Spatial prediction of the probability of liver fluke infection in water resource within sub-basin using an optimized geographically-weighted regression model

• Published in: Frontiers in veterinary science Vol. 11
• Main Authors: Pumhirunroj, Benjamabhorn, Littidej, Patiwat, Boonmars, Thidarut, Artchayasawat, Atchara, Buasri, Nutchanat, Slack, Donald
• Format: Journal Article
• Language: English
• Published: Frontiers Media S.A 07.11.2024
• Subjects: geographically-weighted regression (GWR); liver fluke; Opisthorchis viverrini (OV); Sakon Nakhon; spatial modeling; Thailand
• ISSN: 2297-1769; 2297-1769
• DOI: 10.3389/fvets.2024.1487222

Introduction Infection with liver flukes (Opisthorchis viverrini) is partly attributed to their ability to thrive in sub-basin habitats, causing the intermediate host to remain within the watershed system throughout the year. It is crucial to conduct spatial monitoring of fluke infection at a small basin analysis scale as it helps in studying the spatial factors influencing these infections. The number of infected individuals was obtained from local authorities, converted into a percentage, and visually represented as raster data through a heat map. This approach generates continuous data with dependent variables. Methods The independent set comprises nine variables, including both vector and raster data, that establish a connection between the location of an infected person and their village. Design spatial units optimized for geo-weighted modeling by utilizing a clustering and overlay approach, thereby facilitating the optimal prediction of alternative models for infection. Results and discussion The Model-3 demonstrated the strongest correlation between the variables X5 (stream) and X7 (ndmi), which are associated with the percentage of infected individuals. The statistical analysis showed t-statistics values of −2.045 and 0.784, with corresponding p -values of 0.016 and 0.085. The RMSE was determined to be 2.571%, and the AUC was 0.659, providing support for these findings. Several alternative models were tested, and a generalized mathematical model was developed to incorporate the independent variables. This new model improved the accuracy of the GWR model by 5.75% and increased the R 2 value from 0.754 to 0.800. Additionally, spatial autocorrelation confirmed the difference in predictions between the modeled and actual infection values. This study demonstrates that when using GWR to create spatial models at the sub-basin level, it is possible to identify variables that are associated with liver fluke infection.