"ValueError: Array conditional must be same shape as self"

Question

I am trying to follow a tutorial . Basically, I want to run random forest classification on my 31-band Sentinel 1 and Sentinel 2 stacked image. Also, I want to extract raster values to my training-testing shapefiles. This is what I tried:

from osgeo import gdal
from PIL import Image
import numpy as np
import matplotlib as mtp
import matplotlib.pyplot as plt
import pandas as pd
import geopandas as gpd
import earthpy.plot as ep
import rasterio
from rasterio.plot import reshape_as_raster, reshape_as_image

get_ipython().run_line_magic('matplotlib', 'inline')

pd.options.display.max_colwidth = 89


# In[2]:


#setting the path for image
S1_S2_stack = 'S1_S2_stack.tif'

#path to training and validation data
training_points = 'testing.shp'
validation_points = 'training.shp'


# In[3]:


colors = dict ((
    (0, (0,76,153,255)),  #wheat
    (1, (0,153,0,255)),   #corn
    (2, (255,0,0,255)),   #other
    (3, (255,153,51,255)),
    (4, (255,255,0,255))
    
))


# In[4]:


for k in colors:
    v = colors [k]
    _v = [_v / 255.0 for _v in v]
    colors[k] = _v
    
index_colors = [colors[key] if key in colors else (1,1,1,0) for key in range (0,5)]
cmap = plt.matplotlib.colors.ListedColormap(index_colors, 'Classification', 5)


# In[5]:


src = rasterio.open(S1_S2_stack)
bands = src.read()


# In[6]:


stack =src. read()
print(stack.shape)

fig, (ax1, ax2) = plt.subplots(1,2,figsize= (20,10))
ax1 = ep.plot_rgb(arr = stack, rgb =(3, 2, 1), stretch=True, ax = ax1, title = "RGB Composite - Sentinel-2")
stack_s1 =np.stack ((stack[28],stack[29],stack[29]/stack[28]))
ax2 = ep.plot_rgb(arr = stack_s1, rgb = (1,0,2), stretch=True, ax = ax2, title= "RGB Composite - Sentinel-1 (VV, VH, VV/VH)")
plt.tight_layout()


# In[7]:


img = src.read()
profile = src.profile
with rasterio.io.MemoryFile () as memfile:
    with memfile.open(**profile) as dst:
        for i in range(0, src.count):
            dst.write(img[i], i+1)
    dataset = memfile.open()

And it works okay from here. But when I run this piece of code:

   #read points from shapefile
train_pts = gpd.read_file (training_points)
train_pts = train_pts[[ 'CID','class', 'POINT_X','POINT_Y']] #attribute fields in shapefile
train_pts.index = range(len(train_pts))
coords = [(x,y) for x, y in zip(train_pts.POINT_X, train_pts.POINT_Y)] #create list of point coordinates

#sample each band of raster dataset at each point in the coordinate list
train_pts ['Raster Value'] = [x for x in dataset.sample(coords)] #all band values saved as a list in the Raster Value column
#Unpack the raster value column to separate column for each band - band names retrieved from snappy in the video but I was looking for an option
train_pts[bands] = pd.DataFrame(train_pts['Raster Value'].tolist(), index = train_pts.index)
train_pts = train_pts.drop(['Raster Value'], axis=1) #drop raster value column
#change the values for last three classes 
train_pts['CID'] = train_pts['CID'].replace([7,8,15],[5,6,7])
train_pts.to.csv('train_pts.csv') #save as csv
train_pts.head () #see first column

I get the following error:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Input In [9], in <cell line: 10>()
      8 train_pts ['Raster Value'] = [x for x in dataset.sample(coords)] #all band values saved as a list in the Raster Value column
      9 #Unpack the raster value column to separate column for each band - band names retrieved from snappy in the video but I was looking for an option
---> 10 train_pts[src1] = pd.DataFrame(train_pts['Raster Value'].tolist(), index = train_pts.index)
     11 train_pts = train_pts.drop(['Raster Value'], axis=1) #drop raster value column
     12 #change the values for last three classes 

File ~\.conda\envs\geocomp3_clone\lib\site-packages\pandas\core\frame.py:3643, in DataFrame.__setitem__(self, key, value)
   3641     self._setitem_frame(key, value)
   3642 elif isinstance(key, (Series, np.ndarray, list, Index)):
-> 3643     self._setitem_array(key, value)
   3644 elif isinstance(value, DataFrame):
   3645     self._set_item_frame_value(key, value)

File ~\.conda\envs\geocomp3_clone\lib\site-packages\pandas\core\frame.py:3685, in DataFrame._setitem_array(self, key, value)
   3680 else:
   3681     # Note: unlike self.iloc[:, indexer] = value, this will
   3682     #  never try to overwrite values inplace
   3684     if isinstance(value, DataFrame):
-> 3685         check_key_length(self.columns, key, value)
   3686         for k1, k2 in zip(key, value.columns):
   3687             self[k1] = value[k2]

File ~\.conda\envs\geocomp3_clone\lib\site-packages\pandas\core\indexers\utils.py:428, in check_key_length(columns, key, value)
    426 if columns.is_unique:
    427     if len(value.columns) != len(key):
--> 428         raise ValueError("Columns must be same length as key")
    429 else:
    430     # Missing keys in columns are represented as -1
    431     if len(columns.get_indexer_non_unique(key)[0]) != len(value.columns):

ValueError: Columns must be same length as key

So my questions are as follows:

1. Could there be a problem with the method I used to import the shapefile's bands?
2. Do I need to write all the fields in the code where I enter the attribute information of the shapefile? Or should I edit these fields in a GIS program?

Answer 1

Creating new DataFrame columns from band descriptions

If you say

src = rasterio.open(S1_S2_stack)
bands = src.read()

then bands is a 3-dimensional Numpy ndarray (a series of raster frames). It makes no sense to use it for indexing like train_pts[bands] , where train_pts is a data frame. But I assume you want to refer to the band names. If so, try src.descriptions instead:

band_names = [*src.descriptions]
train_pts[band_names] = pd.DataFrame(train_pts['Raster Value'].tolist(), index = train_pts.index)

Some pitfalls to be aware of

• band_names should be a list so we can use them as indexes in train_pts[band_names] . As far as src.descriptions is a tuple , we have to transform it into a list .
• If there are empty or duplicate values in the band descriptions, we need to deal with them somehow:

band_names = [f'{src.descriptions[i]}, band {i}' for i in range(1, src.count + 1)]

• We can force some default names as an alternative to the previous remark:

band_names = [f'Band {i}' for i in range(1, src.count + 1)]