When forecasts failed, but action didn’t - Start Ready’s experience with basis risk

Emma Cazou, CARF MEAL Advisor, and Mark Weegmann, Start Ready Manager reflect on the system's adaptability and potential.

In April 2025, the flood predictive model used by Start Ready in the Democratic Republic of the Congo failed to forecast devastating floods in Kinshasa preventing the system's automatic release of funds and implementation of anticipatory action activities. However, through stop-gap mechanisms built into Start Ready, the system managed to deliver timely funding-demonstrating both its adaptive potential and the critical need to review and strengthen forecast-based financing mechanisms.

Context

Start Ready operates by prepositioning funds that are released when forecasts of a crisis materialise enabling locally positioned members to implement pre-agreed activities designed to mitigate forecasted impacts. In DRC, a flood system using this approach has been operational since 2022, expanding to include Kinshasa in 2024. The system, led by the DRC Hub as a locally governed network, utilises global forecasting data to provide indication of riverine flooding up to ten days in advance, district-level contingency planning to identify actors, activities, and at-risk communities, and preparedness actions working with communities to ensure readiness to respond and coordination of activities. However, at the start of April 2025, heavy rainfall and river overflow caused devastating flooding across urban areas of Kinshasa impacting over 60,000 people, displacing more than 7,000, and leading to 165 fatalities (more information on this technical brief). In the days and weeks preceding, our flood model did not indicate any immediate flood risk.

Understanding basis risk 

Basis risk arises when there is a mismatch between what predictive models forecast and what happens on the ground. Basis risk is inherent in modelling and whilst we seek to minimise this risk, data gaps and uncertainties always leave us with this risk. While models are central to anticipatory action, they cannot always account for local variability, inaccurate parameters, or data gaps, leading to situations where financing is not triggered despite real need.

A first-of-its-kind activation

With modelling and forecasts so central to Start Ready's approach, this is a known risk and one we have discussed at length with our Governance Committee.

Governance Committee member Chiara Ambrosino (Climate & Resilience Advisor for Plan International UK, Global AA Lead for Plan International) says:

"Start Ready has proved to be a highly effective funding mechanisms in the past 3 years through its successful activations. Its innovative approach relies on models to represent the complexity of the climate system and interactions with the affected social systems. However, we understand that models are simplification of those systems, using assumptions, simplifications and carrying inherent limitations."

"Through accounting for Basis Risk and pre-allocating funding to be readily deployed when model failure arises Start Ready supports action in time of crises."

Start Ready manages this risk but integrating a stop-gap solution to verify and dynamically decide on activating funds for these instances.

This approach helps to ensure that the system setup on protection of communities from climate risks can still function when one critical component does not.

Within one week of these floods first materialising, members on the ground were discussing the situation, sharing information with the Start Ready team, and alerting this model misfire. Taking this data to the Start Ready Committee, the governing group reviewed and approved the release of £400,000 within 24 hours.

Chiara Ambrosino:

"Basis Risk approval comes in times of heightened concern, when members are watching a crisis developing and lives being affected, while the model used to guide disbursements to address those crises is failing. With that in mind, it's key a prompt assessment is conducted to verify whether the model has failed to predict the scale of the crisis in the locations covered by the Start Ready. Once that is verified, Basis Risk funding can be released to help address the impact of the crisis."

Action

Whilst the model did not function as expected, the other components of the system were in place. DRC Hub members ActionAid, CBS (Convention pour le Bien être Social, and PADECO (Promotion et Appui au Développement Communautaire), were already preselected to implement the pre-agreed contingency plan developed over several months in collaboration with local communities, local authorities, and other agencies. This plan, whilst requiring minor modification and update upon activation, was also supported with Start Ready funded National Reserves enabling preparedness activities, community engagement, and pre-selection of participants the few months prior. Thus, the consortium was able to launch rapid response and whilst missing the initial window of opportunity, were able to provide rapid relief and support ahead of recurrent heavy rains that repeated flooding in the weeks and month succeeding, particularly through the reinforcement of riverbanks that had eroded.

Rosette Mokabi, Business Development, ActionAid DRC says:

"Thanks to the implementation of a clear contingency plan and a needs assessment based on the reality on the ground, we were able to consider and position required players and services. This rigorous preparation, supported by the Start Ready's National Reserves, enabled us to quickly formulate a costed proposal and, when the JBA model was not activated, we issued an alert during the crisis with a minor modification. Start Ready was able to mobilise and validate the necessary resources to respond to the crisis and mitigate the impact on the affected community. The result: an immediate, coordinated and effective response to the floods in Kinshasa, because being prepared in advance means saving more lives."

Lessons on model effectiveness

Of course, whilst basis risk cannot be eliminated, it can be minimised. The event provided need for a model review, exposed numerous limitations including calibration of data. As a result, the modelled flows may not accurately reflect observed hydrological conditions. Additionally, the rainfall forecasts used by GloFAS present challenges. The convective processes that drive heavy rainfall typically occur at small spatial and temporal scales, which may not be adequately captured by the underlying meteorological models. This can lead to a poor representation of actual conditions within GloFAS, further contributing to model uncertainty.

These misalignments led to inaction even when real-world conditions warranted response. In response, we have explored modifying the return period estimation method within Flood Foresight in regions where significant trends were observed in the GloFAS reanalysis dataset. Specifically, the return period threshold was adjusted for Risk Pool 4 to account for these anomalies and improve the accuracy of flood risk representation

Members also reiterated that the global nature of the model lacks local nuance, highlighting the need for better integration of contextual data and community feedback. Data availability and quality remain key barriers.

Lessons on impact and implications

The Basis Risk route enabled rapid decision-making and minimising of time lost although members noted limited visibility into who makes decisions and how, which undermines trust and understanding. Still, the process marked progress: clearer support from Start Ready helped overcome earlier hesitations tied to the form's complexity.

Though activated after the anticipatory window, the funding enabled early response, including cash transfers, WASH support, and awareness-raising activities. It also maintained community trust in contingency planning and demonstrated the value of National Reserves investment.

One of our project participants says:

"My name is Mama Séraphine. I live alone with my five children, three of whom are girls. I sell spices at the small market in the Ndanu neighbourhood. On the night of 4 April, it rained incessantly. The water came in so fast that I didn't have time to take anything with me. In the panic, I only had the strength to get my children out. My small business, my possessions... everything was gone."

"But your help came like a light in the night. You gave us hope and, above all, you saved our lives. Today, I am rebuilding my business, and my children and I are wholeheartedly involved in community activities. May God bless each and every one of you who made the effort to quickly reach out to us."

The event strengthened collaboration with external actors like the DRC government, OCHA, and WFP, showcasing how DRF mechanisms can respond flexibly.

Still, the response was underfunded compared to need and the missed forecast meant anticipatory action could not mitigate impacts. Notably, members saw little practical difference between using Basis Risk or Start Fund (not accounting for national reserve inputs and pre-selection of members and plans), except that this activation enabled a subsequent Start Fund alert for a disease outbreak, showing how smart sequencing of funds can add flexibility.

Moving forward

This first Basis Risk activation shows the potential of Start Ready to protect communities even when forecasts fall short. It also underscores the need for:

• Simplified tools for members to access Basis Risk.
• Stronger community-based early warning systems.
• Ongoing technical model review and validation.
• Clearer governance and communication around funding decisions.

Ultimately, this event proved the importance of flexibility, preparedness, and learning in adaptive crisis financing.