This paper presents the results of testing the homogeneity of the basic data used in this study, that is, the mean monthly maximum and minimum temperature (as derived from daily observations) of several long-term climatological stations of Venezuela and Colombia, and an evaluation of the fluctuations and deviations from long-term means of such parameters, as well as of the diurnal temperature ranges (DTR) during the 73-yr period, 1918–90. The 14 longest-term climatological stations in Venezuela and Colombia with available metadata and the most consistent temperature records were used. Homogeneity tests were done by application of the Alexandersson’s Test. Eleven stations (the ones with less inhomogeneities) were finally left for partial data adjustments of their records when needed (only a few of them), following the technique developed by Easterling and Peterson. Then, the temperature trends and deviations from long-term averages (1918–90) were determined. The most remarkable feature of the analysis is a relative and sustained increase of the minimum temperature series and a DTR decrease during the last 25 yr of the period at a significant rate in most of the station records.
1. Introduction and previous works
Most studies of the climate and its fluctuations have focused on the middle latitudes of the Northern Hemisphere (Jones 1988; Jones et al. 1986a,b, 1990; Karl et al. 1986, 1993a,b; Karl and Jones 1990), and few have included the South American continent. The scarcity of long and homogeneous records in this area is a real problem in studying regional climates and their fluctuations.
This paper presents both the results of testing the homogeneity of the basic data used in this study, the mean monthly maximum and minimum temperatures of the northernmost South American countries, Venezuela and Colombia, and after that, an evaluation of the fluctuations and variations of the temperature (maximum, minimum, and temperature range) during the period beginning around 1920 throughout 1990.
Many recent studies have used monthly maximum and minimum temperature data that are derived from the daily observations, and the mean monthly diurnal temperature range (DTR), defined as the difference of the mean monthly maximum and minimum temperatures (Karl et al. 1984; Karl et al. 1993a,b; Easterling et al. 1997). A first indication that there might be important large-scale characteristics related to the change of mean and daily maximum and minimum temperatures was reported by Karl et al. (1984) who pointed out a noticeable decreasing of the DTR at a statistically significant rate at many rural meteorological stations across North America.
In the 1990s, after intergovernmental agreements, more studies were made using maximum and minimum temperature analysis (IPCC 1990, 1992). Such studies have been made in the People’s Republic of China, southeastern Australia, the former Union of Soviet Socialist Republics, and again in the contiguous United States, where a general widespread decrease of the DTR over these huge land masses of the world has been reported. Most recent analyses have included Sudan, South Africa, Egypt, Kenya, and Uganda in Africa; eastern Australia, Japan; Denmark; northern Finland; Pakistan; and some Pacific island stations; as well as in a few long-term individual stations in Europe (Karl et al. 1993b). Easterling et al. (1997) provided an update to increase the analysis to 51% of the global landmass.
Through this study, Venezuela and Colombia, a couple of countries in tropical South America, are analyzed for differential changes of the maximum and minimum temperatures and DTR. This entire tropical region (Fig. 1) counts for somewhat above 2 × 106 km2, lying (roughly) between meridians 59° and 80°W and parallels 4°S and 13°N. Easterling et al. (1997) included Chile and Argentina, also in South America.
The monthly maximum and minimum temperature records were provided mainly by both the Venezuelan and Colombian Meteorological Services: Servicio de Meteorología de la Fuerza Aérea de Venezuela (SMFAV 1993) and the Instituto Colombiano de Meteorología, Hidrologia y Adecuación de Tierras (HIMAT 1993). Other sources of data such as the World Weather Records 1941–60 (1966) and the Great Britain Meteorological Office’s Reséau Mondial 1910–34 (Great Britain Meteorological Office 1957) were also used.
A network of 41 stations was chosen for initial evaluation, and based on the amount of missing data, 14 stations were retained for homogeneity testing (Table 1). The data for these stations were then subjected to the Alexandersson’s (1986, 1995) test of homogeneity, prior to evaluating the temperature and the DTR deviations in the area. Figure 1 shows also the location of the 14 stations. All metadata (station histories) for the selected stations were provided as well, which proved to be extremely useful for verifying the results obtained from the Alexandersson’s test.
3. Procedure and methods of analysis
Very few (if any) long-term datasets are free of inhomogeneities; therefore the term relative inhomogeneity was used in this work. This means that the changes in the biases are small relative to any detected signal of climate variation or change (Karl et al. 1993a).
Several tests have been developed to detect inhomogeneities in climatological time series. Easterling and Peterson (1992) ranked some of these tests; they found that the Alexandersson’s test (1986) was very effective in detecting significant inhomogeneities, but they also noted a limitation of the test in finding multiple discontinuities. To address this issue, a slight modification of the method is introduced.
