted a multi-method approach, combining different methods like neural networks, diagnostic model and dynamical models to improve overall scope and reliability. Here we present experimental forecasts for certain aspects of summer monsoon of 2003.
C-MMACS Environmental Modelling Programme ( CEMP)
Dynamical Forecasts for June and July,2003
The onset, in terms of jump in precipitation over coast of Kerala is likely to take place around June 6.
The month of June will receive more than normal rainfall over almost all parts of central and southern India and north-east India. However, there will be strong deficit over the western coast.
The monthly rainfall for June will be below normal also for most parts of western and northern India and parts of eastern India.
For the month of July, the rainfall will be significantly below normal over the southern India, while the central India will receive significantly above normal rainfall. The western India, especially Gujrat, should receive above normal rainfall.
For all-India summer monsoon is 740mm (+/- 40mm) implying about 15% deficit from long-period mean.
The best candidate for generating forecasts with high spatio-temporal resolution (preferably at user specified scales) is a dynamical model. As the monsoon is a long-scale system, it is necessary to use a global circulation model (GCM) to simulate and forecast monsoon. At the same time, the monsoon dynamics is affected by convective systems which have scales as small as a few kilometers. This makes it necessary to use as high a spatial (and hence temporal, for reasons of numerics ) resolution as possible. Long term integration of a GCM with high spatio – temporal resolution require a lot of efforts and computational resource. Besides, the physics and dynamics of convective processes, especially this interaction with the large-scale systems still very poorly, understood. Owing to these and other factors, high-resolution, long-range forecasting is major scientific challenge, especially over the monsoon region. Our strategy has been to configure and evaluate a GCM for monsoon forecasting and then further development and improvement in parallel with (and through) experimental forecasting.
A rigorous quantification of reliability requires a large statistics of model simulations as well as elaborate analysis. The model climatology and its comparison with observations is only one such test.
While the overall agreement the observed and the simulated climatologies is encouraging it should be noted that this agreement, or the error, is a strong function of geographical locations; reliability over locations with large errors, such as over the Bay of Bengal, is particularly low. It is clear, however, that the question is not a simple one of over or under prediction, as over the simulated field is in good agreement with the observed climatology.
See Observed and Simulated Climatology
The present results may change as we continue to improve the scope and methodology; some of the factors that can bring about changes are:
Multi Lead (Initial condition): The present forecasts have been generated with initial condition for April 1, 2003. However, our policy is to generate forecasts with progressive lead (getting closer to June 1). As the simulations strongly depend on the initial conditions, the subsequent forecasts may differ from the present ones.
Ensemble Forecasting: For the reason that the initial conditions play a strong role in the simulation, it is desirable (and necessary for reliability quantification) to average forecasts generated from several, ‘close’ initial conditions. This procedure, under progress, may change the present results somewhat. The experimental forecast developed one; please contact us for on-going developments and resulting changes in the forecasts.
The sole purpose of the experimental forecasts is an objective (field) evaluation of the proposed methodology. These forecasts are therefore, not included for any operational, or commercial use.
Dr. P Goswami
e-mail: goswami@cmmacs.ernet.in
