Detection of Urbanization signals in Extreme Winter Minimum Temperatures change over Northern China Qingxiang Li, Jiayou Huang, Zhihong Jiang, Liming Zhou, Peng Chu, Kaixi Hu GEV The GEV distribution is a family of continuous probability distributions developed within extreme value theory to combine the Gumbel, Fréchet and Weibull families, with a theoretical distribution function as follows: exp{ [1 ( x ) / a]1 / k }, k 0, x a / k F ( x) exp{ exp[ ( x )]}, 0 exp{ [1 ( x ) / a]1 / k }, k 0, x a / k (1) Where ξ represents the location parameter (referred to as CM) determining the position of the distribution; α is the scale parameter (referred to as CF) determining the range of extension of measurement distribution curve; k is the shape parameter (referred to as CK) determining the tail behavior of the distribution (i.e., the pattern of extreme distribution): when k=0, it is the Gumbel distribution; when k>0, it is the Weibull distribution; when k<0, it is the Fréchet distribution. Therefore, the GEV probability distribution is described by the 3 parameters (characteristic values). Estimating the GEV distribution parameters generally consists of two major methods, maximum likelihood methods and L-moments methods (Hosking, 1990; Coles, 2001). The calculation procedures are as follows: First, the 30 extreme minimum temperature samples for each 10-year window 1 are sorted to form a new sequence in the ascending order: xi (i 1,2, n) Where n is the sample size, and the linear matrix is: 1 b0 (2) 2 2b1 b0 (3) 3 6b2 6b1 b0 (4) where, b0 1 n xi n i 1 b1 1 n (i 1) xi n i 2 (n 1) b2 1 n (i 1) (i 2) xi n i 3 (n 1) (n 2) are the sample probability weight matrix. The shape parameter in the GEV distribution can be estimated as follows: k 7.8590C + 2.9554C 2 (5) Where, C 2 ln2 (3 3 ) 3 2 ln3 The scale parameter can be estimated by, k2 (1 2 )(1 k ) k (6) and the position parameter can be calculated by: 1 [1 (1 k )] / k (7) Where (1 k ) is the function of . 2 References Coles, S. G. (2001), An Introduction to Statistical Modeling of Extreme Values. Springer. Hosking J. R. M. (1990), L-moments: Analysis and Estimation of Distributions using Linear Combinations of Order Statistics. J. R. Statist. Soc. B, 52, 105-124 3