Šiška B. et al.: Environmental changes and adaptation strategies Slovakia, Skalica, 9 th – 11 th September 2013 Objective weather classification for environmental applications Janos Mika 1 , Andras Razsi 1 , Agnieszka Wypych 2 , Zbigniew Ustrnul 2 1 Eszterhazy Karoly College, Eger, Hungary, 2 Jagiellonian University, Krakow, Poland Abstract. An objective statistical classification of daily weather is presented for 8 stations of Poland and 5 stations of Hungary for 30 years periods (1966-1995 and 1961-1990, respectively. Initially, eight weather elements were pre-selected and to four variables, according to factor analysis, based on strong correlation between the elements. The further cluster analysis uses four selected weather elements belonging to different rotated factors. They are the diurnal mean values of temperature, of relative humidity, of cloudiness and of wind speed. The omitted (redundant) elements are (logarithm of) precipitation, sunshine duration, diurnal temperature amplitude and water vapour pressure. The omitted elements will be used in independent validation of classification efficiency. Next, hierarchical cluster analysis (i.e. no a priori number of classes), the method of furthest neighbours is selected, after testing various other approaches. Considering the steep maximum of six for the optimum number of classes this number is fixed, and limits of the types are re-defined by method of K-means for all months and stations of the two investigated countries. The obtained local classification will be assessed in comparison with efficiency of macro-circulation types. In overwhelming majority of the months, stations and variables the local types reduce the variance more effectively than the compared Péczely (1957) types for Hungary and an amalgamated (Mika et al., 1999) version of the original Hess-Brezowsky (1969) types, based on objective classification of average sea-level pressure maps derived by Bartholy and Kaba (1990). These local weather types are important tools in understanding the role of weather in various environmental indicators, in climatic generalisation of short samples by stratified sampling and in interpretation of the climate change. Detection of climate change in terms of the frequency of weather types is another possible application of the local weather types. Key words weather types, climatology, factor analysis, cluster analysis, Hungary, Poland. Introduction Synoptic climatology i.e. classification of the endless variability of the everyday weather states according to the pressure configuration and frontal systems relative to the point, or region of interest has long history in meteorology. Its main advantage is to set a limited number of similar meteorological situations, which is the unavoidable to study any quantity or event of the environment for which its dependence on meteorological conditions should be quantified. Another advantage of this, so called, macro-synoptic classification (Peczely, 1957, Puskas, 2001, Piotrowicz, 2010) is that having the actual class of a given day selected, the same code can be applied for various stations or field-campaign. The price of this convenience is the limited efficiency of such circulation-based classifications for at least two reasons. The first one is that the circulation objects and their frontal systems, related to them, often change their positions within the 24 hours of the most common classifications. Hence, the same code may hide rather different situations, indeed. The second reason is the lack of mezo-synoptic object due to the large-scale nature of synoptic analysis and the otherwise reasonable trial to keep the number of the individual classes limited. (Otherwise too long samples were needed.) The logical alternative, i.e. classification of weather according to the observed local weather elements were less popular until the recent times for various reasons. At first, for long time, the numerical weather forecasts were able to outline the synoptic situation, but not the near-surface meteorological variables. At second, there were no computing facilities to operate with multivariate diurnal samples (order of ten variables, at once). Both problems have been resolved in the recent decades as a result of the rapid development in computer technology. The numerical weather forecasting does not use the synoptic situation any longer, so a local classification may be equally useful, especially if providing better fit of the local weather elements. Materials and methods Eight stations from Poland and five stations from Hungary and 30 years periods (1966-1995 for Poland and 1961-1990 for Hungary) were selected with 8 weather elements (see below). The stations are Łeba, Suwałki, Olsztyn, Warszawa-Okęcie, Zielona Góra, Wieluń, Rzeszów-Jasionka, Bielsko-Biała-Aleksandrowice for Poland and Szombathely, Pécs, Budapest, Szeged, Debrecen for Hungary. The four key weather elements of classification have been selected by factor analysis from the 8 candidates. Considering the skewed distribution of precipitation, its logarithm was further considered (with a 0.1 mm correction to keep the zero precipitation in the sample). All these elements have been standardised against the standard deviation within the monthly samples. Table 1 indicates approximate results of the factor analysis for Budapest. The main conclusion is that 3 or 4 factors are enough, and, except precipitation, the climate elements belong to the same factors in majority of bimonthly sub-samples. The selected elements are diurnal mean temperature (Tm: o C), cloudiness (Cl: % of sky), wind speed (Ws: m/s) and relative humidity (Rh: %). The omitted (redundant) elements are precipitation (Pcp: mm/d), sunshine duration (Sd: hour/d), diurnal temperature amplitude (ΔT: o C), water vapour pressure (Wvp: hPa).