search

hamidrezaomidvar / LINDER

LINDER (Land use INDexER) is an open-source machine-learning based land use/land cover (LULC) classifier using Sentinel 2 satellite imagery
13 stars 2 forks source link

Evergreen vs deciduous #5

Open hamidrezaomidvar opened 5 years ago

hamidrezaomidvar commented 5 years ago

Do we have any land cover data set that distinguish between evergreen and deciduous vegetation? I found a potential method that we might be able to predict these two classes using two images. It is based on combination of NDVI and NDWI

hamidrezaomidvar commented 5 years ago

Some background: There are two separate metrics that can separate vegetation and water. The first one is NDVI as follows:

NDVI=\frac{NIR-RED}{NIR+RED}

NIR and RED are near infrared and Red bands. This is a very good indicator of vegetations in the satellite images. Another indicator is NDWI, which is a good indicator of water bodies:

NDWI_1=\frac{GREEN-NIR}{GREEN+NIR}

where GREEN is the green band in the satellite data. So far I have been using these two metric to predict water, vegetation, and others. However, there is another version of NDWI which is a good indicator of the amount of water (either in vegetation or water):

NDWI_2=\frac{NIR-SWIR}{NIR-SWIR}

where SWIR is the shortwave infrared band. Let's take a look at these three features for Colombo:

bands

You can see that NDVI and NDWI_1 can separate vegetation and water very good. NDWI_2 is a combination of water and vegetation.

So now I am thinking if we look at NDVI-NDWI_2, where NDVI values are from an image during the growing season, and NDWI_2 is from an image during non-growing season, then the value of NDVI-NDWI_2 should be high for deciduous vegetation, while it is small for other. This is what we see here for a forrest I plotted for north America:

QB

So we can give this additional metric to our model to predict these kinds of vegetations. To do so, we need data sets containing these two different vegetations

sunt05 commented 5 years ago

Have a look at this site: http://data.ess.tsinghua.edu.cn

I'm not quite sure if it would be useful.

hamidrezaomidvar commented 5 years ago

Thanks, I will take a look.