Cross-Comparison between Landsat 8 (OLI) and Landsat 7 (ETM + ) Derived Vegetation Indices in a Mediterranean Environment

: Landsat 8 is the most recent generation of Landsat satellite missions that provides remote sensing imagery for earth observation. The Landsat 7 Enhanced Thematic Mapper Plus (ETM + ) images, together with Landsat-8 Operational Land Imager (OLI) and Thermal Infrared sensor (TIRS) represent fundamental tools for earth observation due to the optimal combination of the radiometric and geometric images resolution provided by these sensors. However, there are substantial di ﬀ erences between the information provided by Landsat 7 and Landsat 8. In order to perform a multi-temporal analysis, a cross-comparison between image from di ﬀ erent Landsat satellites is required. The present study is based on the evaluation of speciﬁc intercalibration functions for the standardization of main vegetation indices calculated from the two Landsat generation images, with respect to main land use types. The NDVI (Normalized Di ﬀ erence Vegetation Index), NDWI (Normalized Di ﬀ erence Water Index), LSWI (Land Surface Water Index), NBR (Normalized Burn Ratio), VIgreen (Green Vegetation Index), SAVI (Soil Adjusted Vegetation Index), and EVI (Enhanced Vegetation Index) have been derived from August 2017 ETM + and OLI images (path: 188; row: 32) for the study area (Basilicata Region, located in the southern part of Italy) selected as a highly representative of Mediterranean environment. Main results show slight di ﬀ erences in the values of average reﬂectance for each band: OLI shows higher values in the near-infrared (NIR) wavelength for all the land use types, while in the short-wave infrared (SWIR) the ETM + shows higher reﬂectance values. High correlation coe ﬃ cients between di ﬀ erent indices (in particular NDVI and NDWI) show that ETM + and OLI can be used as complementary data. The best correlation in terms of cross-comparison was found for NDVI, NDWI, SAVI, and EVI indices; while according to land use classes, statistically signiﬁcant di ﬀ erences were found for almost all the considered indices calculated with the two sensors.


Introduction
In recent decades, an increasing number of satellite (and sensor) systems for Earth observation have provided large datasets of remote sensed imagery and indices, contributing to monitoring environmental changes at both regional and global scales. However, despite their increasing availability, this information cannot be easily compared due to slight differences among sensors, and thus, it is essential to define standards for cross-device validation, as well as reliable algorithms for dataset difference reductions [1,2].
To date, the cross-comparison analysis covers most of the different optical-and radar-based satellite systems currently in use for earth observation. These studies involve both the intercalibration among Furthermore, the Landsat 8 OLI bands are narrower compared to the Landsat 7 ETM+, avoiding atmospheric absorption [38]. In particular, the OLI NIR band is more similar to the MODIS NIR band, except for the 0.825 µm wavelength relative to the ETM+ for the absorption of water vapour [39]. The OLI Bands 6 and 7 are narrower compared to the ETM+ bands 5 and 7 respectively, reducing the atmospheric absorption and thus reducing the sensibility to atmospheric changes in terms of water vapour content.
To improve the standardization among different Landsat image generations, it is thus necessary to verify if the substantial differences between the two sensors regarding both bandwidth for the visible and SWIR, and the detection system technology, allowing for an efficient comparison among the images by the two sensors.
The main objectives of this study are: i) to extend the knowledge on the effects that the differences between Landsat 7 ETM+ and Landsat 8 OLI spectral responses may have through the analysis of different land use types through a wide range of derived vegetation indices; ii) to evaluate specific intercalibration functions for the standardization of vegetation indices in order to perform long-term time series analysis.
Up to date, the differences between Landsat 7 ETM+ and Landsat 8 OLI spectral responses have only been analyzed on a single derived vegetation index such as the NDVI, while in the present study, we extended the analysis to seven different indices and different land use types in order to achieve a more comprehensive analysis between the two sensors.

