https://www.youtube.com/watch?v=rdXQNScmXS8 (also watch part-3)

Hi Munkhnaran,

To get some ideas how to assess accuracy of remotely sensed maps you should read this book:

Russell G. Congalton, Kass Green, Assessing the Accuracy of Remotely Sensed Data - Principles and Practices Second edition (2009) CRC Press, Taylor & Francis Group, Boca Raton, FL 978-1-4200-5512-2 183 pp.

The best way to compute precision of thematic map rely on confusion matrics which can be used to assess user's accuracy, producer's accuracy and Kappa statistics. Just read above mentioned book and you find an answer.

Personally I prefer Erdas Imagine module - Accuracy Assessment

Dear Munkhnasan,

For taking accuracy assessment:

1. Take samples unit from tentative desertification map. Each sample unit may consist of one or more pixel as sample (cluster).

2. Go to field and than plot each sample unit using GPS receiver. Note percentage of field condition (may be land use or land cover) within sample unit size (ex: for each pixel size of Landsat has 30 m in spatial resolution) and then validate it using confusion matrix.

3. For detail step by step, please read Jay Gao-Digital Analysis of Remotely Sensed Imagery-McGraw Hill (2009) as additional reference....

You need to explain us what your "desertification" map is; a (u)nsupervised image classification, an NDVI image with the lower values classified as desert (bare ground), a visual interpretation or a change detection (desertification defined as high cover (or NDVI) converted to low cover)?

In case you have a desert class map/image, the accuracy assessment is the same as for any other cover class map; a recent example for an accuracy assessment of remotely sensed forest cover with field data is attached.

Hope it helps

Article A comparison of fine resolution census and image-based natio...

Dear Hein Van Gils, I used 3 kind of variables (NDVI, Albedo, GSI-soil grrain size index). Then i do assessment that used decision tree classification method. 

Getting there: you classified with three imagery-derived variables.

More questions: what input imagery was used for your NDVI, Albedo and GSI?

Desert is presumably defined as a pixel, or group of pixels, with low NDVI (low green biomass), high Albedo and/or high GSI (high sand content) ?

Desertification as decrease in NDVI, increase in albedo and/or GSI?

NDVI and vegetation cover may be statistically tested for correlation, GSI and  proportion of sand at the surface as well; as far I can see you have no proxy for albedo. Albedo and NDVI may be tricky in your desert context as spatial variation in surface lithology (e.g. sand versus rock desert) within your Area-of-Interest may overrule any variation in cover or albedo signal from the "desert"; if so you better stratify your area accordingly. For an example of extraordinary low NDVI values due to calcareous bare rock (arctic-alpine desert) please see attached.

Article Fine resolution distribution modelling of endemics in Majell...

With remote sensing images display in ArcGIS would be a solution.

Dear Munkhnasan Lamchin,

You should use randomly selected points or polygon  (parcels) areas. First, you need help from a statistician in order to find min. point value (statistically acceptable number) or polygon area.  Sure, I suggest you ERDAS and it is more efficient and easy with understandable report summarize. In addition to standart accuracy assesment options available in it. Moreover, you can find good example on websites. Ofcourse, ArcGIS 10.X, Envi, Grass or IDRISI others also good options. It is your choice.

And, I added one paper for you.

Best regards, 

Article Analyzing riparian forest cover changes along the Firniz Riv...

Hi Munkhnasan,

I recommend you visit Professor Pontius website at Clark University (http://www.clarku.edu/~rpontius/). He has worked extensively on accuracy assessment. In his webpage you will find several publications for conducting this analysis for different kind of variables. He has also available a free Excel program and an R program (called diffeR, available from CRAN) that calculates different accuracy assessment metrics.

Hope this helps,

Best,

Ali

Dear Munkhnasan

I don't know for what purpose have you carried out mechanical Composition and Analysis whether your purpose is construction or use of desert soils as construction material. Whatever it may be, if you have done sewing and noted average grain sizes and if in your chemical analysis if you have determined humus content of soils, then you need to go for ground truthing. If humus content is zero or near zero and cation is generally positive and average particle size is about 0.05 mm in diameter you can easily considered such samples as belonging to desert. Using Long and Lat (if you have used GPS when sampling) you can determine whether you have correctly classified desert area or not.

Hi Munkhnasan,

I think using field sampling for assessment the resulting map is essential. Organic Matter (OM) content and soil albedo measured in field or lab(related to sand percentage) or extracted from image fro the soil samples location will be needed. You can used ENVI program wich is more easy and fast to perform the accuracy assessment. On the other hand, I suggest you to contact with Dr. Nasem_Badreldin who is working on soil desertification https://www.researchgate.net/profile/Nasem_Badreldin

Best wishes

You can use either ERDAS or ENVI for accuracy assessment., but for checking the desertification which is a process,  you need to have the temporal data for verification