nasa-nccs-hpda/ethiopia-lcluc-tensorflow

Ethiopia LCLUC using Deep Learning # Ethiopia LCLUC Ethiopia LCLUC using WorldView imagery ## Objectives - LCLUC utilizing random forest algorithm - LCLUC utilizing XGBoost algorithm - LCLUC utilizing CNN algorithm - LCLUC utilizing CNN ensemble algorithm ## Data Catalog ```bash - Project Location: /explore/nobackup/projects/ilab/projects/Ethiopia/LCLUC_Ethiopia - Full Domain Data Location: /adapt/nobackup/people/mwooten3/Ethiopia_Woubet/VHR - Gonji Subset Data Location: /adapt/nobackup/people/walemu/NASA_NPP/CRPld_Map_Pred_and_Forec/EVHR/Gonji_Subset/5-toas ``` ## Structure of this Repository This repository takes care of preprocessing, training, inference, and compositing of WorldView imagery for Ethiopia. The different steps are guided by pipelines. There are two main pipelines available in this repository: - Land Cover: generates GeoTIFF predictions of land cover outputs - Compositing: takes the outputs from the Land Cover pipeline and generates multi-year composites ## Explore/ADAPT Basic Information 1. SSH to ADAPT Login ```bash ssh adaptlogin.nccs.nasa.gov ``` 2. SSH to GPU Login ```bash ssh gpulogin1 ``` 3. Clone above-shrubs repository Clone the github: ```bash git clone github.com ``` 4. Accessing the container To download a clean version of the container, run the following command: ```bash singularity build --sandbox /lscratch/$USER/container/ethiopia-lcluc-tensorflow docker://nasanccs/ethiopia-lcluc-tensorflow:latest ``` An already downloaded version of the container is location in the Explore HPC cluster under: ```bash /explore/nobackup/projects/ilab/containers/ethiopia-lcluc-tensorflow.2025.04 ``` ## Workflow Documentation ### Land Cover Outputs Generation TBD ### Cloud Masking Outputs Generation NOTE: these instructions need to be updated with the new vhr-cloudmask software developed by the team. Overall example to run cloud masking: ```bash for i in {0..64}; do sbatch --mem-per-cpu=10240 -G1 -c10 -q ila …