Current Research

My work is focused on understanding Monsoons and its variability at various scales using models of climate. Being a computationally intensive topic, I have also worked on exploitation of High Performance Computing innovatively. I have also worked on emerging techniques such as AI/ML for climate and weather studies.

• Variability of the Indian Monsoon
  The enigma of Indian Monsoon is yet to be unravelled. In the recent years there has been significant improvements in our understanding of the Monsoon. Yet there are areas that need to be addressed. Some of these are:
  
  
• The mechanism governing the intraseasonal variability is yet to be fully understood. I have worked on understanding the meechanism of poleward propagations (Nanjundiah et al, 1992, Rajendran et al, 2002 etc). With availability of data form various sources and new analytical techniques , my more recent work has focused on understanding the structure of this variability (Karmakar et al 2017) This work has shown that there exists a rich structure of intraseasonal variability involving poleward propagations at various scales, westward movement of cloud bands and interaction with mid-latitudes. These features have been studied individually but not in totality. I would like to build on my previous experience to obtain a combined view of these phenomena. I propose to use Climate Models/Earth System Models for this purpose and observational data. I would like to determine the ability of these models to realistically simulate these variations. I also would attempt modifications to the models to make their simulations more realistic.


• Linkage between Indian Monsoon and the Indian Ocean is an important issue in most climate/seasonal prediction models. While the linkage between Indian Monsoon and Equatorial Pacific is reasonably realistic , it is less satisfactory in simulation of its linkage with the Indian Ocean, and quite often the linkage is of the opposite kind (Nanjundiah et al, 2013) This improper linkage could have impact on prediction of monsoons at seasonal scales. Some of the years when seasonal predictions are less than satisfactory, the linkage is even more unrealistic Using data and models I propose to understand this challenging issue.


• Studies with Cloud system resolving models.
  Using a cloud system resolving model, my colleagues and me have understood the mechanism of southward propagation of cloudbands at quasidiurnal scales (Jain et al, 2018). We further propose to understand convective systems e.g. lows and depressions and their structure using these cloud system resolving models. To understand even finer systems over smaller regions, we propose to use LES.


• AI/ML in climate prediction and studies.
  This is an emerging area. We have built AI/ML based models at seasonal (Saha et al, 2016, 2020) , intraseaonal (Vishwanath et al, 2019) and and daily scales (Kumar et al 2022, Singh et al 2023). While some of these models are satisfactory, much work needs to be done to improve their skills , especially at shorter spatial and temporal scales.


• HPC and Climate Modelling
  architecture of computers is rapidly changing. The move is towards developing more energy efficient and faster machines by using computational accelerators such as GPUs. I had worked on using GPUs (Konwar et al, 2014) and Xeon PHI systems (Kashyap et al 2017) for optimising radiation calculations in atmospheric models and diffusion calculations in ocean models (Aketh et al, 2016). For efficient use, the computational flow may need to be modified without sacrificing the accuracy of the simulation .

Details of Research Experience of Ravi S Nanjundiah

My research has been focussed on trying to understand Indian monsoon and its variability by investigations of a slew of atmospheric models including general circulation models (AGCMs), coupled AtmosphereOcean models (AOGCMs) as well as cloud system resolving models and on-line chemistry models , to assess the role of important factors and processes in the different observed facets of the monsoon. I have also assessed the performance of state of the art climate models in simulating/predicting the monsoon, its variability and its observed teleconnections so as to get an insight into how the skill in prediction could be improved . From the outset, I have also been deeply interested in the computational aspects and numerics of the models as well as use of machine learning methods. I have worked on improving the scalability of climate models, use of High Performance Computing, grid computing and computational accelerators for increasing computational throughput of climate models. All this has been achieved with successful collaboration with mathematicians, atmospheric scientists, oceanographers, hydrologists and computational scientists within and outside IISc, and with students.

A few of the investigations carried out are described briefly here to give a flavour of the problems addressed, the approach adopted and results obtained.