In the works of Alexandersson (1986, 1995), Easterling and Peterson (1992, 1995), and Peterson and Eastering (1994), the authors formed a series of reference station data of mean values by averaging a rather large group of station data. They argued that it is risky to use data for only one “good” station because the information may have inhomogeneities of its own, so that when compared to the station’s information under study, it would be difficult to determine correctly which station’s data have the inhomogeneities. But having a large group of stations with rather complete and long time series is very unlikely in the area under study, northern South America. Hence, using one station at a time as a candidate reference versus each of the rest for a geographically similar area (and picking out the common breaks or inhomogeneities) was more realistic and effective; so each station pair was tested individually. It was not considered convenient to test for homogeneity mixing the Venezuelan and Colombian stations together as these are quite different geographical areas, with particular and regional climate characteristics (Quintana-Gomez 1986, 1991).
The Alexandersson’s test was applied to annual values and month series that best represent the area’s prevailing year-round weather conditions of dry and rainy periods:January and July in Venezuela and February and August in Colombia. Therefore, for each of the six datasets (January, July, Annual) × (Tmax, Tmin), the test was carried out for each of the 9 × 8 pairs of stations for Venezuela’s nine stations. A score of 1 (or 0) was assigned if the test was (or was not) significant at the 95% level. The total scores were then added for each station and recorded in the last column of Table 1 (IH score) as a % of the maximum score (inhomogeneities) possible, 8 × 6, for Venezuela. Stations showing an inhomogeneity score over 25% were discarded (Maracaibo, Coro, and Maracay; Table 1). The stations with the most homogeneous records were Maturin, Colonia Tovar, Ciudad Bolivar, Barquisimieto, San Fernando, and Mérida; they were then selected for further analysis.
A similar scoring was also carried out for the Colombian data. None of this data was discarded as the maximum score was 17% (Table 1).
The stations with fewest inhomogeneities in their records (after the test) were selected. These stations’ time series were then chosen for analyzing the trends and deviations of temperature and DTRs in the target area.
4. Analysis of results
The Alexandersson’s test was used to check for the homogeneity of all possible difference series from each preselected station versus each of the rest for all common years in the region. The inhomogeneities were considered to be significant if the test statistics exceeded the values of 8.40, 8.55, 8.78, and 8.90, which are the test’s 95% critical levels for n = 40, 50, 60, and 70 (years), respectively (see Alexandersson 1986).
According to Alexandersson (1986), series with significant breaks or inhomogeneities that occur either during the first or the last five values (years), may or may not be homogeneous, because the mean comparison is very biased; therefore, all series that were found to have breaks during the starting and ending 5 yr were not considered.
Although the maximum and minimum temperature arrays showed dissimilar results regarding the position of the break points, it was rather common to find that the same break year either in the maximum or minimum groups seems to repeat in more than one set, that is, in the January–July–Annual (or February–August–Annual) arrays. There does not seem to be any relationship between the break point years of maximum and minimum temperatures in the Venezuelan and Colombian stations. A similar situation was also found and reported by Keiser (1994) in the northern Great Plains of the United States.
Since the Venezuelan and Colombian Meteorological Services provided not only the data but the stations’ metadata as well, this information was extremely valuable to verify the Alexandersson test results; for instance, Maracaibo, the Venezuelan station with the greatest number of inhomogeneities, reported three different locations, a transfer from a former meteorological bureau to the current agency, and at least a whole renovation of all its climatological instruments. The 1969 and 1972 breaks coincide closely with one of the reported station relocation changes (1972) and with the year of instrument change (1969). However, a major potential cause of inhomogeneities in climatological records, a transfer or change from one meteorological service to another (change of observers, observation times, procedures, and methods, etc.) that occured at the Maracaibo station in 1949 (as well as at the remaining eight Venezuelan stations during the lapse 1946–50) was not detected at all, either in Maracaibo or in the rest of the Venezuelan stations. Some suggestions as to why this is the case will be discussed later.
Six out of the nine Venezuelan stations have reasonably homogeneous records; five of them have moved only once to nearby airports. These “best quality” stations were then selected to evaluate the trends of temperature in Venezuela (see Fig. 2, ID# 1, 2, 4, 7, 8, and 9 in Table 1). In Colombia, some stations’ records have “gaps” due to missing data in the series, some of them with up 4 or 5 continuous missing years, but since all inhomogeneities in Colombia were detected just after the end of a record gap, precisely at the restarting year, these time series were not considered inhomogeneous and were used for trend analysis of temperature in Colombia. See Fig. 2 for names and locations on map of the 11 homogeneous stations in both countries.
