Neural network modelling for estimating linear and nonlinear influences of meteo-climatic variables on Sergentomyia minuta abundance using small datasets

In recent years, meteo-climatic changes contributed to the geographical expansion and modifications of habitat that become more suitable to phlebotomine vectors. Among these vectors, the role of Sergentomyia minuta in the circulation of mammalian leishmaniases has been recently discussed. Here we apply a neural network model (specifically developed for modelling relationships among variables in small datasets) to estimate the population abundance of S. minuta starting from meteo-climatic variables only, during three capturing seasons (2014–2016) in an Italian site. The results show that we are able to explain a wide majority of the variance in the data of population density (R2 = 0.632). This is obtained through the application of a neural model driven in input by averaged mean temperature, relative humidity and temperature at 10 cm belowground during oviposition, larval and adult stages. A modelling pruning activity shows the major role of humidity in driving the number of captures, but also an important nonlinear role of temperature, which highlights the importance of possible heat waves on population density of S. minuta.