How to exclude/eliminate invalid areas (such as NoData, cloud, water, etc) from raster so that it does not effect the accuracy assessment?

Question

I have two classified raster (snow cover % 10-100) one is prepared from coarse resolution and another is from finer resolution. I am doing accuracy assessment and matrix calculation of snow cover %. However, both classified raster has invalid classes like NoData (-9999), cloud (value=205), water (210), permanent ice(215).

Now both the raster has same pixel size after resampling and covering same area by using gdalwarp

These classes/values are effecting the accuracy assessment. I only want to comapre both raster in terms of snow cover %. How can I eliminate those value from both raster??

My code

#reading in array Snow cover % raster from true raster (finer resolution)
img= rasterio.open(filepath+'20020310_ndsi_reclass.tif')
img_r=img.read(1)

#Reading in array Snow cover % from prediction raster (coarser resolution)
img_pre= rasterio.open(filepath+ '20020310_ndsi_reclass_pred.tif')
img_pre_r=img_pre.read(1)

#making it 1dimesional
y_true= np.ravel(img_r)
y_pred= np.ravel(img_pre_r)
y_true = y_true.astype('int')
y_pred = y_pred.astype('int')

from sklearn import metrics
print(metrics.classification_report(y_true, y_pred))

Matrix calculation

from sklearn.metrics import confusion_matrix

cm=confusion_matrix(y_true, y_pred)


import seaborn as sn
ax= plt.subplot()
sns.heatmap(cm, annot=True, ax = ax); #annot=True to annotate cells

# labels, title and ticks
ax.set_xlabel('Predicted labels');ax.set_ylabel('True labels'); 
ax.set_title('Confusion Matrix'); 
ax.xaxis.set_ticklabels(['nodata','0','10', '20', '30', '40', '50', '60', '70', '80', '90', '100', 'cloud', 'water', 'ice']); ax.yaxis.set_ticklabels(['nodata','0','10', '20', '30', '40', '50', '60', '70', '80', '90', '100', 'cloud', 'water', 'ice']);

How to exclude Nodata, cloud, water, and ice from both the report?

Answer 1

Take a look at np.where and maybe do something like:

y_true= np.ravel(img_r)
y_pred= np.ravel(img_pre_r)
y_true = y_true.astype('int')
y_pred = y_pred.astype('int')

indx = np.where((y_true != -9999) & (y_true != 205) & (y_true != 210) & (y_true != 215))

y_true = y_true[indx]
y_pred = y_pred[indx]

Or apply the np.where call to both true and pred arrays and use logical operators.