Hi,
Just carrying the accuracy assessment for the landuse maps as i have selected the randomly 100 samples over all of the basin (20 each class). The kappa coefficient is 0.86, as i think the number of samples that i collected are small please, let me know any criteria of selection of samples for accuracy assessment.
Of course, as far as I know, there is no definition for the actual number of land locations. But it is better to add points that are evenly distributed over your entire area.
Dear Usama Nazar
The kappa statistic is used to regulate just those events that may have been categorized properly by chance. This may be computed using both the observed (total) and random accuracy. The formula for calculating Kappa is: Kappa = (total accuracy – random accuracy) / (1- random accuracy).
The kappa coefficient is another accuracy statistic. It is a measure of how the categorization results compare to random values. It can take values ranging from 0 to 1. If the kappa coefficient is equal to zero, there is no agreement between the categorized and reference images.
To compute the % accuracy, divide the total number of accurate reference points by the total number of "true" reference points and multiply by 100.
Kind Regards
Qamar Ul Islam
Qamar Ul Islam but how to select the number of samples for kappa coefficient, is there any criteria.
As i have taken 100 samples randomly on 6428 km2 without any referance.
Have a look at our published work
Article Analytical Study on Urban Expansion Using the Spatial and Te...
Actually that is depend on the size of your study area and how many land use classes in your LULC classification .it is better to select random points to equally represent the each landuse classes in your traget /area.
Hi!
The methods for selecting samples for the kappa coefficient are the same as for selecting samples for the confusion matrix. Random sampling and stratified random sampling are the most recommended. To ensure all classes are represented I would recommend stratified random sampling. Regarding the number of samples, usually, a percentage of the estimated area is used (from 1 to 5 % is reasonable). A minimum of 50 samples per class is also recommended.
To summarize, I would recommend using stratified random sampling, number of samples 1% of the estimated areas for each class (usually you can do it automatically in software packages).
Hope it helps, regards
As of now, there is no set criteria to select the number of samples for the accuracy assessment of LULC study using Kappa Coefficient. However, samples should be taken from all the landsue classes throughout the study area. Higher the number of samples more will be the accuracy.
- Badges
- Science topic
- Similar topics
- Cognitive Science
- Thinking