Willibroad G.
Buma, PhD

The assignment

Kenya required a spatially consistent way to identify where irrigation investment could be technically viable and strategically aligned with agricultural development. The work brought together land suitability, water availability, groundwater conditions and agricultural value-chain priorities across a country with significant spatial variation in all of them.

My contribution

• Led national FLID suitability analysis across approximately 800,000 ha, 46 counties and 281 subcounties.
• Developed investment colocation analysis covering 718 locations and nine agricultural clusters.
• Applied hydronomic zoning as a pre-screening framework for investment targeting.
• Designed a service registry for the National Irrigation Informatics Center covering 22 priority services.
• Supported technical engagement with national irrigation institutions and preparation of operational documentation.

Operational use

The analytics support investment design under Kenya's national irrigation program and contribute to the technical design of the National Irrigation Informatics Center, including its analytical services, institutional interfaces and digital architecture.

Google Earth Engine · Python · spatial analysis · groundwater screening · satellite earth observation · hydronomic zoning

The assignment

Participating countries enter the program with different institutional arrangements, data systems and levels of digital maturity. The challenge is to build interoperability and regional intelligence without replacing or duplicating national systems, and to do it in a way governments can actually own and sustain.

My contribution

• Developed the national-to-regional analytical architecture.
• Defined interoperability principles for national data nodes and regional services.
• Developed an 18-month implementation and sequencing roadmap.
• Contributed to terms of reference for geospatial, benchmarking and knowledge-management systems.
• Supported government engagement and technical communication across Francophone and Anglophone countries.

Selected outputs

Regional geospatial architecture, a national system interoperability framework, a benchmarking and accountability framework, and an implementation sequencing roadmap.

Platform architecture · metadata design · digital public infrastructure · monitoring frameworks · institutional design

The assignment

Nigeria's water and irrigation data are distributed across federal and state institutions with limited interoperability. The work explores how a federated national architecture can establish common standards while preserving state-level ownership of operational data and workflows.

My contribution

• Supported field and technical engagement with the Kano State Project Implementation Unit.
• Assessed existing state-level platform concepts and interoperability requirements.
• Applied satellite-based irrigation performance analytics using scheme-level spatial data.
• Developed the federal reference architecture for the Nigeria Irrigation and Water Intelligence Platform.
• Supported engagement on national hosting and infrastructure arrangements.

Operational direction

The design uses state-level implementation experience to shape a scalable national reference architecture while preserving state ownership of operational data and workflows.

Google Earth Engine · satellite monitoring · platform architecture · spatial analytics · institutional data systems

Burundi and Tanzania

Developed irrigation investment analytics using satellite evapotranspiration, precipitation, soil and land-cover datasets. Supported remote sensing-based water accounting along major economic corridors, linking water availability and use with investment planning questions.

Georgia

Developed high-resolution irrigation performance benchmarking across multiple irrigation schemes using satellite-derived indicators and a structured performance framework.

Methods

WaPOR · CHIRPS · Google Earth Engine · Python · R · water accounting · irrigation performance assessment · spatial prioritization

Operational monitoring

• Satellite-based irrigation performance assessment across national and scheme-level scales.
• Flood mapping and surface-water dynamics using Sentinel-1 SAR.
• Reservoir monitoring and water-body change detection.
• Agricultural and land-use monitoring for water planning.

Remote sensing research

• Hyperspectral and multispectral imagery analysis for vegetation and water quality.
• PROSAIL radiative transfer modeling and inversion.
• Leaf area index retrieval and soil moisture estimation.
• Field spectroscopy and model validation.

Research environment

From 2021 to 2024, I conducted remote sensing research at the U.S. Naval Research Laboratory, combining field measurements, satellite observations, physical modeling and machine learning. That work produced more than ten peer-reviewed publications and forms the scientific foundation for the operational work that followed.

Sentinel-1 · Sentinel-2 · Landsat · MODIS · hyperspectral imagery · PROSAIL · Python · machine learning · spectroscopy

Applied AI

• Co-leading work under the WBG Water AI Task Force, covering a $2B+ lending portfolio.
• Supporting development and testing of AskWater, a conversational knowledge system for the water sector.
• Developing evaluation frameworks and testing protocols for LLMs in operational contexts.
• Identifying and scoping AI use cases across water-sector projects and analytical work.

Predictive analytics

Machine learning work includes random forest classification, remote sensing model development, spatial prediction and multi-country environmental analytics. Recent work extends into NLP-based document analysis and structured classification of unstructured project text.

Methods

Python · RAG · LLM evaluation · random forest · machine learning · AI product design · geospatial AI · NLP

The question

Innovation is rarely recorded as a structured field in project documents. It appears in descriptions of technologies, institutional arrangements, financing mechanisms and pilots. This pipeline tests whether those signals can be systematically identified and classified at scale using NLP and LLM-assisted classification validated against expert judgment.

Approach

• Retrieved and processed 60 World Bank PADs via the Documents and Reports API.
• Extracted 484 component and subcomponent analytical units, chunked and embedded using sentence-transformers.
• Used semantic retrieval to identify 65 innovation candidates from 1,781 passages.
• Classified candidates against a structured innovation ontology using an LLM classifier.
• Validated results against 35 expert-labelled records across four innovation classes.

Results

74.3% four-class accuracy. Binary innovation-signal precision: 1.000, meaning no false-positive innovation signals in the validation sample. Macro F1: 0.785. Binary F1: 0.939.

Python · sentence-transformers · Groq · scikit-learn · Streamlit