Homogenization and Quality Control of long time series of Daily Temperature in Uruguay

This paper describes the collection, quality control and homogenization of a Uruguayan daily temperature database. The object of the exercise was to create a database of daily temperature for as many stations as possible for Uruguay as far back as possible. Data sources included the Dirección Nacional de Meteorología (National Weather Service) and the Instituto Nacional de Investigaciones Agropecuarias (National Agricultural Research Institute). This project is the first attempt to gather all the different data sources together, and digitize, quality control and homogenize them. Much of the data was in paper form and had to be digitized. We work with 7 stations; note that Uruguay is a small country (around 177.000 Km2) so we used almost the total of the data available. There are 3 longest series, with start dates in 1930, and are located near the Uruguay river, the rest of the series start around 1950. Because the percentage of missing data at first was important, we try to recover it reading the thermograph bands, when these were available. The rate of missing value is 4% over the total of daily data (maximum and minimum temperature). Although, this percentage is low, the problem is that missing data are grouped in continuous days or months, which affect some homogeneity test. In many stations there are periods of lost data that they can reach to several months. Another big trouble we found was the poor metadata information at the National Weather Service, because this problem affects homogeneity test outputs.

For the quality control we firstly used the RClimdex 1.0 (available at http://cccma.seos.uvic.ca/ETCCDMI/software.html), which detect some common errors such as maximum temperature cooler than minimum temperature and outliers. Because we found some errors that this software doesn't detect, we applied another methodology. We calculate for each station and for the maximum and minimum temperature the 24-hr changes and establish a threshold of four times the standard deviation of the month of study. This procedure found out more erroneous data, these errors are mostly detected in maximum temperature (79%). The most common errors were: typing errors and confusion between maximum and minimum temperature. Particularly in one station we found a complete month with erroneous data in both variables, so perhaps they input another variable. Different homogenization procedures were used together with the aim of comparison. The Standard Normal Homogeneity Test (SNHT; Alexandersson, 1987), the Buishand Range Test (Buishand, 1982) and the Homogeneity tests proposed by Vincent et al, (2002) using regression models. For the SNHT and Buishand test, the selected tested variable was annual mean of the diurnal temperature range (mDTR) (J.B.Wijngaard et al.,2003), and in some cases we applied these test to mean annual maximum and minimum temperature series, the other test use as a tested variable the annual mean of maximum and minimum temperature separately. To better analyze inhomogeneities we broke up the total period in two parts when an inhomogeneity is indicated. Is important to mention that for Paysandú, for instance, the documented change of location in 1967, is no detected for anyone of the applied test but the changes detected around 1945-46 and 1955-57 were coincident for different tests indicating a real inhomogeneity. A change around the year 1945 is detected in two stations and it seems to be because of the maximum temperature. When the series will be used for climate analysis the inhomogeneities enhanced by the tests should be taken into consideration for proper results.