I started research on the Indian Summer Monsoon in 1986 with the development of a simple climate model which could simulate realistically some of the most important features of the observed intraseasonal variability of the Indian monsoon such as the northward propagations of cloud-bands/rain-belts from the equatorial Indian Ocean onto the Indian monsoon zone at iintervals of 2-6 weeks throughout the summer monsoon season. We found that in the model the propagations occur due to gradient in vertical moist static instability (Nanjundiah et al,1992). This could potentially be useful for forecasting such propagations. While working on this model, I also began to use alternative computational techniques such as parallel computing so as to enable faster simulations on the available computers. I have continued work on monsoon modelling as well as on computational aspects and numerics of models throughout my career.

I. Studies of the simulation of the mean monsoon

Role of the spectacular mountain ranges of the Indian subcontinent and Myanmar in the mean monsoon has been emphasized in one of the early theories of the monsoon in the beginning of the last century. A special feature of the Bay of Bengal which sustains the monsoon with cloud systems from the region moving onto the Indian landmass, is the low salinity water near the surface due to river runoff and rainfall. River runoff could decrease due to increased impounding of river-waters. It is thus important to understand the role of river runoff on the monsoon. It has been suggested that increasing aerosols, particularly, black carbon, may decrease the monsoon rainfall and the importance of considering the impact of aerosols, clouds and their interaction has been pointed out.

Prescription of cloud related processes are important for monsoons and its simulation. We examined cumulus closure in two AGCMs and found that simulation of tropical rainfall in general, and Indian summer monsoon in particular , were sensititve to the rate at which clouds reduce moist instability. Lower rates of instability reduction gave more realistic simulations (Jain et al 2013, Mishra et al 2008) .

a. Assessment of the performance of climate models in monsoon simulation and prediction

An analysis of the retrospective predictions by seven coupled ocean–atmosphere models from major forecasting centres of Europe and USA showed that they had reasonable skill in predicting the extremes of the Indian summer monsoon rainfall, particularly those associated with ENSO. However, despite large intermodel differences in the parameterizations, numerics etc., there was also a remarkable coherence between the failures of the models to predict the Equatorial Indian Ocean Oscillation (EQUINOO) and the monsoon in years in which EQUINOO played an important role. It was found that models were able to simulate ENSO–monsoon link realistically, whereas the link of monsoon rainfall with EQUINOO was simulated realistically by only one model – the ECMWF model. Furthermore, in most models the simulated link is opposite to the observed, suggesting that prediction of EQUINOO and its links with the monsoon need to be improved for improving monsoon predictions by these models (Nanjundiah et al, 2013). We found that while the CGCMs used in IPCC AR4 were also able to simulate the ENSO-Monsoon relationship realistically, almost all models failed to simulate the EQUINOO (Rajeevan and Nanjundiah 2009)-Indian monsoon relationship. Prediction of monsoons on seasonal scales analysed using DEMETER data showed similar behaviour (Nanjundiah et al 2013,

b. Impact of orography

Our investigation of what happens to the monsoon when the mountains over different parts of the globe are removed in an AGCM (the NCMRWF GCM) showed that, with the exception of African orography, removal of orography in any other part of the world including proximate and remote regions such as the Rockies and the Andes reduced the monsoon rainfall and delayed its onset, these effects being largest for Western Himalayas (Chakraborty et al 2002, 2006). The absence of African orography strengthened the low level westerly jet and the rainfall over the Indian monsoon region because of non-linear feedbacks between rainfall and circulation that had been ignored in previous studies.

