The analysis and monitoring of dieback and mortality phenomena have been addressed through various methodologies, such as visual analysis of vegetation conditions, field surveys, remote sensing techniques and many others. Some studies in the past have used a visual and qualitative assessment of trees (vitality classes) to evaluate the severity of dieback ^[1][61]. This approach consists of assigning each observed plant a vitality class, i.e., a numerical value in the range from 1 to 6 (healthy to dead plant) ^[1][2][45,61]. However, this method, being a visual and qualitative assessment of the state of the canopy, is not very objective, so it depends on the operator’s ability to distinguish between the different vitality classes. Other studies ^[3][4][43,62], for example, have differentiated between declining and non-declining trees based on the current percentage of crown transparency or defoliation. This is a widely used practical approach to characterise tree vigour; nevertheless, this approach has been subject to some criticism. Indeed, establishing a fixed threshold of defoliation to distinguish trees in decline from those that are not can be questioned because crown transparency can change from year to year. Thus, a defoliated tree may recover, while some non-defoliated trees may start to die back. However, there are defoliation thresholds that, once exceeded, are not reversible.

Other studies have used remote sensing to assess forest cover. Indeed, it has been shown that airborne Lidar (Laser Imaging Detection and Ranging), through the acquisition of point clouds, can detect defoliation in terms of LAI ^[5][63] and thus provide feedback on canopy and forest vigour ^[6][64]. Given the complexity of forest systems and their response to disturbances, visual or remote canopy assessment methods must always be accompanied and validated by quantitative field surveys and measurements to ensure representativeness and correlation between the data obtained.

Qualitative observations of canopies alone, therefore, are not sufficient to best discriminate the state of forests and, consequently, their vulnerability. Quantitative investigations to examine forest dieback phenomena can be obtained using dendroecological data; an example of this type of investigation is that employed by several studies ^[3][7][22,43], in which, in addition to an initial visual assessment of canopy transparency, time series of tree rings were also obtained. Trees under drought conditions show a reduction in the radial increment and area of the vasal lumen and a consequent reduction in hydraulic conductivity ^[7][22]. Following frequent extreme weather events that trigger dieback phenomena, a decline in the growth of trees is observed long before their death, which can vary in intensity and duration. This phenomenon results in divergent growth trends between trees that experience dieback and those that do not ^[8][9][12,20]. Thus, the reduction in growth immediately before death could be due to a generalised water failure and/or secondary stress factors (diseases and pathogens) favoured by a loss of tree vigour, while a slow growth slowdown could be associated with a gradual decline in hydraulic performance and depletion of carbon reserves ^[10][65].

Therefore, tree rings and their anatomical variations are considered important proxies for studying the response of forests to environmental changes by retrospectively analysing, with high temporal resolution, the climatic dynamics permanently recorded in the wood structure ^[11][66].

However, even the growth ring does not always show reductions in growth during a particularly hot and droughty year, e.g., the formation of early wood in porous ring species depends on the remobilisation of stored carbon, thus not exclusively reflecting the climatic conditions during that actual growth period ^[12][13][46,60]. In other words, growth is maintained during drought through the use of stored carbohydrates, but this can cause depletion of non-structural carbohydrate (NSC) reserves and reduce the trees’ resistance to further drought events, making them prone to death ^[14][15]. In addition, drought responses may vary depending on vegetative earliness, i.e., two species in the same area may show different growth responses depending on the time of sprouting, and thus the growth ring may or may not highlight the drought event ^[15][59].

In spite of these difficulties, to date dendrochronological surveys have been the most suitable for providing information and quantifying forest dieback phenomena, but these types of studies can only be applied to single sites on a small scale and require considerable resources, so even these alone do not allow for the study of large areas such as those affected by dieback.

