The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLVIII-4/W6-2022
https://doi.org/10.5194/isprs-archives-XLVIII-4-W6-2022-87-2023
https://doi.org/10.5194/isprs-archives-XLVIII-4-W6-2022-87-2023
06 Feb 2023
 | 06 Feb 2023

SPATIO-TEMPORAL ANALYSIS AND MODELLING OF DENGUE INCIDENCES IN QUEZON CITY USING ORDINARY LEAST SQUARES AND SPATIAL REGRESSION

A. C. Blanco, B. J. J. Harder, and I. A. R. Teh

Keywords: Dengue, OLS regression, Spatial regression, Moran’s I, Factor analysis

Abstract. This research examined monthly dengue incidences per barangay (village) in Quezon City, Philippines over a period of six years (2010–2015) to determine the relative significance of environmental variables on dengue prevalence. The data were subjected to correlation analysis, spatial autocorrelation assessment, multiple factor analysis, ordinary least squares (OLS) and spatial regression. Local Indicators of Spatial Autocorrelation (LISA) Moran’s I cluster maps indicate significant (p=0.05, 0.01) High-High clustering and Low-Low clustering in the northern and southern parts of the city, respectively. Monthly total cases indicated increasing trend staring from May/June, peaking at around August/September, and declining afterwards to lower levels in November/December. This corresponds to the typical temporal rainfall pattern. Dengue cases were found to be positively correlated (α=0.05) with Population (R=0.84), Informal_Settlements (IS) (R=0.716), Very_Low_Density_Residential (VLDR) (R=0.512), Open_Spaces (OS) (R=0.339), Mean_Rainfall (RFM) (R=0.637), and Mean Elevation (EM) (R=0.498). Dengue incidence (DI) was negatively correlated with Mean Air Temperature (ATM) (R=−0.3 to 0.5). Based on factor analysis, the dengue incidences were closely related to these variables, though factors F1 and F2 accounted for only 28% of the data variability. Ordinary Least Square (OLS) regression analysis of DI with general land use (LU) classes (e.g., no subclasses in residential areas) identified only IS and OS, explaining 43% of the variability of DI, with IS having twice as much influence on DI compared to OS. When residential subclasses are considered, VLDR was added to the model, slightly increasing R-squared to 0.452. Considering, in addition, EM, RFM, and ATM, the R-squared improved to 0.589, with RFM and EM considered more influential on dengue compared to IS and OS. ATM was however removed due to multicollinearity. The use of Spatial Error regression (SER) and Spatial Lag regression (SLR) produced improved models relative to the OLS model with R-squared of 0.676 and 0.667, respectively. This indicates the importance of spatial dependence. This can be explained by the fact that mosquitos fly over considerably long distances, traversing across the different barangays. The SER model (AIC=1359.89; SE=26.9584) is slightly better than the SLR model (AIC=1366.05; SE= 27.3399).