Türkiye'de Yıllık Toplam Yağışların Homojenlik Analizi

thumbnail.default.placeholder
Tarih
1996
Yazarlar
Özçelik, Dilek
Süreli Yayın başlığı
Süreli Yayın ISSN
Cilt Başlığı
Yayınevi
Fen Bilimleri Enstitüsü
Institute of Science and Technology
Özet
Türkiye'deki istasyonlar için oluşturulan yıllık toplam yağış verilerinin homojenlik analizi sonuçlan sunulmaktadır. Temel veri, 1951-1990 periyodunda kayıt uzunlukları 40 yıl olan 84 meteoroloji istasyonunun günlük değerlerden aylık ve yıllık toplamların elde edilmesi ile oluşturulmuştur. Burada kullanılan metod, bağıl homojenlik tekniklerini kullanarak seri içerisindeki homojen olmayan kısımları tespit edebilen verimli bir yöntemdir. Metod, parametrik olmayan Kruskal-Wallis homojenlik, Swed-Eisenhart run testlerini ve grafiksel analizi içermektedir. Yöntemin 84 aday istasyonla yüksek korelasyonlu komşu istasyonlarla oluşturulan çiftlere uygulanması neticesinde, 69 istasyonun homojen veriye sahip olduğu gösterilmiştir.
It has long been recognized that inhomogeneous climate time series may lead to biased results in climatic change studies [Kohler, 1949; Conrad and Pollak, 1950]. Inhomogeneities that tend to offset each other when studies are conducted by generating grided data with grid sizes comparable to those used in current GCM studies are much less likely to do so when smaller station networks are used to detect regional climate change. The series with several inhomogeneities reflect a serious source of uncertainty in studies of climatic trends and climatic change. The idea behind the powerful relative homogeneity tests for precipitation is to use logarithmic ratio series between candidate station and its neighboring stations. Neighboring stations which have high correlation of annual anomalies of precipitation and small year-to-year variances of these anomalies are suitable for detecting inhomogeneities at the candidate station. Many homogeneity tests are of a type that gives no information about the probable date for a shift in total and no information about the magnitude of the break. Examples are the Kruskal-Wallis homogeneity test and Swed-Eisenhart runs test which are tested for efficiency in this study and applied to the precipitation series with the graphical relative homogeneity analysis. The homogeneity analyses can be classified in two groups. The fist group includes the analysis of station's climate record without explicitly considering station history information [ Potter, 1981; Mitchell et ai, 1966; Alexandersson, 1986; Gullet et al, 1990]. Others use the available station history information [ Karl and Williams, 1987]. Data are necessary to carry out any study dealing with the climatic changes. Therefor, before any statistical technique can be applied to a dataset, it must be ascertained that the data are homogeneous. The methodology includes non- parametric Kruskal-Wallis homogeneity test, Swed-Eisenhart runs test and graphical analysis application to the annual total precipitation series of the highly correlated stations. Pearson correlation matrix between those stations are formed and for each station the highest correlated ones are taken as reference. In the past, objective statistical evaluation of the homogeneity of climatic time series has been confined largely to the evaluation of relative homogeneity of two or vm more series, considered one pair at a time [Conrad, 1925; Conrad and Pollak, 1950] and the application of a randomness test. Today relatively a wider array of suitable statistical tests are available such as distribution free Kruskal-Wallis homogeneity test and Swed-Eisenhart runs test. An important problem encountered here is how to form station pairs so that relative homogeneity comparisons can easily be made with high confidence level. It is intuitively clear that an inhomogeneous series can be detected with the highest possibility when the stations are geometrically closer to (or rather with largest correlation to and minimum year-to-year variability to) the actual test site. Thus, Pearson correlation matrix between those stations are formed and for each station the highest correlated ones are taken as reference. The main difficulties encountered in homogeneity analysis are: 1) The fact that no series is perfect is always valid and there may be minor inhomogeneities that were not documented and cannot be detected by the statistical tests. 2) Ambiguous conclusions are possible when several neighboring stations do have inhomogeneities themselves. In order to overcome or reduce the effect of these problems one must identify and use as many of the candidate station's nearest neighbors as possible. The use of at least several nearby stations is very labor intensive and requires a substantial quantity of computer compatible data, particularly for cases where station inhomogeneities are not well documented as it is in Turkey. Analyzing Kruskal-Wallis homogeneity and Swed-Eisenhart runs tests with 2 generated time series: The mimic our case studies, we generated pairs of artificial time series of 40 years of length. The first element of the pairs is kept the same. The second element is 'contaminated' with an inhomogeneity of some sort (jumps, trends, or different U levels). These series are tested by Kruskal-Wallis homogeneity test and Swed- Eisenhart runs test, each