I have NDVI images from four different years ( e.g., 2007,2008,2010 &2018) but all are in same month ( e.g., January). The range of all NDVI products are also varies significantly. I need to classify these NDVI as sparse and dense vegetation. How could i make the threshold for my desire classification without using any ground truth information.Before classify do i need to resale the NDVI outcomes and if so can i rescale all the products in a range of -1 to -1 or adjust with the NDVI which has highest values.
shareholding the most difficult i used several threshold and then choose the one with the best results , try estimate threshold you think it useful and start changing it and see the best results.
Teaba Wala Aldeen Khairi thanks for your answer. How did you check the results. Can you tell me the process ?
Dear Islam,
if no ground truth data is used then how do you expect to verify your results?
Anyway NDVI should be between -1 and 1 if not you have to round all values below -1 to 1 and above 1 to 1. For the sake of threshold value I would use the following procedure I would ignore -1 and 0 values and then I take the mean and standard deviation for every month of January for each year for values above 0. Then I will use these statistics as my thresholds for sparse, sparse to dense and dense vegetation. This is one way there are many ways using other different statistics.
Best regards
Dear sir
Yes your assumption is right. But my purpose is a bit different from traditional vegetation classification.
What you have suggested seems reasonable. But I have few confusions regarding threshold value. I just need two classes sparse and dense. In this case what value I need to consider? Is it the mean? Suppose I have got a mean value of 0.23 in case of January 2007, hence 0 to 0.23 would be my spare and > 0.23 would be the dense, isn't it ?
Similary i need to calculate for my other year?
Min additional prior to fixing the threshold statistics I need to rescale the NDVIs?
i don't clearly understand the use of standard deviation value. Can you clarify please.
Lastly I would request you to suggest me some relevant articles regarding this issue
Thanks again and looking forward to hear from you
Dear Islam,
Well it is better to know from what value of NDVI you have vegetation. Whether it is very low in density, medium density or dense or highly dense, it is important to separate vegetation from other classes or at least to know the NDVI value where there is really vegetation. From there I would take the mean and classify the NDVI values where less than the mean as highly sparse to sparse vegetation and another class as dense to highly dense. In this case the mean is the threshold. You can also try other indexes such as EVI, SAVI and others.
Thanks for your further clarification.
Did you mean the "Mean" value of a particular year NDVI after normalising the NDVI values to 0 to 1?
If I want to compare my vegetation condition considering year 2007 as my base level. What do you think individual means for year 2008,2010 and 2018 would give me the logical threshold values for sparse and dense?
Let me clarify my query, suppose I have got mean value 0.13 for year 2007 which indicates that values ~0 to 0.13 classified as sparse veg. and >0.13 as dense veg. For year 2008 let’s assume got the mean 0.15 then my classification would be ~0 to 0.15 as sparse and >0.15 is dense for year 2008. Now the problem is there are some pixels in 2008 having NDVI values ~ 0.13 to 0.15 which eventually classified as dense in year 2007. How could I interpret this particular issue?
Hope you understand and would suggest me accordingly.
Dear Islam,
you can normalize the data between 0 and 1 (remove negative NDVI) and use the standard deviation as the starting point for highly sparse to mean (sparse) and above mean to maximum from dense to highly dense. This way you have two classes
highly sparse to sparse
and
dense to highly dense
Best regards
- Badges
- Science topic
- Similar topics
- Computer Science
- Images