When Climate Meets Disease: Lessons from a Week of Modelling in Kigali

Inside the DHIS2 Workshop on Spatiotemporal Modelling of Climate Sensitive Diseases — what we learned, what still needs to change, and why it matters for health systems across Africa and beyond.

Malaria cases surge after heavy rains. Cholera spreads when flooding overwhelms water systems. Dengue fever tracks the movement of mosquitoes as temperatures rise. These are not coincidences — they are patterns. And for the first time, we have the tools to read them in advance.

That was the central premise behind a week-long workshop held in Kigali, Rwanda, where members of the Climate-Sensitive Infectious Disease Network (CSIDNet) gathered alongside practitioners from ministries of health, research institutions, and partner organizations spanning Nigeria, Mozambique, Ghana, Uganda, Nepal, Lesotho, South Africa, Rwanda, and Kenya.

The event, co-hosted by DHIS2, the HISP Centre at the University of Oslo, and the Rwanda Ministry of Health, was not just a training. It was a reckoning with a simple but difficult question: if we can forecast where and when disease outbreaks are likely to occur, why aren’t more health systems acting on that information?

Why “Where” and “When” Need to Be Asked Together

Traditional disease surveillance is often retrospective — it tells you what happened and where but rarely helps you anticipate what’s coming next. Spatiotemporal modelling changes that calculus by treating space and time not as separate dimensions to be analyzed one at a time, but as interconnected forces that together explain how disease unfolds.

Think of it as capturing the “4D nature” of reality: latitude, longitude, altitude, and time. Disease outbreaks don’t occur randomly. They spread across regions, follow seasonal rhythms, and respond to environmental triggers. A model that can represent all four dimensions simultaneously gives public health practitioners something far more powerful than a map of where cases are — it gives them a window into where cases are going.

“Global climate variability, environmental change, and the increasing spread of infectious diseases demand analytical approaches that can account for both spatial and temporal dynamics.”

For climate-sensitive diseases like malaria, cholera, and dengue, this matters enormously. Rainfall patterns, temperature fluctuations, and land cover changes don’t just correlate with outbreaks — they precede them. A health system that can read those signals early enough has a real chance to respond before case counts start climbing.

That said, spatiotemporal modelling is not always the right tool. Its usefulness depends on the nature of the research question and, critically, on the availability and quality of spatial and temporal data. The workshop was honest about this: analytical sophistication without good data is not sophistication at all.

The Tools: CHAP, DHIS2, and the Power of Integration

Much of the workshop’s technical content centered on the CHAP Modelling Platform and its integration with DHIS2, the open-source health information platform used by dozens of countries. The combination is significant because it connects analytical power to existing infrastructure — rather than requiring health ministries to adopt entirely new systems, it builds on what they already have.

The workflow looks like this: routine disease surveillance data — malaria case counts, cholera reports, dengue notifications — are extracted from DHIS2. They are then combined with climate and environmental covariates such as rainfall and temperature using DHIS2 Climate Tools. Those harmonized datasets train forecasting models that learn the relationships between disease incidence, seasonality, and climate drivers. Validated forecasts are then fed back into DHIS2, where district health officers can view predictions alongside the routine data they already use every day.

Instead of analysis happening months after the outbreak, health authorities can work with near real-time trends and generate predictions as part of standard operations.

The workshop introduced participants to both frequentist and Bayesian modelling frameworks, with particular attention to how uncertainty should be communicated. This is where a metric called the Continuous Ranked Probability Score (CRPS) became important: it measures not only how close a model’s central estimate was to what happened, but whether the uncertainty expressed around that estimate matched observed reality.

For a health officer deciding whether to pre-position medical supplies, that distinction is not academic. Knowing that a model predicts a likely surge is useful. Knowing how confident to be in that prediction is what makes action possible.

What Country Experiences Reveal: The Gap Between Models and Action

The workshop brought together early warning system practitioners from across the network, and their country updates told a strikingly consistent story — one in which technical progress is outpacing institutional readiness.

The countries making it work

Ghana has one of the most advanced deployments in the network. Its malaria early warning system for children under five is fully operational in DHIS2 production, with District Health Officers and the national malaria programme as active users. Malawi is close behind, with active validation underway and a two-tier national-and-district deployment planned. In both cases, the key to progress was co-design: bringing programme staff into model development from the start, including giving them a direct role in defining the thresholds that trigger alerts.

The countries stuck at the last mile

Zimbabwe has a working cholera model. Disease data sits ready at the Ministry of Health. Climate data is accessible through the Meteorological Department. But the system isn’t integrated, and the reason isn’t technical — it’s bureaucratic. In Nepal, dengue predictions are running on a test server with multi-source climate data and ongoing validation, but full deployment has stalled because key health officials were transferred, taking institutional knowledge with them.

