I have developed algorithms for extraction of one or more signals embedded in multivariate time series. The concept of signal-to-noise is defined and optimized over a certain class of models. In this vein, I work closely within and around the domain known as linear dimensionality reduction, which includes methods like Principal Component Analysis and Multivariate Statistical Analysis.

I have also developed methods for extracting signals for incomplete multivariate time series, for example time series with different sampling intervals or missing data, a common feature of real life data sets. Here, I infer the missing data points while extracting a smooth underlying signal. This outperforms traditional algorithms for data imputation because we take advantage of the underlying time structure.

A paper called "Future Climate Emulations Using Quantile Regressions on Large Ensembles" that projects the current climate into the future with code available here.

I used to do research on extreme event characterization, modeling and forecasting using geological data like climate models and observations. For example, are days of extreme temperatures getting more frequent? Or is the annual temperature profile more variable with the current post-industrial climate forcings compared to the pre-industrial era? To answer these questions I look at climate model output, e.g. Global Circulation Models, using different starting conditions and see how their behavior changes statistically.

Related to this is also the desire to incorporate a more statistical framework with the current deterministic climate models. This involves for example adaptive grid sizes and time steps dependent on desired spatio-temporal resolution.