Forests in a semi-arid climate die with a memory: satellite signals predict forest mortality years after drought

Abstract Widespread tree mortality is increasingly associated with extreme drought, yet its mechanisms remain poorly understood. To address this, we investigated which indicators most accurately delineate the relationship between climatic stressors and satellite-derived metrics related to tree crown/forest canopy conditions and forest decline for conifers and evergreen broadleaves. The study was performed between 1990 and 2020 in woodlands of Cyprus dominated by Juniperus phoenicea , Pinus brutia , and Quercus alnifolia (endemic to Cyprus). Landsat 5, 7, 8 and 9 images were used to assess the condition of tree crowns via 8 remote sensing (RS) indicators (NDMI, EVI, GPP, LAI, NBR, NDVI, NDWI, SAVI) correlated, thereafter, with the SPI, SPEI and PDSI drought indicators. Our findings clearly outline that very severe drought conditions <−2 for the SPI-12 and the SPEI-12, or <−5 for the PDSI-12, exceeded the capacity of all 3 species to sustain healthy stands at habitats representing the xeric limits of their natural distribution range in Europe. Very low precipitation appears as the driving force. However, starting from the year of drought induced mortality and including the years that followed, their annual vegetation’s response via decadal monitoring was related to climate averaged over 4 to 7 past years (including the year of monitoring) depending on species. Observed multi-year associations are consistent with a ‘memory’ effect that may reflect cumulative depletion and slow recharge of deeper, root-accessible moisture pools; however, our remote-sensing indicators do not directly perceive subsurface storage. NBR and NDMI were significantly connected with climatic variability, as described by the SPI or SPEI for the J. phoenicea and by the PDSI for the P. brutia , before and at the first years after mortality has occurred. After some years after the mortality year, the decadal response of vegetation to climate was better described by the NDVI. Before oak mortality the RS indicators applied failed to capture the evergreen vegetation dynamic of Q. alnifolia dense stands. Thereafter, the NDVI provides the highest accuracy, as the oaks likely experienced more severe and faster defoliation than the conifers, reducing their vegetation density at levels detectable by the NDVI. Thus, our study highlights the importance of considering long-term drivers for tree foliage-defoliation status, dependent on species type and habitat-specific water availability. This understanding can explain the types of droughts (seasonal to multiyear) that can trigger tree mortality under climate change.