topic Calculate Correlation and P-value between rasters in Python Questions

Calculate Correlation and P-value between rasters

margaritastevia — Wed, 03 Jul 2019 05:15:40 GMT

Hi Xander Bakker‌ I have used this script and got two rasters "correlation" and "pearson". Which one is the P-value?

Branched from https://community.esri.com/thread/200534-re-correlation-between-two-different-rasters

Re: Calculate Correlation and P-value between rasters

DanPatterson_Retired — Sun, 12 Dec 2021 05:01:55 GMT

Neither, p-value is a measure of the strength of the correlation coefficient calculated from sample size and the measured 'r' and a sampling distribution

scipy.stats.pearsonr — SciPy v1.3.0 Reference Guide

example for variants of simple arrays, using this

from scipy import stats

a = np.arange(9)
b = np.arange(9, 0, -1)
c = np.array([9, 7, 8, 6, 5, 4, 3, 2, 1]) # ---- like b, but 2 numbers switched

r, p_val = stats.pearsonr(a, b)  # ---- perfect inverse, r = -1, p = 0.0000
r, p_val
(-1.0, 0.0)

r, p_val = stats.pearsonr(a, c)
r, p_val
(-0.9833333333333333, 1.936196303745927e-06)‍‍‍‍‍‍‍‍‍‍‍‍‍

Re: Calculate Correlation and P-value between rasters

margaritastevia — Thu, 04 Jul 2019 05:43:07 GMT

I want to extract the pixels where correlation is positive and p-value is < 0.05 How i can do that? How i can estimate P as a raster?

Re: Calculate Correlation and P-value between rasters

margaritastevia — Mon, 08 Jul 2019 05:38:37 GMT

Hi Xander Bakker‌ please help me with how i can extract correlation values at 0.05 significance level. I'm running out of time.

Re: Calculate Correlation and P-value between rasters

XanderBakker — Mon, 08 Jul 2019 13:24:31 GMT

Hi margaritastevia

Can you share a sample of your data and the script you have so far?

Re: Calculate Correlation and P-value between rasters

margaritastevia — Mon, 08 Jul 2019 15:39:57 GMT

Hi Xander Bakker,

Thank you very much for your kind response. please find the attachment.

Re: Calculate Correlation and P-value between rasters

XanderBakker — Sun, 12 Dec 2021 05:01:58 GMT

Hi margarita stevia ,

Could you validate if this works for you? I will create the additional p-value raster and a mask raster containing the pixels with value 1 where the condition (p-value <= 0.05 and positive correlation) is met. The calculation of the p-value is based on the code provided by Dan Patterson

It seems that the resulting areas (blue pixels) are near the rivers . Does that make sense?

See code below:

def main():
    print "load modules..."
    import arcpy
    arcpy.env.overwriteOutput = True

    import numpy as np
    template1 = r'C:\GeoNet\Pvalue\Send\NDVI\r{0}_WUE.TIF'
    template2 = r'C:\GeoNet\Pvalue\Send\temp\r{0}_TEM.TIF'
    nodata = -3.4028235e+38
    out_ras_corr = r'C:\GeoNet\Pvalue\Send\Results\correlation_v01.TIF'
    out_ras_pear = r'C:\GeoNet\Pvalue\Send\Results\pearson_v01.TIF'
    out_ras_pval = r'C:\GeoNet\Pvalue\Send\Results\pvalue_v01.TIF'
    out_ras_mask = r'C:\GeoNet\Pvalue\Send\Results\mask_v01.TIF'

