alanacg/cymplEO-umd

(in development) GEOCIF-inspired machine learning crop yield model for maize crop in Kenya (previously named umd-agroml) # Guide to cymplEO ### Steps to modeling: 1. Step 1: Prepare RS Data - (a) Extraction of input features via Google Earth Engine - (b) Python package requirements and installation - (c) Scripts to combine GEE Data and add yield 2. Step 2: Prepare ML models - (a) Select Models To Use - (b) Convert EO data into descriptive, model-compatible features - (c) Perform feature importance and hyperparameter tuning 3. Run ML models 4. Analyze Model Output - Analyze CSV output metrics - create graphics ## Step 1: Prepare Remote Sensing Data ### Part (a) - Extraction of input features via Google Earth Engine This model was created with the intention of relying exclusively on inputs derived from remote sensing (RS) products, to assess their feasibility in producing results without requiring extensive field campaigns. This data has been sourced in the past exclusively using Google Earth Engine (GEE) scripts, however methods are available for extracting data not available on GEE as well. The variables chosen in this example cover both meterological influences to crop growth, and metrics of ecological conditions related to crop growth. These include air temperature, precipitation, soil moisture, evaporative stress index (ESI) and normalized differential vegetation index (NDVI). Copy repository from this link: code.earthengine.google.com (not shared 5/27) Needed files: * Shapefile of regional boundaries (Administrative 1 County-level for Kenya) asset * Shapefile of crop mask (maize for this model) asset * CCI global LC Layers 2015-2020 assets ### Part (b) - Python package requirements and installation Package requirements: - pandas - numpy - scikit-learn - pip install merf - pip install xgboost - pip install glob2 Most code for this model can be executed as scripts without an IDE and just a code editor. However, the scripts have been organized into Jupyter Notebooks for this repository, and can be ran with various local platf …