Study Area
The study area is located in the Basilicata region, in southern Italy ( Figure 1). The Region can be considered representative for the Mediterranean area due to both climate and vegetation characteristics. In particular, the area is characterized by both a Mediterranean humid-type climate in the Apennine Mountains, and a Mediterranean dry-type climate in the hilly and flat areas. Forest vegetation covers 35.6% (355,367 ha) of the total area, represented by deciduous oak forests (51.8%), beech forests (8.4%), Mediterranean Macchia (7.9%), shrublands (6.9%), and other deciduous broad-leaved forests (5.5%) [42]. The analysis was performed in four rectangular plots covering an area of 4536 ha each, for a total area of 18,144 ha. The selected plot areas cover different land uses, climatic and orographic conditions. As shown in Table 1, the main land uses are "Forests" and "Arable lands" classes, representing 90% of the plot area. Other land uses are "Tree crops" (5%) and "Pastures and natural grasslands" (2%). Main differences in the plot areas depend on the distribution of the two main land use classes: i) plots 1 and 3 with predominant "Forests" class distribution; and ii) plots 2 and 4 with predominant "Arable lands" class distribution.

Spatial Data Preprocessing
To perform the cross-comparison analysis, the images from the two Landsat sensor generations have been selected according to the temporal proximity and to the overall quality, using the Global Visualization Viewer (Glovis) developed by the USGS Earth Resources Observation and Science Center (EROS). The selected scenes are the ETM+ SLC-off (Scan Line Corrector) (path = 188; row =  (1,2,3,4) within the territory of Basilicata Region (Southern Italy) presented on Landsat 8 Operational Land Imager (OLI) Normalized Difference Vegetation Index (NDVI) (Tile: 188, 32; image acquired on 18/08/2017).
The analysis was performed in four rectangular plots covering an area of 4536 ha each, for a total area of 18,144 ha. The selected plot areas cover different land uses, climatic and orographic conditions. As shown in Table 1, the main land uses are "Forests" and "Arable lands" classes, representing 90% of the plot area. Other land uses are "Tree crops" (5%) and "Pastures and natural grasslands" (2%). Main differences in the plot areas depend on the distribution of the two main land use classes: i) plots 1 and 3 with predominant "Forests" class distribution; and ii) plots 2 and 4 with predominant "Arable lands" class distribution.

Spatial Data Preprocessing
To perform the cross-comparison analysis, the images from the two Landsat sensor generations have been selected according to the temporal proximity and to the overall quality, using the Global Visualization Viewer (Glovis) developed by the USGS Earth Resources Observation and Science Center (EROS). The selected scenes are the ETM+ SLC-off (Scan Line Corrector) (path = 188; row = 32), WRS2 (Worldwide Reference System2) (10/08/2017) and OLI (18/08/2017), classified as L1T (Level-1 Terrain-corrected) and registered to UTM coordinates, zone 33N, WGS84 Datum. The acquisition dates Remote Sens. 2020, 12, 291 5 of 20 were chosen to minimize potential differences in terms of phenological phases between the ETM+ and OLI images.
The land use information has been derived merging the Regional Agricultural Land Use Map, and the Regional Forest Map of Basilicata [42]. The land use classes in the two maps have been reclassified to reduce land use types (Table 2), as well as corrected through visual analysis based on the visual matching of land use classes from thematic maps compared to digital images form aerial orthophotos at 0.5 m spatial resolution TerraItaly 2008 data set. According to several studies [43][44][45][46], both the radiometric calibration and the atmospheric correction represent a fundamental prerequisite for quantitative analyzes on Landsat data.
In particular, the radiometric calibration is required to convert the signal of quantified energy from digital numbers (DNs) into sensor radiance values, by using: where TOA (top-of-atmosphere) represent the radiance at the sensor for each band (W m −2 sr −1 µ −1 ); DN is the Digital Number value; G and B are respectively Gain and Bias values, calculated as follows: where L min and L max represent minimum and maximum values of spectral radiance, as reported by the ETM+ and OLI image metadata files; Q min and Q max represent minimum and maximum values of DN for the two sensors, ranging from 1 to 255 for ETM+ and from 1 to 65,535 for OLI (16 bit images [47]. The atmospheric correction is necessary to reduce or normalize the modifications of the remotely sensed signal due to the intervening atmosphere between Earth's surface and satellite. Although the atmospheric correction usually follows standardized procedures [48,49], it is a crucial pre-processing step, especially for data acquired by different sensors at different times, as well as to field-based and remote sensing data [49].