c. Monsoon & Ocean-Land-Atmosphere interaction

Comparison of a 100 year run of the Community Climate System Model (CCSM3) in which the river runoff is included with another similar run in which there is no river runoff showed that when the river discharge was shut off, global average sea surface temperature (SST) rose by about 0.5°C and the monsoon rainfall increased by about 10% with a large increase in the eastern Bay of Bengal and along the west coast of India. In addition, the frequency of occurrence of La Ni˜na-like cooling events in the equatorial Pacific increased and the correlation between ISMR and Pacific SST anomalies became stronger. The teleconnection between the SST anomalies in the Pacific and monsoon was effected via upper tropospheric meridional temperature gradient and a shift of the North African Asian Jet axis (Vinayachandran et al, 2015). Further studies with modulating runoff from the Amazon showed that Studies about the role of Amazonian runoff on climate showed that the Atlantic Meridional Overturning Circulation (AMOC) strengthened in its absence. Lack of Amazonian runoff also triggered a bipolar seesaw in SST across the thermal equator with warming to the north and cooling to the south. The boreal summer rainfall in the tropical Atlantic Ocean sector responded to this change in SST (Jahfer et al, 2020)

d. Intraseaonal Variations, Trends and Structure

We have stuided the trends in intraseaonal variations at decadal and longer scales (Karmakar et al, 2017). We found that while the long term trend of mean rainfall showed a decreasing trend, we also found a reduction in strength of low-frequency (20-60 days), while there was an increase in variability at higher frequencies (less than 10 days). We also find that there was an increase in sporadic rainfall events especially during the break phase and a reduction in of such events during the active phase. These were understood using an AGCM with prescibed heating over the Central Indian Region. The rich structure of intraseaonal Variations was studied using rainfall data from TRMM and other sources. Two dominant modes of variability with periodicities of 10–20 and 20–60 days were found, with the latter being strongly modulated by sea surface temperature. The 20–60 day mode showed northward propagation from the equatorial Indian Ocean linked with eastward-propagating modes of convective systems over the tropics. The 10–20 day mode showed a complex space–time structure with a northwestward-propagating anomalous pattern emanating from the Indonesian coast. This pattern was found to be interacting with a structure that emerged from higher latitudes which propagated into the Indian region (Karmakar et al 2017)

We have developed a technique for analysis of space-time variation of tropical precipitation using wavelets (Shanker and Nanjundiah, 2004, Chakraborty and Nanjundiah, 2012). This technique was used to quantify the spatial extent (scale) and centre of these propagating convective bands, as well as the time period associated with the different spatial scales. Results suggested that during a good monsoon year the spatial scale of this oscillation was about 30o centered around 10 oN. During weak monsoon years, the scale of propagation decreased and the centre shifted farther south, closer to the equator. A strong linear relationship was obtained between the center/scale of convective wave bands and intensity of monsoon precipitation over the Indian land on the interannual time scale. Moreover, the spatial scale and its centre during the break monsoon period were found to be similar to an overall weak monsoon year. Based on this analysis, a new index was proposed to quantify the spatial scales associated with propagating convective bands. This automated wavelet-based technique could be used to study meridional propagationof convection in a large volume of datasets from observations and model simulations.

e.Study of Midtropospheric Cyclones

We studied Midtroposhperic Cyclones (MTC) all across the globe (Kushwaha et al, 2021). These are moist synoptic systems with distinct midtropospheric vorticity maxima and weak signatures in the lower troposphere. Composites and statistics of tropical MTCs were constructed and compared with monsoon lows and depressions [together, lower-troposphere cyclones (LTCs)]. Tracking MTC over South Asia, we found that MTCs changed character during their life, i.e., their track comprised of MTC and LTC phases. The highest MTC-phase density and least motion were noticed over the Arabian Sea, followed by the Bay of Bengal and the South China Sea. An MTC-phase composite showed an east–west-tilted warm-core above a deep cold-core temperature anomaly with maximum vorticity at 600 hPa. In contrast, the LTC-phase showed a shallow cold core below 800 hPa and a warm upright temperature anomaly with a lower-tropospheric vorticity maximum. Globally, systems with MTC-like morphology were observed over west and central Africa and the east and west Pacific in boreal summer. In boreal winter, regions that support MTCs include northern Australia, the southern Indian Ocean, and southern Africa. MTC fraction is higher equatorward where there is a cross-equatorial low-level jet that advected oppositely signed vorticity, whereas LTCs were more prevalent farther poleward. We found that relationship between differential vorticity (the difference between middle and lower levels) and the height of peak vorticity for cyclonic centers was bimodal. One peak, around 600 hPa, corresponded to MTCs, while the second, at approximately 900 hPa,were from LTCs.

