泰勒图--MATLAB实现

0.泰勒图介绍

见参考文献[4]

1.准备步骤

（1）taylordiag.m。
在参考文献【2】下载泰勒图的函数脚本。

表1.函数需要的输入变量如下表

输入变量	术语解释
STDs	标准差（Standard deviations）
RMSs	去中心的均方根误差（Centered Root Mean Square Difference 就是减去平均值再求RMS）
CORs	皮尔逊相关系数（Correlation）

（2）allstats.m
在参考文献【1】下载计算输入变量的函数脚本。
为了计算表1.中的参数，可直接利用allstats.m函数计算。如表2，随意设计一组原始数据格式，记为a，第一行为参考序列，第二行为模式1，第二行模式2，依次类推。

表2 原始数据

模式	值1	值2	值3	值4	值5	值6	值7
参考序列	26	47	51	52	28	47	32
模式1	32	37	40	52	46	45	21
模式2	49	58	51	52	38	47	35
模式3	26	47	51	52	28	47	32
模式4	26	47	51	22	28	47	32
模式5	26	47	51	52	38	47	35

下列代码计算表1中的变量

for iserie = 2 : size(a,1)
 		    S = allstats(a(1,:),a(iserie,:));
 		    MYSTATS(iserie,:) = S(:,2); % We get stats versus reference
end       %for iserie
 MYSTATS(1,:) = S(:,1); % We assign reference stats to the first row

2.生成泰勒图

预处理后，直接带入taylordiag函数即可，生成图片如下所示。

taylordiag(MYSTATS(:,2),MYSTATS(:,3),MYSTATS(:,4));

3.后处理

（1）一般泰勒图都要标准化处理。标准差与均方根误差同时除以参考序列的标准差。

（2）根据数值范围，设置坐标轴参数。具体参数设置参考taylordiag.m的492-523行。

（3）不想看代码，可保存后导入Ai、PS后处理。完毕。

%%计算输入参数
for iserie = 2 : size(a,1)
 		    S = allstats(a(1,:),a(iserie,:));
 		    MYSTATS(iserie,:) = S(:,2); % We get stats versus reference
end       %for iserie
 MYSTATS(1,:) = S(:,1); % We assign reference stats to the first row
 
%  taylordiag(MYSTATS(:,2),MYSTATS(:,3),MYSTATS(:,4));
stdev=MYSTATS(1,2);

MYSTATS(:,2)=MYSTATS(:,2)/stdev;% 标准化处理
MYSTATS(:,3)=MYSTATS(:,3)/stdev;%标准化处理

 [hp ht axl] =taylordiag(MYSTATS(:,2),MYSTATS(:,3),MYSTATS(:,4), ...
                 'tickrms',[0:.2:1],'titleRMS', 0 ,'showlabelsRMS',1,... 
                 'widthRMS',1,'colRMS','r',...
                 'tickSTD',[0:.25:1.25],'limSTD',1.25,'styleSTD','-',... 
                 'tickCOR',[.1:.1:.9 .95 .99],'showlabelsCOR',1,'titleCOR',1);

注：
①对下面报错

错误使用 taylordiag (line 108)
taylordiag.m : Something's wrong with ALL the datas
You must have:
RMSs - sqrt(STDs.^2 + STDs(1)^2 - 2*STDs*STDs(1).*CORs) = 0 !

不要自己创建输入变量STDs、RMSs、CORs这些，一定要自己利用原始值算出。
②R语言创建泰勒图
？ TaylorDiagram {openair}（强烈推荐）
？ taylor.diagram {plotrix}
③PeterRochford/Skill​MetricsToolbox
PeterRochford/Skill​MetricsToolbox也可创建泰勒图，但此时注意到的是MATLAB对用的版本要在2016b及以上。否则会报错，"xticks"函数不存在。

4.参考文献

[1]allstats.m下载
[2]taylordiag.m下载
[3]参数设置
[4]Taylor, K. E. (2001). Summarizing multiple aspects of model performance in a single diagram. Journal of Geophysical Research: Atmospheres, 106(D7), 7183-7192.

