Comparison of Normalized Di ﬀ erence Vegetation Index Derived from Landsat, MODIS, and AVHRR for the Mesopotamian Marshes Between 2002 and 2018

: The Mesopotamian marshes are a group of water bodies located in southern Iraq, in the shape of a triangle, with the cities Amarah, Nasiriyah, and Basra located at its corners. The marshes are appropriate habitats for a variety of birds and most of the commercial ﬁsheries in the region. The normalized di ﬀ erence vegetation index (NDVI) has been derived using observations from various satellite sensors, such as the Moderate Resolution Imaging Spectroradiometer (MODIS), Advanced Very-High-Resolution Radiometer (AVHRR), and Landsat over the Mesopotamian marshlands for the 17-year period between 2002 and 2018. We have chosen this time series (2002–2018) to monitor the change in vegetation of the study area since it is considered as a period of rehabilitation for the marshes (following a period when there was little to no water ﬂowing into the marshes). Statistical analyses were performed to monitor the variability of the maximum biomass time (month of June). The results illustrated a strong positive correlation between the NDVI derived from Landsat, MODIS, and AVHRR. The statistical correlations were 0.79, 0.77, and 0.96 between Landsat and AVHRR, MODIS and AVHRR, and Landsat and MODIS, respectively. The linear slope of NDVI (Landsat, MODIS, and AVHRR) for each pixel over the period 2002–2018 displays a long-term trend of green biomass (NDVI) change in the study area, and the slope is slightly negative over most of the area. Slope values ( − 0.002 to − 0.05) denote a slight decrease in the observed vegetation index over 17 years. The green biomass of the marshlands increased by 33.2% of the total area over 17 years. The areas of negative and positive slopes correspond to the same areas in slope map when calculated from Landsat, MODIS, and AVHRR, although they are di ﬀ erent in spatial resolution (30 m, 1 km, and 5 km, respectively). The time series of the average NDVI (2002–2018) for three di ﬀ erent sensors shows the highest and lowest NDVI values during the same years (for the month of June each year). The highest values were 0.19, 0.22, and 0.22 for Landsat, MODIS, and AVHRR, respectively, in 2006, and the lowest values were 0.09, 0.14, and 0.09 for Landsat, MODIS, and AVHRR, respectively, in 2003.


Introduction
The Mesopotamian marshes in Iraq are regarded as one of the most vital wetlands in the world. It is the largest ecosystem in the Middle East and western Asia. During the previous few decades, the area of marshlands has varied between 10,500 km 2 and 20,000 km 2 due to flooding. The marshlands include the Al Hammar marsh, the Central marshes or Al-Amarah marsh, and the Al-Huwaiza marsh [1,2]. The marshes are a source of biodiversity in the Middle East, because of their size, hydrous vegetation, and their detachment from comparative equivalent frameworks [3,4]. The quick drying up of more than The floodplains are an area of low elevation and have been created through the accumulation of alluvial materials carried by water runoff from upstream [48]. The marshes are an essential part of provincial biodiversity in the Middle East and serve as an important habitat for migratory birds and plant species, such as reeds. Additionally, the marshlands act as a natural wastewater treatment system for the water flowing from the Tigris and Euphrates rivers, as they filter out contaminants

AVHRR LTDR V5 Daily NDVI Product
AVHRR is onboard the National Oceanic and Atmospheric Administration (NOAA) polar-orbiting environmental satellites, which are used to calculate the NDVI [52,53]. The latest version (V5) products were made available (http://ltdr.nascom.nasa.gov) on 15 August 2017. The daily NDVI estimates are derived using daily surface reflectance data from the AVHRR instruments onboard the NOAA platforms. There are two spectral channels (Red and NIR) on the NOAA AVHRR sensor that are used to calculate the daily NDVI value. The data set used in this study was compiled from the NOAA-16, -18, and -19 sensors. The LTDR products from the NOAA AVHRR sensors are available at a spatial resolution of 0.05 × 0.05 • with daily temporal resolution. The LTDR NDVI products have been used to map vegetation area and burned areas [54][55][56] and assess droughts [57]. The NDVI has range from −1 to 1, where positive values represent vegetated areas, while negative NDVI values represent water bodies, snow, clouds, and non-vegetated surfaces [58,59] (Table 1).