f. Convecion at Shorter Scales

We have studied convection over Bay of Bengal using a cloud system resolving model (Jain et al, 2018). Equatorward propagating precipitation episodes over the Bay of Bengal were noticed ( as in previous previous observational studies). The mechanism governing this southward propagation was determined. The mechanisms included mean surface to midtropospheric wind shear driving the convection orthogonal to the lower tropospheric winds and the gravity currents generated by outflow from convection initiated by the diurnally varying land-ocean circulations dispersing southwards.

II . Monsoons, Aerosols & Clouds

Aerosols and their interaction with atmosphere especially in modulating radiative heating can have a significant impact on monsoons. Over India black carbon aerosols (from vehicular emissions) and dust can play a significant role in modulating monsoons.

When we incorporated the atmospheric heating effect of black carbon aerosols (and cooling effect on the ground), we found that the strength of the mean monsoon increased. However, the regions where significant changes in precipitation occurred were sensitive to the cumulus parameterization used. (Chakraborty et al, 2004, Chakraborty et al 2012). Our studies have also shown that aerosol radiative forcing over geographically remote regions such as Eastern China could impact the Indian Monsoon (Chakraborty et al 2014). We have also assessed the simulation of aerosol concentrations over India in an on-line chemistry model (Govardhan et al 2015, 2016). We further assessed the regional aerosol radiative effects under the SWAMI campaign (Pathak et al 2019, 2020) and also developed a comprehensive dataset of aersol concentration by merging Satellite and terrestrial data using data assimilation techniques.

Studies with On-line Chemistry Models

While most studies focus on the near-surface abundance and impacts of BC, we examined the implications of sharp and confined layers of high BC concentration (called elevated BC layers) at altitudes of more than 4 km over the Indian region using an online regional chemistry transport model (WRF-Chem). Our study demonstrated that high-flying aircraft are the most likely cause of these elevated BC layers. Furthermore, we showed that such aircraft-emitted BC can be transported to upper tropospheric or lower stratospheric heights ( ∼ 17km) aided by the strong monsoonal convection occurring over the region, which is known to overshoot the tropical tropopause, leading to the injection of tropospheric air mass (along with its constituent aerosols) into the stratosphere. We showed observational evidence for such an intrusion of tropospheric BC into the stratosphere over the Indian region using extinction coefficient and particle depolarisation ratio data from CALIOP Lidar on-board the CALIPSO satellite (Goverdhan et al,2017)

Air Quality Prediction

Modelling and prediction of Air Quality is gaining importance in India. I have been associated with the development of an Air Quality Early Warning Sytem for Delhi and its environs. The highlight of this system is that Chemicalinformation from satellite and terrestrial ground stations have been incorporated using a Data Assimilation System (Kumar et al 2020, Jena et al, 2020). Based on this a Decision support is being developed for authorities to take informed decisions durin extreme air quality events.

III Machine Learning Techniques for Climate Change & Seasonal Prediction:

We have studied the impact of possible anthropogenic climate change on smaller scales (such as that of a river basin or a meteorological subdivision) than those that can be resolved by IPCC scenario models, by using machine learning downscaling techniques (Tripathi et al, 2006, Anandhi et al, 2012). We have shown that forecasts of the Indian summer monsoon rainfall on the seasonal scale using machine learning techniques, such as stacked encoder have skills comparable to the operational statistical models (Saha et al, 2016,2020). Models were also developed for the different homogeneous regions of India (Saha et al, 2017a). We also developed models for early part of the (June and July) and later part of the Monsoon, using deep neural networks such as stacked autoencoder and the ensmeble network tree regression tree technique (Saha et al, 2017b). Also based upon upon Deep Learning techniques, we developed a prediction model for EQUINOO and ENSO. The model for EQUINOO is the first of its kind (Saha and Nanjundiah 2020). Models based on LSTM were developed to study and predict active and break spells of the monsoon (Vishwanath et al, 2019) With enhanced concern about increased localised extreme precipitation events, we have developed techniques using as ConvLSTM (Kumar et al, 2022) and Unet based system on cubed sphere (Singh et al, 2023).

