NCDP Perspectives

By Sumana Palle

Introduction

In December 2025, I traveled to Addis Ababa, as part of a technical capacity-building mission with NCDP’s Financial Instruments (FIST) team, formerly of the International Research Institute for Climate and Society (IRI). Led by Max Mauerman from NCDP and Jeff Turmelle from the Columbia Climate School, the mission was conducted in partnership with the World Food Programme (WFP), as part of its broader effort to scale up anticipatory action for drought in the Horn of Africa.

Ethiopia is highly exposed to climate-induced shocks, particularly drought, with rainfall patterns that vary significantly across regions and seasons. These shocks have cascading impacts on crop and livestock production, water availability, food security, and household resilience. In response, WFP is implementing a multi-year initiative focused on strengthening national systems for climate risk management by integrating tailored seasonal forecasts and drought triggers into an Anticipatory Action Plan (AAP). The aim is to shift the response earlier and reduce humanitarian costs and recovery time. As part of this effort, WFP enlisted NCDP’s expertise in forecasting, especially in the usage of PyCPT, a climate prediction tool that translates global climate model outputs into locally tailored, probabilistic seasonal forecasts.

Over five intensive days, we worked with more than 15 staff members from the Ethiopian Meteorological Institute (EMI), most of whom were regional forecasters and domain experts responsible for producing seasonal outlooks for their respective regions. The focus was on strengthening EMI’s ability to own the entire forecasting and decision-support pipeline, from maintaining data on their own servers to evaluating forecast skill.

Day 1: Introductions and Technical Foundations

Upon our arrival at EMI’s office, we were greeted by EMI’s IT lead, Sinegiorgis, and led to the conference room, where the regional experts gathered. The first day was focused on updating the software and technical infrastructure required to run the latest version of PyCPT. Questions ranged from why a Python command wasn’t resolving in a notebook, to how to properly clone and manage repositories using SSH keys across different operating systems, all of which Max and Jeff were able to answer and follow up with, despite technological constraints such as internet speed.

Day 2: Forecast Skill

The second day continued with more technical training and testing. In the afternoon, as the experts grew more comfortable with the tools and began seeing the outputs of localized forecasts, discussions emerged about forecasting skill and what different skill metrics actually mean in operational contexts.

Much of the discussion was driven by practical insights and suggestions from EMI’s regional forecasters. For example, they emphasized the need for easier ways to define forecast domains at sub-national levels, such as zones, livelihood regions, or woredas, rather than relying on national boundaries that often mask local variability.

Participants also highlighted how forecast skill is closely tied to regional rainfall regimes: areas with a well-defined March-May (MAM) rainy season are seeing improving MAM skill, while regions without that seasonal structure continue to exhibit low skill, underscoring the importance of tailoring forecasts by season and geography. This led to ideas around developing interfaces that explicitly show which regions are in-season at any given time, along with clearer visualizations, such as normal distribution plots comparing forecasts to observed outcomes, to better communicate uncertainty and performance.

EMI staff shared their expertise and experiences with regional performance differences and how shifts in climatology in the Horn of Africa and the Sahel complicate baselines and interpretation. For example, we discussed whether seasonal rainfall alone is sufficient for decision-making, suggesting complementary indicators like vegetation conditions, and stressed the value of monthly forecasts within a season so agriculturalists and pastoralists can act in real time. As several participants noted, information that arrives too late or at too coarse a scale has little practical value for pastoral communities, reinforcing the need for forecasts that align more closely with how decisions are actually made on the ground.

We also noted the limitations of current global climate models, which ultimately constrain our ability to glean actionable information from forecasts.

Day 3: Visiting the United Nations

Midweek, while Max and Jeff continued to work with EMI staff, I stepped away from the training room to visit the UN compound and meet with colleagues working in the United Nations Office for the Coordination of Humanitarian Affairs (UN OCHA) country office in Ethiopia. Kuda, Zelalem, and Willem, who work in OCHA Ethiopia and are seasoned professionals in humanitarian coordination, were kind enough to meet with me to discuss context-specific issues and how they rely on the forecasts generated by EMI to make humanitarian decisions. 

While Ethiopia has a relatively strong early warning and coordination architecture, led by the Disaster Risk Management Commission (EDRMC) and supported by EMI, sector ministries, and humanitarian partners, the scale of risk far outpaces available resources. Multiple hazards often overlap, with drought and flooding occurring simultaneously across different regions and seasons, stretching already limited anticipatory action financing.

Forecast variability complicates humanitarian planning, particularly when conditions improve or deteriorate mid-season, requiring flexibility to pause, stop, or redirect activities.

Colleagues also highlighted opportunities for improvement in how forecasts are communicated: tercile-based categories like “below normal” or “normal” often fail to capture spatial or temporal variability, like slow-onset hazards. Finally, the country team emphasized the need for ongoing collaboration and capacity-building, not just for climate scientists but also for humanitarian coordinators.

Days 4-5: Wrapping Up

As the week progressed, the focus shifted toward operationalization. EMI staff explored how changing predictor and predictand domains could improve local relevance, and how to balance probabilistic and deterministic products depending on audience and use case. We looked at visualizations that compare forecast distributions to historical normals, making uncertainty tangible rather than abstract.

On the final day of the workshop, participants broke into groups and presented their takeaways to EMI leadership, including the Director General. Their slides reflected technical learnings, their own innovations and interpretations, and takeaways from the weeklong session.

After the workshop officially ended, Max and I went to the WFP office to meet with the team there and discuss the proceedings of the workshop.

Conclusion

The week in Addis Ababa was extremely insightful in understanding the end-to-end overview of the climate forecasting chain. I had interacted with the full circle of actors involved: from the experts generating seasonal forecasts at EMI to the humanitarian decision-makers at the United Nations. All of these conversations reinforced why the capacity-building work with EMI mattered: strengthening national forecasting systems ultimately strengthens the quality of humanitarian decisions made downstream, especially in a context like Ethiopia, where forecasting and early warning systems have a life-saving impact.

I am deeply grateful to the EMI staff for their engagement and generosity, to Max and Jeff for their leadership and technical guidance, and to colleagues at WFP and OCHA for sharing their perspectives on how forecasts are used in practice. I am also thankful to the SPS Professional Development Grant for making this experience possible and for supporting opportunities to learn directly at the intersection of climate science and humanitarian decision-

making.