Disease Modelling of Japanese Encephalitis in Taiwan through the Use of Satellite Remote Sensing.

This report describes the development of probability models for occurrence of infectious disease using variables derived from the Landsat series of satellites and other geographic factors. The models are applied to yield a predictive program for the occurrence of Japanese encephalitis on the island of Taiwan. Four different forms of model are evaluated linear regression, log-linear regression, logit models, and discriminant analysis. Observations consisted of disease data from 106 sites in Taiwan, in the period 1968-1975. Of the models employed, it was found that linear regression produces the best results, and a 9-variable linear regression was preferred. The independent variables in this case consist of means and variances of Landsat data over regions surrounding each disease site, plus altitude data. A significant correlation was found between observed and predicted disease incidence rates correlation coefficient 0.75. No systematic residual biases were observed in the final predicted results. The overall form of the model and the use of the data should remain the same for many indectious diseases in many different parts of the world.