Spatio-temporal modelling of COVID-19 incidence in Bulawayo, Zimbabwe

Abstract

The study mapped and evaluated the spatial and temporal patterns of COVID-19 in Bulawayo and identified the spatiotemporal distribution characteristics and changing trends of cases. From April 2020 to August 2021, the spatiotemporal distribution characteristics and shifting trends of cases were found. A time-series analysis of COVID-19 reported cases was done to determine trends, their direction, and size using the Mann Kendal trend test and the Sen slope. For the first three waves of the disease, a hotspot analysis, and a Cluster and Outlier analysis were used to identify the spatial distribution and clustering of COVID-19 cases. A Space-Time Cube of COVID-19 cases was then created to enable a 3D analysis of COVID-19 in space and time. An Emerging hotspot analysis and a Local Outlier Analysis were used to identify the hotspot clusters temporally and determine the trends of the hotspots. COVID-19 occurrences were then forecasted spatially for September 2021 using Exponential smoothing and Forest-based forecasting. An evaluation by location was done to produce a combination of the two prediction models. Hotspots were identified in the eastern parts of Bulawayo throughout the period and no clustering of the cold spot was observed using the 2D hotspot analysis. The 3D Local outlier analysis showed that fourteen (48,28%) of the locations had multiple patterns, nine (31,03%) locations contained low-low clusters, one (3,45%) location was a high-high cluster and five of the wards (17,24%) in Bulawayo were insignificant. The Emerging hotspot analysis identified two sporadic hotspots southeast of Bulawayo and thirteen consecutive hot spots concentrated mainly in the central parts of Bulawayo. Using the RMSE, the Forest-based forecasting was a better prediction model than the exponential smoothing forecast in predicting the spatiotemporal distribution of COVID-19 in Bulawayo. Thirteen out of twenty-nine wards in Bulawayo were predicted using the exponential smoothing forecast and sixteen of the wards used Forest-based forecasting. The findings of this study prove the significance of geospatial mapping which can quickly present and generate maps by highlighting the geographical and temporal variation of vulnerable areas. The study also proves the possibility to predict COVID-19 cases for future time steps using the exponential smoothing forecast and the Forest-based forecasting. The study recommends the use of geospatial techniques in monitoring and modeling the spatial trend and spread of pandemics that could help in targeted testing, prioritizing vaccination areas, and optimizing the allocation of limited resources