Empirical Forecasting of Slow-Onset Disasters for Improved Emergency Response

Size: px

Start display at page:

Download "Empirical Forecasting of Slow-Onset Disasters for Improved Emergency Response"

Nancy Shepherd
5 years ago
Views:

1 Empirical Forecasting of Slow-Onset Disasters for Improved Emergency Response An Application to Kenya s Arid Lands Andrew Mude, Christopher B. Barrett, John McPeak, Robert Kaitho, Patti Kristjanson Pastoralism and Poverty Reduction in East Africa: A Policy Research Conference June Safari Park Hotel -- Nairobi -- Kenya

2 Objective Develop an empirical forecasting model to predict the human impact of slow-onset disasters for early warning. Motivation Increasing frequency of humanitarian crises call for efficient and practical methods of emergency needs assessment.

3 Geographic Focus: Kenya s Arid North Highly vulnerable to recurring shocks such as droughts and floods Largely populated by nomadic pastoralists Data Arid Lands Resource Management Project Livestock Variables Child Nutritional Data Global Livestock CRSP (LEWS/LINKS) Climate Variables

4 Laisamis Loiyangalani Maikona Marsabit Central North Horr Sentinel points Sample Sites Turkana (20) Sentinel points Marsabit (9) Balesa North Horr Kalacha Maikona Loiyangalani Olturot Sagante Karare Korr Logologo Ngurunit Legend Marsabit divisions N 0 70 Kilometers Kapunyany Kapenguria Kollowa Loiwet Maron Nakoko Kinyach Katikit Orus Yatya Komolion Kaptuya Baringo (12) Samburu (14)

5 Severe Child Malnutrition and Food Crisis Forecasting food crisis requires a suitable indicator variable Child anthropometric measures of acute malnutrition typically used Mid-Upper Arm Circumference (MUAC) Superior predictor of child mortality A MUAC Z-Score of -2 often used as an indicator of severe wasting Our Indicator of Food Crisis: Whenever 20% or more of children are suffering severe wasting.

6 Early Warning: Forecasting Food Crisis Effective response requires early warning Two-forecasting models: 1) One Month Forecast: Better accuracy, less response time 2) Three-Month Forecast: Less accurate, more response time

7 Forecast Results Fraction of Children Experiencing Severe Wasting Fraction with MUAC Z-Score < Food Crisis Actual (Smoothed) One-Month Forecast Three-Month Forecast Food-Crisis Threshold Fraction with MUAC Z-Score < Actual Food Crisis One-Month Forecast Three-Month Forecast Food-Crisis Threshold Jan00 Jul00 Jan01 Jul01 Jan02 Jul02 Jan03 Jul03 Jan04 Jul04 Jan05 Date Jan04 Apr04 July04 Oct04 Jan05 Apr05 Date Frequent experience of food crisis Forecasts trace actual values quite well and improve with time Difference between one-month and three-month forecast accuracy not considerable

8 Making Practical Use of the Forecasts Example: Food Security Organization Predictions in the form of intensity of severe wasting Decision making parameters Minimum likelihood of food crisis required before initiating emergency response Defining a correct decision : Initiating response when there is actually a food crisis No response when there is no food crisis Forecast Performance Fraction of correct decisions

9 Forecast Performance Forecast Horizon Confidence Threshold 75% 66% 50% Fraction of Correct Decisions One Month Three Month Fraction of Errors that are Type 1 One Month Three Month Type 1 Errors: Failing to respond when a famine actually occurs.

10 Conclusions/Policy Implications Developed an empirical forecasting model that can predict with reasonable accuracy the expected human impact of slow onset shocks such as drought. Model is based on a non-restrictive set of variables making it quite cost effective Model can be easily and regularly updated with new information that should continuously increase its forecast performance Invaluable for early warning and emergency response to food crisis

11 Acknowledgements Arid Lands Research Management Project Global Livestock CRSP Pastoral Risk Management (PARIMA) Livestock Early Warning System (LEWS) Livestock Information and Knowledge System (LINKS) International Livestock Research Institute World Food Program

Empirical Forecasting of Slow-Onset Disasters for Improved Emergency Response: An Application to Kenya s Arid and Semi-Arid Lands

Empirical Forecasting of Slow-Onset Disasters for Improved Emergency Response: An Application to Kenya s Arid and Semi-Arid Lands Andrew Mude 1, Christopher Barrett, John McPeak, Robert Kaitho, Patti Kristjansen