An evapotranspiration deficit-based drought index to detect variability of terrestrial carbon productivity in the Middle East

The primary driver of the land carbon sink is gross primary productivity (GPP), the gross absorption of carbon dioxide (CO2) by plant photosynthesis, which currently accounts for about one-quarter of anthropogenic CO2 emissions per year. This study aimed to detect the variability of carbon productivity using the standardized evapotranspiration deficit index (SEDI). Sixteen countries in the Middle East (ME) were selected to investigate drought. To this end, the yearly GPP dataset for the study area, spanning the 35 years (1982–2017) was used. Additionally, the Global Land Evaporation Amsterdam Model (GLEAM, version 3.3a), which estimates the various components of terrestrial evapotranspiration (annual actual and potential evaporation), was used for the same period. The main findings indicated that productivity in croplands and grasslands was more sensitive to the SEDI in Syria, Iraq, and Turkey by 34%, 30.5%, and 29.6% of cropland area respectively, and 25%, 31.5%, and 30.5% of grass land area. A significant positive correlation against the long-term data of the SEDI was recorded. Notably, the GPP recorded a decline of >60% during the 2008 extreme drought in the north of Iraq and the northeast of Syria, which concentrated within the agrarian ecosystem and reached a total vegetation deficit with 100% negative anomalies. The reductions of the annual GPP and anomalies from 2009 to 2012 might have resulted from the decrease in the annual SEDI at the peak 2008 extreme drought event. Ultimately, this led to a long delay in restoring the ecosystem in terms of its vegetation cover. Thus, the proposed study reported that the SEDI is more capable of capturing the GPP variability and closely linked to drought than commonly used indices. Therefore, understanding the response of ecosystem productivity to drought can facilitate the simulation of ecosystem changes under climate change projections.


