Satellite-Based AGB/CHM Estimation Model​

Overview.

concept concept

This project is an UNET probabilistic deep learning model for estimating forest structure data such as canopy height/above ground biomass from satellite-based data such as sentinel 1 , sentinel 2 and GEDI data.

Here you can find all relevant information about this project in this repository.

Repository

Source code available here

AGB Output Result , Model Checkpoint , Data

  • AGB baseline output mean , variance result from paper are locate at Varuna GCS

Path : varuna-data-nonprod-analytic/biomass-estimation-project/vm-backup/AGB_model_data/ensemble_final_output

  • Ensemble model checkpoints are locate at Varuna GCS

Path : varuna-data-nonprod-analytic/biomass-estimation-project/vm-backup/AGB_model_data/best_checkpoint/ensem1

  • Sampling training data (train/test/val) are locate at Varuna GCS

Path : varuna-data-nonprod-analytic/biomass-estimation-project/vm-backup/AGB_model_data/train_data

Prerequisites

Attention

Some background in geospatial data analysis is require to work through the project. For new comer who does not have geospatial background, do not skip this step.

PyGIS - Open Source Spatial Programming & Remote Sensing will introduce you to the methods required for spatial programming and make you getting familiar with concept like raster, vector, coordinate reference systems

First-timer

If you don’t have any clue about this project , go through Theorem section to look for brief overview of the project

It will guide you from background and details literature , methodology such as

  • Problem assetment of the project
  • Biomass / Aboveground Biomass definition
  • Remote Sensing process
  • Model Methodology

Dev/ML

Explore Data and Model section for implementation detail of the project

User

  • Prediction Section will walk through step-by-step how to use model to produce outcome.
  • Application Section descibe how to interpret the output prediction and get correct result.