5 allstats.m代码

% STATM Compute statistics from 2 series
%
% STATM = allstats(Cr,Cf)
%
% Compute statistics from 2 series considering Cr as the reference.
% 
% Inputs:
%	Cr and Cf are of same length and uni-dimensional. They may contain NaNs.
%
% Outputs:
% 	STATM(1,:) => Mean
% 	STATM(2,:) => Standard Deviation (scaled by N)
% 	STATM(3,:) => Centered Root Mean Square Difference (scaled by N)
% 	STATM(4,:) => Correlation
%
% Notes:
%	- N is the number of points where BOTH Cr and Cf are defined
%
% 	- NaN are handled in the following way: because this function
% 		aims to compair 2 series, statistics are computed with indices
%		where both Cr and Cf are defined.
%
% 	- STATM(:,1) are from Cr (ie with C=Cr hereafter)
% 	  STATM(:,2) are from Cf versus Cr (ie with C=Cf hereafter)
%
%	- The MEAN is computed using the Matlab mean function.
%
%	- The STANDARD DEVIATION is computed as:
%			          /  sum[ {C-mean(C)} .^2]  \
%			STD = sqrt|  ---------------------  |
%			          \          N              /
%
%	- The CENTERED ROOT MEAN SQUARE DIFFERENCE is computed as:
%			           /  sum[  { [C-mean(C)] - [Cr-mean(Cr)] }.^2  ]  \
%			RMSD = sqrt|  -------------------------------------------  |
%			           \                      N                        /
%
%	- The CORRELATION is computed as:
%			      sum( [C-mean(C)].*[Cr-mean(Cr)] ) 
%			COR = --------------------------------- 
%			              N*STD(C)*STD(Cr)
%
%	- STATM(3,1) = 0 and STATM(4,1) = 1 by definition !
%
% Created by Guillaume Maze on 2008-10-28.
% Rev. by Guillaume Maze on 2010-02-10: Add NaN values handling, some checking
%				in the inputs and a more complete help
% Copyright (c) 2008 Guillaume Maze. 
% http://codes.guillaumemaze.org




%
% This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by
% the Free Software Foundation, either version 3 of the License, or any later version.
% This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of
% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for more details.
% You should have received a copy of the GNU General Public License along with this program.  If not, see <http://www.gnu.org/licenses/>.
%

function STATM = allstats(varargin)
	
Cr = varargin{1}; Cr = Cr(:);
Cf = varargin{2}; Cf = Cf(:);

%%% Check size:
if length(Cr) ~= length(Cf)
	error('Cr and Cf must be of same length');
end

%%% Check NaNs:
iok = find(isnan(Cr)==0 & isnan(Cf)==0);
if length(iok) ~= length(Cr)
	warning('Found NaNs in inputs, removed them to compute statistics');
end
Cr  = Cr(iok);
Cf  = Cf(iok);
N   = length(Cr);

%%% STD:
st(1) = sqrt(sum(  (Cr-mean(Cr) ).^2)  / N );
st(2) = sqrt(sum(  (Cf-mean(Cf) ).^2)  / N );
%st(1) = sqrt(sum(  (Cr-mean(Cr) ).^2)  / (N-1) );
%st(2) = sqrt(sum(  (Cf-mean(Cf) ).^2)  / (N-1) );

%%% MEAN:
me(1) = mean(Cr);
me(2) = mean(Cf);

%%% RMSD:
rms(1) = sqrt(sum(  ( ( Cr-mean(Cr) )-( Cr-mean(Cr) )).^2)  /N);
rms(2) = sqrt(sum(  ( ( Cf-mean(Cf) )-( Cr-mean(Cr) )).^2)  /N);

%%% CORRELATIONS:
co(1) = sum(  ( ( Cr-mean(Cr) ).*( Cr-mean(Cr) )))/N/st(1)/st(1);
co(2) = sum(  ( ( Cf-mean(Cf) ).*( Cr-mean(Cr) )))/N/st(2)/st(1);


%%% OUTPUT
STATM(1,:) = me;
STATM(2,:) = st;
STATM(3,:) = rms;
STATM(4,:) = co;
	
end %function	

6 拓展 MVIETool: 改进的多变量集合评估方法及多模式比较评估工具

Zhang Meng-Zhuo, Zhongfeng Xu*, Ying Han, and Weidong Guo, 2021: An improved multivariable integrated evaluation method and tool (MVIETool) v1.0 for multimodel intercomparison, Geosci. Model Dev., 14, 3079–3094, https://doi.org/10.5194/gmd-14-3079-2021