Vegetation Indices
The ETM+ and OLI derived vegetation indices used for the cross-comparison analysis are reported in Table 3.
The NDVI is the first and most known vegetation index to explore and detect vegetated areas and plant canopies [61]. The NDWI, based on the combination of the green band with NIR or SWIR, is a good indicator for vegetation liquid water content and in the same time is less sensitive to atmospheric scattering effects than NDVI [62]. Its usefulness for drought monitoring and early warning has been demonstrated in different studies [63,64].
The LSWI (Land Surface Water Index) is sensitive to the total amount of liquid water in vegetation and in its soil background [62,[65][66][67][68][69]. It is based on the SWIR1 and on the NIR regions of the electromagnetic spectrum. It is particularly used to detect the vegetation water content during the different phenological stages [70,71], and the inter-annual changes in vegetation water content related to climatic variations [66,[72][73][74].
The NBR (Normalized Burn Ratio) index is used to detect the burnt areas and fire intensity [75][76][77][78][79]. The NIR band is sensitive to vegetation chlorophyll content, whilst the SWIR2 band is sensitive to soil and vegetation water content [80,81] providing information on typical conditions that occur after the fire events [82,83].
The EVI (Enhanced Vegetation Index) represents an improvement of NDVI, showing a reduced saturation in high vegetation cover regions, a reduction in atmospheric influences and a de-coupling of the canopy background signal [35,84,85]. These improvements are based on the introduction of the blue band to reduce the effects of the atmospheric aerosols in the red band, and on some correction coefficients to reduce the effect of soil reflectance. According to these differences, the NDVI is more sensitive to the chlorophyll content, whilst the EVI is more sensitive to the structural characteristics of the vegetation cover [86]. Thus, it has been widely used for Land Cover and Land Use/Land Cover Change detection [87], for the evaluation of vegetation bio-physical parameterization [88,89], phenology [85,90,91], evapotranspiration [73], biodiversity assessment [92], and gross primary production [93][94][95].

Index
Formula Author The VIgreen (Vegetation Index Green) is sensitive to chlorophyll content and particularly to the photosynthetic active vegetation [99]. It has been used to detect the physiological conditions of the vegetation, the vegetation water content implemented in fuel models [100], the structural characteristics of the tree crops [101], and the phenology in agricultural crops [102].

Preliminary Analyzes
Due to the 8-day time lag between the Landsat ETM + and OLI image, a preliminary analysis was carried out on climatic data derived from the local meteorological stations in each considered plot. The occurrence of climatic events (especially with regard to precipitations) was verified, in order to remove any significant influence on the NDVI values, during the period between images. Over the 8-day time period (data processed from the Hydrological Annals from 10 to 18 August 2017) no anomalies were recorded that could significantly impact the NDVI values (Table 4). However, in order to account for possible phenological effects that occurred over the 8-day time span, the vegetation growth rate was calculated using the MOD09Q1 MODIS product (surface reflectance 8 days composite): Subsequently, to correlate the Landsat 8 NDVI of 18 August, the growth rate [5] between the 10 and 26 August ETM + images has been used: where  Thus, the comparison analysis between the two sensors has been performed for the all selected indices. Thus, the comparison analysis between the two sensors has been performed for the all selected indices.

