ML Hydrologist, Flood Forecasting

About the Role Global ML flood models now outperform traditional approaches on large rivers, but they still break down in the small, fast catchments where flash floods happen. Riverkin, a VC-backed ETH Zürich Spin-off, is closing that gap. We start from hydrological models trained across thousands of catchments worldwide and make them accurate locally, using a sensor network we deploy ourselves in exactly the places public datasets don't reach. We are looking for an ML hydrologist to own that work. You will report to the CTO and work directly with our other hydrologist, who covers rainfall radar and short-term nowcasting. Why this is interesting Data nobody else has: Our own field-deployed sensors give you live streamflow observations in small, previously ungauged catchments. An open problem: Whether large-sample models can be made reliable at flash-flood scale is unsolved, and you would be working on it with real data and real users. Built to protect lives and critical infrastructure: Your models will power early flood warning systems safeguarding hydropower plants and giving first responders more time to protect communities Publishing and open source: We encourage publishing results and contributing to open tools such as NeuralHydrology and Caravan. Compute: GPU resources are available for training and experimentation. What You'll Do Bring global flood models to the local scale Take state-of-the-art hydrological ML models (LSTMs, Transformers) trained on large-sample datasets and adapt them to the catchments our customers care about. Use our own sensor data alongside public gauge, reanalysis and radar data to fine-tune and validate those models locally. Tackle prediction in ungauged and data-scarce basins through transfer learning and regionalisation. Own hydrological modelling end to end: set the technical direction, decide when to adapt pre-trained models or train our own, and lead the work from experimentation through validation to production. Work as one modelling team Make practical trade-offs between forecast accuracy, model complexity and development effort to build a reliable product that meets customer needs. Work side by side with our hydrologist specialising in rainfall radar and nowcasting to directly integrate rainfall nowcasts into your flood models. Work with software engineers to put models into production and monitor their performance. Identify and test advances in ML flood forecasting, turning promising research into more accurate and reliable forecasts for our customers. Stay close to the real world We believe the best modellers understand the world behind their models. We encourage you to join field deployments and customer conversations to build that understanding firsthand. You are an opinionated contributor to key technical decisions. Who You Are You’ve built and validated hydrological ML models across many catchments and have experience taking modelling work into real-world use. You combine scientific rigour with product judgement: you test assumptions, understand uncertainty and focus on improvements that matter to customers. You take ownership from an open question to a working solution. You set priorities, make decisions and keep moving without waiting for perfect information. You care deeply about the quality of your work. You dig into why models fail, follow problems through and stay invested in how your forecasts perform after deployment. You communicate clearly in writing and in person, can explain your reasoning and trade-offs, and work closely with people across hydrology, software, hardware and field operations. Technical Requirements Hands-on experience training or fine-tuning ML streamflow or flood models across many catchments at once (regional, continental or global), for example on CAMELS or Caravan. Experience with single-catchment models alone is not sufficient. Experience transferring such models to new or data-scarce catchme