How we did a accuracy assessment

Yesterday, I wrote up the accuracy assessment method we used for a landuse/landcover change detection project we used to do. Figured that some of you have been asking how to conduct an accuracy assssment, I'm putting it here. Might be of some use.

**************************************************************************
Accuracy Assessment for the Change Detection Program
A Brief Description

Zhanfeng Liu, zliu@spf-davis.info


Sampling Method

* Stratified Random Sampling


Reference Data

* Aerialphotography (color, hardcopy)
* Digital flight photos
* DOQQ


Results

* Standard Error Matrix Analysis
o Overall Accuracy
o Omission Error/Producer's Accuracy
o Commission Error/User's Accuracy
* ** Be noted that theoretically speaking, the Standard Error Matrix was designed for a Simple Random Sampling, therefore when applied to data obtained via a Stratified Random Sampling could produce biased estimates.


Procedure

1. Sample sites and polygons for the accuracy assessment were generated based on stratified random sampling strategy using a customized ArcView program . (manual for the ArcView 3.2 extension -- ACCP) The ArcView ACCP Extension was later implemented in ArcGIS 9.1.
2. For a sample site where reference data existed in digital format, the reference image was clipped out using the sample site polygon (location shift performed when necessary based on visual inspection). Then, canopy coverage was estimated using the ForestEye software.
3. For a sample site where reference data existed in hardcopy aerialphotography, the sample polygon was first printed onto a mylar sheet which was then used to pin-point the site on the aerial photo. Canopy coverage was estimated using a measurement grid system specifically designed for the project.
4. Canopy coverage estimates were input into a customized MS ACCESS database which could automatically calculate all the standard error matrix results.


Read the Original document page here.