New Study: Algorithm Reveals Warmest & Coolest Climates
New study presents an algorithm that helps scientists identify coldest and hottest time periods. The algorithm is applied to a case study performed by Dr Darko Butina of winters recorded at Armagh Observatory (pictured) over a 161-year period between 1844 and 2004.
The study will feature in the International Journal of Chemical Modeling ISSN: 1941-3955 Volume 7, Number 3.
Reference:
International Journal of Chemical Modeling ISSN: 1941-3955 Volume 7, Number 3 © Nova Science Publishers, Inc.
NEW ALGORITHM TO IDENTIFY COLDEST AND HOTTEST TIME PERIODS. CASE STUDY: COLDEST WINTERS RECORDED AT ARMAGH OBSERVATORY OVER 161 YEARS BETWEEN 1844 AND 2004
Darko Butina* Chemomine Consultancy (Looking for Patterns in Instrumental Data) Letchworth Garden City, Hertfordshire, UK
ABSTRACT
A novel algorithm has been developed that ranks cold and/or hot annual time periods using daily maximum and/or minimum air temperature for a given archived dataset. The author used a version of the algorithm that he developed and extensively used to find similar structural patterns in databases containing millions of different molecules. However, in this paper, the method is applied to assess the similarity of ‘winter’ daily tmax patterns to the cold boundary pattern, wMIN, of an ultimately cold winter.
The fundamental problem with the current practices of weather forecasting is the use of the mean as a predictive variable which has no usable predictive power. The paper clearly shows that very small variations in the mean values are totally lost in very large natural variations of daily temperatures measured by a calibrated thermometer.
The key feature of this algorithm is to identify two extreme boundaries, the coldest one and the hottest, from a given archive dataset of daily tmax and/or tmin observed temperatures, and then calculate the distance between each annual time period chosen against those two reference patterns.
In our case, the ranking of the coldest winters recorded at Armagh Observatory (UK) identified the two most unusually cold winters of 1963 and 1895. Definition of winter is a variable that is chosen by the user, in this case it was first 60 days of the year, and therefore each winter was treated as a pattern or if you wish a fingerprint, consisting of 60 daily tmax readings. Since the objective was to rank the winters from the coldest to the hottest, the cold reference pattern or a cold boundary was obtained by identifying the coldest daytime, tmax, reading for each day of the winter.
The ranking algorithm then calculates the Euclidean Distance between each winter and the cold reference boundary and then sorts the winters by that distance, from the smallest to the largest, i.e., in ascending order. The results of the ranking algorithm were then compared with the observations and found that not only did the findings reflect the real observations for Armagh, but that the findings for Armagh can be applied to the whole of the UK.
A brief summary of the paper for non-specialists in the field of pattern recognition, machine learning and statistics will be sent to you within next couple of weeks. To highlight the main power of this algorithm, observe that the algorithm starts with summary graph of 161 winters, each consisting of 60 tmax readings
Figure 7. Each differently coloured line represents one winter at Armagh Observatory between 1844 and 2004 (page 218)
After the algorithm is applied, the coldest and the hottest winters are clearly separated:
Figure 10. Second coldest winter, 1895 (light blue) and the hottest, 1846 (dark red) are plotted against wMAX (top in red), wMIN (bottom in dark blue) and the mean of all winters (green) (page 221)
Home page for Dr Darko Butina: www.l4patterns.com
Trackback from your site.