Statistical Analysis
According to the main studies on cross-comparison statistical analysis between different sensors [1,10,29,37,59,103], we performed for each test plot: i) the analysis of differences of the spectral responses between ETM+ and OLI for different land uses by estimating the differences between the reflectance average values of the two sensors and their dispersion. The analysis has been separately conducted for the main land uses in order to highlight potential spectral differences between them and the two sensors. ii) The analysis of differences in terms of descriptive statistic between different vegetation indices. In this ambit we analyzed the distribution of the different vegetation indices using indices of position and dispersion, highlighted through the box and whisker plots. Thus, the mean values and the measures of dispersion (mean + − SE and mean + − 2 SD) have been evaluated for the values of the indices differences and in order to examine the possible effects for different land use and environmental characteristics of each plot. The vegetation indices statistical significance difference has been also evaluated using the Student's t-test, by considering the values of the indexes both on the whole image and on the stratified values, as a function of the land use. iii) the regression analysis between vegetation indices for the two sensors and for each plot, in order to evaluate a possible linear relationship between the two sensors, and consequently to assess the estimated intercalibration functions.

Results
Figure 3 shows significant differences in the average reflectance values between the ETM+ and OLI bands by land use classes. In particular, the OLI NIR band seems to be more sensitive (higher average reflectance value compared to ETM+) in the forest class, while the reflectance in the visible band (and especially in the red one) is greater in ETM+ than OLI, determining the higher values of OLI NDVI than ETM+ NDVI.
The SWIR1 and SWIR2 infrared bands show few differences for the forest land use class with a slight absorption for the SWIR1 OLI band. The NIR OLI band shows higher average reflectance value than ETM+ for the land use classes characterized by high vegetation cover than for low vegetation cover areas (arable lands, urbanized areas). The SWIR (SWIR1 e SWIR2) OLI bands show higher absorption values than ETM+, although these differences tend to decrease in the area with high vegetation cover. The highest differences between the two sensors refer to the water bodies with significantly higher average reflectance values for all the ETM+ bands, and particularly for the SWIR1 band. The sensor sensitivity to visible bands shows an opposite behaviour for the NIR, in all the land use classes, with ETM+ average reflectance values constantly higher than OLI (especially for the red band) with significant differences for pastures, arable lands, and water bodies. To compare the ETM+ and OLI derived vegetation indices, for the 4 plots, the NDVI, LSWI, NDWI, NBR, VIgreen, SAVI, and EVI have been calculated from the ETM+ and OLI images. The Student's t-test results (Table 5) show a significant difference between the ETM+ and OLI derived vegetation indices (NDVI, LSWI, NDWI, NBR, VIgreen, SAVI, and EVI) for the 4 plots.
As shown in Figure 4, there are relevant dissimilarities between the two sensors in all indices. The NDWI, SAVI, and LSWI indices show the higher correlation between the two sensors with a mean difference of −0.0500, −0.0725, and 0.0819, respectively. Instead, the standard deviation is smaller than the other indices. The other indices (NDVI, VIgreen, and EVI), with the exception of NBR (which has the difference average value equal to −0.8446, and the standard deviation of 1.27) show difference average values higher than 0.1. Among these, the EVI shows a greater similarity between the two sensors (with very low deviations), followed by the NDVI with an average difference value equal to 0.1223. The VIgreen shows the higher differences between OLI and ETM+ (average difference value equal to 0.1367). Figure 3 shows significant differences in the average reflectance values between the ETM+ and OLI bands by land use classes. In particular, the OLI NIR band seems to be more sensitive (higher average reflectance value compared to ETM+) in the forest class, while the reflectance in the visible band (and especially in the red one) is greater in ETM+ than OLI, determining the higher values of OLI NDVI than ETM+ NDVI. The SWIR1 and SWIR2 infrared bands show few differences for the forest land use class with a slight absorption for the SWIR1 OLI band. The NIR OLI band shows higher average reflectance value than ETM+ for the land use classes characterized by high vegetation cover than for low vegetation cover areas (arable lands, urbanized areas). The