Geographical Clustering of Climatic Effects

            Geographical clustering of the magnitude of the R² values, the amount of variability in the malaria prevalence rates that was predicted by the best combination of climatic variables alone, is shown in the following two maps.  For both API and SPR, the districts whose malaria prevalence rates were predicted the most by climatic variables were in the center of the state, and the ones that were the least predicted were in the outer parts, both in the north, south and west. 

R² values for the Best Multi-Linear Regression with API for each District in Karnataka, India

R² Values for the Best Multi-Linear Regression with SPR for Each District in Karnataka, India

            This geographical clustering effect also occurred for R² values from simple linear regressions with mean temperature versus precipitation.  Maps of the degree of correlation between SPR and January mean temperature and SPR and November precipitation shown below demonstrate that malaria rates in the southern interior districts were better predicted by precipitation than by mean temperature whereas the malaria rates in the northern interior districts were better predicted by mean temperature than by precipitation.  January mean temperature and November precipitation were chosen because they were the ones with the largest number of districts that significantly correlated with lower p-values than the other months for mean temperature and precipitation.  In addition, the R² values for regression models for January mean temperature and SPR were higher than those for November precipitation and SPR.

R² Values for Linear Regressions between SPR and January Mean Temperature by District in Karnataka, India

R² Values for Statistically Significant Linear Regressions Between SPR and November Precipitation by District in Karnataka, India

The fact that the coastal districts did not have as high R² values, may be attributed to the fact that these are wetter and hotter districts than the interior districts.  Therefore, variations in mean temperature and precipitation are not as likely to be limiting factors in malaria transmission.  The interior districts are cooler and drier, and therefore would likely be more susceptible to changes in mean temperature and precipitation affecting malaria transmission.  The clustering effect may also be explained by the specific climatic or ecosystem type that predominates in a district, by the implementation of malaria control efforts, by the extent of irrigation, deforestation, migration, and urbanization in an area, or by any number or combination of such confounding factors.  

The difference between which districts that correlated stronger with mean temperature versus those that correlated stronger with precipitation could also be explained due to climatic differences between districts.  Because both the south and north interior regions are semi-arid to arid regions, they would both be expected to be affected by variability in precipitation since it is likely a limiting factor for vector breeding.  From 1970 to 1998, Bellary, Chitradurga and Hassan, Tumkur, Kolar, and Dharwad were the driest districts, and all are located in the center of the state where there was a higher correlation with mean temperature than with precipitation.  It is possible that irrigation was influential as malaria vectors in areas with more irrigation would be less affected by variability in rainfall for creation of breeding areas.  However, without district irrigation data, I was not able to assess this relationship in this study.  It is also possible that the northern areas were too dry and therefore even higher than normal precipitation would not have created enough surface waters for vector breeding (Guha, 2000, personal communication).

            Shimoga and Dharwad, two districts whose malaria rates were highly predicted by variability in mean temperature and precipitation, had low average API and SPR levels in comparison to the rest of the districts.  Whereas Bellary, Mandya, Chikmagalur, and Chitradurga and Hassan, districts that also had high R² values for the best multi-linear regression model, had high average API and SPR levels in comparison to the other districts.  Therefore, both districts with high and low malaria prevalence correlated strongly with mean temperature and precipitation.  This implies that both districts with current high and low levels of malaria prevalence had strong correlations between malaria and climate and will likely be strongly influenced by changes in those climatic conditions.