Project Overview

A full-stack geospatial platform for precision agriculture, integrating satellite-derived indices, soil analysis, and machine learning to deliver per-farm yield predictions and actionable agronomic recommendations.

Problem Statement

Smallholder and commercial farm managers lacked an integrated, spatially explicit tool to monitor crop health, assess soil conditions, and forecast yields — decisions were largely reactive and not data-driven.

Methodology

A Vue 3 + OpenLayers frontend was paired with a Python/Flask REST API backed by PostGIS for spatial data management. Farm features were managed as dynamic vector layers with full geometry CRUD. Satellite-derived NDVI time-series and zonal statistics were computed per farm polygon. Soil moisture and property mapping fed a composite risk index. Yield prediction was implemented using an XGBoost regression model (R² = 0.954, MAE ≈ 3.38 bags/acre) trained on annotated farm sample data. JWT-secured endpoints managed project, layer, and feature lifecycles.

Key Findings

The platform achieved 95.4% model confidence on yield prediction across test features. NDVI zonal stats and change detection reliably identified crop stress zones. The recommendations engine generated context-aware agronomic guidance per farm feature based on soil and vegetation outputs.

Project Overview

An evaluation of spatial access to primary healthcare facilities across a target region in Kenya, identifying coverage gaps and optimal facility placement scenarios.

Problem Statement

A significant proportion of the target population faced extended travel times to the nearest healthcare facility due to inadequate spatial planning of facility locations.

Methodology

A GIS-based accessibility model was developed using network analysis combined with population distribution layers derived from WorldPop data. Travel time isochrones were generated to delineate service catchment areas.

Key Findings

Analysis identified zones where over 35% of the population exceeded recommended travel thresholds. Optimal locations for two additional facilities were modeled to reduce underserved population by 60%.

Project Overview

A decade-long satellite-based analysis of land use and land cover change to quantify urban expansion, agricultural encroachment, and vegetation loss.

Problem Statement

Rapid and unmonitored land use change was creating environmental risks including soil degradation, water stress, and loss of ecosystem services in the study region.

Methodology

Multi-temporal Landsat imagery was classified using a Random Forest supervised classification algorithm in Google Earth Engine. Post-classification change detection was applied to quantify transition matrices across land cover classes.

Key Findings

Urban land expanded by 28% over the study period. Forest cover declined by 14% with concentrated loss in buffer zones. Agricultural land increased in previously forested areas.

Project Overview

Spatial assessment of climate-induced agricultural vulnerability across smallholder farming zones using satellite-derived vegetation and rainfall indices.

Problem Statement

Smallholder farmers in the target region lacked spatially explicit risk information to guide adaptive agricultural planning in response to increasing rainfall variability.

Methodology

MODIS-derived NDVI time-series data was integrated with CHIRPS rainfall anomaly datasets. Vulnerability zones were classified using a composite risk index combining vegetation stress and precipitation deficits.

Key Findings

Approximately 40% of the agricultural zone was classified as high to very high climate risk. Vulnerable zones showed a strong spatial correlation with areas of declining seasonal NDVI trends.