SWIR (SWIR1 e SWIR2) OLI bands show higher absorption values than ETM+, although these differences tend to decrease in the area with high vegetation cover. The highest differences between the two sensors refer to the water bodies with significantly higher average reflectance values for all the ETM+ bands, and particularly for the SWIR1 band. The sensor sensitivity to visible bands shows an opposite behaviour for the NIR, in all the land use classes, with ETM+ average reflectance values constantly higher than OLI (especially for the red band) with significant differences for pastures, arable lands, and water bodies. To compare the ETM+ and OLI derived vegetation indices, for the 4 plots, the NDVI, LSWI, NDWI, NBR, VIgreen, SAVI, and EVI have been calculated from the ETM+ and OLI images. The Student's t-test results (Table 5) show a significant difference between the ETM+ and OLI derived vegetation indices (NDVI, LSWI, NDWI, NBR, VIgreen, SAVI, and EVI) for the 4 plots.  The indices based on non-visible bands combinations (LSWI and NBR) are therefore sufficiently similar and stable between the two sensors, with few fluctuations in the average difference over the four plots. However, the NDVI and NDWI show the greater stability in the different plots with very few fluctuations. The NDWI also shows the lower differences between the two sensors in all the plots. The VIgreen index shows the lower stability, confirming that indices using only visible bands show higher variation due to the SBDE (spectral difference between bands) [2,29].
In particular, these results are related to the main differences in the red band between the two sensors for all the land uses with the exception of urbanized areas. Figure 4 was characterized by small differences in every index. This can be attributed to the morphological characteristics of the area depending on climatic conditions and on land use classes. In particular the plot 4 (Table 2), located in the Southeast of the regional survey area, is characterized by low altitude and flat terrain. As for the land use classes, Figure 4 shows a predominance of arable land (about 50% of the area), followed by forests, for about 1/3 of the entire area, and tree crops that reach here the greater extension (12% of the plot area) among all the plots.
As shown in Figure 4, there are relevant dissimilarities between the two sensors in all indices. The NDWI, SAVI, and LSWI indices show the higher correlation between the two sensors with a mean difference of −0.0500, −0.0725, and 0.0819, respectively. Instead, the standard deviation is smaller than the other indices. The other indices (NDVI, VIgreen, and EVI), with the exception of NBR (which has the difference average value equal to −0.8446, and the standard deviation of 1.27) show difference average values higher than 0.1. Among these, the EVI shows a greater similarity between the two sensors (with very low deviations), followed by the NDVI with an average difference value equal to 0.1223. The VIgreen shows the higher differences between OLI and ETM+ (average difference value equal to 0.1367).  The results of the analysis of the difference variability between the vegetation indices for each of the test area and for each land use class are reported in Figure 5 and Table 6.
The results in Table 6 show that, for almost all the considered indices, statistically significant differences exist between the two sensors by land use class. Only for water bodies class, NDWI (often used for the identification of wet areas), SAVI and EVI indices do not show significant differences between the sensors, and SAVI also does not show significant differences for the Urbanized and Mixed agricultural areas classes. This result also confirms the robustness of the correction of the soil brightness introduced in the index for both sensors, for the discrimination of the classes in which the vegetation is absent (Urbanized areas) or it is interspersed with bare soils (Mixed agricultural areas).
According to the box-plot results showed in Figure 5, for forest land use, some indices based on the combination of NIR and Red (SAVI and EVI) show a remarkable difference between the two sensors with significant variability. The same performance for the forest land use has been found in NBR, based on the combination of infrared bands, and VIgreen, based only on visible bands. NDVI and NDWI for forest land use show good results with low average difference values and low variability. Tree crops show average difference values generally close to zero and low variability with respect to the other land uses for all indices.
Even arable land classes show a similar trend to the tree crops with generally low average difference values between the two sensors, both for the vegetation indices based on the combination of infrared bands (NBR, LSWI), and based only on visible bands (VIgreen).
