The monthly malaria data is correlated with NOAA daily ARC estimated precipitation. This precipitation dataset covers rather well the period for which we have malaria data. The period therefore covered by the study is February 1995 to December 2005, leaving aside the two first years with available monthly malaria data (1993-4). In order to have comparable datasets, the gridded 10th degree precipitation has been averaged over Madagascar districts. Then daily precipitation has been cumulated to monthly one.

Correlating the total field shows a good correlation between precipitation and malaria, especially when performing a two-month lag. It means that malaria seasonal cycle is highly correlated to 2-month in advance rainfall seasonal cycle (cf. Figure 1).

It is comforting to see how seasonal cycles correlate, however, it is more useful to detect if unusual rainfall events can be correlated to unusual malaria incidence events. Therefore, the seasonal cycle is now removed and the same analysis is performed on anomalies. Only the best result is shown here, which is still the 2-month lag correlation. It suggests that there is little inter-annual correlation between malaria and rainfall and therefore that cumulated precipitation can not be used to monitor nor forecast malaria incidence (cf. Figure 2).

This interesting but disappointing result invites us to make more research to look for another precipitation-related variable to try to correlate with malaria incidence. A better understanding of the malaria transmission process could help define such a variable which could be for instance a numnber of days per month when precipitation has reached a certain range, or focus on months when malaria peaks, or combine with other environmental/climate variables.

Previous studies show well the correlation between environmental factors (precipitation, temperature, vegetation) seasonal cycle and malaria one. However, the monitoring and forecasting of malaria epidemics require inter-annual variability correlation to be useful. Previous studies had trouble establishing such relationships, except for one (Bouma, 2003) where Bouma found correlation between a temperature variable and malaria incidence in Antananarivo. I therefore tried to find such a relationship with the data available but failed to do so. I present samples of the result here, illustrating the different variables and parameters I tested.

The longest common time period is from January 2000 to December 2005, using daily weather station data from Madagascar. Three of those stations fit into the highlands: Anatananarivo, Antsirabe and Fianarantsoa. The different analysis were performed between those stations and the associated malaria district respectively. Figure 3 shows monthly mean minimum temperature and malaria incidence anomalies in Fianarantsoa on the same graph.

Exceptionnaly high minimum temperature doesn't seem to relate to malaria incidence peaks. In order to take advantage of the daily temperature data, and taking into account previous studies based on general biologically plausible variables, I compute the monthly number of days when minimum temperature is above 15°C and maximum temperature doesn't exceed 32°C. Figure 4 shows the anomalies of such count of days along with malaria incidence anomalies in Fianarantsoa.

Again, nothing seems to indicate predictability of malaria incidence from suitable temperature conditions. Bouma had much more descent available time period to analyze inter-annual variability. He found that average December-January minimum temperatures was a good predictor to malaria incidence. I can't perform such a comparison because I have only four Dec-Jan couples in my record. It would be nice however to look at a variable that focuses only in the season of interest. I will therefore perform the analysis with the months January-February. Figure 5 shows this seasonal averaged anomaly of minimum daily temperature with malaria incidence anomalies in Antananarivo.

Nothing too exciting about that... Another way to caractarize unusually hot months is to count the days 1.5°C above the montlhy climatology. Figure 6 shows the anomaly of such a count and malaria incidence anomalies in Antsirabe.

Even though that diagnostic seems a little less discouraging, it is not satisfying enough.

I've not been successful in linking temperature to malaria so far, even though I tried different ways and variables.

Bouma is looking at a period of time (before 1990) and finds that the temperature explaining malaria variance is highly correlated to the distribution of An. gambiae. That vector may have played a major role in this time and location (the study looks at Anatanarivo).

On the other hand, another study (Guintran WHO report) looking at post 1990 data can't find either good relation between temperature and malaria. The WHO report suggests that this period is marked by a rather epidemic free period in the middle of 90s and then a return of epidemics with the decrease of pulverizations. This return of the 2000s epidemics is associated with the return of another mosquito: An. funestus. That species is expected to be less sensible to climate conditions (as opposed to An. gambiae) because developing in the environment of ever more numerous irrigated paddy fields.

The period I have data for (2000-5) might therefore present little correlation to temperature or other climate data. In light of this, I'd have three comments. First, further inverstigate to find an appropriate climate parameter/variable to link with malaria. Even though I tried many, there is more to take advantage of with daily data. For instance, daily data allow to calculate hot spells or wet spells that might be meaningful to describe the conditions of the growth of the vector population. This requires the definition of a hot/wet day and the length of a spell. As well, instead of working on the incidence of malaria, I could work on epidemics events, that also require a definition. Figure 7 shows on the left hand side an example of the visualization of spells. The black curve is the monthly malaria incidence anomaly in Fianarantsoa; the red bars are days above 15°C; and the blue bars show days above 5mm. On the right hand side is a graph showing all districts of the Highlands. In green we have months with monthly precipitation above 152mm, in orange we have malaria epidemics which is defined as an incidence above the mean plus two sandard deviation, in purple we have both events (wet and epidemic), and in blue none. So typially in such a grah we expect green preceding orange or purple, which is not happening very often. Second, it seems difficult to identify the impact of climate inter-annual variability without finding mechanisms to ponder other preponderant factors such as paddy fields locations and differentiation between cases due to different vectors, but also pulverizations periods and locations, or seasonal population movements between low lands and high lands, for instance. To illustrate the potential impact of pulverization, Figure 8 shows in black areas with an elevation between 1000m and 1500m. It has been reported that those thresholds have been used to determine where the pulverizations were to take place. Third, if working on temperature and precipitation data in Madagascar (in-situ or remotely sensed) is not producing results, it would be good to have a look at the correlation with ENSO and maybe also the sea surface temperature of the tropical Indian Ocean (I would have to ask a climate person about the role of the Indian Ocean in this part of the world). Previous studies highlight a correlation between malaria and ENSO without much exploring the prediction capacities.