Mâamora forest is located in the north west of Morocco between 6˚ and 6˚45' west longitude and 34˚ and 34˚20' north latitude (Figure 1). From the west to the East, five cantons are distinguished in this forest: A, B, C, D and E. The forest witnesses a gradual decrease of rainfall from canton A to canton E and it varies between 400 to 600 mm. Conversely, the average temperature increases from the West to the East. Bioclimate is sub-humid in the Western part and semi-arid in the Central and Eastern part.

Sentinel 2 images of 2015 and 2020 were acquired from the United States Geological Survey (USGS) in the beginning of summer, mid-summer and at the end of this season (Table 1). This season was chosen in order to avoid the confusion of classification due to the presence of herbaceous and shrub vegetation out of summer. Image processing was carried out using QGIS 3.16. Since all retrieved images were of level 1C, radiometric and geometric corrections are included

(USGS, 2018). As for the atmospheric correction, it was done using the empiric algorithm of QUick Atmosphic Correction (QUAC). Given that the vegetation indexes were calculated on different years and for a good comparison of these spectral indexes (Chander et al., 2007; Chander & Markham, 2003), it was of essential importance to convert digital numbers DNs of each pixel to radiance and top-of-atmosphere (TOA) reflectance (Young et al., 2017). In the present study, the Semi-Automatic Classification Plugin (SCP) with QGIS was used to perfom this conversion.

The selection of spectral indexes was based on two criteria. The decline of cork oak forests is the result of the presence of pests favored by water deficit (Mattson & Haack, 1987a, 1987b) constituting, therefore, the main cause of trees’ dieback (El Abidine, 2003). Leaves dry out drastically in dry conditions or due to pest infection (Wang et al., 2007). Hence, the first index should take into account the vegetation moisture. Furthermore, among the symptoms of cork oak dieback are recorded defoliation and branch mortality; this phenomenon may lead to trees mortality and thus to natural vegetation loss (de-densification). In this respect, the second index should take the vegetation density into consideration. The Normalized Difference Water Index (NDWI) fulfills the first criterion related to water stress and uses the band of Short-Wave Infrared (SWIR) (between 1500 and 1750 nm) which is the most sensitive to the variation of leaves water content (Jensen, 2000) and is negatively correlated to this parameter (Tucker, 1980). The combination of this channel with that of the Near Infrared (NIR) enhances the precision regarding vegetation moisture (Ceccato et al., 2001) and therefore provides information about its hydric status (Wang et al., 2007).

NDWI = R NIR − R SWIR R NIR + R SWIR (1)

Bands 8 (NIR) and 11 (SWIR) of Sentinel-2 images were used in Equation (1).

Normalized Difference Vegetation Index (NDVI) is used as an indicator of vegetation stress (Recanatesi et al., 2018). Some studies have indicated its effectiveness to discriminate health status of vegetation (Adams et al., 2000). Hence, NDVI meets the second criterion and presents an exponential relationship with vegetation density (Holben, 1986).

NDVI = R NIR − R R R NIR + R R (2)

Bands 4 (R) and 8 of Sentinel-2 images were used in Equation (2).

Aiming at mapping cork oak dieback in Mâamora, Wang et al. (2007) method was utilized. First, the differential NDWI of 2015 and 2020 was computed as follows:

Δ NDWI = NDWI 2020 − NDWI 2015 (3)

Negative values of ΔNDWI reflect dying back trees indicating low leaf moisture. On the contrary, positive values represent trees recovery from previous stresses. Secondly, a statistical method was adopted based on thresholds to distinguish dying back cork oak, those recovered and those with no changes:

· Strong dieback: ΔNDWI < −2σ;

· Slight dieback −2σ < ΔNDWI < −σ;

· Cork oak with no changes: −σ < ΔNDWI < σ;

· Slight recovery: σ < ΔNDWI < 2σ;

· Strong recovery: ΔNDWI > 2σ.

