Inferring drivers of changing land surface phenology from Landsat time series
Date created: | Last Updated:
: DOI | ARK
Creating DOI. Please wait...
Description: Phenology is an important ecosystem property, and monitoring and modeling of phenology is particularly important for understanding climate change impacts on vegetated ecosystems. However, in-situ measurements are frequently confined to a few specific observation sites and species, and are thus limited for fully understanding the drivers of changing phenology at broader scales. Moderate resolution remote sensing time series from the Landsat archive can help overcome this limitation by delivering a consistent estimate of land surface phenology over the past 30 years. Yet, methods for inferring the drivers of variation in land surface phenology from these data remain scarce. We here present a new model for inferring drivers of changing land surface phenology from Landsat time series. We demonstrate our model using a case study comprising broadleaved and coniferous forests and estimating the effects of pre-season temperature and winter-chilling on inter-annual variation in the start of season. We identified significant effects of pre-season temperature on inter-annual variation in start of season, with a -3.74 d °C-1 earlier start of season for broadleaved and a -2.68 d °C-1 earlier start of season for coniferous forests, respectively. This relationship, however, was modulated by the number of chilling days, with a decreasing effect of pre-season temperature with decreasing number of chilling days. The inter-annual variation in start of season predicted from our model – i.e., calibrated solely from Landsat satellite time series – showed good agreement with in-situ observations of bud-break (Pearson’s r = 0.79/RMSE = 4.88 d for broadleaved forests and r = 0.62/RMSE = 6.57 d for coniferous forests). Our model thus allows for inferring drivers of changing land surface phenology directly from Landsat time series, opening up phenological research in areas where in-situ measurements are unavailable, and at spatial and temporal scales difficult to tackle with field and coarse-scale remote sensing data.