As for the land use classes, Figure 4 shows a predominance of arable land (about 50% of the area), followed by forests, for about 1/3 of the entire area, and tree crops that reach here the greater extension (12% of the plot area) among all the plots.
The results of the analysis of the difference variability between the vegetation indices for each of the test area and for each land use class are reported in Figure 5 and Table 6.  The plot 4, characterized by a lower percentage of forest land use cover and combined with the higher percentage of tree crops cover and high arable lands cover, shows a low variability of the average difference values for all the vegetation indices.
Plots 1 and 2, characterized by a higher percentage of forest land use cover (more than 60% of the plot area), by a lower percentage cover of arable land, and by a significant reduction, compared to the fourth plot, of tree crops land use (respectively 0.1% and 2% of the surface of the plot), show the highest variability for all vegetation indices.
Finally, the regression analysis between the vegetation indices ETM+ based and OLI based has been conducted in order to identify the intercalibration functions between the two sensors datasets. The goodness of fit of the OLS regressions were defined by the coefficient of determination (R 2 ) and the significance of the OLS regressions was defined by examination of the regression overall F-statistic p-value. To provide simple overall measures of similarity, the Root Mean Square Deviation between the OLI and ETM+ data has derived: The identified functions and the related indices (Table 7) show a good linear relationship between the dependent variables (indices calculated with OLI) and the independent variables (indices calculated with ETM+). The intercalibration functions, however, show significant dissimilarities among vegetation indices: the indices based on the combination between infrared and visible bands (NDVI, NDWI, SAVI, and EVI) show a better correlation with a determination coefficient always higher than 0.8. Instead, the indices based only on infrared (NBR and LSWI) or on visible bands (VIgreen) show a lower correlation. In particular, the VIgreen shows a determination coefficient less than 0.5, highlighting the differences SBDE in the range of the visible waveband, and, thus, confirming the previously described analysis. NBR and LSWI also show a poor correlation, particularly for NBR, which highlight the existence of a non-linear component between the two sensors. In addition, in this case, the differences are most likely related to SBDE, as for the bands SWIR highlighting a significant difference in the bandwidth between the two sensors.
From the regression analysis, it has been found that lowest values are reported in plot 4, whilst plot 2 and 3 show higher correlation values. In particular, plot 4 is characterized by the lowest average values for all the calculated vegetation indices, whilst plot 2 and 3 are characterized by highest average values for all the calculated vegetation indices.
These differences also suggest that the higher variability within vegetation indices values occur within areas characterized by low vegetation indices values and thus for areas with a low percentage of vegetation cover. Figure 6 shows different trends according to the land use type and vegetation index. The OLI NDVI is always higher than the ETM+ NDVI in all land uses, with higher differences for pasture and arable land and a decrease for tree crops and forests. This may suggest that the greatest differences of index values are related to low biomass content (pastures and arable land), whilst lower differences are related to the high value of biomass content. In particular, pasture, arable lands and tree crops show a regression function with a slope almost parallel to the 1:1 line, and forest land use shows the largest differences for lower NDVI values. With the increase of the index value, there is a constant reduction of the differences between OLI and ETM+ derived indices. The EVI shows the same trends described for NDVI and for indices based on the normalized difference of infrared bands (LSWI in Figure 6). VIgreen, based on visible bands (green and red), shows the greatest differences for land use classes characterized by high values of vegetation cover (forests) with differences nearly constant. This confirms the existence of differences due to SBDE between the two sensors in the visible range, especially with respect to the red band. For arable land, the trends of the index are different because OLI is higher than ETM+. For low index values, the increase of the VIgreen values ETM+ is higher than OLI ones due to a greater absorption in OLI red band together with the intensification of the vegetation cover.

