Evidence of Rapidly Warming Rivers in the UK from an Extensive Additive Modelling Study at the National Scale Using R

River water temperature data exhibit non-linear behaviour over the past 50 or so years. Standard techniques for identifying and quantifying trends have centred around the use of linear regression and Mann-Kendall and Thiel-Sen procedures. Observational data from UK rivers suggest that temperatures are far more variable then assumed under these statistical models. In a national-scale assessment of the response of riverine systems to global climatic change, an additive model framework was employed to model patterns in water temperatures from a large database of temporal observational data. Models were developed using R, which allowed for the deployment of cutting-edge additive modelling techniques to describe trends at 2773 sites across England and Wales, UK. At a subset of sites, additive models were used to model long-term trends, trends within seasons and the long-term variation in the seasonal pattern of water temperatures. Changes in water temperature have important consequences for aquatic ecology, with some species being particularly sensitive even to small shifts in temperature during some or all of their lifecycle. While there are many studies reporting increasing regional and global air temperatures, evidence for changes in river water temperature has thus far been site specific and/or from sites heavily influenced by human activities that could themselves lead to warming. Here I present selected results from a national-scale assessment of changing river water temperatures, covering the whole of England and Wales, comprising data from 2,773 locations. Positive trends in water temperature were observed at 86% of sites. At a subset of sites, seasonal trend models were developed, which showed that 90% of locations demonstrated statistically significant increases in water temperature during Autumn and Winter periods. Multivariate smoothers, that allow for within-year and longer-term trend interactions in time, suggest that periods of warmer waters now extend earlier into spring and persist later in the year during autumn. Here I (i) present details of the modelling exercise undertaken, (ii) discuss the deployment of sophisticated statistical models at such a large scale , (iii) note the issues involved with presenting and summarising the results of a large-scale modelling study to the responsible agencies in the UK, and (iv) comment on the role of open-source software in general, and R in particular, in the feasibility of undertaking modelling at such large spatial and temporal scales.