Monitoring Pollen Counts and Pollen Allergy Index Using Satellite Observations in East Coast of the United States

Allergic diseases have become increasingly common over the world during the last four decades, and they are affecting millions of people. Pollination is an important process in the life cycle of plants. However, pollen exposure is associated with allergic diseases such as asthma and seasonal allergic rhinitis (hay fever). As a result, the total annual expenditure for asthma-associated morbidity is about $56 billion in the United States, and the overall cost of allergic diseases is over $18 billion annually. For allergic rhinitis, the annual medical cost is approximately $3.4 billion. The intensity and frequency of the pollen exposures can be easily affected by many factors such as climate, vegetation, and topography, which are difficult to predict in large scales. Vegetation is very important as a pollen source, and the amount and time of pollinations depend on the flowering and growth of plants. With optimal water and temperature, vegetation can reach a maximum growth and flowering during a growing season, which means that maximum amount of pollen can be released from the plants. However, if the requirements of water and temperature cannot be met in the specific times within the growing season, pollen dispersal will be affected negatively. It is an urgent need to develop models or systems for predicting pollen events at large scales and providing early warning to prevent pollen effects on people. Unlike manual pollen counting at local sites, remote sensing facilitates the pollen estimates at large scales with temporally and spatially distributed observations, which significantly reduces the time and labor costs. With remotely sensed observations, Artificial Neural Network (ANN) helps us fill the gaps in understanding of the relationships between environmental variables and pollen concentration. At this point, I investigated pollen estimates from satellite observations in the states of East Coast United States with short and long-term data. This region is highly populated with a population of 104 million. In addition, this region has a great variety of temperature, precipitation, and vegetation. The final goal of this project is to investigate the relationships between satellite-derived variables (precipitation, land surface temperature (LST), and enhance vegetation index (EVI2)) and pollen count and further to generate a model for the prediction of pollen counts at high temporal and spatial resolutions. For this purpose, to predict pollen concentration using environmental variables, a Neural Network Analysis was performed. The results showed that strong correlations existed between pollen counts and environmental variables, except for precipitation in most locations. The validation analysis using regression models revealed strongly significant relationships between the observed and predicted pollen concentrations obtained for short and long-term data. The R squares (R2) for long term pollen counts were mostly higher than 0.5, ranging from 0.5542 for Olean, NY to 0.8589 for Savannah, GA. For short term predictions of pollen allergy index, R2 ranged from 0.53 to 0.966 except for a few sites, especially in southern Florida. The pollen distribution was mostly affected by precipitation in the southern part, whereas it was influenced by temperature in the northern part. Moreover, results demonstrated that ANN is a suitable tool for complicated statistical analysis and EVI2 combining with LST and precipitation is a reliable predictor of pollen variation. Overall the results provide a better understanding of pollen variation with vegetation seasonality and climate variables, which could assist an approach towards the establishment of an early warning system for allergy patients