Filling missing meteorological data with Computational Intelligence methods

Joanna Kajewska-Szkudlarek,Justyna Stańczyk,W. Zielinski,A. Michalski,B. Kazmierczak,L. Kuchar

doi:10.1051/itmconf/20182300015

Abstract

Estimates of temperature and humidity values at a specific time of day, from hourly to monthly profiles, are needed for a number of environmental, ecological, agricultural and technical applications, ranging from natural hazards assessments, crop growth forecasting to designing solar energy systems. In climatology, they constitute the basis for drawing conclusions about climate variability. Data used in such analyses should be complete and reliable. Therefore, effective methods for filling missing values are sought. The initial scope of this research is to investigate the efficiency of computational intelligence methods in filling missing daily temperature and humidity parameters values. For this reason, a number of experiments have been conducted with Artificial Neural Networks and Support Vector Regression using meteorological data from the city of Wroclaw in Poland. The performance of these methods has been evaluated using standard statistical indicators, such as Correlation Coefficient and Root Mean Squared Error. Finally, certain computational intelligence techniques are proposed that can be used to predict daily temperature and humidity values more accurately in order to fill the missing data.

Highlights

The automation of meteorological stations and the use of electronic sensors result in collecting vast databases, whose potential is often not used properly, as their analysis is strenuous and time-consuming
Such data are still more commonly analysed with use of various Data Mining methods, including Computational Intelligence methods, such as Artificial Neural Networks, Support Vector Machines or Evolutionary Algorithms
The aim of the study was to assess the possibility to use selected Computational Intelligence methods (Artificial Neural Networks, Support Vector Regression) for the homogenisation and completion of incomplete data series obtained from meteorological measurements

Summary

Introduction

The automation of meteorological stations and the use of electronic sensors result in collecting vast databases, whose potential is often not used properly, as their analysis is strenuous and time-consuming. In meteorology and climatology, such methods are used to develop weather forecasts, and to estimate the value of parameters that are difficult to measure (such as evapotranspiration) or to generate predictions of future values of meteorological elements based on the possessed data. One of their advantages is the fact that they do not require knowledge about correlations between data or about the existence of such correlations.

Objectives

Methods

Findings

Conclusion