The aim of the present study was the development of an improved weather-driven malaria model, which is able to simulate malaria transmission in both epidemic as well as endemic malaria areas. This section provides a detailed discussion with regard to various aspects of the present study. The present-day climate performance of the new model version are discussed relative to results of former studies and the model calibration is evaluated.
Comparison of LMM2010 runs with those performed by the original formulation of the model reveals a significant improvement (see Table 4) of the model performance both for epidemic as well as endemic malaria areas. In contrast to the LMM2004, the LMM2010 is able to reproduce the low transmission rates in the northern part of the Sahel. This enables, therefore, an improved detection of epidemic areas, in particular, in the Sahel. The LMM2010 validation also shows that the model can now also be applied for endemic malaria areas. The usage of the fuzzy distribution model enables the simulation of realistic sizes of the mosquito population under humid rainfall conditions resulting in reasonable transmission rates. Moreover, the lag in the malaria seasonality has disappeared in the new version of the LMM.
Various published malaria distribution maps [43, 48] correspond well with the simulated spread of malaria by the LMM2010. However, in certain parts the simulated intensity of malaria transmission differs considerably from published malaria maps. The LMM2010 in general seems to predict higher transmission rates than satellite-derived predictions of EIR
from Rogers et al. . The maps of transmission intensity provided by Gemperli et al.  are fairly patchy. In fact, the prediction from Gemperli et al. significantly suffers from the neglected interannual variability of malaria. Based on the few available EIR
observations it is difficult to judge which estimates are closer to reality. However, the validation of the LMM2010 under different climatic conditions provides evidence that the present study generated realistic biting rates and a reasonable interannual variability.
The calibration of the LMM was performed in West Africa for different atmospheric conditions of epidemic and endemic malaria regions. Realistic temperature and precipitation time series were reconstructed from various synoptic weather stations. The comparison with observations from eleven entomological and parasitological variables finally enabled the setting of undetermined model parameters.
The databases (including meteorological, entomological, and parasitological observations) for the LMM calibration are not optimal. There is a mismatch between the scales at which a disease vector responds to hydrologic variability and the scales at which hydrologic variability is actually observed. Systems should be developed that monitor hydrologic variability at scales corresponding to disease system ecologies . In this study, the generation of realistic meteorological station time series enabled the comparison with atmospheric conditions from malaria field studies, which were not conducted directly at the weather stations. These sites therefore in any case exhibit a different temporal variability of rainfall and temperatures. This might be one reason, amongst other factors such as environmental conditions, why year-to-year comparisons between observation and simulation were weakly correlated at single locations. In order to circumvent this problem, the present study refrained from looking at paired annual correlations at single stations but applied a problem-adapted scoring system.
The required historical entomological and parasitological data are rarely available with sufficient coverage. Most locations show only one, two, or even no field measurements. It is therefore likely that a larger set of observations would have an impact on the result of the model calibration. Ideally, model simulations and malaria observations should be compared from year-to-year. However, this would require the simultaneous monitoring of long-term malaria data and meteorological measurements. Such long time series are available for the area of Ndiop/Senegal (S. Louvet, personal communication, 2007), but these data sets are not publicly available.
The close ranking of diverse model runs as well as the lack of sufficient validation data further restricted an objective formal fitting of the model. In fact, various steps of the calibration procedure were subjective. Due to high computational costs it was furthermore not possible to fit all remaining model parameters simultaneously. However, because various settings compensate each other it is likely that the final model formulation conforms as much as possible to reality.
The calibration and validation of the model should also be ideally not only restricted to West Africa. However, such an extension to, for example, East Africa was beyond the scope of this study. The Malaria Atlas Project intends to provide access to various malaria studies . This might provide an efficient access to malaria data beyond that of West Africa. Such an extension would ideally include East African highlands and an estimation of the sporogonic temperature threshold.
This study was naturally not able to account for all processes involved in the spread of malaria. Some factors might be included in a future extension of the LMM. The simulation of the parasitological malaria variables by the LMM2010 is a simplification of real processes. The validation of the LMM2010 by means of parasitological measurements in West Africa revealed shortcomings of the new model version. Lower skill scores were achieved by the three parasitological variables when compared to the results from the eight entomological variables.
In addition to the lack of immunity, the LMM2010 does not account for other malaria factors such as chemoprophylaxis and human activities. However, this could be implemented by means of a variable parameter setting. Observations suggest a greater variability of the parasite ratio. At Bobo-Dioulasso, for example, the ten observed annual mean asexual parasite ratios (PR
) range between 29.1 and 77.5%. In contrast, the 34 annual values of the LMM2010 only span values between 50 and 70% .
Due to the lack of long-term observations, Kleinschmidt et al.  and Gemperli et al.  were forced to neglect the interannual variability of PR. This fact might again partly be responsible for their projected irregular PR pattern in West Africa. Their maps also show a sharp decrease of PR
north of about 15°N. In contrast to the LMM2010 runs and the Garki model simulations from Gemperli et al. , PR
is frequently lower than 50% south of 15°N. Only few regions exhibit higher PR
values than 70%, which are simulated by the LMM2010.
It should be pointed out here that climate is rarely the only important driver of malaria. Numerous other studies showed [21, 22, 53, 54] that in particular human activities are crucial for the transmission and prevention of malaria across Africa. For example, the modification of the landscape by irrigation , forest clearing , or urbanization  can significantly alter malaria transmission. The present study assessed only the malaria risk for rural areas without the influence of permanent breeding places. The applicability of this analysis is therefore limited when permanent water bodies or urban centres are present. In principle, the calibration of the LMM2010 could also be performed for urban areas. However, such an undertaking seems to be hampered by the lower number of available observations (see Additional file 2).