Marina Knight



I completed a PhD in Statistics in 2006 at the School of Mathematics, University of Bristol, and then held a post-doctoral research position sponsored by GCHQ also at Bristol, until early 2008.  From early 2008 until mid 2011, I took a career break to fit with my very young family, and worked part-time as a medical statistician for NHS Blood and Transplant. In August 2011 I returned to academia as a post-doctoral researcher, working on an exciting EPSRC funded project on the modelling and forecasting of energy-related non-stationary time series. In September 2013 I joined the Department of Mathematics at the University of York as a lecturer in Statistics.

Departmental roles

University roles

Member of the University Impact Leads for Science Departments Group



My main research interests are second generation wavelet techniques for analysing irregularly sampled data (often termed lifting schemes) and modelling of time series whose characteristics evolve through time (often referred to as locally stationary time series).

Work in the second generation wavelet area led to improved precision in nonparametric regression estimation and found applications in the field of bioinformatics, as well as offering a practical, viable solution to the often encountered question of `Which wavelet should I use?’. The interested reader can use the associated R packages ‘adlift’ and ‘nlt’ on CRAN at, respectively

Work on locally stationary time series introduced new methods for spectral estimation in the presence of missing data with practical applications, e.g. paleoclimatic data. Both real- and complex-valued wavelets have been proposed, with the complex-valued construction allowing for phase and coherence information to be represented when dealing with nonstationary bivariate time series sampled over non-uniform designs. A recent wavelet lifting strand is the lifting-based reliable estimation of long memory parameters for processes that exhibit large-scale dependence (its associated R package ‘liftLRD’ can be found on CRAN at The forecasting of non-stationary time series has been explored through a time-scale localised partial autocorrelation function that encodes non-stationarity and accurately assists the difficult task of forecasting locally stationary time series.

Recent co-authors: Seth Davis, Idris Eckley, Piotr Fryzlewicz , Rebecca Killick, Guy Nason, Matt Nunes, Jon Pitchford, Jean Sanderson.

Current PhD student (joint with Pitchford and Davis): Jess Hargreaves.

Research group(s)

Statistics and Probability Research Group

Available PhD research projects

My projects span the fields of multiscale methods, nonparametric statistics, analysis and forecasting of non-stationary time series, in particular for data collected on irregular and spatial structures. Particular application fields of interest include energy and biology.




Generalised Linear Models


Advanced Regression Analysis

Contact details

Dr Marina Knight

Tel: +44 1904 32 4166