What was done
The 2004 summer monsoon season of India experienced a 13% deficit that was not predicted by empirical or dynamical models used in making rainfall forecasts.  As a result of this failure, the authors performed an historical analysis of such models' forecast skill over the period 1932 to 2004.

What was learned
Despite numerous model changes and an ever-improving understanding of monsoon variability, Indian monsoon model forecast skill has not improved since 1932.  Large differences are generally observed when comparing monsoon rainfall measurements with empirical model predictions.  In addition, these models often fail to correctly predict even the sign of the precipitation anomaly, frequently predicting excess rainfall when drought actually occurs and drought when excess rainfall is eventually received.

Dynamical models fared even worse.  In comparing observed versus predicted monsoon rainfall from 20 "state-of-the-art" atmospheric general circulation models and one supposedly superior coupled atmosphere-ocean model, the authors report that none were able "to simulate correctly the interannual variation of the summer monsoon rainfall over the Indian region."  And once again, they frequently failed to correctly capture not only the magnitude, but even the sign of the observed rainfall anomaly.

What it means
In spite of the billions of dollars that have been spent on developing and improving climate models, taxpayers have achieved essentially no return on their investment in terms of the models being able to correctly simulate one of the largest and regionally important of earth's atmospheric phenomena - the tropical Indian monsoon.  And for those who think that all we need is a finer resolution model to solve the problem, think again.  There has been one recent study of the Indian monsoon using a high-resolution GCM (Brankovic and Molteni, 2004), and as Gadgil et al. note, that model proved to be "not realistic."

Reference
Brankovic, C. and Molteni, F.  2004.  Seasonal climate and variability of the ECMWF ERA-40 model.  Climate Dynamics 22: 139-155.