Scottish Sensor Systems Centre

University: Prof E Marian Scott, Professor of Environmental Statistics, University of Glasgow, Dr M F Villoria, University of Glasgow

Collaborative Partner: Scottish Environment Protection Agency (SEPA)

Project Type: Investigation (£18K Grant)

Project Duration: Five months, Feb - June 2013

Project Abstract

This project will embed statistical tools for the modelling of high temporal frequency data derived from recent deployment of semi-continuous water quality monitors installed by SEPA. It is a significant extension of an earlier project on modelling water quality, part of which was also supported by S3C, in a river network using routinely collected data (bi-monthly/monthly intervals), but without focus on events or interventions. This project will make use of new technology delivering data at 15 minute intervals.

Continuous and semi-continuous monitoring systems generate data with a very high temporal sampling frequency, where unusual events and extremes, are characteristics of interest to both parties in this project. Modelling such data to address questions concerning the effects of interventions is of particular importance to SEPA, and requires complex statistical tools to account for the strong temporal dependence in the system.

Modelling extreme values (whether very high or very low) is a specialised area, with specific models developed for dealing with block maxima (minima) or peak over threshold approaches. Dealing with the temporal and spatial extremes is computationally intensive: the University of Glasgow has considered several approaches, and will explore one in this project, namely quantile regression.

Data analytic tools of quantile regression allow estimation of the relationship between response and explanatory variables at any percentile of the distribution of the response (conditioned on the explanatory variables). As a result, rates of change in the response variable can be estimated for the whole distribution and not only on the mean.

These techniques of quantile regression and intervention analysis being developed at Glasgow are very powerful, computationally intensive and ideally suited for data of this nature, but currently are not routinely in use within SEPA.

The focus of this project is therefore the modelling and visualisation of the data from a prototype series of semi-continuous monitors across the network of Scottish rivers, focusing on river flow, water quality and temperature, which have then been replicated at multiple locations (the spatial ‘frequency’ is of very much lower resolution than the temporal frequency) and for which there is a known intervention. Event analysis and drivers of and impacts resulting from such events will also be explored.

S3C Project Impact

Increasingly more environmental monitoring is being undertaken using automated sensor technologies collecting digital data at relatively high temporal frequencies (15 minute data vs monthly sampling), and with the potential for increased spatial coverage in sampling.

SEPA is keen to develop improved approaches to capturing, processing and synthesising these large data volumes in order to extract useful information from these monitoring networks, and is specifically interested in:

•Where its measures aimed at improving environmental quality are most effective and why
•What constraints and controls the physical environment may place upon the effectiveness of SEPA’s intervention measures.
•How to improve the ways in which SEPA models such high frequency data

These kinds of data handling challenges are particularly important in environmental reporting, but have wider relevance and applicability. The data analytic tools associated with this project are generic and transferable across many sectors, offering potential for impact in a number of broader fields, e.g., addressing ‘big data’ challenges in space, weather, energy, and power systems.