MODIS Data
In this study, we used the MOD13A2 Version 6 Vegetation Index (VI) values at 1 km spatial resolution [60] (https://lpdaac.usgs.gov/products/mod13a2v006/) ( Table 1). MOD13A2 was processed from the MODIS L3 daily surface reflectance product (MOD09) that provided red and near-infrared surface reflectance, corrected for the effect of atmospheric gases, thin cirrus clouds, and aerosols. The atmospheric correction of MOD09 bands 1-7 had been based on the 6S radiative transfer model used to generate the surface reflectance [61]. The NOAA Center for Environment Prediction (NCEP) provided the surface pressure, ozone, and water vapor content for the atmospheric corrections for the MOD09 product [17].

Landsat
Landsat 7 ETM+ is the seventh satellite of the Landsat program. Launched on 15 April 1999, Landsat 7's goal is to refresh the global archive of satellite images, offering updated and cloud-free images. The Landsat 7 data products are generated at 30 m spatial resolution and 16

Additional Data
The USGS Earth Explorer provides digital elevation model (DEM) elevations at a 30 m spatial resolution from the Space Shuttle Radar Topography Mission (SRTM). In this research, we used DEM data to examine the correlation between NDVI changes and elevation in the study area.

Methodology
The NDVI values were derived from the three sensors, AVHRR, MODIS, and Landsat 7 ETM+, for the Mesopotamian marshes corresponding to the maximum biomass period (June) over 17 years (2002-2018). According to [2], the growing season over the Mesopotamian marshes spans from April to August, with the maximum in the month of June. We used the month of June due to the minimum presence of clouds during the year. We mosaicked the Landsat images ( Row 39). We used 34 images (two/year) and mosaicked them together. The MODIS and AVHRR were re-projected from the sinusoidal projection to the UTM projection, the same as the Landsat data. For the direct comparisons between LTDR AVHRR, Terra MODIS, and Landsat ETM+, NDVI data were resampled to the spatial resolution of AVHRR NDVI (0.05 × 0.05 • ). The boxcar method was used for resampling when changing the pixel sizes. We took the average of all the 30 m or 1 km pixels for the NDVI from Landsat and MODIS, respectively, that fell inside a 5 km pixel of AVHRR as the 5 km pixel value. Spatial resolution alluded to the size of one pixel on the ground. In our examination, we utilized Landsat, MODIS, and AVHRR data that had 30 m, 1 km, and 5 km, respectively, that implied one pixel on an image corresponded to a square of 30 m by 30 m, 1 km by 1 km, and 5 km by 5 km on the ground, respectively. Sensors that have high spatial resolution cover a smaller area, while temporal repeat describes how frequently data for a region were gathered, also referred to as Revisit Time. The temporal repeat of the data was 16 days, 16 days, and daily for Landsat, MODIS, and AVHRR, respectively ( Table 1). The three different sensors were intercompared over the 17-year period (2002-2018). We analyzed the NDVI time series by examining the linear trends over the 2002-2018 time period. The linear trend is one of the most essential descriptors of vegetation dynamics for time series, to determine an increase and/or decrease in vegetation by using linear regression analysis [8]. Simple linear regression was carried out for the NOAA AVHRR, MODIS, and Landsat ETM+ time collection, to observe the temporal changes in NDVI. The linear trend model was quantified using the p-value (tested for statistical significance at the confidence levels of 95% (p < 0.05) and 90% (p < 0.1)), the correlation coefficient (R 2 ) and the standard deviation (SD). We also used the DEM at 30 m to identify the relationship between elevation and changes in vegetation over the 17-year period.

