Spectral signature wise you classify mangrove vegetation and minimum 30 plot survey height measurement of the tree

Landsat didn't measure height. Try ICESat or ICESat-2 instead.

https://icesat.gsfc.nasa.gov/

They had lidar sensors (GLAS & ATLAS resp.) and measured tree height. Point density was very low and the footprint large but with ICESat this forest heights map was generated:

https://visibleearth.nasa.gov/images/76901/forest-canopy-height

Data from ATLAS/ICESat-2 L3A Land and Vegetation Height, Version 2 is here:

https://nsidc.org/data/ATL08/versions/2

The new spaceborne lidar sensor is GEDI. Some data products from the GEDI mission are already available: https://gedi.umd.edu/

Canopy height is in product L1A-2A and I believe it is still not available.

Try with sentinnel 2 (10m resolution), using an object based approach (segmentation), the try to combine the two variables the "shadow" class and texture indexes (entropy, homogenity...)

Good luck

Landsat 30m will not be very helpful here as its spatial resolution is too coarse and really not necessary since Sentinel 2 imagery has better resolution and is also freely available.

I would consider doing a shadow analysis using object-oriented/ machine learning approaches such as exemplified here with ESRI ArcGIS products : https://github.com/Esri/raster-deep-learning , you can check them out and choose the one most appropriate for you. I have personally used the 'Object Classification implementation with PyTorch' on Sentinel 10m imagery with satisfactory results. You can find the python Jupyter notebook here : https://developers.arcgis.com/python/sample-notebooks/building-damage-assessment-using-feature-classifier/

I also came across this article that uses a different methodology and might be useful to you: https://www.mdpi.com/2072-4292/10/7/1134/htm

Success!

Zahrah.