    # settings
    max_pval = 0.05
    min_corr = 0
    mask_val = 1
    no_mask_val = 0

    print "create nested numpy array list..."
    lst_np_ras = []
    for i in range(1, 5):
        ras_path1 = template1.format("%03d" % (i,))
        print " - ", ras_path1
        ras_np1 = arcpy.RasterToNumPyArray(ras_path1)
        ras_path2 = template2.format("%03d" % (i,))
        print " - ", ras_path2
        ras_np2 = arcpy.RasterToNumPyArray(ras_path2)
        lst_np_ras.append([ras_np1, ras_np2])

    print "read props numpy raster..."
    ras_np = lst_np_ras[0][0]  # take first numpy array from list
    rows = ras_np.shape[0]
    cols = ras_np.shape[1]
    print " - rows:", rows
    print " - cols:", cols

    print "create output numpy array..."
    ras_path = template1.format("%03d" % (1,))
    raster = arcpy.Raster(ras_path)
    ras_np_res_corr = np.ndarray((rows, cols))
    ras_np_res_pear = np.ndarray((rows, cols))
    ras_np_res_pval = np.ndarray((rows, cols))
    ras_np_res_mask = np.ndarray((rows, cols))
    print " - out rows:", ras_np_res_corr.shape[0]
    print " - out cols:", ras_np_res_corr.shape[1]

    print "loop through pixels..."
    pix_cnt = 0
    for row in range(rows):
        for col in range(cols):
            pix_cnt += 1
            if pix_cnt % 5000 == 0:
                print " - row:", row, "  col:", col, "  pixel:", pix_cnt
            lst_vals1 = []
            lst_vals2 = []
            try:
                for lst_pars in lst_np_ras:
                    lst_vals1.append(lst_pars[0][row, col])
                    lst_vals2.append(lst_pars[1][row, col])
                lst_vals1 = ReplaceNoData(lst_vals1, nodata)
                lst_vals2 = ReplaceNoData(lst_vals2, nodata)
                # perform calculation on list
                correlation = CalculateCorrelation(lst_vals1, lst_vals2, nodata)
                ras_np_res_corr[row, col] = correlation
                pearson = CalculatePearsons(lst_vals1, lst_vals2, nodata)
                ras_np_res_pear[row, col] = pearson
                pvalue = CalculatePvalue(lst_vals1, lst_vals2, nodata)
                ras_np_res_pval[row, col] = pvalue
                # compare correlation and p-value
                if pvalue == nodata:
                    mask = nodata
                elif (correlation >= min_corr) and (pvalue <= max_pval):
                    mask = mask_val
                else:
                    mask = no_mask_val
                ras_np_res_mask[row, col] = mask

            except Exception as e:
                print "ERR:", e
                print " - row:", row, "  col:", col, "  pixel:", pix_cnt
                print " - lst_vals1:", lst_vals1
                print " - lst_vals2:", lst_vals2

    pnt = arcpy.Point(raster.extent.XMin, raster.extent.YMin) #  - raster.meanCellHeight
    xcellsize = raster.meanCellWidth
    ycellsize = raster.meanCellHeight

    print "Write output rasters..."
    print " - ", out_ras_corr
    ras_res_corr = arcpy.NumPyArrayToRaster(ras_np_res_corr, lower_left_corner=pnt, x_cell_size=xcellsize,
                                 y_cell_size=ycellsize, value_to_nodata=nodata)
    arcpy.CopyRaster_management(ras_res_corr, out_ras_corr)
    arcpy.DefineProjection_management(in_dataset=out_ras_corr, coor_system="GEOGCS['GCS_WGS_1984',DATUM['D_WGS_1984',SPHEROID['WGS_1984',6378137.0,298.257223563]],PRIMEM['Greenwich',0.0],UNIT['Degree',0.0174532925199433]]")

    print " - ", out_ras_pear
    ras_res_pear = arcpy.NumPyArrayToRaster(ras_np_res_pear, lower_left_corner=pnt, x_cell_size=xcellsize,
                                 y_cell_size=ycellsize, value_to_nodata=nodata)
    arcpy.CopyRaster_management(ras_res_pear, out_ras_pear)
    arcpy.DefineProjection_management(in_dataset=out_ras_pear, coor_system="GEOGCS['GCS_WGS_1984',DATUM['D_WGS_1984',SPHEROID['WGS_1984',6378137.0,298.257223563]],PRIMEM['Greenwich',0.0],UNIT['Degree',0.0174532925199433]]")