Time Series Analysis
For each pixel, we calculated the average NDVI over the 17-year period (2002-2018) for the three different sensors, at their original spatial resolution-5 km, 1 km, and 30 m for LTDR AVHRR, MODIS, and Landsat 7 ETM+, respectively ( Figure 2). Figure 2 shows a good match, where NDVI values greater than 0.2 were mostly located in the northeast of the Al-Huwaiza marsh and the southwest and west parts of Al-Hammar and the Central marshes of the study area. If we examine Figure 2, we can see the "bulls-eye" feature on each of the images (for Landsat, MODIS, and AVHRR) that is present in the northeast corner of the area. There is a high vegetation area in the southwest corner of the area, which is seen in all three images. The only difference in both these features between the images is that in the AVHRR image the features are blurry (low resolution at 5 km) and in Landsat, the features are sharp (high resolution at 30 m). MODIS displays the same feature with a resolution in between the two. Overall, the three datasets display similar average NDVI variability. values greater than 0.2 were mostly located in the northeast of the Al-Huwaiza marsh and the southwest and west parts of Al-Hammar and the Central marshes of the study area. If we examine Figure 2, we can see the "bulls-eye" feature on each of the images (for Landsat, MODIS, and AVHRR) that is present in the northeast corner of the area. There is a high vegetation area in the southwest corner of the area, which is seen in all three images. The only difference in both these features between the images is that in the AVHRR image the features are blurry (low resolution at 5 km) and in Landsat, the features are sharp (high resolution at 30m). MODIS displays the same feature with a resolution in between the two. Overall, the three datasets display similar average NDVI variability. The Landsat and MODIS NDVI data were resampled to match the spatial resolution of the AVHRR NDVI (0.05 × 0.05°). We employed the boxcar method for this process. Each raster value was a temporal-average of 17 months (month of June). The peaks in the NDVI plots from the different sensors appeared at approximately the same time through the 17 years of overlap ( Figure 3). Figure  3 shows that there is a good correspondence between the area-averaged NDVI derived from Landsat and AVHRR. In addition, the MODIS time series show a positive bias by about 0.02 relative to Landsat and AVHRR. The Landsat and MODIS NDVI data were resampled to match the spatial resolution of the AVHRR NDVI (0.05 × 0.05 • ). We employed the boxcar method for this process. Each raster value was a temporal-average of 17 months (month of June). The peaks in the NDVI plots from the different sensors appeared at approximately the same time through the 17 years of overlap ( Figure 3). Figure 3 shows that there is a good correspondence between the area-averaged NDVI derived from Landsat and AVHRR. In addition, the MODIS time series show a positive bias by about 0.02 relative to Landsat and AVHRR. values greater than 0.2 were mostly located in the northeast of the Al-Huwaiza marsh and the southwest and west parts of Al-Hammar and the Central marshes of the study area. If we examine Figure 2, we can see the "bulls-eye" feature on each of the images (for Landsat, MODIS, and AVHRR) that is present in the northeast corner of the area. There is a high vegetation area in the southwest corner of the area, which is seen in all three images. The only difference in both these features between the images is that in the AVHRR image the features are blurry (low resolution at 5 km) and in Landsat, the features are sharp (high resolution at 30m). MODIS displays the same feature with a resolution in between the two. Overall, the three datasets display similar average NDVI variability. The Landsat and MODIS NDVI data were resampled to match the spatial resolution of the AVHRR NDVI (0.05 × 0.05°). We employed the boxcar method for this process. Each raster value was a temporal-average of 17 months (month of June). The peaks in the NDVI plots from the different sensors appeared at approximately the same time through the 17 years of overlap ( Figure 3). Figure  3 shows that there is a good correspondence between the area-averaged NDVI derived from Landsat and AVHRR. In addition, the MODIS time series show a positive bias by about 0.02 relative to Landsat and AVHRR. There is a good spatial match after the resampling process ( Figure 4). The same features discussed for Figure 2 can be seen in Figure 4. The only difference between the three images for these two features is that both the features appear sharper in the Landsat and MODIS as compared to AVHRR. This clearly demonstrates that the observation at a higher spatial resolution is important, and that averaging up (from 30 m and 1 km to 5 km) does not degrade the features displayed in the aggregated data.
There is a good spatial match after the resampling process ( Figure 4). The same features discussed for Figure 2 can be seen in Figure 4. The only difference between the three images for these two features is that both the features appear sharper in the Landsat and MODIS as compared to AVHRR. This clearly demonstrates that the observation at a higher spatial resolution is important, and that averaging up (from 30 m and 1 km to 5 km) does not degrade the features displayed in the aggregated data.