To validate the method used, three descriptors were considered to assess the vegetation health in the field. These parameters were examined in areas classified as severely and moderately affected by dieback in the generated map. This validation concerned 60 plots and in each one 10 cork oak trees were observed and rated. The number of plots was restricted to 60 due to the limited time and resources. The descriptors were: 1) branches mortality in the tree crown (the upper part of the crown excluding the lower and lateral parts undergoing natural competition phenomenon), 2) deficit in leaves as compared to a reference cork oak tree and 3) inappropriate pruning practiced by the human. The dieback index (ID) of Bouvarel (1984) was calculated for every plot taking into consideration the degree of defoliation of all the rated trees. This index was calculated as follows:

I D = ( n 1 P 1 ) + ( n 2 P 2 ) + ( n 3 P 3 ) + ( n 4 P 4 ) + ( n 5 P 5 ) N

where n₁, n₂, n₃, n₄ and n₅: the number of trees rated in each class (1, 2, .., 4)

P₁, P₂, P₃, P₄ and P₅: weight of each class (Table 2).

N: Number of trees in the plot (=10).

Depending on the value of the ID, three dieback degrees were distinguished (Table 3).

The calculation of NDWI of 2015 and 2020 enabled to produce histograms of images of the corresponding years (Figure 2). The image difference of the two

years generated a ΔNDWI histogram that is normally distributed with a mean and a standard deviation of −0.05 and 0.04 respectively (ΔNDWI~ N (−0.05, 0.04)) (Figure 2(c)). The statistical method used allowed to generate a map describing the evolution of cork oak dieback, between 2015 and 2020, of Mâamora forest (Figure 3). Areas occupied by each class were calculated, and it turns out that more than the half of the area of cork oak did not record a change in terms of health status between 2015 and 2020 and it is of the order of 36,892.9 ha, constituting 54.63% of the total area (Table 4). Oak stands affected by a severe decline represent 8757.7 ha, making 12.97% of the total area and those afflicted by a slight dieback form 21,002.4 ha, representing 31.10%. Surfaces of cork oak witnessing slight and strong recovery remain minimal and represent only 1.04%

and 0.25% respectively.

In cantons B, C and D, a large proportion of the area occupied by cork oak has not experienced a change between 2015 and 2020 (Figure 4). Canton A and E are the most afflicted by the decline phenomenon for the reason that 19.33% and 18.46% of stands are strongly dying back, and 3.29% and 39.96% are moderately dying back respectively. Conversely, B and D cantons are the least affected by the decline with a rate of the order of 9.71% and 6.61% respectively.

According to the NDVI map (Figure 5), the maximum value of this index is 0.89 while the minimum is −0.19. High values correspond to dense and healthy vegetation with a high reflectance, appearing light in the image. From the results, it is obviously clear that from the west to the east the pixel tends to be dark. This means that the density and the sanitary state of the forest exacerbate as per continentality. From the field, cork oak canopy density less than 10% was assigned to pixels with a value of NDVI between 0 and 0.2, between 10% and 40% to 0.2 and 0.4 and the density higher than 40% was attributed to pixels with a value larger than 0.2 (Figure 6). Stands with the latter density account for less than one-third of the total area occupied by cork oak (Table 5). The dominant class of the canopy density in Mâamora is between 10% and 40%, making 62.13% of the total surface.

Going from the west towards the east, meaning from canton A to E, the class of density higher than 40% curbs. More than the half of the area occupied by cork

oak has a density larger than 40% in canton A. The same class of density constitutes only 26% in the neighboring canton B (Figure 7). Canton E, with its eastern position is individualized regarding stands density.

Investigations on the field have exhibited that in the severe dieback class, 48.89% of the rated trees were severely pruned afflicting 1/4 to 1/2 of the whole tree, 35% experienced branch mortality affecting between 1/4 to 1/2 of the whole tree branches and 34.44% have lost between 1/2 and 3/4 of their leaves. Plots observed in the moderate dieback class have indicated that 55.33% of cork oak trees were slightly pruned, 71.33% have branch mortality in less than 1/4 of the tree branches and 47% have lost between 1/4 and 1/2 of their leaves. In order to determine the contribution of remote sensing in monitoring health status of cork oak, a comparison between data collected in the field, used to calculate the index of Bouvarel, and those obtained by satellite images was made resulting in an overall accuracy of 66.66%.