Abstract
We have proposed a new method of forecasting pollen dispersal by combining the non-linear least squares method with neural networks. Because making highly accurate estimates of pollen count in cedar forests is critical for improving the accuracy of forecasting pollen dispersal, we propose a method to estimate pollen count by using observed aerial pollen concentrations measured in living areas and using convection-diffusion equations to calculate aerial pollen concentrations that reproduce the observed values. To forecast pollen dispersal, we made use of neural networks’ learning and decision-making functions. We created neural networks that could use their learning function to estimate the pollen count from the weather condition in areas of interest. To estimate the amount of pollen dispersed in an area of interest on a specific date, we used the most suitable neural network that can estimate the pollen count to estimate the pollen count on that specific date, and then calculated the aerial pollen concentration in the area of interest. We confirmed that the calculated concentrations tended to match the observed concentrations. Thus we confirmed the feasibility of a new method of predicting aerial pollen concentration that combines the non-linear least squares method with neural networks.