Comparison of Linear NDVI Trend Components
Pearson's correlation analysis was used for each pixel to investigate the consistency between the NDVI dataset derived from the LTDR AVHRR, Terra MODIS, and Landsat 7 ETM+ products from 2002 to 2018 ( Figure 5). In particular, the Landsat NDVI-MODIS NDVI (Figure 5c) shows a higher correlation than the Landsat NDVI-AVHRR NDVI and MODIS NDVI-AVHRR NDVI ( Figure 5 a  and b). The correlation coefficients between Landsat NDVI-AVHRR NDVI range from −0.2 to 0.6. About 10% of the area shows no significant agreement for Landsat and AVHRR NDVI. We can clearly see the lack of the dark red (correlation > 0.8) in Figure 5a (as compared to Figure 5c), signifying lower and some with no significant correlations between Landsat NDVI and AVHRR NDVI. The spatial pattern of the correlation between MODIS NDVI and AVHRR NDVI (Figure 5b) is quite similar to the spatial pattern of the correlation between Landsat NDVI and AVHRR NDVI (Figure 5a). The coefficient of correlation ranges between −0.3 and 0.6 (similar to Figure 5a). The correlation between Landsat NDVI and MODIS NDVI ranges between 0.4 and 0.9 (Figure 5c). This is much higher than the range of correlation between Landsat NDVI and AVHRR NDVI (

Comparison of Linear NDVI Trend Components
Pearson's correlation analysis was used for each pixel to investigate the consistency between the NDVI dataset derived from the LTDR AVHRR, Terra MODIS, and Landsat 7 ETM+ products from 2002 to 2018 ( Figure 5). In particular, the Landsat NDVI-MODIS NDVI (Figure 5c) shows a higher correlation than the Landsat NDVI-AVHRR NDVI and MODIS NDVI-AVHRR NDVI (Figure 5a,b). The correlation coefficients between Landsat NDVI-AVHRR NDVI range from −0.2 to 0.6. About 10% of the area shows no significant agreement for Landsat and AVHRR NDVI. We can clearly see the lack of the dark red (correlation > 0.8) in Figure 5a (as compared to Figure 5c), signifying lower and some with no significant correlations between Landsat NDVI and AVHRR NDVI. The spatial pattern of the correlation between MODIS NDVI and AVHRR NDVI (Figure 5b) is quite similar to the spatial pattern of the correlation between Landsat NDVI and AVHRR NDVI (Figure 5a). The coefficient of correlation ranges between −0.3 and 0.6 (similar to Figure 5a). The correlation between Landsat NDVI and MODIS NDVI ranges between 0.4 and 0.9 (Figure 5c). This is much higher than the range of correlation between Landsat NDVI and AVHRR NDVI (Figure 5a) and MODIS NDVI and AVHRR NDVI (Figure 5b). discussed for Figure 2 can be seen in Figure 4. The only difference between the three images for these two features is that both the features appear sharper in the Landsat and MODIS as compared to AVHRR. This clearly demonstrates that the observation at a higher spatial resolution is important, and that averaging up (from 30 m and 1 km to 5 km) does not degrade the features displayed in the aggregated data.

