- Forest ecosystems function under increasing pressure due to global climate changes, while factors determining when and where mortality events will take place within the wider landscape are poorly understood. Observational studies are essential for documenting forest decline events, understanding their determinants, and developing sustainable management plans. A central obstacle towards achieving this goal is that mortality is often patchy across a range of spatial scales, and characterized by long-term temporal dynamics. Research must therefore integrate different methods, from several scientific disciplines, to capture as many relevant informative patterns as possible. We performed a landscape-scale assessment of mortality and its determinants in two representative Pinus halepensis planted forests from a dry environment (~300 mm), recently experiencing an unprecedented sequence of two severe drought periods. Three data sources were integrated to analyze the spatiotemporal variation in forest performance: (1) Normalized Difference Vegetation Index (NDVI) time-series, from 18 Landsat satellite images; (2) individual dead trees point-pattern, based on a high-resolution aerial photograph; and (3) Basal Area Increment (BAI) time-series, from dendrochronological sampling in three sites. Mortality risk was higher in older-aged sparse stands, on southern aspects, and on deeper soils. However, mortality was patchy across all spatial scales, and the locations of patches within “high-risk” areas could not be fully explained by the examined environmental factors. Moreover, the analysis of past forest performance based on NDVI and tree rings has indicated that the areas affected by each of the two recent droughts do not coincide. The association of mortality with lower tree densities did not support the notion that thinning semiarid forests will increase survival probability of the remaining trees when facing extreme drought. Unique information was obtained when merging dendrochronological and remotely sensed performance indicators, in contrast to potential bias when using a single approach. For example, dendrochronological data suggested highly resilient tree growth, since it was based only on the “surviving” portion of the population, thus failing to identify past demographic changes evident through remote sensing. We therefore suggest that evaluation of forest resilience should be based on several metrics, each suited for detecting transitions at a different level of organization.