Abstract
İklim ve hidroloji açısından yağışın çok önemli bir parametre olduğu düşünüldüğünde, bu parametrenin
mekãnsal ve zamansal dağılımının ve değişiminin incelenmesi gelecekteki iklim koşulları ve su kaynakları
hakkında faydalı fikirler verebilir. Bu nedenle yağışın zamansal, mekãnsal ve mekãnsal-zamansal
dağılımlarının doğru bir şekilde haritalanması hidrolojide, iklim biliminde, tarım biliminde, ekolojide ve
diğer çevre bilimlerinde birçok uygulamada önemlidir. Bu çalışmada Türkiye’nin toplam yıllık ve uzun
yıllar toplam yıllık yağış değerlerinin mekãnsal-zamansal dağılımları ve değişimleri analiz edilmiştir.
Çalışmanın ana veri kaynağı meteorolojik istasyonlarda ölçülmüş aylık yağış değerleri ve bununla mekãnsal
olarak ilişkili geniş kapsamlı veri setleridir. Bunlar yükseklik, yüzey pürüzlülüğü, deniz kıyısına mesafe,
akarsu yoğunluğu, bakı, arazi kullanımı ve ekolojik bölge olarak belirlenmiştir. Fırat havzasının yıllık yağış
değerlerine “mekãn-zaman Sıradan Kriging” ve “mekãn-zaman Evrensel Kriging” yöntemleri
kullanılmıştır. Enterpolasyon yöntemlerinin karşılaştırılması 10 gruplu çapraz sağlama yöntemi ile
yapılmıştır. Doğruluk değerlendirmesi işlemi Kare Kök Ortalama Hata (RMSE) ve R-kare (r2) istatistiksel
ölçütler kullanılarak yapılmıştır.
Abstract
Methods to define and estimate the spatial
variability of hydrologic, climatic and other
environmental variables and perform spatial
interpolation using the quantified spatial variability
are abundant in the environmental sciences.
Recently, the extension of these methods to variables
that vary both in space and time has received
increasing attention. Environmental variables like
precipitation, temperature, air quality, soil
characteristics, etc. are temporally variable
parameters which mean that they may have different
values at the same place through consecutive time
periods. So rather than defining the distribution of
environmental variables at a unique date, if
sufficient amount of data are available at different
time periods and different locations, space-time
interpolation may be more beneficial than spatial
techniques that use only location information.
Snepvangers et al. (2003) compared two Space-time
kriging techniques: Space-time Ordinary kriging
(ST-OK) and ST kriging with external drift (STKED)
on soil water content interpolation. They
found out that predictions are more realistically
obtained from ST-KED, and prediction uncertainty
of this method is lower compared to ST-OK. Jost et
al. (2005) performed a study about spatio-temporal
distribution of soil water storage by using spacetime
kriging methods in a forest ecosystem. Hengl et
al. (2012) used ST Regression kriging to predict
daily temperatures for 2008 obtained from 159
meteorological stations in Croatia.
Accurate mapping of the temporal, spatial and
space-time distributions of precipitation is important
for many applications in hydrology, climatology,
agronomy, ecology and other environmental
sciences. In addition, it is an environmental
parameter which can be analyzed in space-time
context since it has variability in time and space.
In space-time (ST) kriging all observations in the
past, present and future are used to predict the
present situation because of temporal correlation as
quantified by the space-time variogram. So spacetime
kriging makes use of all observations from all
years and locations and if there is indeed temporal
correlation then the observations from other times
(other years) will be included in making the
prediction.
In this study, the Euphrates Basin which is the
biggest and one of the most important basins of
Turkey is selected as a study area to implement
space-time interpolation techniques. For this
purpose, ST-OK and ST-UK methods are applied to
total annual observations for the period of 1970-
2008. The former uses only observed values, the
latter uses observed values and secondary
information as well. Main data source of the study is
point observations of monthly precipitation at
meteorological stations and spatially exhaustive
covariate data sets. These are elevation, surface
roughness, distance to coast, river density, aspect
and land use. Comparison of interpolation methods
are made with ten-fold cross-validation
methodology. Accuracy assessment is done by
calculating the Root Mean Squared Error (RMSE),
R-square (r2
) values.
ST-UK method was applied twice to precipitation
data. At first application elevation, surface
roughness, distance to coast, river density, land
cover, Year and elevation-distance to coast
interaction were used. According to performance
assessment results of cross-validation, R-square is
calculated as 0.73 and RMSE is 107 mm. In the
second application of ST-UK; elevation, surface
roughness, distance to coast, river density and
elevation-distance to coast interaction were used.
The obtaining results are more reliable and
accurate. This time R-square is calculated as 0.85
and RMSE is 78 mm.
For ST-OK the results of R-square is 0.86 and
RMSE is 75 mm. Contrary to expectations, ST-OK
method resulted in more accurate prediction values
than ST-UK according to R-square and RMSE.
Since most of the meteorological stations are located
at lower elevations compared to basin’s mean
elevation, the secondary variables may not be
representative parameters to precipitation
prediction in the basin. However prediction maps of
ST-UK can be regarded as more realistic than STOK
since maps are not so smooth. The prediction
maps of ST-OK have smooth appearance as details
have disappeared during interpolation.
Details
| Other ID | JA24KR86GU |
|---|---|
| Journal Section | Articles |
| Authors | |
| Publication Date | December 1, 2016 |
| Submission Date | December 1, 2016 |
| Published in Issue | Year 2016 Volume: 7 Issue: 2 |