We propose a revised map of plant geographical regions for the Southern Levant (most of Israel, the Palestinian Territories and Jordan). Since this region has been heavily impacted by human activity over millennia we have created a detailed natural vegetation map using GIS, remote sensing and a species distribution model (MAXENT). We created an extensive database of over 1800 historical and field observations for the region, incorporating plant species presence-only data, as well as environmental variables, including temperature and precipitation. Vegetation was modeled according to plant geographical regions and vegetation types. Our revised map generally agrees with previous vegetation maps done for the region, but provides more detail regarding the distribution of various forested communities that have been significantly disturbed by millennia of human activity. Given its digital format, this new map can be used for conservation and other applications that require a vegetation surface. It can also be used as a basis with which to model vegetated environments in past and future scenarios. Mariela Soto-Berelov is a post doctoral research fellow in remote sensing at RMIT, Australia. She specializes in land-use change science (LUCC), vegetation mapping/ modelling, characterization of forest ecosystems to validate biophysical products derived from coarse resolution satellite remote sensing, and the use of geographic information science and remote sensing to study landscape change. She also investigates ways through which to model past vegetation and climate in the eastern Mediterranean. Patricia L. Fall is a Charles La Trobe Professorial Fellow of Geography at the School of Historical and European Studies in La Trobe University, Victoria, Australia. She studies past environments in a variety of geographic settings, including tropical rain forests on Pacific and Caribbean islands and the semi-arid and arid environments of the eastern Mediterranean basin. She specializes in the analysis of pollen and plant macrofossils from small lakes and bogs to reconstruct past environments, as well as carbonized seeds and wood from archaeological deposits. She also models modern and past vegetation and climate in the eastern Mediterranean and uses remote sensing technologies to study environmental change. Steven Falconer is Professor and Chair of the Archaeology Program at the School of Historical and European Studies in La Trobe University, Victoria, Australia. He studies the rise and collapse of urbanized societies in the Eastern Mediterranean and Near East, turning particular attention to the interactions of small agrarian villages with their larger social, political and natural environments.
This study presents a revised map of plant geographical regions for the Southern Levant, a region where much of the vegetation has been altered over thousands of years. Our approach to mapping the natural vegetation using a maximum entropy model (MAXENT) is important for five reasons: (1) it utilizes presence-only vegetation observations to reconstruct the natural vegetation landscape of the region where little remains of the natural vegetation; (2) it provides a standardized vegetation map of the region where previous maps had focused on either the vegetation of Jordan or the vegetation of Israel and the Palestinian Territories; (3) it produces a digital vegetation map that can be used in applications that require digital vegetation surfaces (e.g., management applications); (5) it provides a basis for modelling more detailed vegetation maps (e.g., different forest types, different vegetation associations found throughout the desert region) and (5) it provides the foundation for comparing modern vegetation communities to future and past vegetation maps.
The authors are specifically interested in using this modern map as a training dataset to create paleo vegetation models and recreate the vegetation of this region throughout the Holocene (the geologic epoch that includes the last 12,000 years ago and in the Levant, a period that experienced dramatic climatic fluctuations as well as the rise and collapse of some of the earliest agrarian and urbanized societies) as well as project into future climatic scenarios. The trajectory of past vegetation in this region is important given the millennial history of complex human habitation in a setting that at present bears almost no resemblance to a supposedly heavily forested area during the early Holocene. Given the overwhelming reduction of forest areas (due to both climate change and human pressure), it is also important to model the vegetation under different climatic scenarios projected in the future, which can help to manage forest resources that are currently under pressure from grazing and other human activities (Al-Eisawi 2012) . MAXENT was chosen because it is the algorithm that best suits this objective of creating a present day model that can be trained to project into other geographic/temporal regions.
