A FRAMEWORK FOR SPATIOTEMPORAL WEATHER SENSOR DATA QUALITY by

In this prospectus, we investigate the impact of various data quality factors on the problem of determining data quality for observations from a given weather sensor data stream. Our problem is an offshoot of various research and development projects conducted at the Western Transportation Institute for the California Department of Transportation (Caltrans) in relation to their Road-Weather Information Systems (RWIS). It has been challenge for Caltrans to assess the quality of the data from their RWIS units, in addition to field calibration and ground-truthing. We have generally employed an approach of using data from third-party providers for comparison against the RWIS data. However, this approach is dependent on multiple quality factors which one of our project champions characterizes as the trio of accuracy, timeliness and reliability. Is the data correct? I.e., is it an accurate measure of the real condition it represents? Do we receive it in a timely fashion? I.e., can we obtain an observation soon after it was recorded in the field? And, can we reliably gain access to this data? For instance, do the communication networks and computer systems that provide the data perform reliably or are they prone to outages? All of these quality factors can have an impact on quality control processes that we implement. Specifically in our comprehensive exam presentation we will investigate the example application of ordinary-kriging to the problem of assessing and quantifying the quality of individual sensor observations. On the surface, ordinary-kriging appears to be a good choice for this task. Given a set of known observations and associated locations, we can estimate values for additional locations using ordinary-kriging as an interpolator. In a typical application of ordinary-kriging, one would estimate values at locations for which observations are not known. In our application, we will hold-out our known observation and compare it to the prediction given by ordinary-kriging. Further, since ordinarykriging provides confidence intervals, our comparison can be expressed in terms of these confidence intervals, and if the observation falls outside of the confidence interval, it might be rejected as being bad. However, as we will demonstrate, the quality factors mentioned above all can contribute performance challenges to our application of ordinary-kriging. As our proposed research thesis, we will present the components of a general framework for spatiotemporal sensor data quality.