Author(s)

G Latini, G Passerini & S Tascini

Abstract

In this work we present a new software tool we developed, aiming to collect and systematise our advances on statistic filling of time-series gaps. Starting from considering our air quality data sets, collected by a poor monitoring network, we found the necessity of solving ambiguity and non-homogeneity problems due to inconstant sample number as well as data fragmentation among time series. Considering literature protocols, we tried to define a gap classification based on its time lapse. Problems have been found for gaps larger than 4 hours since they are not generally discussed. In order to approach this aspect we tried applications of spatial algorithms (such as nearest neighbour or smooth-fill) and neural networks which seemed to give excellent results for gaps wider than 8 hours. On this preliminary study we achieved a classification based upon three branches: small gaps (up to 4 hours); medium gaps (5-8 hours) and large gaps (9-50 hours). Once a gap has been recognised, a filling algorithm is selected and applied by the main module choosing among: linear interpolation, time matrix interpolation and neural networks based algorithms. Since a data base interface has been developed, we are able to use this software to fill up the most part of our incomplete series, and we can actually consider to extend the width of recoverable gaps.

Keywords