Sacpy, a repaid Statistical Analysis tool for Climate or Meteorology data.

Author : Zilu Meng

e-mail : [email protected]

github : https://github.com/ZiluM/sacpy

gitee : https://gitee.com/zilum/sacpy

examples or document : https://github.com/ZiluM/sacpy/tree/master/examples or https://gitee.com/zilum/sacpy/tree/master/examples

version : 0.0.8

For example, Sacpy is more than 60 times faster than the traditional regression analysis with Python (see speed test).

Compatible with commonly used meteorological calculation libraries such as numpy and xarray.

You can use pip to install.

    pip install sacpy

As a comparison, we use the corr function in the xarray library, corrcoef function in numpy library, cdist in scipy, apply_func in xarray and for-loop. The time required to calculate the correlation coefficient between SSTA and nino3.4 for 50 times is shown in the figure below.

It can be seen that we are four times faster than scipy cdist, five times faster than xarray.corr, 60 times faster than forloop, 110 times faster than xr.apply_func and 200 times faster than numpy.corrcoef.

Moreover, xarray and numpy can not return the p value. We can simply check the pvalue attribute of sacpy to get the p value.

All in all, if we want to get p-value and correlation or slope, we only to choose Sacpy is 60 times faster than before.

Calculate the correlation between SST and nino3.4 index

import numpy as np
import scapy as scp
import matplotlib.pyplot as plt


# load sst
sst = scp.load_sst()['sst']
# get ssta (method=1, Remove linear trend;method=0, Minus multi-year average)
ssta = scp.get_anom(sst,method=1)
# calculate Nino3.4
Nino34 = ssta.loc[:,-5:5,190:240].mean(axis=(1,2))
# regression
linreg = scp.LinReg(np.array(Nino34),np.array(ssta))
# plot
plt.contourf(linreg.corr)
# Significance test
plt.contourf(linreg.p_value,levels=[0, 0.05, 1],zorder=1,
            hatches=['..', None],colors="None",)
# save
plt.savefig("./nino34.png")

Result(For a detailed drawing process, see example):

multiple linear regression on Nino3.4 IOD Index and ssta pattern

import numpy as np
import scapy as scp
import matplotlib.pyplot as plt


# load sst
sst = scp.load_sst()['sst']
# get ssta (method=1, Remove linear trend;method=0, Minus multi-year average)
ssta = scp.get_anom(sst,method=1)
# calculate Nino3.4
Nino34 = ssta.loc[:,-5:5,190:240].mean(axis=(1,2))
# calculate IODIdex
IODW = ssta.loc[:,-10:10,50:70].mean(axis=(1,2))
IODE = ssta.loc[:,-10:0,90:110].mean(axis=(1,2))
IODI = +IODW - IODE
# get x
X = np.vstack([np.array(Nino34),np.array(IODI)]).T
# multiple linear regression
MLR = scp.MultLinReg(X,np.array(ssta))
# plot IOD's effect
plt.contourf(MLR.slope[1])
# Significance test
plt.contourf(MLR.pv_i[1],levels=[0, 0.1, 1],zorder=1,
            hatches=['..', None],colors="None",)
plt.savefig("../pic/MLR.png")

Result(For a detailed drawing process, see example):

Thank Prof. Feng Zhu (NUIST,https://github.com/fzhu2e) for his guidance of this project