To obtain information on forest vulnerability on a large scale and save the time and resources needed for field surveys, remote sensing can assist. Indeed, satellite-based vegetation indices have made it possible to switch from individual- to forest-scale studies. Therefore, the combination of the dendrochronological approach and remote sensing is promising for assessing forest decline ^[16][17][51,67]. A widely used remote sensing index on which many other indices are based is the Normalized Difference Vegetation Index (NDVI) ^[18][68]. This index is widely used as a proxy for forest photosynthetic activity ^[2][16][19][45,51,52] and productivity in drought-prone Mediterranean biomes ^[20][50]. Thus, after drought events or heat waves, which lead to a reduction in photosynthetic activity, the NDVI tends to assume lower values, while higher values of the index indicate favourable conditions for plant health. Therefore, this index has been used to study mortality phenomena or increases in biomass related to climatic conditions ^[21][44]. For example, some studies ^[22][53] have shown the existence of a positive correlation between resilience indices ^[23][39], obtained using growth in terms of tree ring width indices (TRWi) and forest productivity in terms of NDVI. However, it must be remembered that the response of forests, as expressed by the remote indices, can differ depending on the type of forest site, tree species and the degree of stand mixing ^[24][69].

Normalized Difference Vegetation Index (NDVI) VI is an index that measures photosynthetically active biomass (canopy of trees and greenness), but its relationship to growth is complex ^[25][26][47,70]. Indeed, changes in carbon allocation may favour foliage over woody biomass, leading to a weakening of the relationship between tree-ring growth and the remote sensing signal. This can cause inconsistency phenomena between trends ^[27][54], i.e., the presence of positive NDVI trends when negative tree-ring trends are observed ^[28][55]. To study these relationships, Vicente-Serrano et al. ^[29][56] compared tree ring data with NDVI time series on a global scale, finding a high spatial and temporal divergence in forest growth responses. In fact, growth rates and vegetative recovery between coexisting species may differ and, respectively, carbon sequestration may vary and influence the growth of rings with respect to NDVI. Therefore, the different phenology of wood and leaf formation could explain the decoupling between NDVI and growth ^[29][56].

Other cases where NDVI may not provide an accurate account of radial growth are surveys in ecotone areas ^[30][57], i.e., at forest edges where there is a transition from tree to shrub vegetation, greening trends with increasing shrub biomass could alter vegetation indices. Furthermore, it has been shown that not only vegetation composition, but also slope ^[16][51], exposure and altitude, influence the climatic response, which means that in an area, different vegetation types or trends may confound the NDVI–ring width relationship.

Consequently, low image resolution, changes in resource allocation in trees and site characteristics may interact to limit the correlation between NDVI and annual radial growth ^[25][47]. These limitations increase with the complexity of the landscape, such as for highly heterogeneous Mediterranean ecosystems that manifest articulated responses to extreme events, so response patterns and tree-ring growth on NDVI time scales may not be fully representative ^[20][50]. On the other hand, if species composition is homogeneous or if the proportion of dominant species responds similarly to climatic variations, then there should be a positive correlation between NDVI and trends in ring width ^[31][58].

In order to use these two indicators (NDVI and tree rings) congruently, one could consider the observed positive correlations between MXD (maximum latewood density), NDVI and temperature during the growing season ^[30][57] and perform satellite analyses at a higher spatial and temporal resolution that could allow for a better investigation. Certainly, an examination of the relationship between NDVI and processes at the tree/species level, date of sprouting or root growth may lead to a better understanding of these dynamics ^[25][47]. Thus, appropriate research is needed to understand the physiological and phenological processes that explain the dependence between wood formation and photosynthetic processes underlying NDVI and the relative time intervals in which these processes occur ^[29][56].

In addition, the use of high-resolution satellite data could improve remote sensing information; in fact, Sentinel-2 10 m × 10 m space resolutions have given good results in small-scale monitoring ^[2][45] of the effects of extreme weather events on mixed Mediterranean forests in southern Italy, showing a good correspondence between NDVI and qualitative data collected in the field. To obtain highly detailed resolutions, an alternative could be remote sensing with drones, which allows lower material and operational costs and greater flexibility in spatio-temporal resolution than satellites ^[32][71]. Recent studies ^[33][72] have used unmanned aerial vehicles (UAVs) to monitor mixed coniferous and deciduous forests in northern Mexico with excellent results. Using specific sensors, they calculated tree height, canopy area and number of trees, and with a multispectral camera (PM4), with a resolution of up to 10 cm per pixel, they accurately estimated a number of multispectral indices related to vegetation activity. However, even then, seasonal monitoring is recommended to obtain an accurate estimate of photosynthetic activity and determine the seasonality of plant response. Furthermore, higher-quality mapping requires new research paradigms and the need to adapt algorithms according to forest stand characteristics ^[33][72].