time the second series including a different errant data property. One important property of two series is that they are not composed of the identical random numbers ( but note that both series have 0 and 1 standard deviation). Thus, when the second series includes no impurities, the results of both tests, note also that the results may very a little bit depending on the relative place of introduced jump in the series. Series production can be accomplished in two steps by: 1) Forming series by generating 'good' random numbers. For the sake of the testing scheme, the 'good' numbers generated by the computer program must be entirely free from any squential correlation and must be entirely unpredictable. A 'portable' random number generator ( which can be programmed in a high level language, and IX ÖZET Çalışmada, Türkiye'deki istasyonlar için oluşturulan yıllık toplam yağış verilerinin homojenlik analizi sonuçlan sunulmaktadır. Temel veri, 1951-1990 periyodunda kayıt uzunlukları 40 yıl olan 84 meteoroloji istasyonunun günlük değerlerden aylık ve yıllık toplamların elde edilmesi ile oluşturulmuştur. Burada kullanılan metod, bağıl homojenlik tekniklerini kullanarak seri içerisindeki homojen olmayan kısımları tespit edebilen verimli bir yöntemdir. Metod, parametrik olmayan Kruskal-Wallis homojenlik, Swed-Eisenhart run testlerini ve grafiksel analizi içermektedir. Yöntemin 84 aday istasyonla yüksek korelasyonlu komşu istasyonlarla oluşturulan çiftlere uygulanması neticesinde, 69 istasyonun homojen veriye sahip olduğu gösterilmiştir. vıı more series, considered one pair at a time [Conrad, 1925; Conrad and Pollak, 1950] and the application of a randomness test. Today relatively a wider array of suitable statistical tests are available such as distribution free Kruskal-Wallis homogeneity test and Swed-Eisenhart runs test. An important problem encountered here is how to form station pairs so that relative homogeneity comparisons can easily be made with high confidence level. It is intuitively clear that an inhomogeneous series can be detected with the highest possibility when the stations are geometrically closer to (or rather with largest correlation to and minimum year-to-year variability to) the actual test site. Thus, Pearson correlation matrix between those stations are formed and for each station the highest correlated ones are taken as reference. The main difficulties encountered in homogeneity analysis are: 1) The fact that no series is perfect is always valid and there may be minor inhomogeneities that were not documented and cannot be detected by the statistical tests. 2) Ambiguous conclusions are possible when several neighboring stations do have inhomogeneities themselves. In order to overcome or reduce the effect of these problems one must identify and use as many of the candidate station's nearest neighbors as possible. The use of at least several nearby stations is very labor intensive and requires a substantial quantity of computer compatible data, particularly for cases where station inhomogeneities are not well documented as it is in Turkey. Analyzing Kruskal-Wallis homogeneity and Swed-Eisenhart runs tests with 2 generated time series: The mimic our case studies, we generated pairs of artificial time series of 40 years of length. The first element of the pairs is kept the same. The second element is 'contaminated' with an inhomogeneity of some sort (jumps, trends, or different U levels). These series are tested by Kruskal-Wallis homogeneity test and Swed- Eisenhart runs test, each time the second series including a different errant data property. One important property of two series is that they are not composed of the identical random numbers ( but note that both series have 0 and 1 standard deviation). Thus, when the second series includes no impurities, the results of both tests, note also that the results may very a little bit depending on the relative place of introduced jump in the series. Series production can be accomplished in two steps by: 1) Forming series by generating 'good' random numbers. For the sake of the testing scheme, the 'good' numbers generated by the computer program must be entirely free from any squential correlation and must be entirely unpredictable. A 'portable' random number generator ( which can be programmed in a high level language, and IX ÖZET Çalışmada, Türkiye'deki istasyonlar için oluşturulan yıllık toplam yağış verilerinin homojenlik analizi sonuçlan sunulmaktadır. Temel veri, 1951-1990 periyodunda kayıt uzunlukları 40 yıl olan 84 meteoroloji istasyonunun günlük değerlerden aylık ve yıllık toplamların elde edilmesi ile oluşturulmuştur. Burada kullanılan metod, bağıl homojenlik tekniklerini kullanarak seri içerisindeki homojen olmayan kısımları tespit edebilen verimli bir yöntemdir. Metod, parametrik olmayan Kruskal-Wallis homojenlik, Swed-Eisenhart run testlerini ve grafiksel analizi içermektedir. Yöntemin 84 aday istasyonla yüksek korelasyonlu komşu istasyonlarla oluşturulan çiftlere uygulanması neticesinde, 69 istasyonun homojen veriye sahip olduğu gösterilmiştir.
Açıklama
Tez (Yüksek Lisans) -- İstanbul Teknik Üniversitesi, Fen Bilimleri Enstitüsü, 1996
Thesis (M.Sc.) -- İstanbul Technical University, Institute of Science and Technology, 1996
Anahtar kelimeler
Yağışlar, Rainfall
Alıntı