About
Water-driven extremes (floods, droughts, landslides) concentrate their damage in the places with the weakest forecasting and risk-mapping capacity. I combine machine learning with physics-based models to close that gap. Three running threads: mountain hazards, arid streamflow, and earth observation for sustainable development. Below is a short tour of each, with the relevant publications and a way to get in touch.
1. Mountain hydrology and hazards
Mountains concentrate water-driven disaster deaths and starve forecasters of the data needed to prevent them. I combine machine learning with physics-based models to map and forecast cascading hazards: rainfall that triggers landslides that block rivers that flood downstream towns. Selected work: cascading-hazard characterization on mountain terrain (Talchabhadel et al., 2023); rainfall-triggered landslide zonation for critical infrastructure (Gnyawali et al., 2023); explainable ML for groundwater potential in data-scarce mountainous regions (Dahal et al., 2023); flood-exposure assessment for schools (Bishui et al., 2026).
2. Arid hydrology and AI streamflow forecasting
Arid streamflow is hard to forecast: the systems are bursty, the between-event signals are weak, and the calibration records are short. I develop transformer-based and ensemble forecasting approaches that learn from satellite observations directly; the thread started during my PhD on Arizona basins and continues at KU. Selected work: ensemble streamflow forecasting with diverse loss functions (Dahal et al., 2026).
3. Earth observation for sustainable development
Climate-change mitigation gets the headlines; for most of the world, development is the live question: clean water, food, infrastructure, livelihoods. I use earth observation and machine learning to inform that agenda directly. Selected work: carbon and biodiversity stakes of climate-driven wildfires in Nepal (Dahal et al., 2025); urban agriculture as a sustainability lever (Pradhan et al., 2024); IPCC science applied to the SDG agenda (Pradhan et al., 2025).
Currently
Building data-driven streamflow forecasting and hazard tools at the University of Kansas (May 2026 – present). Active fellowship: AGU Thriving Earth Exchange Community Science Fellow on the Lumberton, NC flood project.
Open to collaborations on:
- mountain hazards and cascading-disaster modelling.
- arid streamflow forecasting with transformer-based and physics-informed models.
- earth observation for water, food, and infrastructure work in developing economies.
News
- May 12, 2026: Started as a Postdoctoral Researcher at the University of Kansas.
- April 10, 2026: Defended my PhD at Arizona State University.
- February 22, 2026: Coverage in The Times of India on school flood-exposure research (article).
- January 24, 2026: Lumberton Flood Dashboard launched (dashboard).
Find your way
- Publications: full list with links and citations.
- CV: appointments, awards, talks, teaching.
- Resources: open course materials and code (Stats, py4all, azwaters).
- News: updates and selected media coverage.
Reach me at geokshitij [at] gmail [dot] com or via Google Scholar.