The study area lies at the junction of three continents (Asia, Europe and Africa) and four major plant geographical regions. As a
result of its geographic position, the vegetation is extremely rich
The study area encompasses about 42,650 km2 between latitude 29°30’ N to 33°N and longitude 34°17’ E to 36°15’E (Figure 1). The region incorporates high topographic and climatic diversity. It extends from the Mediterranean Sea in the west across coastal plains and then ascends to the Central Hills (up to 1200 m). The study area then descends into the Rift Valley which extends from the Sea of Galilee in the north to the lowest place on Earth, the Dead Sea (- 410 m), to the Wadi Araba and then the Red Sea in the south. East of the Rift Valley lies the highest elevation landscape on the Jordanian Plateau (up to 1650 m), which is deeply incised by wadi valleys flowing west towards the rift. Beyond the plateau the landscape gives way to the Eastern Desert. The region’s climate is Mediterranean, with long, dry and hot summers and wet, cool winters. Rainfall is most prevalent in winter and spring (November to May), although short and intense rain can occur during summer months, mainly affecting the more arid southern portion of the study region that receives less than 100mm/yr. The amount of precipitation in the study area decreases from west to east and from north to south, away from the Mediterranean Sea and south of the cyclonic storms that bring precipitation from the west. A second precipitation gradient is caused by topography, which contributes to the lowest amounts of precipitation being found in the Rift Valley and in the Eastern Desert. Temperature variation across the study area relates mainly to elevation and latitude. The coolest temperatures are found along the southern highlands on the Jordanian Plateau. Snow accumulation and freezing temperatures also can occur in the higher elevations. The warmest temperatures are found in the Wadi Araba, where they can reach 50°C (Al-Eisawi 1996).
The study area includes four of the major vegetation zones found in the Middle East. These are the Mediterranean, IranoTuranian, Saharo-Arabian and Sudano-Decanian, following Eig (1931/32, 1938) and Zohary (1962, 1968). Mediterranean vegetation grows in areas that experience the highest amounts of rainfall. Forests of pine (Pinus halepensis) and deciduous and evergreen oak (Quercus ithaburensis and Q. calliprinos) grow in the wettest areas near the Mediterranean and at higher elevations, particularly surrounding the Sea of Galilee and in the northwestern coastal region. More open woodlands are found in areas that receive less moisture both west of the Rift Valley where carob and pistacia (Ceratonia siliqua and Pistacia lentiscus) are found surrounding and penetrating the Central Hills, and east of the rift on the Southern Highlands of the Jordanian Plateau, which supports open woodlands of juniper (Juniperus phoenica) and evergreen oak (Quercus calliprinos). The Irano-Turanian region is distinguished by steppe vegetation that consists of grasses and shrubs. Artemisia herba-alba is the most common shrub species; other taxa include Noaea mucronata, Retama raetam, Salsola vermiculata, and Anabasis syriaca. Scattered Pistacia atlantica and Juniperus phoenicia trees can also be found with an understory typical of steppe vegetation. This region, which sits along a transition zone between the more humid Mediterranean and the more arid Saharo-Arabian regions, receives 150 to 350 mm/yr of rainfall. Temperatures in the Irano-Turanian region are more extreme, with hotter summers and colder winters than are found in the Mediterranean region. Mean annual maximum and minimum temperatures range from 1225°C and 5-20°C (Al-Eisawi et al. 2000).
The Saharo-Arabian region, where precipitation rarely exceeds 200 mm/yr, is characterized by drought tolerate plant species. It is
found along the Rift Valley, in the Negev Desert and in the Eastern Desert on the plateau. This region experiences the most
extreme annual temperature fluctuations with mean annual minimum and maximum temperatures in the Rift Valley ranging
between 10-20 and 20-35°C, respectively (Al-Eisawi et al. 2000). Summers are extremely dry, although seasonal storms from
Africa may bring sudden storms with high amounts of runoff. The driest area occurs in the Wadi Araba in the southern part of the
Rift Valley, where annual precipitation is less than 50 mm/yr, and in the Eastern Desert, which receives around 100 mm/yr. The
vegetation typically found in this region is manifested in diffused (e.g., spread out on slopes or depressions) or contracted (for
example along wadis) patterns, and is ultimately dependent on water availability. Common plants found throughout this region
include Zygophyllum dumosum, Haloxylon articulatum, Anabasis articulata, Anabasis syriaca, Astragalus spinosus, Suaeda
palaestina, Salsola tetrandra, S. asphaltica, and Achillea fragrantissima. Riparian species like Tamarix sp., Phragmites sp., Salix
sp., and Nerium oleander are also found along wadis. For instance, a riparian forest of Populus euphratica Tamaricetum jordanis
grows along the banks of the Jordan River
Sudano-Decanian vegetation is restricted to pockets along the Rift Valley from the Gulf of Aqaba north along the Jordan Valley to Deir ‘Alla, which are the hottest locations in the region. The most common species include Acacia spp., Ziziphus spina-christi, Balanites aegyptica, Moringa aptera, Ocradenus baccatus, Salvadora persica and Calotropis procera, and represent the northern extent of tropical African vegetation.