Introduction
In the last few decades, humanity has faced many challenges in the environmental and agricultural sectors, such as the need for reductions in carbon dioxide emissions (Jardine 2003), climate change (Saavedra et al 2009, Dai 2011, drought and water scarcity (Mishra and Singh 2011), and food insecurity (Margulis 2013). The concept of drought as a temporal dynamic phenomenon differs from aridity as a spatial dynamic phenomenon, which results from a decrease in precipitation in a specific area and a remarkable increase in the potential evapotranspiration (PET) rate , Elbeltagi et al 2021. The time factor or the length of the drought period is considered one of the most important criteria for determining the severity of drought risk (Mokhtar et al 2021a). The issue of drought and its spatial consequences have often been the focus of studies carried out in the Middle East (ME) ecosystem (Kaniewski et al 2012, Hameed et al 2020. In recent decades, drought has turned into one of the most severe natural disasters for the ME's ecosystem in the context of global climate change (Lelieveld et al 2012, Mohammed et al 2020b. Moreover, drought has significantly affected terrestrial ecosystems, economic, social and political systems in the ME, including food security (Hameed et al 2020), hydrological processes (Bozkurt and Sen 2013), vegetation growth (Zaitchik et al 2007, Karakani et al 2021 and the extinction of many plant species (Belgacem and Louhaichi 2013). Generally, the ME's ecosystem is classified as a transitional climate pattern, located between a hot-dry and cold-humid climate pattern, which is most exposed to drought events as a result of prolonged water shortages. The ME is a crucial region for comprehension of drought globally (Barlow et al 2016).
As a result of the close relationship between ecosystem dynamics and available water, a water deficit can be restrictive to ecosystem growth (Yi et al 2010, Mokhtar et al 2021b. Thus, ecosystem conditions can reflect drought risk; for instance, both precipitation and temperature changes significantly affect the ecosystem (Mokhtar et al 2020, Mohammed et al 2020a. Precipitation plays a vital role in controlling the productivity of the grass; increasing precipitation results in increasing grass productivity while, in contrast, the grass lands are critically impacted by temperature , Lei et al 2015. Indeed, biogeochemical carbon cycles of the ecosystem reflect the atmosphere conditions and serve as an indicator of climate change (Piao et al 2005, Xu et al 2009. Drought is one of the main factors impacting ecosystems' distribution patterns and types (Vicente-Serrano et al 2012. Several studies have documented that ecological programs can reduce ecosystem degradation and enhance vegetation coverage, although these programs did not work well (Deng et al 2014, Wu et al 2014, Zhang et al 2015b. Further, afforestation of semiarid areas results in ecosystem degradation due to the impact of drought and human activity on ecosystem change (Cao 2008, Huang et al 2016a.
Consequently, vegetation conditions are positively and negatively correlated with precipitation in dry and wet regions, respectively (Prasad andStaggenborg 2008, Jiao et al 2019a). In previous studies, experimental methods, satellite observations, and carbon process models have documented that largescale droughts reduce the vegetation activity, and severe temperatures lead to negative terrestrial carbon productivity even under mild drought conditions , 2013, Stocker et al 2019. Droughts impact ecosystem productivity by restricting vegetation growth stages, including a wide range of tree mortality and ecosystem fire, and can influence the global CO 2 balance (Chen et al 2013, Reichstein et al 2013, Yu et al 2017, Jiao et al 2021. Understanding the response of terrestrial ecosystems to drought remains challenging because of the biochemical and physiological activities of vegetation growth and cultivated crops, especially their CO 2 assimilation and ecosystem carbon dioxide fluxes that are constrained by varying degrees of drought at various timescales (Sun et al 2021).
Several evapotranspiration-based indices were established to quantify drought and its impacts on terrestrial ecosystems, including Palmer drought severity index (Palmer 1965), Reconnaissance drought index (RDI) (Tsakiris and Vangelis 2005), the standardized precipitation evapotranspiration index (SPEI) (Vicente-Serrano et al 2010a) and the evaporative drought index (EDI) (Yao et al 2010), the drought severity index (Mu et al 2013). Also, there are dozens of drought indices, such as the standardized precipitation index (McKee et al 1993), vegetation growth anomaly-based drought indices e.g. vegetation condition index (Liu and Kogan 1996) and vegetation health indices (Kogan 2002), soil moisture deficit index (Narasimhan and Srinivasan 2005), scaled drought condition index (Rhee et al 2010), the process-based accumulated drought index (Zhang et al 2017a), integrated drought index (IDI) (Jiao et al 2019b), and microwave IDI (Zhang et al 2019a).
In light of the findings of previous studies, the AET and PET-based drought indices significantly highlight the intensity of water deficits and the influence on vegetation activities more than the individually based indices (Zhang et al 2019c). As such, Kim and Rhee (2016) suggested a standardized evapotranspiration deficit index (SEDI), and concluded that the SEDI helps detect agricultural drought events with strong land-atmosphere interactions. Zhang et al (2019c) applied the SEDI to investigate the impact of water stress on vegetation growth under global warming and documented the strong interaction between vegetation and the evapotranspiration deficit (ED). Consequently, the SEDI is based on the evaporative stress perspective, which is more directly linked to vegetation water stress than other drought indices (Zhang et al 2019c). The PET reflects the atmospheric potential to receive water or evaporative demand. The PET regulates soil water stress conditions, while the actual evapotranspiration (AET) is the amount of water lost from an ecosystem induced by evaporation and transpiration. From eco-physiological and agricultural perspectives, the ED is the difference between the AET and the PET (Vicente-Serrano et al 2018). Under climatic stress, a high ED results in stomatal closure, which decreases the photosynthetic process, carbohydrate accumulation, and terrestrial ecosystem production (NPP and GPP) (Stephenson 1998, Eamus 2003. Based on the above, we used a correlation analysis between the standardized GPP residual series and SEDI to investigate the drought variability and ecosystem response. This research is focused mainly on 35 years for both GPP and SEDI over the ME. Consequently, this research aims to improve the understanding of the correlation between evaporation deficit-based drought and ecosystem vegetation and detect the response of national-level GPP to drought in different land-cover types. As such, the principal aims of this research are: (a) quantify the spatial-temporal variability of GPP dynamics in response to drought at the pixel scale and for each land cover type separately, (b) analysis of ecosystem resilience to drought events of several types, and assessment of an ecosystem's capability to tolerate drought event disturbances, and (c) prove that the ED-based index is more capable of capturing the standardized GPP residuals (sGPPR) variability and closely linked to drought than commonly used indices based on variables other than ED.