Comparison of Linear NDVI Trend Components
Pearson's correlation analysis was used for each pixel to investigate the consistency between the NDVI dataset derived from the LTDR AVHRR, Terra MODIS, and Landsat 7 ETM+ products from 2002 to 2018 ( Figure 5). In particular, the Landsat NDVI-MODIS NDVI (Figure 5c) shows a higher correlation than the Landsat NDVI-AVHRR NDVI and MODIS NDVI-AVHRR NDVI ( Figure 5 a  and b). The correlation coefficients between Landsat NDVI-AVHRR NDVI range from −0.2 to 0.6. About 10% of the area shows no significant agreement for Landsat and AVHRR NDVI. We can clearly see the lack of the dark red (correlation > 0.8) in Figure 5a (as compared to Figure 5c), signifying lower and some with no significant correlations between Landsat NDVI and AVHRR NDVI. The spatial pattern of the correlation between MODIS NDVI and AVHRR NDVI (Figure 5b) is quite similar to the spatial pattern of the correlation between Landsat NDVI and AVHRR NDVI (Figure 5a). The coefficient of correlation ranges between −0.3 and 0.6 (similar to Figure 5a). The correlation between Landsat NDVI and MODIS NDVI ranges between 0.4 and 0.9 (Figure 5c). This is much higher than the range of correlation between Landsat NDVI and AVHRR NDVI (Figure 5a) and MODIS NDVI and AVHRR NDVI (Figure 5b). The most likely reason for the higher agreement between Landsat and MODIS is the closer spatial resolution of the two sensors (30 m and 1 km) as compared to MODIS and AVHRR (1 km and 5 km) or Landsat and AVHRR (30 m and 5 km). Additionally, the coefficient of determination (R 2 ) for the three different sensors over 17 years (2002-2018) shows a positive and high R 2 ( Figure 6). Figure 6a shows a positive linear correlation between Landsat NDVI and AVHRR NDVI (R 2 = 0.79), Figure 6b represents R 2 = 0.77 between MODIS NDVI and AVHRR NDVI, and Figure 6c shows R 2 = 0.96 between NDVI derived from Landsat 7 ETM+ and MOD13A2 V6. As expected from the correlation results displayed in Figure 5, the highest correlation is between Landsat NDVI and MODIS NDVI.
The most likely reason for the higher agreement between Landsat and MODIS is the closer spatial resolution of the two sensors (30 m and 1 km) as compared to MODIS and AVHRR (1 km and 5 km) or Landsat and AVHRR (30 m and 5 km). Additionally, the coefficient of determination (R²) for the three different sensors over 17 years (2002-2018) shows a positive and high R² ( Figure 6). Figure 6a shows a positive linear correlation between Landsat NDVI and AVHRR NDVI (R² = 0.79), Figure 6b represents R² = 0.77 between MODIS NDVI and AVHRR NDVI, and Figure 6c shows R² = 0.96 between NDVI derived from Landsat 7 ETM+ and MOD13A2 V6. As expected from the correlation results displayed in Figure 5, the highest correlation is between Landsat NDVI and MODIS NDVI. We calculated the spatial standard deviation (SD) for each NDVI pixel of datasets at the LTDR AVHRR resolution over 17 years (Figure 7). The SD of the time series from the three sensors shows similar spatial patterns. We observed the same area of high standard deviation in the northeast portion of the study area as well as in the southwest portion (visible in the light blue color). Figure 7 shows that the standard deviation is generally less than 0.15 for all three sensors. This means that all sensors showed similar spatial variation of standard deviation NDVI around the average standard deviation of NDVI. The NDVI time series used to statistically analyze was based on linear regression, to identify the areas that showed a change in vegetation index over time. The temporal trend of vegetation greenness (NDVI) was measured using the slope. We calculated the spatial standard deviation (SD) for each NDVI pixel of datasets at the LTDR AVHRR resolution over 17 years (Figure 7). The SD of the time series from the three sensors shows similar spatial patterns. We observed the same area of high standard deviation in the northeast portion of the study area as well as in the southwest portion (visible in the light blue color). Figure 7 shows that the standard deviation is generally less than 0.15 for all three sensors. This means that all sensors showed similar spatial variation of standard deviation NDVI around the average standard deviation of NDVI. The most likely reason for the higher agreement between Landsat and MODIS is the closer spatial resolution of the two sensors (30 m and 1 km) as compared to MODIS and AVHRR (1 km and 5 km) or Landsat and AVHRR (30 m and 5 km). Additionally, the