Intense rainfall often leads to floods and landslides in the Himalayan region even with rainfall amounts that are considered comparatively moderate over the plains; for example, ‘cloudbursts’, which are devastating convective phenomena producing sudden high-intensity rainfall (∼10 cm per hour) over a small area. Early prediction and warning of such severe local weather systems is crucial to mitigate societal impact arising from the accompanying flash floods. We examine a cloudburst event in the Himalayan region at Shillagarh village in the early hours of 16 July 2003. The storm lasted for less than half an hour, followed by flash floods that affected hundreds of people. We examine the fidelity of MM5 configured with multiple-nested domains (81, 27, 9 and 3 km grid-resolution) for predicting a cloudburst event with attention to horizontal resolution and the cloud microphysics parameterization. The MM5 model predicts the rainfall amount 24 hours in advance. However, the location of the cloudburst is displaced by tens of kilometers

Performance of four mesoscale models namely,the MM5,ETA,RSM and WRF,run at NCMRWF for short range weather forecasting has been examined during monsoon-2006.Evaluation is carried out based upon comparisons between observations and day-1 and day-3 forecasts of wind,temperature,specific humidity,geopotential height,rainfall,systematic errors,root mean square errors and specific events like the monsoon depressions.

It is very difficult to address the question of which model performs best over the Indian region? An honest answer is ‘none ’.Perhaps an ensemble approach would be the best.However, if we must make a final verdict,it can be stated that in general,(i)the WRF is able to produce best All India rainfall prediction compared to observations in the day-1 forecast and,the MM5 is able to produce best All India rainfall forecasts in day-3,but ETA and RSM are able to depict the best distribution of rainfall maxima along the west coast of India,(ii)the MM5 is able to produce least RMSE of wind and geopotential fields at most of the time,and (iii)the RSM is able to produce least errors in the day-1 forecasts of the tracks,while the ETA model produces least errors in the day-3 forecasts.

There have been very few mesoscale modelling studies of the Indian monsoon, with focus on the verification and intercomparison of the operational real time forecasts. With the exception of Das et al (2008), most of the studies in the literature are either the case studies of tropical cyclones and thunderstorms or the sensitivity studies involving physical parameterization or climate simulation studies. Almost all the studies are based on either National Center for Environmental Prediction (NCEP), USA, final analysis fields (NCEP FNL) or the reanalysis data used as initial and lateral boundary conditions for driving the mesoscale model.

Here we present a mesoscale model forecast verification and intercomparison study over India involving three mesoscale models: (i) the Weather Research and Forecast (WRF) model developed at the National Center for Atmospheric Research (NCAR), USA, (ii) the MM5 model developed by NCAR, and (iii) the Eta model of the NCEP, USA. The analysis is carried out for the monsoon season, June to September 2008. This study is unique since it is based entirely on the real time global model forecasts of the National Centre for Medium Range Weather Forecasting (NCMRWF) T254 global analysis and forecast system. Based on the evaluation and intercomparison of the mesoscale model forecasts, we recommend the best model for operational real-time forecasts over the Indian region.

Although the forecast mean 850 hPa circulation shows realistic monsoon flow and the monsoon trough, the systematic errors over the Arabian Sea indicate an easterly bias to the north (of mean flow) and westerly bias to the south (of mean flow). This suggests that the forecasts feature a southward shift in the monsoon current. The systematic error in the 850 hPa temperature indicates that largely the WRF model forecasts feature warm bias and the MM5 model forecasts feature cold bias. Features common to all the three models include warm bias over northwest India and cold bias over southeast peninsula. The 850 hPa specific humidity forecast errors clearly show that the Eta model features dry bias mostly over the sea, while MM5 features moist bias over large part of domain. The RMSE computed at different levels clearly establish that WRF model forecasts feature least errors in the predicted free atmospheric fields. Detailed rainfall forecast verification further establishes that the WRF model forecast rainfall skill remains more or less same in day-2 and day-3 as in day-1, while the forecast skill in the MM5 and Eta models, deteriorates in day-2 and day-3 forecasts.

The operational medium range rainfall forecasts of the Met Office Unified Model (UM) are evaluated over India using the Contiguous Rainfall Area (CRA) verification technique. In the CRA method, forecast and observed weather systems (defined by a user-specified rain threshold) are objectively matched to estimate location, volume, and pattern errors. In this study, UM rainfall forecasts from nine (2007–2015) Indian monsoon seasons are evaluated against 0.5$^{\circ }\times$ 0.5$^{\circ }$ IMD–NCMRWF gridded observed rainfall over India (6.5$^{\circ }{-}$38.5$^{\circ }$N, 66.5$^{\circ }{-}$100.5$^{\circ }$E). The model forecasts show a wet bias due to excessive number of rainy days particularly of low amounts (<1 mm d$^{-1}$). Verification scores consistently suggest good skill the forecasts at threshold of 10 mm d$^{-1}$, while moderate (poor) skill at thresholds of <20 mm d$^{-1}$ (<40 mm d$^{-1}$). Spatial verification of rainfall forecasts is carried out for 10, 20, 40 and 80 mm d$^{-1}$ CRA thresholds for four sub-regions namely (i) northwest (NW), (ii) southwest (SW), (iii) eastern (E), and (iv) northeast (NE) sub-region. Over the SW sub-region, the forecasts tend to underestimate rain intensity. In the SW region, the forecast events tended to be displaced to the west and southwest of the observed position on an average by about 1$^{\circ }$ distance. Over eastern India (E) forecasts of light (heavy) rainfall events, like 10 mm d$^{-1}$ (20 and 40 mm d$^{-1}$) tend to be displaced to the south on an average by about 1$^{\circ }$ (southeast by 1$-2^{\circ }$). In all four regions, the relative contribution to total error due to displacement increases with increasing CRA threshold. These findings can be useful for forecasters and for model developers with regard to the model systematic errors associated with the monsoon rainfall over different parts of India.

In recent times, instances of intense heat waves have increased over the Indian subcontinent. This increase in temperature has an adverse effect on human health and the economy. Over India, such high temperatures are usually seen during the months of March–May (summer). For weather forecasters, it is a challenging job to accurately predict the timing and intensity of this anomalous high temperature. The difficulty in the accurate prediction of weather is increased because of the presence of systematic biases inthe models. These biases are present because of improper parameterizations or model physics. For increasing the reliability or accuracy of a forecast it is essential to remove these biases by using a process called post-processing. In this study the biases in the surface temperature maximum are corrected using two methods, namely, the moving average and the decaying average. One of the main advantages of both the methods is that they do not require a large amount of past data for calibration and they take into account the most recent behaviour of the forecasting system. Verification, for maximum surface temperature during March–May 2017, was carried out in order to decide upon the method giving the best temperature forecast. It was found that both the bias correction methods lead to a decrease in the mean error in maximum surface temperature ($T_{max}$). However, the decaying average method showed a higher decrease in the mean error. Scores obtained from a contingency table like POD, FAR and PSS, showed that for $T_{max}$, the decaying average method outperforms the forecasts, i.e., raw and moving average in terms of having high POD and PSS and a low FAR.