A comparison of least squares regression and geographically weighted regression modeling of West Nile virus risk based on environmental parameters

Abstract

BackgroundThe primary aim of the study reported here was to determine the effectiveness of utilizing local spatial variations in environmental data to uncover the statistical relationships between West Nile Virus (WNV) risk and environmental factors. Because least squares regression methods do not account for spatial autocorrelation and non-stationarity of the type of spatial data analyzed for studies that explore the relationship between WNV and environmental determinants, we hypothesized that a geographically weighted regression model would help us better understand how environmental factors are related to WNV risk patterns without the confounding effects of spatial non-stationarity.MethodsWe examined commonly mapped environmental factors using both ordinary least squares regression (LSR) and geographically weighted regression (GWR). Both types of models were applied to examine the relationship between WNV-infected dead bird counts and various environmental factors for those locations. The goal was to determine which approach yielded a better predictive model.ResultsLSR efforts lead to identifying three environmental variables that were statistically significantly related to WNV infected dead birds (adjusted R2 = 0.61): stream density, road density, and land surface temperature. GWR efforts increased the explanatory value of these three environmental variables with better spatial precision (adjusted R2 = 0.71).ConclusionsThe spatial granularity resulting from the geographically weighted approach provides a better understanding of how environmental spatial heterogeneity is related to WNV risk as implied by WNV infected dead birds, which should allow improved planning of public health management strategies.

Exploring the socio-economic and environmental components of infectious diseases using multivariate geovisualization: West Nile Virus

“…WNV was first detected in California in 2003 ( Reisen et al, 2004 ), and then received national attention for the high rates of the disease during the following two years ( Jean et al, 2007 ). Results of WNV vector-borne environmental modeling in California ( Kala et al, 2017 ) let to this study of combining socio-economic data with the results of the environmental model using multivariate geovisualization. This study utilized coarse-scale data (county level) of reported cases of WNV human incidence along with infected dead bird counts as the basis for estimating WNV risk.…”

“…The dataset contained seven variables for each census tract. A single environmental variable (referred to in this study as “mosquito risk”) that represented the results of our earlier GWR model ( Kala et al, 2017 ) was derived from analysis of environmental eight parameters (stream density, surface temperature, surface slope, cultivated land, developed land, road density, vegetation type, evapotranspiration rate). Mosquito risk was found to be statistically significantly related to annual WNV-infected dead birds sentinel data, averaged for the 2004–2010 ( Kala et al, 2017 ).…”

“…In our earlier study ( Kala et al, 2017 ) geostatistical and spatial analysis techniques were used to build a spatially explicit model after exploring multiple environmental factors (i.e., factors directly or indirectly related to known mosquito determinants such as vegetation, elevation, evapotranspiration, streams, land use and temperature) that linked mosquito habitat suitability to the number of WNV-positive dead birds, which was used as a surrogate for human WNV risk. That study concluded that including spatial heterogeneity in the modeling improved predictive ability in understanding WNV risk.…”

“…A geographically weighted regression (GWR) was applied to a statistically significant ordinary least squares (OLS) model to improve model fit from 61% to 71%. The resulting WNV disease risk surface was created using multi-criteria decision analysis approach (detailed process can be referred to our previous article, Kala et al (2017) ). This modeling process was based upon four steps: (1) establishment of the environmental factors, (2) standardization of the factors, (3) establishment of relative weights for each factor, and (4) a Simple Additive Weighting (SAW) method to construct the disease risk surface.…”

Efficiency of Geographically Weighted Regression in Modeling Human Leptospirosis Based on Environmental Factors in Gilan Province, Iran

“…As a result, local variations would be ignored in this way and may conclude unreliable results. Therefore, the role of approaches such as GWR becomes highlighted, which capability has been confirmed in comparison to various methods [40,41]. Moreover, it is the first study on leptospirosis performed in Gilan province that uses GIS approaches, and especially GWR.…”