These are not isolated failures. They reflect a structural reality that the workshop named directly: technical readiness and operational readiness are not the same thing.

Resource allocation: the most immediate dividend

When early warning systems do reach operational status, smarter resource allocation is the clearest benefit. Togo is designing malaria forecasts six months ahead specifically to feed into procurement cycles for vaccines, rapid diagnostic tests, and drugs. Nigeria envisions predictions that position the right commodities in the right locations before case numbers surge. The logic is consistent across all contexts: forecasts reduce guesswork in supply chain planning, and preparation is almost always cheaper than emergency response.

The Real Bottleneck Is Not the Algorithm

Perhaps the most important insight from the workshop is also the most humbling: across every country represented, the binding constraint is almost never the model.

The challenges that keep functional forecasting tools from becoming operational policy tools are human and institutional. They include:

• Bureaucratic friction: data exists, needs are understood, yet administrative structures slow the flow of information to decision-makers.
• Capacity gaps: district and subnational health systems are structurally under-resourced, limiting their ability to act on forecasts even when they receive them.
• The buy-in catch-22: ministries require demonstrated value before committing, but demonstration requires operational deployment, which requires commitment.
• Coordination failures: personnel transfers, fragmented stakeholder ownership, and unclear data governance are endemic features of resource-pressured health systems.

The gap between a model on a test server and a health manager acting on a forecast is, in the end, a human and institutional gap.

This is not a reason for pessimism. It is a clear signal about where investment needs to go next: joint validation processes, district-level capacity building, clear data governance frameworks, and sustained ministry engagement. The technical foundations are solid. Institutional architecture needs to catch up.

What CSIDNet Can Do That No Single Country Can Do Alone

The country experiences documented through this workshop, read collectively, reveal something important: the bottlenecks are remarkably consistent across very different contexts. Zimbabwe, Nepal, Nigeria, and Sri Lanka are not dealing with the same diseases or the same climate pressures, but they are struggling with the same institutional friction. That convergence is not just a problem — it is an opportunity.

It means that solutions developed in one context are very likely to transfer to others. And it points to a clear role for CSIDNet as the connective tissue that allows those solutions to travel.

Peer learning and regional collaboration

Countries at more advanced stages of deployment — Ghana, Malawi — have hard-won insights that countries at earlier stages — Ethiopia, Nigeria — urgently need. CSIDNet can formalize what is currently happening informally by establishing structured peer-learning mechanisms and regional working groups organized around shared disease burdens.

Shared modelling infrastructure

Rather than every country building from scratch, CSIDNet can develop and maintain a common model library with validated approaches for malaria, dengue, and cholera. Shared computing infrastructure, a flexible roster of modelling expertise, and openly available methodological advances would dramatically lower the barrier to entry for countries still building foundational capacity.

Institutional brokering

Perhaps most importantly, CSIDNet can play a role that no individual country team can play alone: serving as a neutral convener that helps broker relationships between technical teams and ministries of health. The buy-in catch-22 is fundamentally a coordination failure — and an established network with credibility and convening power is exactly the kind of entity that can help break it.

The immediate next steps are concrete: a readiness tiering of member countries to target support more precisely; a shared validation protocol that gives country teams a ministry-endorsed standard to apply; and a data governance working group focused specifically on the bureaucratic access barriers that are stalling progress in multiple countries simultaneously.

The Bottom Line

For the participants who spent a week in Kigali working through time-series forecasting, Bayesian models, machine learning approaches, and uncertainty metrics, the experience was more than a skills upgrade. It was “a paradigm-widening” encounter with how different analytical traditions are converging around the same urgent challenge.

Climate change is not a future threat to health systems — it is reshaping disease patterns right now. The analytical tools to respond to that are increasingly mature. The integrated platforms to deploy them at scale exist. The data, in many countries, is already being collected.

What the Kigali workshop made clear is that the final distance between a working model and a health manager acting on its output is not technical. It is institutional, relational, and political. Closing that distance is the work ahead — and it is work that no single country, no matter how technically advanced, can do alone.

That is what networks are for. And that is what CSIDNet is for. CSIDNet brings together researchers, practitioners, and policymakers committed to building climate-resilient health systems. Whether you work in a health ministry, a research institution, or a partner organization, there is a role for you in this network. Follow our work, share your experience, and help turn country-level efforts into a network whose impact is greater than the sum of its parts.

A collective reflection by CSID Network members:
Daniel Adediran, Elias Dogbatsey, Henok Tadessa Bireda, Henry Njoku, Robert Selemani, Sisay Wondaya Adall, Verrah Otiende, and Yusuf Suleiman