Study area
The ME is located in the western part of Asia and northeast of Africa, between 12 • 06´N and 42 • 07´N latitudes, and 24 • 41´E and 63 • 17´E longitudes (figure 1). The ME officially includes 16 countries: the Arabian Peninsula region includes Bahrain, Kuwait, the Sultanate of Oman, Qatar, Saudi Arabia, the United Arab Emirate, and Yemen, and the Levant includes Jordan, Lebanon, Palestine, and Syria. Included as well are Egypt, Iran, Iraq, Israel, and Turkey. The ME covers an area of 6928 thousand km 2 , inhabited by about 357.23 million people according to World Bank's 2019 statistics (World Bank 2020) and according to the FAO Global Land Cover SHARE database (FAO 2014). The ME's land cover type is mostly bare soil lands, at 61%. However, the whole area of Turkey, northeastern Iraq, western Iran, western and northwestern Syria, most parts of Lebanon, and around the Nile River in Egypt are dominated by grasslands, shrublands, and croplands. In contrast, forests cover northern and southern Turkey and the north and western parts of Iran (see figure 2).

Ecosystem gross primary productivity (GPP) data
To quantify the sensitivity of GPP against SEDI, we evaluated four state-of-the-art GPP datasets based on Earth observation data as following: (a) The yearly GLASS-GPP (the Global Land Surface Satellite) dataset, specifically the latest version

AET and PET
To quantify the SEDI, the Global Land Evaporation Amsterdam Model (GLEAM, version 3.3a) dataset, spanning 35 years , was used (Miralles et al 2011, Martens et al 2017. The GLEAM dataset gives estimates based on remotely sensed observations to set algorithms that predict the various components of terrestrial evapotranspiration separately. They include transpiration, open water evaporation, bare soil evaporation, sublimation, and interception loss. We used annual AET and PET estimates (mm yr −1 ) to calculate the SEDI. The GLEAM datasets are currently available within the GLEAM portal at daily, monthly and annual temporal resolution, and at 0.25 • arc degree spatial resolution for 1980-2018 (NETCDF files) (www.gleam.eu).

The SEDI calculation
For all gridded points of more than 28 000 pixels for 1982-2017 at the annual scale, the SEDI was calculated (Alsafadi and Bi 2021). In this study, the SEDI was implemented as the standardized difference between AET and PET. This was similar to Zhang et al (2019c), as shown: where ED is the AET and PET difference (mm yr −1 ) and ED std and ED avg denote the standard deviation and multi-years mean, respectively. This can highlight dry and wet conditions by tracking local water storage changes within the soil, compared with direct evaporation.

Standardized GPP residual series (sGPPR)
Normally, the contributions of human activities to variations in vegetation are removed to detect climate elements impacts independently, calculated by the residuals of the vegetation trend models by computing the de-trended analysis. Since the GPP series is affected by many variables besides climate factors, often, the annual GPP series has a positive trend specifically over agricultural systems and forests. The sGPPR series was obtained from the mean µ and standard deviation value σ of deference between the GPP value and its de-trended value for each year separately, from 1982 to 2017 at the pixel scale (Alsafadi and Bi 2021). The sGPPRS was calculated as: where (y 0 i ) is the observed GPP and (y (τ ) i ) is the value of the de-trended GPP in a separate year. The GPP, de-trended during the period of 1982-2018, was calculated for each pixel (more than 23 000 series) using a simple linear regression (SLR) analysis, by using the ordinary least square method, OLS, which was calculated as: Herein, we assumed the temporal evolution (x i ) from 1982 to 2017 is an independent variable, and the GPP series data a dependent variable (y 0 ), to fit the SLR or the value of the de-trended GPP in a separate year where n is the length of the studied period, while β is the slope ratio or annual change of GPP (g m 2 yr −1 ) acquired by the ordinary least square method. β < 0 indicates that the GPPs tended to decrease over the studied years, and vice versa. The sGPPR dataset is openly available in Zenodo repository (Alsafadi and Bi 2021) Pearson's correlation coefficient (r) was used to assess the temporal consistency of the observed GPP and SEDI across the time series.

