D. Cheruiyot (Kenya) and I.O. Osunmakinde (S. Africa)
Rainfall, Environment, Food Security, Statistical and Temporal Probabilistic Models
Recent studies have developed many approaches to learn onset of rainfall with the help of highly skilled domain experts. This research investigates an alternative approach to automatically evolve the hidden temporal distribution of onset of rainfall directly from multivariate time series (MTS) data in the absence of domain experts. Temporal probabilistic modelling of the emergent situation awareness (ESA) is proposed to reveal hidden variability and dependencies over time for the onset of rainfall. Several weather parameters such as sea surface temperature, 700hPa wind anomalies, and climate indices such as El-Niño/Southern Oscillation (ENSO), etc. are analysed using the ESA technology to evolve model of temporal dependencies among these parameters. The target parameter, onset of rainfall is meant to reveal the degrees of beliefs for false, early, normal, late or failed state. Using rainfall observations from Botswana, this work has shown that three month lead time of Southern Oscillation Index (SOI); geopotential height anomalies at 500hpa level and wind anomalies at 700 hpa parameters are better indicators for the onset of summer rains in Southern Africa. Our experimental results give promising insights to approaches to sustainable food security, water conservation and early warning systems in Southern Africa.
Important Links:
Go Back
