In my study area, there are three components: vegetation, soil and water. I want to unmix or decompose their reflectances in mixed pixels. I read the help of the Spectral Unmixing but didn’t understand where to get the endmembers (vegetation, soil and water) and related information.
You cannot run the unmixing without a table containing the pure spectral response of your classes (endmembers).
you have to import the endmembers as a CSV table on this button 
Please have a look at this post: Weird result from the Spectral Unmixing of a Sentinel2 It includes a sample CSV file. Another example on a file comes a couple of posts later.
Given that soil moisture or vegetation influence spectral reflection, what do you mean by pure spectral response? At what level of moisture?
in case your surface considerably changes with different moisture content, you might want to add an endmember for its dry and moist condition. But this is not easy to get reliably without field measurements.
An endmember for dry and an enmember for moist?
I don’t know what surface we are talking about, so it could also not make any difference.
Spectral unmixing deals with the proportional fraction of a defined reflectance within a pixel. If moisture makes a difference here and it’s important to you, you need to include it. If the surface reflectance stays the same, you can use one endmember for this class.
The surface is soil and vegetation. And since moisture affects the spectral reflectance of both of them, it’s important for us. The question is, do we have to determine four members? Dry soil, wet soil, dry vegetation and wet vegetation?
of course, you need to provide an input to the unmixing module in any case.
But I am not sure if unmixing is the best choice here because dry and wet endmembers probably are too similar.
Maybe a supervised classification might be the better option?
In fact, we want to determine the contribution of soil and vegetation to the reflection of a mixed pixel (pixels containing soil and vegetation)