Assessment of SEDI impacts on GLASS-based sGPPR across terrestrial biomass
In order to quantify the sensitivity of GPP to the SEDI in different land-cover types, the yearly GLASS-GPP at 0.05 • arc degree spatial resolution was used, since it has sufficient spatial resolution unlike than that of the other products.
Based on the ecosystem map analysis, we further evaluated the impact of the SEDI on ecosystem productivity from 1982 to 2017. The correlation between the GLASS-based sGPPR and annual SEDI for the ME (figure 2) and each ecosystem type is shown in figure 3. The strongest significant positive correlation was presented in the central portion of the study area, especially in the north of Iraq and central area of Turkey, as well as some portions of northern Iran which reached more than 0.5 (p < 0.05). Based on table B1, the significant positive correlation areas of crops, grass, tree covered area and shrubs area in Iraq covered 29.6%, 30.6%, 32.4%, and 28.3% of the total area for each cover type, respectively.
The highest significant positive correlation was recorded in the crops ecosystem at 34% of its total area in Syria. The spatial correlation between the GLASS-sGPPR and annual SEDI for each ecosystem type is shown in figure 3. The croplands and grasslands' productivity had the highest positive correlation with the SEDI for several of the study area's countries, especially Iraq, Israel, Syria, and Turkey. Hence the cropland and grassland were more sensitive to SEDI variability.
In contrast, a significantly negative correlation between the sGPPR and annual SEDI was observed over Israel and Jordan for bare land, and the sGPPR of crops vegetation and the annual SEDI was negatively correlated over Egypt and Yemen; 27.2% of this area in Egypt had a significant negative correlation (figure 2). Table B1 and figure 3 demonstrate that the tree-covered area had a significantly positive correlation for the studied countries, especially in Iraq and Turkey at 32.4% and 21.3% of the total area. Moreover, Turkey, Iraq, and Lebanon achieved the highest percentage of shrubs covered area, with a significantly positive correlation of 31.8%, 28.3%, and 23.8%, respectively. In contrast, the KSA recorded the highest values of significantly negative correlations for the same ecosystem type of 15% (table B1  and figure 3). For sGPPR of the sparse vegetation land in Palestine, Iraq, Turkey, Syria and Iran generated significantly positive correlations with the SEDI of 32.5%, 20%, 18.8%, 16.7%, and 16.1%, respectively. In contrast, the KSA recorded a significantly negative correlation of 30.9% compared to other countries. The aquatic vegetation versus the SEDI produced a significantly positive correlation of 29.4% and 10.4% in Turkey and Iran, respectively, while Iraq reached 5.4% for a highly negative correlation. The sGPPR in the soil bare ecosystem was highly positively correlated with the SEDI in Lebanon and in the UAE, at 22.5% and 20.1%. On the other hand, 7.1% and 7% of the soil bare was highly negatively correlated in Egypt and Jordan (table B1).

Performance of SEDI in detecting response of the sGPPR dataset to dry and wet climatic condition
This section aimed to prove that the ED-based index is more capable to capture the sGPPR variability and closely linked to drought than commonly used indices based on variables other than ED. Herein, we assessed the performance of SEDI against SPEI in detecting drought effects on terrestrial carbon productivity, using four state-of-the-art GPP dataset. At the regional scale, we assessed the sensitivity of sGPPR data to the evolution of drought for each index separately. Figure 4 indicates the spatial distribution of correlation values between the four sGPPR datasets and the SEDI from 1982 to 2016. The correlation values between sGPPR and the SEDI in the study area were 0.5, 0.41, 0.3, and 0.2 for the FluxCom, VPM, GIMMS and GLASS models, respectively. In contrast, the correlations between the SPEI and sGPPR were lower with the values of 0.35, 0.23, 0.15, 0.16 for the same models respectively. The spatial distribution patterns of correlations between the SEDI and sGPPR were similar between the SPEI and sGPPR, but higher in the first pattern, which indicates that SEDI can detect stronger response from terrestrial processes, specifically the northern part. The strongest significant positive correlations were presented in the central portion of Turkey, north Iraq, and the western part of Iran and some portions of northern Iran, which reached more than 0.7 (p < 0.05) for the FluxCom and VPM models against the SEDI.

