I am using landsat 7 & 8 imagery to detect the forest vegetation and land cover change in a tropical swamp forest over 20 years time period. I have choosed two vegetation indices -NDVI & EVI. I have the multitemporal mean values of NDVI & EVI for the same study area & period. In other hand, i am using maximum likelihood supervised classification to classify the NDVI and EVI derived map. I am using confusion or error matrix to assess the accuracy of the classification with ground truth data. I want to know that, can i employ any further statistical test to compare the mean NDVI and EVI value? Is there any suitable test to compare NDVI and EVI mean value ?
you can you t test to compare means if the data is normally distributed, otherwise use Mann Whitney test :)
An Independent t-test can be used to compare means If the data set is normally distributed.
Furthermore, you can go for a "Trend Analysis" due to your multi-temporal approach. It is actually not a mean comparison of NDVI or EVI, but an extra analysis for your data
I think EVI and NDVI data can be similar in responses to land cover and vegetation. That means you can see whether they are correlated or not. You can also create temporal profile (time series/trend lines) for both EVI and NDVI. Time series will show perfect situation happening for all 20 years. From that you can choose which time sector (section) you want to analyze.
As mentioned by others, if the data is normally distributed, you can use simple t-test to test the means from chosen time sector (or whole 20 years).
There are many published article you can search online to follow through. Here are few..
Article A comparative analysis of multitemporal MODIS EVI and NDVI d...
https://ieeexplore.ieee.org/abstract/document/4293049
Article Analysis of Normalized Difference Vegetation Index (NDVI) mu...
Given the nature of the data / type of changes studied, it does indeed seem to make the most sense to study changes over time. Of course, they can be analyzed globally for the research area, but such information will say little, and only indicate a general trend. Therefore, I would suggest using geostatistical analyzes offered by GIS. Then you will get a spatial distribution. After all, you can attach tables with data from statistical analyzes that you will perform in external statistical packages to the GIS database you have.
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification