使用python，numpy重新分类栅格值

Question

I have a raster which contains the values [0, 1, 2, 3, 4, 255] and I want to reclassify it. It worked fine doing it pixel by pixel but that took forever, so I tried to change it... :

# working
for j in range(cf.RasterXSize):
    for i in range(cf.RasterYSize):
        if cf_array[i, j] <= 1:
            cf_array[i, j] = 5  # 5 = No Clouds
        elif 1 < cf_array[i, j] <= 4:
            cf_array[i, j] = 6  # 6 = Clouds
        elif 4 < cf_array[i, j]:
            cf_array[i, j] = 7  # 7 = NoData

# Not working:
cf_array[np.where(cf_array <= 1)] = 5
cf_array[np.where((1 < cf_array) & (cf_array <= 4))] = 6
cf_array[np.where(cf_array > 4)] = 7

values = list(numpy.unique(cf_array))
print (values)

And this is what I get when I print the value list with list(numpy.unique(cf_array)) :
original raster: [0, 1, 2, 3, 4, 255]
after reclassify attempt: [7]

So, why is that not working? I found several existing threads regarding the reclassification of a raster but all suggested the numpy.where method which is apparently not working for me...

Answer 1

As has been noted, you don't need the where , however, I believe the issue is the order of your statements. You are first setting some elements to five and six, respectively, and then finally everything bigger than four to seven - this will include all those elements previously set to five and six. It should work by changing the order:

cf_array[cf_array > 4] = 7
cf_array[cf_array <= 1] = 5
cf_array[(1 < cf_array) & (cf_array <= 4)] = 6

Perhaps worthwhile mentioning: Since the ordering makes such a difference here and can introduce subtle problems which can be difficult to find, a safer approach might be to first collect the indices and afterwards change the target array:

# collect index arrays
one_or_less = cf_array <= 1
from_2_to_4 = (1 < cf_array) & (cf_array <= 4)
greater_4 = cf_array > 4

# now modify target array
cf_array[one_or_less] = 5
cf_array[from_2_to_4] = 6
cf_array[greater_4] = 7

Answer 2

You don't need numpy.where . Try:

cf_array[cf_array <= 1] = 5
cf_array[(1 < cf_array) & (cf_array <= 4)] = 6
cf_array[cf_array > 4] = 7

Answer 3

You can simply use np.searchsorted for such a binning operation, like so -

np.searchsorted([1,4],cf_array)+5 # [1,4] act as the bin boundaries

Sample run -

In [55]: cf_array
Out[55]: 
array([[4, 4, 2, 5, 6],
       [0, 5, 5, 6, 0],
       [8, 3, 6, 1, 5],
       [6, 5, 8, 1, 2]])

In [56]: np.searchsorted([1,4],cf_array)+5
Out[56]: 
array([[6, 6, 6, 7, 7],
       [5, 7, 7, 7, 5],
       [7, 6, 7, 5, 7],
       [7, 7, 7, 5, 6]], dtype=int64)