Extreme drought-induced GPP anomaly
As presented in figure A1, the ME's terrestrial ecosystems have experienced drought events, as presented  Crescent in Iraq and Syria, by 20%-60%, and higher than that in some parts. The GPP accounted for a high decline of >60% during the 2008 extreme drought (peak of drought) in the north of Iraq, the northeast of Syria, and southwest of Iran (figure 5).
The strongest reductions in annual GPP during 2008 extreme drought were presented for the GLASS and VPM models for the study area. They constituded 20% and more than 60% in some parts, while the GIMMS and FLUXCOM models-based GPP negative anomalies during the drought stress were lower than that of other GPP data and had low spatial variability. The anomalies in annual GPP from 2009 to 2012 might have resulted from a notable decrease in the annual SEDI at the 2008 extreme drought event (figures A1 and A2). After that, the moderate and slight drought from 2009 to 2012 resulted in persistent stress to vegetation, as observed via negative anomaly values of the GPP. Ultimately, this led to a long delay in restoring the ecosystem in terms of its vegetation cover.

Sensitivity of GPP data to the evolution of SEDI
The results of this study have shown that spatialtemporal variability of SEDI significantly impacted the GPP models at the pixel scale in the ME. Moreover the SEDI is more capable of capturing the GPP variability than the SPEI. Regarding the SEDI structure, the PET regulates the soil water stress conditions, while the AET is the amount of water lost from an ecosystem induced by both evaporation and transpiration. This is considered an indicator for the physiological activities of vegetation. Regarding agronomic and eco-physiological terrestrial systems, the ED can explicitly account not only for the atmospheric evaporative demand, and also the water transferred to the atmosphere from the soil and vegetation which physiologically explains the vegetation behavior and activity (Kim and Rhee 2016, Vicente-Serrano et al 2018, Zhang et al 2019c. The soil moisture deficits could result in a stomatal closure to avoid additional water deprivation. If the deficits become high enough to reduce soil moisture below the wilting point, plants will be under stress and may die as a result of vascular damage (Anderegg et al 2015). Therefore, the SEDI helps detect agricultural drought events in areas that have a strong land-atmosphere interaction and has a proven high performance in vegetation-drought interactions (Zhang et al 2019c).

Responses of vegetation activity to SEDI
The GPP for the ecosystem in the ME at annual scale showed a moderate correlation with the SEDI during the studied period. However, cropland and grassland were more sensitive to the droughts than other vegetation ecosystem types, these results consistent with previous studies (Pei et al 2013, Xiaobin et al 2014, Sun et al 2016. The maximum correlation for grass and croplands' GPP against the SEDI was in Iraq, Iran and Syria. In a similar context, Huang et al (2016b) found that the semi-arid ecosystem types have an essential role in inter-annual NPP variability during long-term droughts. The highest positive correlations between yearly NPP and SPEI were recorded in shrubcovered land, followed by cropland. The results identified that a slight correlation with the SEDI during the studied period was detected in the tree-covered area throughout the northern and northeastern of the study area. It has been suggested that forests productivity may be more resilient to drought events than other ecosystem types due to deeper forest root system compared with grasslands and can access water from a deeper soil profile (Teuling et al 2010, van den Hoof andLambert 2016).
On the other hand, the results indicated that ED driven-drought events positively impacted the GPP of irrigated cropland and herbaceous vegetation aquatic over a large area on both sides of the Nile river and Egyptian Delta as presented in figure 6. Under several management practices of irrigation, it is demonstrated that the variation caused by the role of vegetation respiration in the terrestrial carbon sink. Moreover the higher increment in GPP may be coupled with the higher rate of respiration to GPP during drought conditions, heat waves, Figure 6. The regional mean of sGPPR and SEDI for different regions (semi-arid ecosystem like north Iraq and irrigated croplands on both sides of the Nile river and Egyptian Delta). R 2 is the determination coefficient. and increament of temperature in the atmosphere all these increases the rate of respiration to GPP (Schwalm et al 2010, Williams et al 2013. Hence, a slight water deficit under irrigation practices will not have an effective role in vegetation productivity losses compared to a high ratio of vegetation respiration. As presented previously, the vegetation productivity in semi-arid lands was more negatively affected by drought than arid and desert land (Zhu et al 2021). Example is the Western Desert in Egypt, central part of Iran and the Empty Quarter desert in the KSAs.

