Predicting reproductive phenology of wind-pollinated trees via PlanetScope time series

Abstract

Airborne pollen triggers allergic reactions which result in public health consequences. A better understanding of flowering and pollen phenology could improve airborne pollen predictions and reduce pollen exposure. Data on the timing of flowering and pollen release are needed to improve models of airborne pollen concentrations, but existingin-situdata collection efforts are expensive and spatially sparse. Satellite-based estimates of plant phenology could potentially enable large-scale data collection, but it is difficult to detect the reproductive phenology of wind-pollinated flowers from space. Here, we infer the reproductive phenology of wind-pollinated plants usingPlanetScopetime series with a spatial resolution of 3 m and a daily revisit cycle, complemented byin-situflower and pollen observations, leveraging the correlation between vegetative and reproductive phenology. On the individual tree level, we extracted PlanetScope-derived green-up time and validated its correlation to flowering time using flower observations in a national-scale observatory network. Scaling up to the city level, we developed a novel approach to characterize pollen phenology from PlanetScope-derived vegetative phenology, by optimizing two tuning parameters: the threshold of green-up or green-down and the time lag between green-up/down and flowering. We applied this method to seven cities in the US and seven key wind-pollinated tree genera, calibrated by measurements of airborne pollen concentrations. Our method characterized pollen phenology accurately, not only in-sample (Spearman correlation: 0.751, nRMSE: 13.5%) but also out-of-sample (Spearman correlation: 0.691, nRMSE: 14.5%). Using the calibrated model, we further mapped the pollen phenology landscape within cities, showing intra-urban heterogeneity. Using high spatiotemporal resolution remote sensing, our novel approach enables us to infer the flowering and pollen phenology of wind-pollinated plant taxa on a large scale and a fine resolution, including areas with limited priorin-situflower and pollen observations. The use of PlanetScope time series therefore holds promise for developing process-based pollen models and targeted public health strategies to mitigate the impact of allergenic pollen exposure.