coefficient of determination (R²) for the three different sensors over 17 years (2002-2018) shows a positive and high R² ( Figure 6). Figure 6a shows a positive linear correlation between Landsat NDVI and AVHRR NDVI (R² = 0.79), Figure 6b represents R² = 0.77 between MODIS NDVI and AVHRR NDVI, and Figure 6c shows R² = 0.96 between NDVI derived from Landsat 7 ETM+ and MOD13A2 V6. As expected from the correlation results displayed in Figure 5, the highest correlation is between Landsat NDVI and MODIS NDVI. We calculated the spatial standard deviation (SD) for each NDVI pixel of datasets at the LTDR AVHRR resolution over 17 years (Figure 7). The SD of the time series from the three sensors shows similar spatial patterns. We observed the same area of high standard deviation in the northeast portion of the study area as well as in the southwest portion (visible in the light blue color). Figure 7 shows that the standard deviation is generally less than 0.15 for all three sensors. This means that all sensors showed similar spatial variation of standard deviation NDVI around the average standard deviation of NDVI. The NDVI time series used to statistically analyze was based on linear regression, to identify the areas that showed a change in vegetation index over time. The temporal trend of vegetation greenness (NDVI) was measured using the slope. The NDVI time series used to statistically analyze was based on linear regression, to identify the areas that showed a change in vegetation index over time. The temporal trend of vegetation greenness (NDVI) was measured using the slope. Figure 8 shows a computed linear slope (for each pixel) for the period from 2002 to 2018. This illustrates the locations of increase (positive slope) or decrease (negative slope) in vegetation. The color green (dark high, light low) (in Figure 8 (NDVI) of the Mesopotamian marshes over 17 years (2002-2018) increased in 33.2% of the total are, i.e., one-third of the area has a positive slope for the trend of the vegetation index. It can be seen from Figure 8 that the slope is slightly negative over most of the area (−0.002 to −0.05), indicating an overall slight decrease in the vegetation index over time. The areas of positive (green) and negative (brown) slopes were located in the same areas in each image, even though Landsat, MODIS, and AVHRR have different spatial resolutions (30 m, 1 km, and 5 km). According to the linear slope analysis, the three sensors are consistent. The reason for the increase in vegetation index in the west of Al-Hammar marsh, the northeast of Al-Huwaiza marsh, and the southern and middle parts of the Central marshes is due to the difference in elevation, where rivers and rainfall water that feed the marshes remain for a longer period due to the lower elevation. The increase in vegetation coverage is concentrated in the lowest elevations, which ranges between 0 and 6 m (Figure 9). According to the linear slope analysis, the three sensors are consistent. The reason for the increase in vegetation index in the west of Al-Hammar marsh, the northeast of Al-Huwaiza marsh, and the southern and middle parts of the Central marshes is due to the difference in elevation, where rivers and rainfall water that feed the marshes remain for a longer period due to the lower elevation. The increase in vegetation coverage is concentrated in the lowest elevations, which ranges between 0 and 6 m (Figure 9). These low elevation areas agree well with the areas of positive slope in the NDVI. Table 2 and Figure 10 show the inverse relationship between the increase in the green biomass over the time series and elevations. The increase in vegetation coverage, which is more than 2000 km², is concentrated in the lowlands (0-6 m) until it reaches as low as 0.4 km² in the highlands (>17 m).  These low elevation areas agree well with the areas of positive slope in the NDVI. Table 2 and Figure 10 show the inverse relationship between the increase in the green biomass over the time series and elevations. The increase in vegetation coverage, which is more than 2000 km 2 , is concentrated in the lowlands (0-6 m) until it reaches as low as 0.4 km 2 in the highlands (>17 m).  However, we tested the linear slope for statistical significance at the level of 95% (p < 0.05) and 90% (p < 0.1) (Figure 11). Figure 11 shows that the three datasets have a good match, where the three data clearly show the major areas are statistically significant, with p < 0.05 and p < 0.1.