Extreme drought-induced GPP anomaly in the ME
Overall, extreme drought is a critical driver of annual and inter-annual variability in continental and regional terrestrial gross and net primary productivity (Ciais et al 2005, Schwalm et al 2012. This research indicated that the ME drought spells and heat waves in 1989 and 2008 over northern and northeastern reduced the carbon cycle, with strong anomalies at regional scale (0.3 and 0.12 PgC yr −1 ), i.e. a reduction by 20% and 9% of GPP respectively. Notably, the 1989 and 2008 GPP anomalies have been coupled with high evaporation deficits, as presented by the SEDI, and were somewhat correlated with the annual SEDI in the same years, resulting in a clear reduction in the annual GPP specifically in the Fertile Crescent in Iraq and Syria by 40%-60% in 2008. The GPP accounted for a high decline of >60% during the 2008 extreme drought (peak of drought) in the north of Iraq, the northeast of Syria, and the southwest of Iran, as presented in the GLASS and VPM models (figure 6). In general, the GPP is highly sensitive to drought conditions than ecosystem respiration, whereas it is less sensitive than NPP (Schwalm et al 2010). For instance, over the Southern United States, the drought intensification is resulted in a significant decrease in NPP, with the highest decrease of 40% occurring during drought of -2004(Chen et al 2012. The incompatibility between the response of GPP and NPP is due to the role of vegetation respiration in the terrestrial carbon sink. Vegetation respiration is less impacted by drought spells than photosynthesis (Schwalm et al 2010), and the higher reduction in NPP may be coupled with the higher rate of respiration to GPP during drought conditions, the land and atmosphere increase in temperature increases the rate of respiration to GPP (Green et al 2019).

Uncertainties in analysis and finding
In our study, comparisons between GPP models and SEDI suggested that all of the GPP datasets significantly responded to the dry-wet climatic condition of the SEDI at an annual time-scale. Even though the patterns of the correlations between SEDI and various GPP dataset were similar, but the proportions of responses were extremely varied among the GPP data, could be caused by uncertainties in the GPP data rather than in the SEDI data. It is still hard to simulate GPP at global and regional levels. This limitation in GPP data may cause uncertain interpretations when evaluating the impacts of drought stress on GPP using a few or different GPP products (Liu et al 2019). Many remotely sensed GPP datasets are presently available and have been extensively utilized at different spatial-temporal scales (Zheng et al 2020). However, either GPP estimates from remote sensing, processbased models, or machine learning methods should be carefully used when assessing GPP response to drought.
The correlations between FLUXCOM-GPP and the SEDI were higher than that of other GPP data in the ME. At the same time, the FLUXCOM modelbased GPP negative anomalies during the drought stress were lower than that of other GPP data and had low spatial variability. The flux tower observationbased upscaling data (e.g. the FLUXCOM GPP based on machine learning) can be regarded as observationbased GPP estimates and are often used to estimate GPP from remotely sensed data and processbased models. While, it suffers from an insufficient representation of several mechanism processes such as nitrogen deposition and CO 2 fertilization, and may imperfectly capture GPP inter-annual variability (Jung et al 2019). On the other hand, Schewe et al (2019) indicated that current terrestrial ecosystem models underestimate the effects of droughts on GPP due to the insufficient fitting of both human management and natural processes in the algorithms. Stocker et al (2019) found that satellite-based GPP estimates underestimate the effects of drought on GPP due to a lack of consideration of soil moisture impact on light use efficiency. As presented in our study, some investigations demonstrated that the VPM (Xiao et al 2004) and modified VPM (Zhang et al 2015a) generally outperform the other remotely sensed GPP estimates in capturing the effect of drought on GPP, due to including the impact of soil moisture on photosynthesis in the models (Zhang et al 2019, Pei et al 2020. As such, the combination of GPP dataset from diverse and independent models can facilitate more reliable conclusions when evaluating the responses of GPP to drought at global and regional levels (Wu et al 2018, Sun et al 2021. Additionally, other mechanism processes such as CO2 fertilization and nutrients (i.e. phosphorus and nitrogen) on photosynthesis in the models should also be included (Du et al 2020).

