A comparative analysis of seven methods for the estimation of values for observations missing from temperature climate data series

Simple item page
dc.contributor.authorBenton, Ross Andrewen_US
dc.date.accessioned2024-08-13T00:08:09Z
dc.date.available2024-08-13T00:08:09Z
dc.date.copyright1990en_US
dc.date.issued1990
dc.degree.departmentDepartment of Geography
dc.degree.levelMaster of Science M.Sc.en
dc.description.abstractThe advent of low cost electronic data recorders for the collection and compilation of weather data for research and climate database purposes has brought with it the ability to collect vast amounts of relatively inexpensive data from areas where it is uneconomical to maintain manned stations. While these dataloggers are generally very reliable they will inevitably break down or will suffer from sensor failure. The result is a loss of data which must be estimated for many practical applications of the data. There are several methods available for the estimation of missing climate data. The selection of method is dependent on the nature of the estimated variable. This thesis compares the results of the application of seven different methods of estimating missing mean daily temperature data. All methods are applied to four data sets; one synthetically generated and three from different physioclimatic regions of British Columbia. Ten, twenty, and thirty percent of the data is removed from complete data series and the estimated data are compared to the original data. The seven methods used in this thesis can be grouped into three general categories: methods based on the series mean, ordinary least squares regression methods, and time series methods. The means based methods are the difference and ratio methods with the the absolute temperature being used with the latter method. Ordinary least squares regression and polynomial modelling are least squares regression methods. Lagged dependent regressor, autoregressive integrated moving average (ARIMA), and modified Kalman Filter models are used as examples of time series methods. Each of the various methods has advantages and disadvantages both in terms of ease of use and restrictions for application. Method application results were compared using several criteria. These included mean square error for least squares and time series methods and the Akaike's Information Criteria (AIC) for time series models. Final comparison for all methods was based on a comparison of estimated and actual data for all series. The difference between these two values was calculated for each estimated point and data were grouped into range classes. The distribution of estimated values within each class was used to compare the methods for each of the data sets. The results of the application of these seven methods to the different data sets showed no definitive 'best' method while a few methods proved to be significantly better than others. The commonly used difference method provides ease of use and average estimation results relative to the other methods. The methods based on ordinary least squares regression provide moderate to good estimation results with relative ease of use but the data violate presumptions upon which the methods are based. The three time series methods provide mixed results. The ARIMA and Kalman Filter method provide generally poor results and require sophisticated computer software to apply. The functional response of the ARIMA method is dependent on the experience of the modeller and thus further experience with this method may provide better results. The lagged dependent regressor method, on the other hand, provides good estimates with ease of use comparable to the ordinary least squares methods. The lagged dependent regressor method also accounts for the spatially and temporally correlated nature of the temperature data which the ordinary least squares methods do not.
dc.format.extent155 pages
dc.identifier.urihttps://hdl.handle.net/1828/17225
dc.rightsAvailable to the World Wide Weben_US
dc.titleA comparative analysis of seven methods for the estimation of values for observations missing from temperature climate data seriesen_US
dc.typeThesisen_US

Files

Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
BENTON_Ross_Andrew_MSC_1990_523060.pdf
Size:
35.2 MB
Format:
Adobe Portable Document Format
Download

Collections

Electronic Theses and Dissertations (ETD)
 
University of Victoria Libraries

We acknowledge and respect the Lək̓ʷəŋən (Songhees and Esquimalt) Peoples on whose territory the university stands, and the Lək̓ʷəŋən and W̱SÁNEĆ Peoples whose historical relationships with the land continue to this day.