Satellite prediction of forest flowering phenology

• DOI: 10.1016/j.rse.2020.112197
• Publication Date: 2021-01-29T01:18:42Z
• Authors: Dan J. Dixon; J. Nikolaus Callow; John M.A. Duncan; Samantha A. Sette...; Natasha Pauli
• Keywords: 15. Life on land; 01 natural sciences; 0105 earth and related environmental sciences

Knowledge of flowering phenology is essential for understanding the condition forest ecosystems and responses to various anthropogenic environmental drivers. However, monitoring spatial temporal variability in at landscape scales challenging (e.g. current often highly localized in-situ or single dates). This study presents a method that combines drone satellite images (PlanetScope) can produce landscape-scale maps dynamics. demonstrated landscapes dominated by eucalypt Corymbia calophylla (red gum marri) Western Australia. Drone-derived canopies, available restricted extents, are used label image pixels with proportion pixel footprint flowering. The labelled proportion, response variable, combined metrics characterize time series spectral indices sensitive presence green vegetation cream-colored flowers, predictor variables. A machine learning model then predicts daily pixel-level proportions. trained data from two sites tested three dates throughout season. able accurately predict season (RMSE <4% across all dates), dense sparse canopy, different background soil covers, robust not detecting false positive when no events occurring. Due spatiotemporal coverage images, this be deployed generate regional dynamics ecosystem supporting research into drivers phenology.