TY - JOUR
T1 - Satellite prediction of forest flowering phenology
AU - Dixon, Daniel
AU - Callow, Nik
AU - Duncan, John
AU - Setterfield, Samantha
AU - Pauli, Natasha
PY - 2021/3/15
Y1 - 2021/3/15
N2 - Knowledge of flowering phenology is essential for understanding the condition of forest ecosystems and responses to various anthropogenic and environmental drivers. However, monitoring the spatial and temporal variability in forest flowering at landscape scales is challenging (e.g. current monitoring is often highly localized and in-situ or for single dates). This study presents a method that combines drone and satellite images (PlanetScope) that can produce landscape-scale maps of flowering dynamics. This method is demonstrated in forest landscapes dominated by the eucalypt Corymbia calophylla (red gum or marri) in Western Australia. Drone-derived images of flowering eucalypt canopies, available for restricted temporal and spatial extents, are used to label satellite image pixels with the proportion of a pixel footprint that is flowering. The pixels labelled with flowering proportion, the response variable, are combined with various metrics that characterize time series of spectral indices sensitive to the presence of green vegetation and cream-colored flowers, the predictor variables. A machine learning model then predicts daily pixel-level flowering proportions. The model is trained with data from two sites and is tested with data from three sites and various dates throughout the Corymbia calophylla season. The model is able to accurately predict pixel-level flowering proportion throughout the flowering season (RMSE <4% across all sites and dates), across sites with dense to sparse canopy, different background soil covers, and is robust to not detecting false positive flowering when no flowering events are occurring. Due to the spatiotemporal coverage of satellite images, this model can be deployed to generate regional maps of flowering dynamics in forest ecosystems that can be used for monitoring forest ecosystem condition and supporting research into drivers of eucalypt forest phenology.
AB - Knowledge of flowering phenology is essential for understanding the condition of forest ecosystems and responses to various anthropogenic and environmental drivers. However, monitoring the spatial and temporal variability in forest flowering at landscape scales is challenging (e.g. current monitoring is often highly localized and in-situ or for single dates). This study presents a method that combines drone and satellite images (PlanetScope) that can produce landscape-scale maps of flowering dynamics. This method is demonstrated in forest landscapes dominated by the eucalypt Corymbia calophylla (red gum or marri) in Western Australia. Drone-derived images of flowering eucalypt canopies, available for restricted temporal and spatial extents, are used to label satellite image pixels with the proportion of a pixel footprint that is flowering. The pixels labelled with flowering proportion, the response variable, are combined with various metrics that characterize time series of spectral indices sensitive to the presence of green vegetation and cream-colored flowers, the predictor variables. A machine learning model then predicts daily pixel-level flowering proportions. The model is trained with data from two sites and is tested with data from three sites and various dates throughout the Corymbia calophylla season. The model is able to accurately predict pixel-level flowering proportion throughout the flowering season (RMSE <4% across all sites and dates), across sites with dense to sparse canopy, different background soil covers, and is robust to not detecting false positive flowering when no flowering events are occurring. Due to the spatiotemporal coverage of satellite images, this model can be deployed to generate regional maps of flowering dynamics in forest ecosystems that can be used for monitoring forest ecosystem condition and supporting research into drivers of eucalypt forest phenology.
UR - http://www.scopus.com/inward/record.url?scp= 85100024073&partnerID=8YFLogxK
U2 - 10.1016/j.rse.2020.112197
DO - 10.1016/j.rse.2020.112197
M3 - Article
SN - 0034-4257
VL - 255
JO - Remote Sensing of Environment
JF - Remote Sensing of Environment
M1 - 112197
ER -