Discussions
The comparison between Landsat 7 ETM+ and Landsat 8 OLI to identify the main differences between the two sensors and the spectral responses for different land uses presented in this study,

Discussions
The comparison between Landsat 7 ETM+ and Landsat 8 OLI to identify the main differences between the two sensors and the spectral responses for different land uses presented in this study, showed significant differences among spectral ranges and vegetation indices derived from the two sensors. However, the 8-day difference between the two Landsat 7 ETM+ and Landsat 8 OLI scenes and the period of the analysis minimizes the effects of vegetation phenology and thus increasing the reliability of the results of the present study. The comparison among the spectral bands provided by the two sensors showed a good similarity. Nevertheless, significant differences are evident in the various spectral ranges: i) high reflectance values of OLI in the band of NIR for all the land use classes, particularly for high values of vegetation cover, ii) and lower reflectance in the SWIR bands, especially for low vegetation cover land uses as for the pastures and grasslands. There are decreasing differences in SWIR band with the increase of biomass values up to maximum values represented by forest cover.
For the visible bands, the ETM+ sensor shows reflectance values greater than OLI ones. The analysis of vegetation indices based on the two sensors shows low differences between the indices based on the combination of NIR and visible bands (NDVI, EVI, NDWI, and SAVI), whilst the indices based only on the visible bands (VIgreen) or only on the infrared bands such as NBR, show higher values of difference. A different result is showed by the analysis of LSWI which, according to the results showed by Li et al. [29], is not only more similar but also more stable between the two sensors, with low fluctuations in the various plot areas. The regression analysis among the vegetation indices calculated with the two sensors highlights a good correlation, especially for NDVI, NDWI, SAVI, and EVI, with an index of determination higher than 0.8. On the other hand, both VIgreen and NBR show a high dispersion, probably due to the differences in the spectral response of corresponding bands.
With regard to the study of the intercalibration functions between ETM+ and OLI derived vegetation indices for the Mediterranean region is fundamental to ensure an effective operational continuity between the current and previous Landsat missions.
As showed by the literature and confirmed by the results of this study, the two sensors derived vegetation indices are not directly comparable. Differences between Landsat 7 ETM+ and Landsat 8 OLI vegetation indices are directly related to the different width of spectral bands and to the different radiometric resolution between the two sensors [104].
Different studies showed how the differences for high values of NDVI are directly related to the spectral bandwidth; other studies highlighted the inverse relationship between the increases in bandwidth and mean NDVI values [46].
Another factor affecting the differences of vegetation indices between the two sensors, and in particular, the dynamic range of vegetation indices, is the difference in spectral band resolution [105]: the higher is the number of bits, the higher is the variability of the estimated index. According to this, the Landsat 8 OLI derived vegetation indices represent an effective tool to study the ecosystem variability in the Mediterranean region, characterized by a very high land use and land cover variability, due to the combination of both land management practices and physical characteristics mainly related to high climatic variability.

Conclusions
Landsat satellite systems represent a fundamental tool for the study of long-term environmental and land use/land cover dynamics at a global scale. The long-term dataset availability combined with the increased processing capacity, contributed to an increase of the number of studies, observation programs and land cover monitoring tasks. In order to emphasize and standardize the available dataset provided by different missions and to allow reliable and site specific multi-temporal analysis, it is important to define intercalibration algorithms between the different missions. The high ecosystem spatial and temporal variability, characterizing the Mediterranean region, represents a real challenge to develop reliable intercalibration algorithms between the two sensors. The results of this study showed an overall good correlation between indices calculated with the two sensors; while according to single land use classes, statistically significant differences was found for almost all the indices analyzed. According to these observations, a greater attention needs to be addressed in long-term environmental and land use/land cover dynamics analysis in the Mediterranean region. In particular, the very fast land use/land cover changes characterizing this region under the ongoing climate change scenario, are mainly due to the combination of both natural and anthropogenic drivers. Such variability combined with sensor's differences may lead to wrong estimates and thus wrong interpretation of environmental patterns of main vegetation indices. In particular, more attention needs to be addressed to differences between the two sensors in the evaluation of infrared (NIR and SWIR) based indices for the estimation of both vegetation and soil water content, which represent highly climate sensitive key indicators in the Mediterranean environment.