In this study, plant geographical regions were mapped using species distribution modelling or SDM. This allows the distributions
of species to be mapped across geographic space using a set of occurrences and environmental or predictor variables that
characterize them (Franklin 1995). As a result of innovative and more accessible statistical and GIS tools, as well as more
available data from museums, herbarium collections and online databases (for example, GBIF http://data.gbif.org/), SDM has
been growing in popularity during recent years in fields like conservation science, planning, evolution biogeography, and ecology
To model plant geographical regions, indicator species for each region were sourced from the literature
Except for the Tropical Sudanian Zone, which includes plants that are mainly influenced by edaphic factors, most indicator plants
are related to their vegetation zone based on climatic and topographic factors. On this basis we selected the following
environmental variables at a 1 km spatial resolution: geology, elevation, a series of climatic variables described below, rivers, and
springs. For similar applications of these variables see Austin et al. 1994;
Elevation, slope and aspect were obtained from high resolution digital elevation models (DEM) derived from 2002 Terra ASTER satellite imagery (http://asterweb.jpl.nasa.gov), which includes a 15m resolution band (Band 3) taken with forward and backward cameras that can be processed into DEM. The DEM were resampled from a 90-m spatial resolution to 1km. Regional geology was digitized from a 1:200,000 geological map of Israel (Sneh et al. 1998) and a 1:750,000 geological photomap of Israel and adjacent areas (Bartov 1994). The climatic variables were derived from a Macrophysical Climate Model or MCM (Bryson & DeWall 2007) that is being used to investigate long term social, environmental and ecological interactions in the Mediterranean Basin during the Holocene by reconstructing and modeling past environments (e.g., Barton et al. 2004, Barton et al. 2010). The variables were assessed for spatial autocorrelation using the MAKESTACK command in ArcGIS 9.3. Variables that were highly correlated (>0.85) were removed (for instance, average summer temperature and average winter temperature proved to be highly correlated with the temperature of the warmest and coldest month, and were therefore not utilized; the latter two were used because they represent extremes and can better capture the species that fall in this range).
Supplemental literature indicates that the vegetation found within the Tropical Sudanian region is mainly restricted to areas in the
Rift Valley that experience extremely hot temperatures and water (given the presence of rivers and springs). A shapefile of spring
locations was created by querying ‘springs’ from the Geographic Names Server (GEOnet Names Server) for Israel, Jordan and the
West Bank. These were then buffered and a grid file was created showing the area surrounding these springs at several distances
(2, 4, 6, 8, greater than 8km). The rivers were also buffered at 1, 2, 3, and 5 km (see below for a description of how this layer was
created). Since this vegetation is only found in the Araba Valley and Lower Jordan Valley, a mask was used outside these areas.
MAXENT was chosen over other modeling tools because it performs as well as or better than other methods for modeling species
distributions
Maxent version 3.3.3e was calibrated using the settings recommended by
e n o Z n a e n a r r e t i d e M
n on ian Ira rau
T lca ian ip an roT udS
rao ian Random 75% used ahS rabA for testing.
for training, 25% used
Resampling, 10-fold crossvalidation # Training samples # Test samples 351 294 225 43 234 98 74 5
No Yes Yes 490
geol3(categorical) mask Regularization values: linear/quadratic/product: 0.050, categorical: 0.250, threshold: 1.000, hinge: 0.500
geol3(categorical) mask Regularization values: linear/quadratic/product: 0.050, categorical: 0.250, threshold: 1.000, hinge: 0.500
Environmental layers used: annprec anntemp elev geol3(categorical) mask Regularization values: linear/quadratic/product: 0.050, categorical: 0.250, threshold: 1.000, hinge: 0.500 Feature types used: P L Q H T Environmental layers used: alt annprec anntemp julytemp mask rivers1(categorical) springs(categorical) Regularization values: linear/quadratic/product: 0.213, categorical: 0.250, threshold: 1.570, hinge: 0.500
The Maxent outputs are given as a continuous surface, in which each cell has a value that is part of the distribution found by the
model and indicates likelihood of presence. To create a binary map showing where the presence is most likely to occur, a
threshold or cut-off value must be chosen. Depending on what value is chosen, the output can be extremely different, and impact
the area over which a species is predicted (this may be expressed as prevalence, which can be calculated by dividing the cells in
which the species is present by the total number of cells). This has obvious implications for both errors of omission and errors of
commission. The smaller the threshold, the smaller the omission error will be (but at the expense of greater commission error). As
the threshold becomes bigger, errors of omission increase whilst those of commission decrease
After the vegetation zones were modeled, they were combined to produce one plant geographical region map for the southern
Levant. For the most part the distributions of the different zones don’t overlap. However, when a particular cell does overlap it is
assigned to a transition zone or ecotone (Mediterranean-Irano Turanian or Irano Turanian-Saharo Arabian transition zone).