IV Numerical Techniques for AGCMs

We have investigated the impact of numerics and developed a framework for deriving splitting methods for semi-linear ordinary differential equations and partial differential equations (Murthy and Nanjundiah, 2000). We developed a technique for grid refinement for the spectral technique using reparameterisation maps tailored for studying tropical convection and teleconnections. Grid refinement in spectral models is a much more challenging task than in grid-point models and only a handful of such efforts have been attempted (Janakiraman et al, 2012). The technique so developed was used in a bartropic model.

V HPC & Climate Modelling:

Climate models are computationally intensive and using conventional computers have always been expensive. I implemented a simple climate on India’s first parallel computer viz Flosolver (at NAL). This implementation in 80s was the first such implementation in the country and perhaps one of the earliest such implementations in the world.

Scalability and throughput are important issues while using High Performance Computing. Any improvements in these implies a faster and more efficient computer at no additional cost. My research (in collaboration with computational scientists) has shown that these can be improved through load-balancing, message compression and use of computational accelerators. Our work on load balancing (Sundari et al, 2009) for a climate-system model showed that moving high computational load related to radiation to less-loaded processors computing ocean and other components, led to about 15% improvement in overall throughput. This perhaps was the first of its kind for a multi-component climate system model.

The trend in computational architecture is to move towards accelerators such as GPUs (for lowering of energy requirement) and increasing throughput at minimal cost. I have worked on exploiting GPUs for climate modelling (Korwar et al, 2014). Radiation computations can be a significant part of a climate model’s execution time. By offloading the most time-consuming routines of radiation onto GPU’s and exploiting asynchronous computing due to the fact that radiation is a slowly varying function of time, we could completely offset the time consumed in this part of execution. In the process these parts of the codes could be called more frequently and hence could result in higher accuracy. The asynchronous mode of execution was replicated on Xeon-Phi multicore sytems (Kashyap et al,2017)

Interprocessor communication is one of the biggest bottlenecks in parallel computing. Bigger the size of the communicated message, higher could be the overhead. My work (Kumar et al, 2008) on messagecompression exploited the fact that the model state does not change significantly between two consecutive timesteps and this could be exploited to reduce inter-processor communication. By sending information only about the changed parts of the message the speedup increases by about 18%. This was the first application of message-compression to a climate model.

Grid Computing for Climate Modelling

Grid computing helps to agglomerate resources at geographically distinct locations. In a resources-starved country like India, it is a good alternative to installing a large computer at a single location. We built a computational grid consisting of clusters at CAOS and SERC that could be used for climate-system model simulations (Sundari et al 2012, 2011) . This grid exploited the feature that a climate-system model consists of loosely coupled sub-components running concurrently with inter-component communication being low (in contrast to intra-component communication which could be much higher). The middleware developed for this purpose could reconfigure computations whenever new/existing systems became available/unavailable. The ideas thus developed can also be used on future exa-scale computers .

Highlights of My Recent Work as Director IITM

During most of this period, I was Director of the Indian Institute of Tropical Meteorology(IITM), Pune. IITM is an institute devoted to the study of Tropical Meteorology addressing a wide range of problems in weather, climate variability and change and for developing forecasting tools to be used by Indian Meteorological Department and other agencies. It has about 150 tenured scientists, about 150 project scientists and about 90 students.The areas of work include development of models at various scales, field observations to study clouds and their interactions with aerosols and studying climate change using models and observations. As Director of IITM I was also the Mission Director of Monsoon Mission - a mission mode project to improve forecasts at all scales and their applications to various sectors. Some of the highlights during this period are:

