Climate-Resilient ICT: How Digital Systems Are Being Designed for Climate Shocks in the Global South

Abstract data-network visualization overlaid with storm-pattern imagery, representing climate-resilient digital infrastructure

Development informatics has spent three decades learning that digital systems fail when they ignore the conditions of the field: intermittent power, unreliable connectivity, low digital literacy, multilingual user bases. Climate shocks are now forcing a parallel reckoning. A mobile money agent network, a digital health record system, or an early-warning platform that performs well in ordinary conditions can fail exactly when it is needed most — during the flood, the drought, or the heat event that disrupts the power grid, damages cell towers, or displaces the population it was built to serve.

This is no longer a marginal design consideration. The 2025 UN climate conference in Belém, Brazil (COP30) placed adaptation and the Global Goal on Adaptation at the center of its negotiating agenda, a shift from the mitigation-dominated framing of earlier COP cycles. Heading into the next cycle — reportedly split between Turkey as formal host and Australia leading Pacific-focused pre-negotiations, reflecting continued disagreement over the COP31 presidency — adaptation financing and implementation are positioned to remain a central agenda item. For the ICT4D field, that shift raises a direct question: what does it actually mean to design digital development systems for climate shocks, rather than simply hoping they hold up?

Early Warning Systems: The Most Mature Climate-ICT Intervention

Of all the places where climate adaptation and digital development intersect, early warning systems have the longest track record and the clearest evidence base. The UN Secretary-General’s Early Warnings for All initiative, launched in 2022 with a target of universal coverage, is the most visible recent effort to close a persistent gap: roughly a third of the world’s population, concentrated in low-income and small island states, still lacks access to adequate early warning systems for extreme weather.

The underlying ICT architecture is not new. The Famine Early Warning Systems Network (FEWS NET), a decades-old USAID-funded program, has long combined satellite rainfall and vegetation data with household-level food security surveys to project famine risk months in advance. The World Food Programme’s mobile Vulnerability Analysis and Mapping (mVAM) system extended this model with phone-based surveys that let field staff monitor market prices and food security indicators in areas too insecure or remote for in-person data collection. More recently, AI-based flood forecasting — including Google’s Flood Hub, deployed in partnership with government agencies in Bangladesh and India — has extended forecast lead times in river basins that previously had little or no hydrological monitoring infrastructure.

What the research literature increasingly emphasizes is that the forecasting model is rarely the binding constraint. The harder problem is “last mile” translation: getting a probabilistic flood or drought forecast into a form that a smallholder farmer, a pastoralist, or a municipal disaster office can act on in time, and building the trust relationships that make people willing to act on a warning before the damage is visible. Systems that skip this translation step — treating the technical forecast as the deliverable rather than the behavior change — have a long history of high technical accuracy and low real-world impact.

Climate-Smart Agriculture: Advisory Services Under a Changing Baseline

Agricultural ICT has been a mainstay of the field since mobile-based market price and extension services proliferated in the 2000s. Climate change complicates a design assumption those systems were built on: that historical weather patterns are a reasonable guide to future planting decisions.

The Participatory Integrated Climate Services for Agriculture (PICSA) approach, developed by the University of Reading and partners and now used across multiple African and South Asian countries, addresses this directly by pairing localized, probabilistic seasonal forecasts with structured decision-support tools that extension workers use in dialogue with farmers — rather than pushing a single forecast number and leaving interpretation to the recipient. Index-based weather insurance programs, such as those coordinated by ACRE Africa across Kenya, Tanzania, and Rwanda, take a related but distinct approach: rather than forecasting to inform a planting decision, they use weather-station or satellite data to trigger payouts automatically when rainfall or temperature crosses a defined threshold, reducing the claims-adjustment burden that made earlier indemnity-based crop insurance unworkable at smallholder scale.

Both approaches illustrate a broader pattern in climate-adaptive ICT4D design: the technology succeeds or fails based on how well it is embedded in an existing decision-making and trust structure, not on the sophistication of the underlying model. A precise index-insurance trigger that farmers do not understand or trust generates as much frustration as it prevents.

Health Systems Under Climate Stress

Climate change alters disease burden in ways that stress health information systems built around historical epidemiological baselines. Shifting rainfall and temperature patterns are altering the geographic range of malaria transmission at higher altitudes in East Africa, and are widely documented as a contributing factor in cholera outbreaks following flooding, as seen repeatedly in Mozambique, Malawi, and Yemen in recent years.

The health information systems most widely deployed across low- and middle-income countries — DHIS2, developed by the University of Oslo’s Health Information Systems Programme, is the dominant platform, used as the national health information backbone in more than 70 countries — were originally designed around routine facility reporting rather than climate-linked outbreak surveillance. Adapting them to integrate climate and environmental data streams (rainfall, temperature, vegetation indices) alongside case reporting is an active area of applied research, with the goal of enabling earlier detection of climate-sensitive disease outbreaks rather than reactive response after caseloads spike.