Because there is no consensus in the literature regarding the best way to evaluate SDM models based on presence-only data
The resulting map of present day plant geographical regions is shown in Figure 3. The combination of the plant geographical regions into one map created two additional transition zones between the Mediterranean and Irano Turanian regions (M-IT), and between the Irano Turanian and Saharo Arabian (IT-SA) regions.
The region of Mediterranean vegetation occupies the wetter, higher elevation portions of the study area on both sides of the Rift Valley. On the west, the Mediterranean region is more continuous and spreads south until it reaches the northern Negev. East of the Rift Valley, the Mediterranean region is not as extensive as its coastal counterpart, although it is still the dominant region in the northern portion of the Jordanian Plateau. Farther south, a few patches of Mediterranean vegetation are found in the Southern Highlands of the Plateau.
The M-IT zone lies alongside the larger Mediterranean region with a few additional patches in the Negev Highlands, at the edge of the study area. The Irano Turanian region (which represents steppe vegetation and is characterized by small shrubs and bushes and a lack of trees) sits between the Mediterranean and Saharo Arabian regions. On the Jordanian Plateau it interrupts the longitudinal Mediterranean belt in some of the valleys (e.g., Wadi Mujib) that drain west into the lower elevations of the Rift Valley. This region is also prominent on the higher elevation areas of the Negev.
The Saharo Arabian zone spreads throughout the Wadi Araba, north into the Negev and the Jordan Valley, as well as along the easternmost fringe of the study area, marking the beginning of the Eastern Desert of Jordan. The IT-SA zone is found in the Rift Valley between the Saharo Arabian and Irano Turanian zones, as well as on the southern Jordanian Plateau region in Wadi Mujib and Wadi Hasa, and in patches in the Negev Highlands and southern Negev. Tropical Sudanian vegetation is found in pockets around the Dead Sea and along the Jordan Valley.
4.1. Model evaluation
4.2. Variable importance The relative importance of the variables used in this modeling may be assessed using assessments produced by MaxEnt (e.g., percent variable contribution table, jackknife plots of variable importance, variable response curves). The most important variable in determining suitable areas for all plant geographical regions is mean annual rainfall. This variable is ranked at the top in percent contribution and permutation importance. The jackknife test of variable importance also shows that this variable allows a good fit to the training data when used alone. When omitted, it decreases the gain for the Mediterranean and Saharo Arabian categories. The response curve plots show how this variable relates to each category. For the Mediterranean category, the gain starts to increase steadily above 300 mm/yr. For the Irano Turanian region, it peaks between 200 and 400 mm/yr. For the Saharo Arabian region, on the other hand, suitability decreases dramatically as rainfall increases. This suggests that this variable will be crucial when creating models of vegetation in past and future scenarios.
The second most important variable, however, was different for each category. For the Mediterranean region this variable was geologic substrate (limestone), which was followed closely by annual temperature. For the Irano Turanian and Saharo Arabian categories, the second most important variable is annual temperature, followed by elevation (Irano Turanian) and geology (Saharo Arabian). Elevation also was important for Tropical Sudanian vegetation, in accordance with its distribution in the lower elevation areas of the Rift Valley. However, when omitting this variable, the gain of the model was reduced only modestly, suggesting that the information contributed by this variable is captured by other variables. Average annual temperature and distance to rivers and springs also were important variables when modelling this category.