Discussion
The Mesopotamian marshes had been subjected to drying operations almost for two decades (the late 1980s to late 2003) due to anthropogenic activities, through the construction of a series of dams upstream of rivers in Turkey, Syria, Iran, and local canals and dams [2,5,63]. However, we tested the linear slope for statistical significance at the level of 95% (p < 0.05) and 90% (p < 0.1) (Figure 11). Figure 11 shows that the three datasets have a good match, where the three data clearly show the major areas are statistically significant, with p < 0.05 and p < 0.1. However, we tested the linear slope for statistical significance at the level of 95% (p < 0.05) and 90% (p < 0.1) (Figure 11). Figure 11 shows that the three datasets have a good match, where the three data clearly show the major areas are statistically significant, with p < 0.05 and p < 0.1.

Discussion
The Mesopotamian marshes had been subjected to drying operations almost for two decades (the late 1980s to late 2003) due to anthropogenic activities, through the construction of a series of dams upstream of rivers in Turkey, Syria, Iran, and local canals and dams [2,5,63].

Discussion
The Mesopotamian marshes had been subjected to drying operations almost for two decades (the late 1980s to late 2003) due to anthropogenic activities, through the construction of a series of dams upstream of rivers in Turkey, Syria, Iran, and local canals and dams [2,5,63]. In late 2003, after the fall of Saddam's government, the marshes experienced re-flooding. In this paper, we have focused on the period of rehabilitation of the marshes (2002-2018) to examine the increase in vegetation index through the use of statistical analysis of the NDVI from the three different sensors (Landsat, MODIS, and AVHRR). We examined NDVI derived from the three different datasets-AVHRR LTDR V5, Terra MODIS, and Landsat 7 ETM+-for the Mesopotamian Marshes. The statistical analysis revealed that, although there are differences in spatial and temporal resolution between the three NDVI datasets, the overall trends and spatial variability in NDVI were similar over the 17-year period (2002-2018). The NDVI time series exhibited similar lowest and highest values of NDVI in 2003 and 2006, respectively ( Figure 3). The coefficients of determination were positive between the three datasets. The linear correlations were 0.786, 0.774, and 0.959 between AVHR and Landsat, AVHRR and MODIS, and Landsat and MODIS, respectively. AVHRR has a much lower spatial resolution (5 km) as compared to MODIS (1 km) and Landsat (30 m). This is the reason for the low similarity. There have some previous studies that found similar results to ours. Gallo et al. [13], concluded that, despite the fact that distinctions exist in several factors which may impact the composite NDVI values, the 16-day composite NDVI values observed with the Terra and Aqua MODIS, and LTDR AVHRR, were comparable when examined over comparable time intervals and land cover. Bédard et al. [64], compared NDVI composites from NOAA AVHRR and Terra MODIS for pastureland in Alberta. They studied the statistical relationship between NDVI values derived from both sources and compared to pasture productivity field data. Correlation coefficients of 0.74 and 0.71 were obtained while correlating pasture productivity with NDVI composites from Terra MODIS and NOAA AVHRR, respectively. According to Brown et al. [12], the AVHRR NDVI dataset offers a bridge between the historical record and the newer satellite sensors, permitting an extension of the short records and assisting global change research that require the use of vegetation data. Tsuchiya et al. [65] studied the observations from three spectral bands (green, red, and near infrared) of the four sensors-Terra/ASTER, Terra/MODIS, NOAA/AVHRR, and Landsat/ETM+-and showed that the sensors were highly cross-correlated. We acknowledge that there are substantial differences between NDVI products, due to their spatial resolution and temporal repeat. However, we are interested in the differences in spatial patterns, as we have the high-resolution Landsat (30 m) at low temporal repeat (bi-weekly) and has been available for a long time period (1980s-present); the AVHRR has a low spatial resolution (5 km) but a good temporal repeat (daily), and has been available for a long time (1980s-present); and in between these two extremes is the MODIS, with moderate spatial resolution (1 km), good temporal repeat (daily), and a time interval of 20 years (2000-present). However, the standard deviations of the time series of NDVI that have been calculated from the three different sensors show a similar spatial pattern (Figure 7). The high values of standard deviation occur in the area that has a positive slope for the trend of vegetation index. The long-term trend of NDVI showed an increase in vegetation coverage in the western area of Al-Hammar marsh, the northeast of Al-Huwaiza marsh, and the south and middle parts of the Central marshes, in the three sensors. According to Guo et al. [17], the spatial distributions of the significantly positive trends of NDVI were similar between the two datasets (MODIS and LTDR AVHRR) for the Loess Plateau and Northeast Plain. The vegetation coverage of the marshes increased by 33.2% in the past 17 years (2002-2018). This increase in vegetation is due to the return of marsh dwellers, who demolished most of the canals and dams that had been built during Saddam's regime [1,2], which prevented water from reaching the marshes. Additionally, the increase in vegetation is linked to the difference in elevation. At low elevations (0-6 m), rivers and precipitation feed the marshes, which provides a favorable environment for plant growth.

