By loài mô hình phân phối theo những cách được phác thảo trong việc xem xét này, chúng ta có thể để ước tính mức độ nhạy cảm của bệnh bất kỳ biến đổi khí hậu bằng cách kiểm tra các giới hạn của phong bì môi trường của nó trong không gian đa biến. | 32 D. J ROGERS environments change. Rogers and Randolph this volume pp. 345-381 show that we are presently far from understanding which type of change to expect. Their great sensitivity to environmental conditions suggests that these diseases may be among the first of all diseases to show distribution and intensity changes as climates change but as Rogers and Randolph this volume pp. 345-381 points out we must not let this sensitivity lead us to conclude that any change in a vector-borne disease is due to climate. By modelling species distributions in the ways outlined in the present review it should be possible to estimate the sensitivity of any disease to climate change by examining the limits of its environmental envelope in multivariate space. When this is matched to the predicted changes in climate also in multivariate space it is possible to map the areas on the ground which will fall within the environmental envelope in the future. It is these areas that are at risk of disease invasion and spread. A common feature of many vector-borne and other diseases is the paucity of hard data we have for their precise geographic distributions. This did not particularly matter in the days of equally sparse climatic data also from point sources or of fairly coarse spatial climate surfaces that were produced from such data. One set of coarse data could be related to another set of coarse data to produce a risk map of such poor spatial resolution as to be almost useless. Today however environmental data from satellites are available at unprecedented spatial resolutions and these reveal the inadequacies both of past maps and of the data on which they are based see Hay et al. this volume pp. 37-77 . This therefore presents us with a new problem of how to deal with sparse distribution data. While it is possible to produce risk maps from such data using the techniques outlined in this review one must nevertheless ask whether the current best model is describing simply the data
