Potential of predicting rainfall onset and cessation dates in Kenya using tropospheric moisture accumulation

Abstract

Rainfall portrays high variability in space and time in Kenya. The severity of rainfall variability demonstrated as excessively heavy precipitation (floods) or excessively light rainfall (droughts) affects the economy of the country, which largely depends on seasonal rainfall. High variability of intra-seasonal rainfall characteristics such as the start, the end and the length of the season leads to adverse effects on the socio-economic activities in the country. The main objective of this study was to determine the potential of predicting rainfall onset and cessation dates over Kenya using tropospheric moisture accumulation during the long and short rainfall seasons. Daily rainfall dataset used were Climate Hazards Infrared Precipitation with Stations (CHIRPs) which is a blend of satellite estimates with rain gauge rainfall covering the period 1981-2018 with a resolution of 0.05º; tropospheric daily temperature and relative humidity circulation variables derived from the 5th generation of European Centre for Medium-Range Weather Forecasts Re-Analysis (ERA-5) spanning from 1989-2018 at a resolution 0.25º. The CHIRPS data was analyzed by means of the scores from the Principal Component Analysis (PCA). Dates of the start and end of seasonal rainfall was determined using the first PCA score (PC1). Accumulated minimum (maximum) values equate to the start (end) date of the rainy season. Clausius Clapeyron, Poisson and specific humidity equations were some of the equations used to determine the equivalent potential temperature and saturation equivalent potential temperature and their anomalies computed in order to assess the moisture supply in the atmosphere. These anomalies were the basis for the determination of the time lags which determined the predicted start and end dates of rainfall. Over the study region the start date of rains occurred between 15th March to 08th April while the cessation ranged from 20th April to 29th May during the long rainy (MAM) season. Short rainy (OND) season on the other hand recorded onset (cessation) which ranged from 23rd September to 11th November (28th November to 27th December). Identification of the dates of onset and withdrawal of moisture accumulation were performed using time series plots. The antecedent months were considered, in this case January and February for MAM season and July and August for OND season. The start of moisture build up and withdrawal come early before the exact dates of the seasonal rainfall. The mean start (end) dates of moisture build up ranged from pentad 3 to pentad 6 (pentad 11 to pentad 15) during the MAM season and pentad 4 to pentad 7 (pentad 14 to pentad 24) during the OND at the 700 hPa level. Using the data at 850 hPa level, the mean start (end) of moisture build up ranged from pentad 4 to pentad 8 (pentad 9 to pentad 17) for the March-April-May season while during the OND season the mean start and end of moisture build up ranged from pentad 41 to pentad 49 and pentad 53 to pentad 69 respectively. Time lags obtained are the lead time used to obtain the predicted onset and cessation of rainfall. The time lags were added to the pentads of moisture build up and withdrawal to give the onset and cessation of rains prior the start and end of rainfall. It ranged from 7 to 15 (10 to 19) pentads for the onset (cessation) at the 850 hPa, during the MAM season. The 700 hPa level obtained 8 to 15 (5 to 19) pentads for the onset (cessation). Conversely, the OND season recorded 7 to 21(2 to 11) pentads for the onset (cessation) at the 850 hPa whereas 10 to 22 pentads and 2 to 11 pentads were recorded for the onset and cessation of moisture build up respectively at the 700 hPa level. Interannual variability of the actual and predicted onset and cessation portrayed non-uniform pattern in the time series. The relationship between actual and predicted onset, and the actual and predicted cessation was insignificant for most stations. Kisumu, Moyale, Voi and Wajir recorded some significant correlation. Kisumu for instance recorded significant relationship with a correlation coefficient of 0.424 (-0.523) between actual and predicted onset (cessation) at 850 hPa level during the October-November-December season. Moyale recorded a correlation coefficient (CC) of 0.401 (-0.372) between the actual and predicted cessation for 850 hPa (700 hPa) level while a CC of -0.375 (1) between the actual and predicted onset (cessation) during the OND season at 850 hPa. The results of the study show the prospective of utilizing tropospheric moisture variables in predicting rainfall start and end dates over the study domain. The method (PCA method) used in determination of the start and end of rainfall dates in this study should be operational since it is able to detect co-variabilities in station rainfall.