Vienna, 2009-04-28
Scientific Report of the COST UBRIS/WG4 Meeting
23. Apr. 2009, Vienna
The meeting was dedicated to problems of urban series (UBRIS) and methods to correct daily climate series (WG4). Xiaolan L. Wang an international expert on homogenization agreed to participate on the occasion of her venue at EGU.
- UBRIS
O. Mestre presented the most important results of the common work with ZAMG about urban effect in Viennese series, only sub-urban ones exhibit a significant urban trend. The group decided to continue working on this important topic.
- MONTHLY BENCHMARK
Monthly benchmark afforded a slight modification in respect to the number of breaks and the overall trend.J. Guijarro, P. Stepanek, O. Mestre, T. Szentimrey will verify this new version.
- DAILY DATA
Methods for correction
Existing methods were presented and briefly introduced. The methods can be divided into three categories: “delta” methods which apply fixed amount of a shift, variable correction methods based upon one element and variable correction methods based on multi-element approach. Available are: Interpolation of monthly factors, MASH, Vincent et al (2002) - cublic spline interpolation , nearest neighbor resampling models by Brandsma and Können (2006), Higher Order Moments (HOM) by Della Marta and Wanner (2006), Two phase non-linear regression by O. Mestre, Distribution Adjusting by Percentiles (DAP) by P. Stepanek, Using weather types classifications (HOWCLASS) by I. Garcia-Borés, E. Aguilar,
Petr Štěpánek showed that variable correction methods work well in respect to changing distribution of candidate series before a break to have exactly what we expect from such a correction, nonetheless sometimes such correction gives misleading results when compared with reference series: even if the (distribution) correction is right, it is not given in right time positions. It means that for some of climatological analysis such series can be applied without problems, but for the other ones, such as impact studies utilizing the number of consecutive days above some threshold etc. (i.e. when exact timing is required) it is not appropriate enough.It was notified by Xiaolan Wang that one has to be very careful about the fact that distribution can change throughout time (it may not stay stationary).
T. Szentimrey emphasized the fact that clear mathematical formulation was needed for daily homogenization. O. Mestre added that understanding the physics behind may help.
Creation of a daily benchmark dataset
Created surrogate daily data are available both for air temperature and precipitation for 15 networks, each containing 9 stations for the period 1908-2007 with statistical properties of measured data.To introduce inhomogeneities it is required to know about the nature of inhomogeneities in daily data. Therefore it is needed to start analyzing measured daily data: comparative measurements where available, but also near neighbor stations series since comparative measurements are available only in very few cases and moreover such measurements can be very similar.
Real data availability and analysis
It has been discussed which datasets (comparative measurements - e.g. manual and automate measurements, and also neighbor stations) are available for studying differences between two sides measurements. A list of possible data providers has been made:CzechRepublic, France, Romania, Norway, Catalonia, and Italy. The provided data should be available only within WP4 not to violate data policies of some of the providing institutions.Olivier Mestreoutlinedhow the introduced inhomogeneities should look like (* shift in mean (constant, season dependent, "regime" dependent), * shift in variance (constant, season dependent, "regime" dependent), * shift in skewand* add change in mean to any change (detectable), * shift only extremes, * shift quantiles).
Comparison and rankingof methods of daily data correction
It was discussed, which statistical characteristics should be applied to describe differences between “true” and adjusted (corrected) datasets. The following list was proposed by Petr Stepanek and participants:Spearman rank correlation, Min. difference (Bias), Max. difference (Bias), Abs. range of difference (Bias), Bias (mean), MAE (mean absolute error), RMSE (root mean squared error), NRMSE (normalized RMSE by the range ofSTD of differences
p-value of t-test for Paired Values, F-test (STDs) and its p-value, t-test (AVGs, INDEPENDENT samples) and its p-value
K-S test, and for each of the series (candidate and “true” onebasic statistical characteristics including percentiles and return periods.
Ingeborg Auer
Central Institute for Meteorology and Geodynamics, ZAMG
Local Organizer