This is also where the offline-first design tradition in ICT4D becomes directly relevant to climate resilience: health facilities in flood- or cyclone-affected areas frequently lose connectivity and power precisely during the period when accurate, timely reporting matters most. Systems designed to function fully offline and synchronize opportunistically — a design pattern with roots in ICT4D’s long engagement with intermittent-connectivity environments — are better positioned to keep functioning through the disruption rather than going dark exactly when demand for accurate data peaks.

The Infrastructure Layer: Keeping Networks Up When It Matters Most

Underneath every application-layer intervention sits a harder infrastructure question: does the connectivity and power infrastructure survive the event at all? Cyclones, floods, and heat waves damage cell towers, cut power to base stations, and sever fiber and microwave backhaul links — often in exactly the low-income and rural areas where redundancy is weakest and restoration is slowest.

The International Telecommunication Union has worked on telecom disaster resilience for years through its work on disaster risk reduction and network restoration planning, and its post-pandemic Connect2Recover initiative extended this focus to connectivity resilience more broadly, supporting national regulators in building redundancy and rapid-restoration capacity into telecom infrastructure planning. The Sendai Framework for Disaster Risk Reduction, the global policy reference adopted by UN member states in 2015, explicitly names critical infrastructure — including telecommunications — as an area requiring resilience investment, giving national regulators and donors a policy hook for infrastructure-hardening projects that might otherwise compete poorly for funding against more visible interventions.

The practical measures that show up repeatedly in the literature are unglamorous: solar backup power at cell sites and health facilities, satellite backhaul as a fallback when terrestrial links fail, low-bandwidth-first application design so that critical functions (emergency alerts, basic financial transactions) degrade gracefully rather than failing outright when bandwidth drops. None of this is exciting relative to AI-driven forecasting models, but the evidence is consistent: a sophisticated early-warning or health-surveillance system built on top of infrastructure that goes dark during the event it was designed for delivers little value when it matters most.

What the Evidence Base Still Doesn’t Establish

For all the activity in this space, the ICT4D research literature on climate-resilient digital systems remains earlier-stage than the parallel literatures on mobile money or digital identity. Several gaps are worth naming plainly rather than glossing over.

Attribution is difficult. Program monitoring data from early-warning and climate-advisory deployments typically report reach (number of farmers reached, number of alerts issued) rather than rigorously evaluated behavior change or loss-avoidance outcomes, and the counterfactual — what would have happened without the intervention — is rarely established with the kind of evaluation design used in, for example, the mobile money or cash-transfer literatures.

Equity of access to these systems is underexamined. Early-warning and climate-advisory systems delivered primarily through smartphones, structured SMS, or radio reach populations unevenly, and the populations most exposed to climate shocks — informal settlement residents, pastoralist communities, populations in active conflict-affected areas — often overlap substantially with populations these delivery channels reach least well. Program evaluations do not consistently disaggregate outcomes by this kind of exposure-versus-access gap.

Sustainability beyond the pilot and donor-funding cycle remains a familiar and unresolved ICT4D problem, now imported directly into the climate-adaptation space: many of the systems described above began as donor-funded pilots, and the institutional and financing pathways for sustained operation once initial funding cycles end are not well established across the sector.

Frequently Asked Questions

Is climate-resilient ICT a new field, or an extension of existing ICT4D work? Mostly the latter. Early warning systems, agricultural advisory services, and health information systems all predate the current climate framing. What is comparatively new is the explicit design requirement that these systems keep functioning through the climate shock itself, rather than simply providing useful information under normal operating conditions.

Why do early warning systems often fail to produce protective action even when the forecast is accurate? The research literature consistently points to the “last mile” translation problem: converting a technical forecast into a warning that a specific recipient trusts, understands, and can act on in time. Trust relationships, message format, and lead time all matter more than raw forecast accuracy in determining whether a warning changes behavior.

What is the biggest infrastructure vulnerability for digital systems during climate disasters? Power and backhaul connectivity at the network edge. Cell towers and health facilities in the areas most exposed to cyclones, floods, and heat events frequently lose grid power and terrestrial connectivity during the event, which is precisely when demand for accurate, timely data is highest.

Do weather-index insurance programs actually protect smallholder farmers from climate losses? The evidence is mixed and context-dependent. Automated, threshold-triggered payouts solve the claims-adjustment problem that made earlier indemnity insurance unworkable at smallholder scale, but “basis risk” — a gap between the index trigger and an individual farmer’s actual loss — remains a documented limitation, and uptake depends heavily on farmers’ trust in and understanding of how the trigger works.

Is DHIS2 designed specifically for climate-related health surveillance? No. DHIS2, the dominant health information system platform across low- and middle-income countries, was built around routine facility-level health reporting. Integrating climate and environmental data streams to support earlier detection of climate-sensitive disease outbreaks is an active area of applied adaptation, not an original design feature of the platform.

Development Informatics