Using spectral indices as early warning signals of forest dieback: The case of drought-prone Pinus pinaster forests

Abstract

Forest dieback processes linked to drought are expected to increase due to climate warming. Remotely sensed data offer several advantages over common field monitoring methods such as the ability to observe large areas on a systematic basis and monitoring their changes, making them increasingly used to assess changes in forest health. Here we aim to use a combined approximation of fieldwork and remote sensing to explore possible links between forest dieback and land surface phenological and trend variables derived from long Landsat time series. Forest dieback was evaluated in the field over 31 plots in a Mediterranean, xeric Pinus pinaster forest. Landsat 31-year time series of three greenness (EVI, NDVI, SAVI) and two wetness spectral indices (NMDI and TCW) were derived covering the period 1990?2020. Spectral indices from time series were decomposed into trend and seasonality using a Bayesian estimator while the relationships of the phenological and trend variables among levels of damage were assessed using linear and additive mixed models. We have not found any statistical pieces of evidence of extension or shortening patterns for the length of the phenological season over the examined 31-year period. Our results indicate that the dieback process was mainly related to the trend component of the spectral indices series whereas the phenological metrics were not related to forest dieback. We also found that plots with more dying or damaged trees displayed lower spectral indices trends after a severe drought event in the middle of the 1990s, which confirms the Landsat-derived spectral indices as indicators of earlywarning signals. Drops in trends occurred earlier for wetness indices rather than for greenness indices which suggests that the former could be more appropriate for dieback detection, i.e. they could be used as early warning signals of impending loss of tree vigor.