Geosciences Seminars

This presentation reveals that Himalayan hot springs are key to understanding the region's carbon cycle, with geochemical analysis showing that fault-controlled meteoric water circulation drives metamorphic CO2 degassing. Water penetrates to depths of ~5 km, creating pathways for deep-sourced CO2 to reach the surface. The hot springs release significant amounts of CO2, which is comparable to the region's silicate weathering sink. The degassed CO2 is primarily derived from metamorphic decarbonation (~78%) reactions, and can turn the orogen into a net source of CO2 on million-year time scales.

Transformers have rapidly become the backbone of modern day forecasting - from the language models reshaping how we interact with information, to Earth system predictors that now rival decades of numerical weather modelling. Yet despite this power, most transformer architectures remain agnostic to the physical structure of geoscientific data. They lack awareness of sensor identity, carry no representation of directed physical influence between variables, and impose no built-in sense of the diverse rhythms that characterise Earth system processes - from diurnal cycles and seasonal variability to longer climate modes such as ENSO. Despite this, the fact that models like GraphCast and Pangu-Weather achieve remarkable forecasting skill even without this inherent physical awareness speaks to the framework's fundamental strength. It also raises a pointed question: what becomes possible when that structure is built in deliberately?
We introduce CRISP, a 22 million parameter transformer architecture specifically designed in-house at PRL, for multivariate geoscientific forecasting. CRISP models observations as collections of distinct sensors - each with its own dynamics and directed, time-lagged influence on others, and constructs faithful context vectors so that every token encodes a single variable’s structured, lagged history. This guarantees physical homogeneity of tokens and simultaneously widens the temporal receptive field before any attention is applied. Temporal structure and sensor identity are encoded jointly through dual rotary positional encodings with learnable frequencies that adapt to dominant periodicities in the data. Finally, an explicit channel-mixing block learns asymmetric, directed cross-sensor interactions without imposing artificial symmetry, while sparse ProbAttention provides scalable long-range interaction across time.
Across multiple standard benchmarks, CRISP achieves state-of-the-art predictive skill relative to leading transformer, MLP, and state-space models and crucially, produces interpretable influence structures that allow Earth scientists to diagnose the physical relationships driving each prediction.

The diverse geochemical composition of Ganga-Brahmaputra floodplain sediments found in the inland aquifers of the Bengal Basin, and their impact on groundwater chemistry over a seasonal timescale, has been poorly investigated yet. This study combines seasonal groundwater geochemistry with isotope-based mass-balance modelling to constrain the controls on dissolved Sr and 87Sr/86Sr in depth-bound inland groundwater from the western Bengal Basin (West Bengal, India) located within the Ganga (Hooghly) floodplain. To further understand the spatial variations in the inland groundwater composition across the Bengal floodplain, the present groundwater database has been compared with available inland groundwater data from the eastern Bengal Basin (Bangladesh) located in both the Ganga and Brahmaputra drainage basins. The shallow groundwater in western Bengal Basin exhibits variable hydrogeochemistry and radiogenic 87Sr/86Sr with limited seasonal variability; however, showing higher solute loads during pre-monsoon period. Comparing with available geochemical data on the Ganga River bedload sediments-water, we suggest detrital radiogenic calcite dissolution (74 - 93%) from the Ganga floodplain sediments and a subordinate localized contribution from silicate mineral weathering may drive heterogeneous Sr release with radiogenic 87Sr/86Sr in the seasonal shallow groundwater, which may further undergo secondary interactions with exchangeable clay-sediment fractions.

The challenge in predicting monsoon is difficult to overcome by relying on instrumental data from only the past few decades. Palaeomonsoon reconstructions help us understand and predict the sensitivity and response of monsoon to forcings on multi-decadal-centennial timescale. Numerous palaeomonsoon reconstructions based on oxygen-18 in speleothems have been done in India to investigate the drivers of Indian Summer Monsoon Rainfall (ISMR). However, it can be difficult to deconvolve the δ18O signal into individual components, including possible kinetic isotope effects, temperature, precipitation amount, moisture source and transport, which can lead to an overstated climatic signal. In such a case, the extent to which the variability in ISMR in the instrumental period reflects natural variability, still remains debatable. In this seminar, I will discuss how triple oxygen isotope systematics can provide a way to identify kinetic effects and delineate the influencing processes. I will present results from in-house setup that have resolved the existing theoretical-experimental gap and interlaboratory inconsistencies, crucial in standardising triple oxygen measurements in carbonates. I will also present preliminary investigations of paired speleothem-dripwater samples from Indian caves.

The Pliocene was the last significant sustained warm period on Earth. Atmospheric carbon dioxide and global temperatures during this interval can be comparable to those modeled and proposed for the near future. Considering a similar continental and oceanic positioning to the present, it is possible to assume that the oceanic and atmospheric circulation patterns were also comparable to those of today. Current data on the Pliocene, and the transition to the cooling conditions of the Pleistocene, mostly arrive from marine archives, thus leaving the continental regions mostly barren of reliable and continuous information. Therefore, continental climate archives from this interval are highly valuable for comprehending the impact of climate change on terrestrial areas and serve as good analogues for understanding present conditions without the influence of humans. 
In the present study, a multi-proxy approach was applied to both sediment cores and outcrop samples retrieved from three different lacustrine formations exposed in the Near East, which are chronologically constrained to the Pliocene and Early Pleistocene. The multi-proxy analyses indicate major fluctuations in the lakes hypsometry, transitions from anoxic to oxic conditions in the sediment-water interface, and major changes in the limnic states, indicating a response of the lake systems to changing conditions in the precipitation patterns through time. It appears that the different lakes responded to orbital-scale forcing, which may have played a key role in governing the dry-wet climate cycles in the Near East. Results from this study provide an important understanding of the hydrological conditions that may have dominated the region during a warmer climate phase, challenging previous estimations, while concurrently providing clues to the role of the climate system in greening the path of early hominin migrations out of Africa through the region.