Concluding remarks
Based on a remote sensing-driven GPP dataset, flux tower observation-based upscaling GPP, and components of terrestrial evapotranspiration (AET and PET) driven by GLEAM dataset estimates, we comprehensively examined the impacts of the SEDI on long term GPP variability over the ME for the period of 1982-2017. Focusing on dynamic changes of ecosystems' GPP, coupling analysis between the sGPPR and SEDI effect, and spatial heterogeneity. An ecosystem's ability to tolerate droughts was considered, along with whether various ecosystem types have varying responses. The main findings are as follows: (a) The ecosystem productivity is sensitive to drought in semi-arid ecosystems, and the GPP's croplands and grasslands recorded the highest positive significant correlations and were more sensitive to the SEDI variability. Moreover, the ecosystem's GPP recorded a high decline during the 2008 extreme drought in the north of Iraq and the northeast of Syria. (b) The proposed study reported that the ED-based index is more capable to capture the GPP variability and closely linked to drought than commonly used indices based on variables other than ED (e.g. SPEI). (c) The VPM and FLUXCOM-based GPP correlated against the SEDI more closely than that of other GPP data in the ME. At the same time, the VPM and GLASS models-based GPP negative anomalies during the drought stress were higher than that of other GPP data.
However, there are some limitations to this study. First, the SEDI was implemented as the standardized difference between AET and PET at annual scale. More attention is required on the impacts of the SEDI and drought intensity and duration at the monthly and seasonal scale to explain this variance clearly. Secondly, vegetation productivity and ecosystem GPP could be indirectly and directly impacted by drought events. Besides drought-related solar radiation and precipitation, temperature, and wind speed factors, also other artificial variables, such as land use type could caused critical changes in vegetation GPP. This study did not investigate the effect of several crop types due to the lack of detailed information about the distribution of crop types spatially, which may drive to inconsistencies and uncertainty in the results. Thirdly, it is still difficult to precisely reproduce GPP. This limitation in GPP data may cause uncertain results when assessing the drought impacts on GPP via a few or divergent GPP parameters. The uncertainties in GPP by the selected models mainly produce from forcing parameters, structures of the models which simulating the stress of drought on photosynthesis and parameterization.
Further improvement on the ability of the models to estimate GPP under numerous conditions is required. This will improve the confidence in the assessment of drought hazards and their impacts on terrestrial carbon productivity. More attention is required on the impacts of drought intensity and duration at the monthly and seasonal scales to explain this variance clearly. Therefore, further studies should analyse the effects of droughts, besides anthropogenic activities and fire events, on terrestrial ecosystems. Those studies can consider ED-based drought indices to understand the potential dynamics affecting the spatial-temporal changes of the GPP in the ME. Moreover, they can consider the importance of detecting the direct concurrent and lagged impacts of droughts.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: http://doi.org/10.5281/zenodo.4540832. Figure A2. Spatial-temporal evolution of the annual GLASS-based sGPPR series over the ME between 1982 and 2017 at 0.05 • arc degree spatial resolution. Table B1. Percentage of area (%) from each land cover separately which has a significance correlation between the GLASS-based sGPPR and annual SEDI series from 1982 to 2017 at p < 0.05 (i.e. 95% confidence interval).