Conclusion
In our recently published manuscript, Albarakat et al. [2], we calculated the statistical analysis (trend of NDVI) from 1982 when the Iraqi marshes were not affected by any anthropogenic activities. From 1982 to 1985, most of the Mesopotamian marshes were covered by vegetation and water bodies.
Between the late 1980s until 2003 the marshlands had been subjected to drying operations by the construction of the series of dams and canals that decreased the flow of the Tigris, Euphrates, and Karkha rivers through the marshlands. This led to the transformation of the marshes to barren land, and the vegetated areas significantly deteriorated compared to pre-drying years. During this research, we employed statistical analysis from 2002-the year which is widely considered to be the end of the drying operations. The Mesopotamian marshlands experienced re-flooding in late 2003, after the fall of Saddam's regime, due to the return of the marsh dwellers, who destroyed many of the canals and dams that had been previously built. In this study, we found the vegetation coverage of the marshes increased by 33.2% compared to 2002 (pre-re-flooding). However, as our analyses demonstrated, there was a similar performance of all three datasets examined here (AVHRR, MODIS, and Landsat). AVHRR can be considered an attractive source of information for studying long-term variations and trends in the land surface vegetation. In addition to AVHHR's long-term availability (almost four decades), AVHRR LTDR data are global and proven at a daily time repeat. The MODIS sensor temporal coverage is limited to less than 20 years (2000-present), and data are provided as 8-or 16-day composites. Landsat images are provided at 30 m spatial resolution and offer a long-term temporal record (30 years). However, the Landsat data are susceptible to cloud cover and only a few images are available per year (with a 20-day repeat visit). In this study, we found that the sensors are cross-correlated with high accuracy. This paper explores the utility of matching low spatial resolution with good temporal repeat and long-term coverage, and a high spatial resolution with low temporal repeat and similar long-term coverage. Finally, this study also proves that we can use the low-resolution AVHRR data for land cover change studies, as it shows a similar trend to the high-resolution Landsat over the same time period.