Sussex Sustainability Research Programme

Prediction of food security crises and effective responses

Working towards early warning systems that forecasting impacts of drought on food security.

SDGs

SDG 1SDG 2SDG 13

The team

Principal Investigator (PI) and Co PI details

Principal Investigator

Co-Investigators

Project team

Post-doctoral Researcher: Dr Monika Novakcova, Global Studies, (M.Novackova@sussex.ac.uk)

Partners

Where we worked

The project was in Kenya.

Project title

Towards prediction of food security crises and effective responses.

Overview

Food security remains a major challenge for many communities around the world, and weather and climate hazards play a significant role in affecting food production. This is certainly the case in East Africa and the Horn of Africa. In this project, we worked with the national agencies in Kenya to develop improved forecasts of impending food security crises. Early warning of such events can allow improved preparation and greater resilience to climate events.

Full project description

The main aims of the project were to lay the foundations of statistical models that can help predict food security crises in advance, so that mandated agencies in Kenya can act earlier to reduce the impacts on communities of food production and access for Kenya and Ethiopia (see WP1). Currently most food security early warning systems rely on real-time monitoring of food production and other measures of food insecurity. This means our responses to food security issues tends to be reactive, i.e. after the event has already started to affect livelihoods and lives. The main activity of the project was analysing the dependence of food security on drought events in Kenya. The aim was to develop a statistical model of food security outcomes (food production and access) based on historical data. Then using climate forecasts combined with our statistical models we aim to be able to provide predictions with greater lead times of food insecurity events. We worked closely with the main agencies in Kenya to ensure that our outputs were developed to align with, and complement, the existing drought early warning systems.

Timeline and funding

The project began in April 2017 and was funded until November 2019. The total amount of funding received was £97,459.

Methods 

Fig 6: Workshop in Kitui in which stakeholders evaluated forecasts and early actions based on forecastsWorkshop in Kitui in which stakeholders evaluated forecasts and early actions based on forecasts.

The overall approach was one of co-production. We worked with:

  1. Forecast producers: Kenya Met Department (KMD) and UK Met office
  2. Drought risk management agencies: National Drought Management Authority (NDMA) and Kenya Red Cross Society (KRCS) and the county-level drought contingency planning process in Kitui County pilot case study

To co-produce:

  1. Improved forecasts: Forecasts tailored to meet the information flows in the existing drought EWS with improved skill compared to existing forecasts
  2. Better use of forecasts: Systematic approaches to use forecasts to trigger drought mitigation actions based on an explicit knowledge of action ‘hit rate’ and ‘false alarms’.

Findings

This project has:

  • Identified entry points: Mapped drought EWS information flows and decision-making processes (Figs. 1 & 2).

  • Co-produced new prototype forecasts of:

    • Vegetation Condition Index with 2-6-week lead time using a statistical model (Fig. 3) which could be used to trigger new ‘early alert’ drought phase (Fig. 1)
    • Rainfall amounts with lead times of 2 weeks to ~5 month, from optimised  combination of international models (example in Fig. 4). The timing of  these new forecasts is shown in red in bottom row of Fig. 2 showing the additional information provided compared to existing forecasts (orange boxes second row from bottom in Fig. 2) and can constitute new Early Alert or early alarm phase classifications (Fig. 1).
    • End of season crop production forecasts with lead times of up to 5 months (Fig. 5)
    • All forecasts are probabilistic in nature with explicit expressions of the skill of the forecast, and hence the ‘hit’ and ‘false alarm’ rate of resulting actions.

  • Assessed potential early actions for drought mitigation:

    • Assessment of ‘early Alert’ classification of county drought phase (Fig 1.)
    • Pilot new prototype forecasts with Kitui County Steering group (Fig. 6)
    • New prototype forecasts are better aligned with the timing of drought management decisions (see Figure 2).

Fig 1: The existing Kenya Drought EWS phase classification (issued monthly for each county). Grey boxes show how the new Sussex forecast information can contribute to early ‘alert’ and  ‘alarm’ phase classificationFig 1: The existing Kenya Drought EWS phase classification (issued monthly for each county). Grey boxes show how the new Sussex forecast information can contribute to early ‘alert’ and ‘alarm’ phase classification.

Fig 2: Kitui County calendar of farming, drought management. Bottom row shows new Sussex forecast products. Note the timing of new forecast information compared to existing forecasts (second bottom row) and how these align better with drought management activities (3rd row from bottom)Fig 2: Kitui County calendar of farming, drought management. Bottom row shows new Sussex forecast products. Note the timing of new forecast information compared to existing forecasts (second bottom row) and how these align better with drought management activities (3rd row from bottom).

Fig 3: Example output from a new Sussex tool for forecasting (3-month averaged) vegetation condition index (VCI, used in Kenya’s drought EWS, see Fig. 1). Top: Time series of observed VCI and predicted VCI (using Gaussian Processes). Bottom left: Contour plot of observed VCI against six-week lead time forecast VCI. Bottom right:  Skill of actions tFig 3: Example output from a new Sussex tool for forecasting (3-month averaged) vegetation condition index (VCI, used in Kenya’s drought EWS, see Fig. 1). Top: Time series of observed VCI and predicted VCI (using Gaussian Processes). Bottom left: Contour plot of observed VCI against six-week lead time forecast VCI. Bottom right: Skill of actions taken based on a forecasted VCI value of less than 35 (the threshold used to trigger drought ‘alert’ phase, see Fig. 1) for lead times of 2, 4 and 6 weeks.

Fig 4: Example New long lead time rainfall index (SPI) forecasts for Kitui county for Oct-Dec 2019 issued July 2019. Shows probability of SPI failing below the 2 key thresholds (-0.09 and -0.98) used in drought phase classification (see Fig 1)Fig 4: Example New long lead time rainfall index (SPI) forecasts for Kitui county for Oct-Dec 2019 issued July 2019. Shows probability of SPI failing below the 2 key thresholds (-0.09 and -0.98) used in drought phase classification (see Fig 1)

Fig 5: Example of new forecasts of end of season crop failure risk issued and updated over the growing season. Colours show the probability of getting soil moisture well below average (i.e. less than the 25th percentile), used as a proxy of crop failureFig 5: Example of new forecasts of end of season crop failure risk issued and updated over the growing season. Colours show the probability of getting soil moisture well below average (i.e. less than the 25th percentile), used as a proxy of crop failur

Conclusion

Potential for anticipatory drought early warning systems in Kenya is clear:

  • Products from Sussex projects may lead to modifications in EWS systems and protocols including the monthly county drought bulletins.
  • For legacy and sustainability, commitment to co-production is necessary.
  • Ending Drought Emergencies (EDE) vision 2030 is the key policy document.
  • SDG synergies: Clear potential for improved EWS to contribute to multiple SDGs. (No obvious trade-offs in this case).

Barriers to uptake include:

  • Lack of finance to support early action
  • Inertia in existing systems and institutions
  • Limited technical capacity at county level
  • Need new understandings in risk management to understand concepts e.g. probabilistic forecasting, forecast skill and long-term evidence base for decision-making.

Related work

Read the Kenyan news article 'Daily Nation' 'Major Boost for Drought and Flood Forecasting Efforts' about the research project.

Useful links

  • ForPAc: Towards Forecast-based Preparedness Action:www.forpac.org