Ibrahim Kabore

Designed and implemented a reproducible environmental ML pipeline combining Sentinel-2, climate datasets, and geospatial features for cereal yield prediction. The project focused on using radar remote sensing to monitor agro-environmental dynamics, correlating vegetation changes with climatic factors.

• Sensor Data Acquisition & Preprocessing: Downloaded, filtered (VV and VH bands), and rigorously preprocessed imagery from the Sentinel-1 radar satellite, ensuring consistent observation regardless of cloud cover or sunlight, overcoming optical sensor limitations.
• Feature Engineering & Indicator Calculation: Calculated the Radar Vegetation Index (RVI) from the VV and VH polarization bands, a key indicator of green aerial biomass and active vegetation cover.
• Multi-Source Data Integration: Integrated time-series of external meteorological data, including CHIRPS (for precipitation) and ERA5 (for temperature), to contextualize RVI evolution and analyze hydric stress factors.
• Correlation Analysis & Interpretation: Conducted correlation analysis between RVI, precipitation, and temperature (2022-2024) , revealing a clear link between declining vegetation and irregular precipitation/rising temperatures. Demonstrated that the irrigated areas showed partial resilience to these constraints.
• Technical Environment: Utilized the Google Earth Engine (GEE) platform for processing, time-series analysis, and thematic mapping (ArcGIS) of the results (RVI, precipitation, temperature).

Built end-to-end preprocessing and ML pipelines for large soil spectral datasets to support SOM (%) prediction and spectral harmonization efforts.

• Developed a complete spectral preprocessing workflow: wavelength trimming, 5 nm resampling, log10(1/R), Savitzky–Golay smoothing, SNV/MSC normalization, GapDer derivatives, modpoly baseline correction, PCA compression (99% variance), and outlier removal.
• Trained and compared PLSR, PCR, RF, SVR, Cubist, and QRF regression models, reporting standardized metrics (R², RMSE, MAE) and uncertainty estimates.
• Deployed a production Shiny application for soil organic matter prediction with quality flags and uncertainty layers.
• Contributed to spectral library harmonization (Moroccan & Mediterranean soils) through calibration, standardization, and QA/QC.
• Delivered fully documented workflows and trained team members on preprocessing order and evaluation standards.