Since the test did not show any significant discontinuity in the Venezuelan stations’ data prior to 1950, in spite of the stations transfer to the current meteorological agency (SMFAV), it was considered expedient to examine the records for singular subperiods under other criteria. A simple way was by comparing decadal averages of the series mean values. Tables 2 and 3 are the decadal averages of the series mean values of annual maximum, minimum, and DTR for the selected stations of Venezuela and Colombia. From examination of the averages, it is evident in all six Venezuelan stations that the temperatures are more extreme before 1950. Therefore, it is perhaps reasonable that the test did not show significant discontinuities since it is designed to detect breaks in time series that “run” differently along common periods, and if all selected stations in Venezuela had a similar pattern prior to 1950, the test would not detect any discontinuity. . .
Several t-test trials were then made to determine whether the differences in the annual means from those decadal subperiods were significant at the 0.01 critical level. The comparisons showed that all the differences in means were significant. However, this could have been due to inhomogeneities in the time series or climate “anomalies” or both.
To ensure that no biasing occurs when testing and using the stations’ data, some records were adjusted when needed. The method of adjustment was similar to Easterling and Peterson’s (1995); that is, once the breaks were determined in some particular stations’ records (in accordance with the introduced slight modification of the Alexandersson’s test described earlier), the means of the difference series created between the candidate station (each station) and a like form of the average obtained from the nearest and most highly correlated stations (in both Venezuela and Colombia, separately), immediately before and immediately after the discontinuity, were computed. Then the difference between these two means (adjustment factor) was added to or substracted from each value of the original (candidate) series before or after the discontinuity. Finally, the deviations in each station for all recorded years from the reference period (mean values), 1918–90 for Venezuela and 1923–90 for Colombia, were calculated from the adjusted annual and monthly series.
The annual departures from the averaged maximum, minimum, and DTR are shown in Figs. 3 and 4. One of the most remarkable features of the graphs is the sustained increase of the minimum temperature in all of the stations, as well as a decrease of the DTR during the last 25 yr of the period. A reduction of the maximum temperature values is also observed in most of the stations (especially in Colombia). The significance of these tendencies (t test) and the values of the deviations are presented in Table 4, which gives the difference between mean annual maximum and minimum temperatures and DTR at each of the selected 11 stations for the subperiod 1966–90 and that for that for the previous rest of the period starting around 1920 and ending in 1965.
5. Results and conclusions
The Alexandersson’s method to detect inhomogeneities in climate time series was useful in this research. But not all possible discontinuities (from the stations’ metadata) were detected with the test. In the process of adjusting values, although the adjustments could be mathematically correct, it is possible that some physical inconsistences have been introduced in the stations’ records. In reality, the unadjusted time series represents the conditions (changes) due to local influences (natural or artificial) at each individual station and the adjusted time series represents the regional climate signal (see Easterling et al. 1996).
The annual mean minimum temperature series in both Venezuela and Colombia shows a statistically significant increase (t test) starting approximately by the middle of the 1960s throughout the end of the series in 1990. This trend is higher in all stations of Colombia compared to Venezuela’s. The DTR series shows a sustained decrease during this interval in 10 out of the 11 stations, which is also more pronounced in Colombia than in Venezuela, mainly in the stations of higher elevation (mountain) of both countries.
The relative warming of the last 25-yr period of the series could be the result of a number of factors. One possibility that can lead to a rise in minimum temperature and decrease in the DTR is an increase in cloud cover (Karl et al. 1993b). In fact, according to the author’s most recent investigations on evaporation and horizontal visibility in part of Venezuela (Quintana-Gomez 1998), an increase in cloudiness year-round has occurred, particularly over mountainous regions. Another important factor concerns urban heat islands due to changes in the environment around the stations, with special reference to the urban growth and land use, as reported by Jones et al. (1990). This situation is likely to have occurred in some of the stations since they are not located in predominantly rural areas. Other factors like aerosols and greenhouse gases, which have been observed to increase in many areas of the world, might have also been affecting Colombia and Venezuela since both countries have developing industrial zones with several million vehicles.
The author thanks the Department of Meteorology of Texas A&M University for the opportunity and facilities to achieve this work; special thanks go to Prof. John F. Griffiths for his valuable suggestions and comments. This gratitude goes also to the Servicio de Meteorología–FAV de Venezuela and the Instituto Colombiano de Meteorología, Hidrología y Adecuación de Tierras–HIMAT de Colombia. Particular thanks are also given to the reviewers of the manuscript, especially reviewer A and Dr. Peter Lamb, editor emeritus, for their sincere help, valuable suggestions, and comments on behalf of this article; all are highly appreciated by the author.
Corresponding author address: Dr. Ramon A. Quintana-Gomez, Universidad Nacional Experimental de los Llanos Ezequiel Zamora, UNELLEZ, Barinas, 5201 Edo. Barinas, Venezuela. E-mail: email@example.com