|
Titel |
Penalized Maximal F Test for Detecting Change Points of Temperature and Wind Speed Data Series |
VerfasserIn |
L. Cao, X. Liu, Q. Li |
Konferenz |
EGU General Assembly 2009
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 11 (2009) |
Datensatznummer |
250022457
|
|
|
|
Zusammenfassung |
The homogeneity of the climate record continues to receive considerable attention. Time
series are commonly contaminated by non-climatic discontinuities that result from
station relocations, observation time changes, and station specific trends related to
environmental changes in the proximity of the observation sites. Several statistical
methods have been proposed for detecting undocumented shifts. Wang Xiaolan et
al. proposed the penalized maximal F test (PMFT) for detecting undocumented
mean shifts that are not accompanied by any sudden change in the linear trend
of time series. This method is based on the penalized maximal F test, which are
embedded in a recursive testing algorithm, with the lag-1 autocorrelation (if any) of the
time series being empirically accounted for. In this research the PMFT method is
used for detecting the shifts of long time series of temperature and wind speed
data series over China. The monthly average temperature data of about tens of
meteorological observing stations and the annual average wind speed data of 753
meteorological observing stations have been detected. The results show that this
homogeneity detection method works well for these two meteorological data series over
China.
1. Results of monthly average temperature data series
To create a reference time series is sometimes very useful for homogeneity detection,
while it is difficult to get a good reference time series especially for the hundred-year long
temperature data with a lot of year data lost over China. The PMFT method are used
without building a references series to detecting the change points of the monthly
average maximum temperature of twelve meteorological stations and the monthly
average minimum temperature of twenty-nine stations. The results show that this
method is fit for the homogeneous detection and we needn’t interpolating the data and
building a reference time series before the detection. Although discontinuities in
temperature time series can be caused by any number of changes in, for example,
sensor type, and even the observation schedule, station relocations are the likely
cause of the majority of abrupt shifts identified in the temperature series evaluated
here.
2. Results of annual average wind speed data series
52 of the data series are too short to be detected among 753 meteorological observing
stations. A total of 356 change points over 271 stations are detected of the annual average
wind speed time series, which accounts for 38.7% of the evaluated stations. The
homogeneous data series are of 231 stations and another 199 stations are not significant that
can be considered as homogeneous, which accounts for 61.3% of the evaluated stations. It is
found that the data of 61.3% stations are homogeneity among the detected 701 stations,
which shows that the homogeneity of the annual average wind speed is good. The change
points of the annual average wind speed range from 1 to 2. The changes of instrument
and location are the main reason for the non-homogeneity, while the change of
the type of the observation instrument for the wind speed is the most important
reason for the non-homogeneity of the annual average wind speed over China. The
environment change seems not so remarkable, because the relocation and the instrument
change may take place at the same time to conceal the effect of the environment
change.
All of the works we have done are the preliminary experiments of using this method.
Although we get some results, there are still a lot of works need to do because the wind
speed data are so special and the probability distribution are not the exact Gaussian
distribution. At the same time the data of the wind speed are affected mostly by the
topography and the barriers aside of the observation fields. The use of reference
series can help to diminish departure from Gaussian distribution. We will do more
experiments on detecting of the wind speed data. What the important thing is that rely
on most detail metadata information to help the work of homogeneity detection. |
|
|
|
|
|