Real-Time Mapping Systems for Routine and Emergency Monitoring

Defining Boundaries between Fairy Tales and Reality

A. Brenning (1) and G. Dubois (2)

(1) Leibniz-Centre for Agricultural Landscape Research (ZALF), Institute of Soil Landscape Research, Müncheberg, Germany (alexander.brenning@alumni.hu-berlin.de)

(2) Institute for Environment and Sustainability, Joint Research Centre Directorate, European Commission, Ispra, Italy (gregoire.dubois@jrc.it) 1. Introduction

Real-time analysis of data reported by environmental monitoring networks poses a number of challenges, one of which is the handling of point measurements of phenomena that display some spatial continuity. This is the case for many variables, such as atmospheric and aquatic pollutant levels, background radiation levels [Fig. 1], rainfall fields, temperature and seismic activity, to name but a few.

What these variables have in common is that the observations are usually interpolated, a step that is necessary to obtain maps showing continuous information in time and space. These maps can then be used for modelling or decision making. In order to allow real-time assessments and minimize human intervention in case of hazards and emergencies, these maps should be established quickly and thus automatically. In such extreme cases, one also would expect from the environmental monitoring system to trigger early-warnings.

Figure 1. Interface of the European Radioactivity Data Exchange Platform (EURDEP) showing monitoring stations reporting dose rates. Outliers can highlight errors or emergencies but not distinguish these without expertise: this is where automatic mapping for early warning is at stake.

2. From Interpolation to Hazard Classification

One of the main challenges of environmental monitoring for detecting emergency conditions [Fig. 2] consists of turning from interpolation to decision making, or from a continuous map to an early-warning. Technically, this step results in a classification problem. This can be formulated either non-spatially (are there any abnormal conditions?) or spatially (where are abnormal conditions?). In both cases, the performance of an environmental monitoring system has to be measured in terms of the probability of correctly detecting an abnormal situation (sensitivity) and of correctly identifying normal situations (specificity) [Fig. 3]. These have to be weighted according to the risk or cost associated with each kind of misclassification, false positive or false negative.

Figure 2. Estimated values of the monitored variable in situation of routine (above) and in the situation of an extreme event (below) characterised by outliers at the early stages of the event.

3. Conclusions

Real-time mapping systems for routine and emergency conditions need to be designed to predict ... the unpredictable, a task that is very challenging in a geostatistical point of view (e.g. automatic identification of anisotropies and statistical distributions). In addition to this, there is a need for a clear definition of performance criteria to be optimized for. While earlier studies mostly focused on the interpolation task behind emergency monitoring, we argue that stronger emphasis has to be put on the classification problem of hazard detection and decision making, which results in different performance measures.

Real-time mapping for emergencies, not a fairy tale but still science fiction?

In situations with a high risk in an emergency case, the monitoring system has to be optimized to produce the highest possible specificity at a fixed, high sensitivity of, for example, 99%, in order not to miss any situation with abnormal environmental conditions. However, in many applications of emergency monitoring systems, little data from abnormal situations is available, because these occur with a low frequency. Consequently, the estimation of the sensitivity at a high specificity will remain uncertain. Simulation studies may alleviate this problem if prior knowledge of the distribution of the data is available.

Figure 3. Schematic representation of processing steps in an emergency monitoring system

Figure 3