Although the region suitable for Mediterranean vegetation shows general agreement with Eig’s (1931/32) and Zohary’s (1962,
1966) delineation of this territory, our modelling shows several noteworthy differences. For instance, Zohary includes the entire
coastal plain within the Mediterranean vegetation region, whereas our results propose Mediterranean vegetation only for the
northern portion of the coastal plain. This area has been heavily impacted by urban development. Nevertheless, the communities
that are found throughout this area (now quite fragmented and as remnant stands) include both evergreen and deciduous oak
maquis as well as pistacia and carob woodlands. The modeled distribution of Mediterranean vegetation for the remaining coastal
plain is fragmented, particularly in the far south. Throughout this area, the vegetation acquires more of a Mediterranean savannoid
character, with Ziziphus spina Christi and Ziziphus lotus trees becoming more abundant. In his map of natural environments,
Horowitz shows a thin strip (about 5 km wide) adjacent to the coastal sand dunes as Saharo Arabian
The Irano Turanian vegetation region produced by our modelling resembles the area delineated by Eig (1931/32) and Zohary
(1962, 1966) to a large extent, with some variations. MaxEnt modelling predicts this region along the northern portion of the Rift
Valley, from the Sea of Galilee to the Lower Jordan Valley. As with the Saharo Arabian region, most of the disagreement between
vegetation studies arises on the Jordanian Plateau. Our modelling predicts a larger area of Irano Turanian vegetation on the
northern Plateau than proposed by
Our modeled Saharo Arabian plant geographical region is quite similar to the area delineated in other maps. For instance,
In keeping with our modelling, other authors identify Tropical Sudanian enclaves surrounding the Dead Sea and along the Lower
Jordan Valley. Al-Eisawi notes this vegetation between Aqaba and Deir ‘Alla, but adds that it is most prominent in the area
surrounding the Dead Sea (hotspots include Ghor Safi and Ghor Faifa). Danin’s map attributes only small areas to this category.
Like Al-Eisawi, Zohary (1947) mentions oases with this vegetation along the Dead Sea and Lower Jordan Valley. In its northern
distributions, Tropical Sudanian vegetation commonly includes Ziziphus spina Christi, Balanites aegyptica, Callotropis procera,
Acacia tortilis and Acacia albida. In addition to these taxa, southern Tropical Sudanian distributions include Phoenix dactylifera,
Salvadora persica, Moringa aptera, Occradenus baccata and Nerium oleander. Our application of Maxent utilized the
northernmost observations of tropical oasis vegetation
Maxent modelling predicts the potential modern distribution of four major vegetation types across the southern Levant with high rates of model performance. The contributions of key environmental and climatic variables differ considerably between the models for these vegetation types. While predicting vegetation distributions in keeping with previous vegetation mapping studies, our modeled potential vegetation patterns diverge in a variety of sometimes counterintuitive ways that hold implications for further modelling of present and past vegetation in this region.
Part of this research was funded by the Mediterranean Landscape Dynamics Project (NSF Grant # 0410269). We also wish to thank the anonymous reviewers who kindly made time to provide feedback.
Al-Eisawi, D. 1985. Vegetation in Jordan. In Studies in the History and Archaeology of Jordan, II, edited by A. Hadidi. Amman: Department of Antiquities.
Al-Eisawi, D. 1996. Vegetation of Jordan. Cairo: UNESCO.
Al-eisawi, D. (2012). CONSERVATION OF NATURAL ECOSYSTEMS IN JORDAN. Pak. J. Bot., 44(Special Issue May), 95– 99.
Al-Eisawi, D., El-Oqlah A., Oran S., and Lahham J. 2000. Jordan Country Study on Biological Diversity: Plant Biodiversity and Taxonomy, edited by U. Publication. Amman.
Anderson, R.P., M. Gomez-Laverde, and A.T. Peterson. 2002. Geographical distributions of spiny pocket mice in South America: Insights from predictive models. Global Ecology and Biogeography 11:131-141.
Austin, M.P., A.O. Nicholls, M.D. Doherty, and J.A. Meyers. 1994. Determining species response functions to an environmental gradient by means of a beta function. Journal of Vegetation Science 5: 215–28.
Barton, C. M., I. I. Ullah, and S. Bergin. 2010. Land use, water and Mediterranean landscapes: modeling long-term dynamics of complex socio-ecological systems. Philosophical transactions. Series A, Mathematical, physical, and engineering sciences 368 (1931): 5275-5297.
Bartov, Y. 1994. The geology of the Arava Valley. Israeli Geological Survey Report GSI/4/94, 16. BioGIS 2000 BioGIS (Israel Biodiversity Information System Database) [http://habitat.bot.huji.ac.il/biogis/static/en/index.html]. BioGIS. 2000. A. Danin's grid database. Israel Biodiversity Information System. [http://www.biogis.huji.ac.il] . BioGIS. 2002. Israel Nature and Parks Authority database. Israel Biodiversity Information System. [http://www.biogis.huji.ac.il]. Bryson and DeWall 2007
Davies, C.P., and P.L. Fall. 2001. Modern pollen precipitation from an elevational transect in central Jordan and its relationship to vegetation. Journal of Biogeography 28:1195-1210.