Remote sensing therefore has great potential in forest monitoring, but most satellites have a low to moderate spatial resolution, which means that a pixel contains a mixture of tree vegetation, undergrowth, soil, shade, etc. ^[34][35][48,49]. This could lead to anomalous index values, particularly in sparse forests and those affected by climate-change-induced mortality ^[35][49].

Therefore, it is necessary to estimate the fractional coverage of photosynthetic vegetation, non-photosynthetic vegetation and bare soil. Guerschman et al. ^[36][73] developed a very interesting approach, i.e., they used the NDVI and the Cellulose Absorption Index (CAI) to distinguish the different cover types. Analysing large areas of Australia characterised by different cover types (Closed Forest > 80% cover, Non Forest < 20% cover, Open Forest 50%–80% cover and Woodland 21%–50% cover) and using data from the EO-1 Hyperion satellite, with a hyperspectral sensor (30 m spatial resolution), they showed that green vegetation is represented by high NDVI values and an intermediate CAI; dry vegetation and litter by low NDVI values and a high CAI; and bare soil by low NDVI values and a low CAI. In other words, CAI increases linearly with increasing non-photosynthetic vegetation ^[37][74]. Furthermore, the work of Guerschman et al. ^[36][73] showed that the ratio between the SWIR3 and SWIR 2 bands of MODIS (bands 7 and 6 at 500 m resolution) is linearly correlated with NDVI and CAI derived from Hyperion. Therefore, fractional vegetation cover can be analysed with satellite data (Hyperion and MODIS satellites), but it is still a moderate resolution.

Over time, in order to solve the surface discrimination problem, attempts have been made to reduce the soil signal in the presence of low vegetation cover by adding soil correction factors, resulting in indices such as the Soil-Adjusted Vegetation Index (SAVI) ^[38][75], Modified Soil Adjusted Vegetation Index (MSAVI) ^[39][76], Optimisation of Soil-Adjusted Vegetation Index (OSAVI) ^[40][77] and Generalized Soil-Adjusted Vegetation Index (GSAVI) ^[41][78]; alternatively, weighting coefficients were added to improve vegetation signals, as in the case of the indices Enhanced Vegetation Index (EVI) ^[42][79], Wide Dynamic Range Vegetation Index (WDRVI) ^[43][80] and Near-Infrared Reflectance of terrestrial vegetation (NIRv) ^[44][81]. However, these satellite-derived indices are not yet able to accurately capture surface phenological changes due to their limited spatial resolution ^[35][49]. In addition, shading causes alterations in indices values, as with NDVI, reducing the accuracy of land cover classification ^[45][82].

An approach that could improve this problem could come from comparing NDVI values obtained from satellites, results obtained from radiometers attached to field towers and field data obtained from drones. Wang et al. ^[35][49] conducted such an approach in Israel, analysing a Pinus halepensis Mill. forest located between the Mediterranean Sea and the Dead Sea, using drones with multispectral cameras with high spatial resolution (around 5 cm at a flight height of 50 m), have improved the accuracy of pine canopy segmentation, vegetation indices and shaded area classification. It was also determined that the satellite data (Landsat 8) were dominated by soil signals (70%), while the tower data were dominated by canopy signals (95%). With these results, discrepancies in NDVI values were recovered and corrected.

Therefore, once again, the use of drones, with the possibility of obtaining high-resolution images, can solve some of the problems encountered by remote sensing with satellites. Of course, in order to use these devices, one must perform a series of systemic time flights over the affected area to obtain an adequate time series. In this way, proximal remote sensing could become increasingly important for forest monitoring, both for the acquisition of remote data and for the calibration/correction of coarser data.