Diagnostics des prédicteurs atmosphériques pour une prévision intra-saisonnière opérationnelle en Afrique Centrale : application à l’agriculture et à la sécurité alimentaire

Abstract

This work is part of the operationalization of intra-seasonal forecasting applied to agricul- ture and food security in Central Africa. The PYCPT interface was used for forecasting and statistical/probabilistic analysis. Rainfall observation data were obtained from CHIRPS over the period 1990 to 2019 at a resolution of 5km. MJO, ENSO, QBO indices and precipitation from different models (NMME and UE-C3S model suite) are used as atmospheric predictors in order to detect the link between these parameters. The Liebmann method was used to determine the start and end of the rainy season. An evaluation of the forecast was made using the deterministic and probabilistic method respectively : Pearson’s product-moment correlation coefficient and Spearman’s rank correlation coefficient, the Ranked Probability Skill Score (RPSS) and the Generalization of Relative Operational Characteristics Curve (GROC). All these methods have helped to detect the correlation between forecast and observation, or between forecast and climatology. The results show an average correlation between forecast and observation, with values sometimes reaching 0.6 or more in certaine area at certain periods. Probabilistic methods show that it is worth using forecasting in most of the study area, providing positive values for the RPSS score and values above 50% for the GROC score in several periods (seasons, weeks). The choice of a model to make a forecast will depend on the individual forecaster’s judgement, combined with the methods used for a given area and season to be studied.