• Development of an ensemble based short range forecasting system.
  This system was developed and operationalised during my tenure. It has a global resolution of 12.5 km and was the highest resolution global forecast system at the time of its launch, and still continues to be one of the highest resolution global ensemble forecast sytems in the world. It has 21 members in its ensemble. This system was operationalised and is now being extensively utilised for forecasting high intensity events such as cyclone - their prediction & tracking , high rainfall events, movement of monsoon lows and depressions , etc The quality of forecast is comparable to the best in the world , especially over the Indian region. An Improved version using a new dynamical core (the octahedral dynamical core) at a resolution of about 6km is under development. Also as part of Monsoon Mission, I had initiated efforts for development of a new dynamical core based on icosahedral techniques. This research is being conducted in collaboration with scientists of IIT Kanpur and and University of Tokyo.


• Establishment of India’s first multi-petaflop High Performance Computing systems-
  a 4PF at IITM and a 2PF system at the National Centre for Medium Range Weather Forecasting (NCMRWF). This system , established in 2018, was the first Multi-petaflop system in the country and still is the fastest general purpose system in India. It has only recently been surpassed in speed by Param Siddhi (a special purpose computer for AI/ML applications). Work was initiated during my tenure for upgrading the systems to more than twice their present capacity. The present systems are being utilized for operational forecasts, for development of new forecast systems and research on various phenomena related to earth sciences.


• Establishment of the Atmospheric Research Testbed (ART) facility near Bhopal :
  This facility is dedicated to study clouds and thier properties. The measurements will help in understanding the structure and behaviour of lows and depressions passing over the core monsoon zone. Other meteorological and climate parameters will also be measured . This facility, set up over 100 acres was initiated during my tenure and established in 2019. It is equipped with a C-Band radar, a Ka-Band radar and other instruments. Many more instruments are being procured and installed. This is the first such Research Testbed dedicated to atmospheric studies established in the country and perhaps the first of its kind in the tropics. The data obtained from experiments conducted here will be valuable inputs that can be used for improving weather and climate forecast models.


• Weather Modification studies:
  An experiment was conducted to study weather modification through seeding of clouds over arid regions. The experiment titled ‘ Cloud Aerosol Interaction and Precipitation Enhancement Experiment (CAIPEEX) was conducted for two years, 2018 and 2019 at Sholapur (an arid region in south western Maharashtra). Two aircrafts , a seeder aircraft and a research aircraft were used. These were supported by a C-Band radar, a network of raingauges, wind profilers and other instruments to study cloud and aerosol properties over the region. . Both low-level seeding (hygroscopic seeding) and high-level seeding (glaciogenic seeding) were attempted to establish their efficacy in improving rainfall over arid regions.More than 480 hours of flying time was utilised. CAIPEEX also addressed issues related to randomization, physical evaluation, and numerical experiments of the cloud seeding effort. The analysis of these results is underway and will lead to a protocol for weather modification that can be used by state governments and other agencies.


• For the first time, India took part in the Climate Modelling intercomparison experiments - CMIP6, with an earth system model, the IITM-ESM. These were an input to IPCC-AR6 assessment report.


• A climate change assessment report for the Indian Region was prepared and published by the Ministry of Earth Sciences as an open source book through Springer-Nature. This document discusses issues related to climate change with specific reference to the Indian Region ..


• A high resolution air-quality forecast model with chemical data assimilation was developed and operationalised for the National Capital Region (NCR). It gives forecasts at a resolution of 400m and uses information from satellite and terrestrial sources for chemical data assimilation. This is being used extensively by various operational agencies. Recently a decision support for operational agencies has been developed to facilitate decision-making, especially during poor-air quality conditions. • Establishment of a Virtual Centre for AI/ML: A Virtual Centre for AI/ML specifically in the field of earth sciences was established in 2021. The centre will facilitate research in all fields of earth sciences for innovative use of AI/ML. I initiated the establishment of this virtual centre and currently research is being conducted to develop forecast models for rainfall etc.