    print " - ", out_ras_pval
    ras_res_pval = arcpy.NumPyArrayToRaster(ras_np_res_pval, lower_left_corner=pnt, x_cell_size=xcellsize,
                                 y_cell_size=ycellsize, value_to_nodata=nodata)
    arcpy.CopyRaster_management(ras_res_pval, out_ras_pval)
    arcpy.DefineProjection_management(in_dataset=out_ras_pval, coor_system="GEOGCS['GCS_WGS_1984',DATUM['D_WGS_1984',SPHEROID['WGS_1984',6378137.0,298.257223563]],PRIMEM['Greenwich',0.0],UNIT['Degree',0.0174532925199433]]")

    print " - ", out_ras_mask
    ras_res_mask = arcpy.NumPyArrayToRaster(ras_np_res_mask, lower_left_corner=pnt, x_cell_size=xcellsize,
                                 y_cell_size=ycellsize, value_to_nodata=nodata)
    arcpy.CopyRaster_management(ras_res_mask, out_ras_mask)
    arcpy.DefineProjection_management(in_dataset=out_ras_mask, coor_system="GEOGCS['GCS_WGS_1984',DATUM['D_WGS_1984',SPHEROID['WGS_1984',6378137.0,298.257223563]],PRIMEM['Greenwich',0.0],UNIT['Degree',0.0174532925199433]]")


def CalculateCorrelation(a, b, nodata):
    import numpy
    try:
        coef = numpy.corrcoef(a,b)
        return coef[0][1]
    except:
        return nodata


def ReplaceNoData(lst, nodata):
    res = []
    for a in lst:
        if a < nodata / 2.0:
            res.append(None)
        else:
            res.append(a)
    return res


def CalculatePearsons(a, b, nodata):
    import numpy as np
    from scipy.stats import pearsonr
    try:
        x = np.array(a)
        y = np.array(b)
        r, p = pearsonr(a, y)
        return p
    except:
        return nodata

def CalculatePvalue(a, b, nodata):
    import numpy as np
    from scipy.stats import stats
    try:
        x = np.array(a)
        y = np.array(b)
        r, p_val = stats.pearsonr(x, y)
        return p_val
    except:
        return nodata



if __name__ == '__main__':
    main()‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍‍

Re: Calculate Correlation and P-value between rasters

margaritastevia — Wed, 10 Jul 2019 06:03:16 GMT

print("Thank you")

Thank you Sir Xander Bakker. I'm vary thankful to you for your kind response. The script is running perfectly. Yes the pixels are near the river.

Re: Calculate Correlation and P-value between rasters

XanderBakker — Wed, 10 Jul 2019 12:19:54 GMT

Hi margaritastevia ,

Glad to hear that the script is producing the expected results. Good luck with you investigation!

Re: Calculate Correlation and P-value between rasters

YingyingZhang — Fri, 06 Sep 2019 14:14:53 GMT

Hi Xander Bakker,

Glad to get the script of the correlation and P-value between rasters, but now I only have two rasters(ndvi and occurence of flooding). I can't solve my question using the script above. I need calculate the correlation and p-value between two rasters not sets of rasters.Xander Bakker Calculate Correlation and P-value between rasters#My data is attached, need your help.

Best wishes

Re: Calculate Correlation and P-value between rasters

XanderBakker — Fri, 06 Sep 2019 22:34:16 GMT

Hi zhangyywhigg ,

If you only have two data values (comparing each pixel between 2 rasters) you will have a perfect linear relationship and will have a result of -1 or +1, so that would not make much sense. What are you trying to obtain?

Re: Calculate Correlation and P-value between rasters

YingyingZhang — Sat, 07 Sep 2019 02:58:54 GMT

Dear Xander Bakker,

I have calculated the coefficient of correlation between two rasters using arcgis spatial analyst tool——multivariate——band collection statistics,and the result of coefficient wasn't the -1 or 1, but it got a .txt result. The problem I want to solve is to find the pixel-by-pixel correlation coefficients and P-values between two layers. I don't know whether the methods of band collection statistics is related to the direct calculation of the correlation between the two rasters. If the only two rasters will have a result of -1 or +1, what methods can visualize the correlation between the two rasters?