Monitoring the Spatiotemporal Evolution of the Green Dam in Djelfa Province, Algeria

: Green walls and green dams are increasingly being considered as part of many national and international desertiﬁcation initiatives. This paper studies the spatiotemporal evolution of the green dam in the Moudjbara region (Djelfa Province, Algeria), from 1972 to 2019, by using Landsat imagery, Land Change Modeler, and OpenLand package. The future evolution of pine plantations, for the year 2029, was also forecasted, based on an anthropogenic scenario (i.e., anthropogenic pressure is the main driver of the green dam destruction). Our ﬁndings revealed that the green dam project was successful for a few years, but, after that, pine plantations deteriorated signiﬁcantly, due to forest harvesting, livestock overgrazing, and the proliferation of the pine caterpillar processionary, which destroyed most of the reforestation. Land change modeler predicted a huge degradation of pine plantations for the year 2029, and if the deforestation continues at the same rate, the green dam in the Moudjbara region will disappear during the next few decades. Being aware of this threat, the Algerian authorities are now planning to reforest more than 1.2 million ha under the latest rural renewal policy, by introducing new principles related to sustainable development, ﬁghting desertiﬁcation, and climate change adaptation. We strongly recommend moving away from the singular tree planting focus, to diversifying desertiﬁcation control methods.


Introduction
Land degradation and desertification (LDD) are the main reasons for the rapid changes that are affecting the majority of the earth's drylands, resulting in an overall loss of habitats and changes in vegetation cover, plant composition and diversity, hydrologic conditions, and soil properties [1]. Moreover, LDD, in the developing countries, in particular, is considered one of the key drivers of many social, ecological, and economic problems [2] Today, regions such as North Africa, Sahel, the Middle East, and China are suffering from LDD that is decreasing land productivity and exacerbating poverty, which could jeopardize the livelihoods of the local people in these countries [3]. The assessment of land degradation mitigation programs, therefore, looks to be essential, as it may offer land managers and decision-makers crucial insights [2].
Anthropogenic revegetation has been regarded as the most effective way to reverse LDD worldwide [4,5]. Particularly, green walls and green dams are increasingly being considered as part of many national and international desertification initiatives [6][7][8], with the main aim of preventing sand encroachment in dryland. However, while the afforestation programs that have been carried out in China (e.g., Grain for Green program,

Study Area
The study area "Moudjbara" is located in the southeastern part of Djelfa Province, which in turn is situated in the Saharan Atlas, nearly 300 km south of the capital Algiers ( Figure 1). Between the latitude 34 • 36.64 E and 3 • 23 0.26 E with a surface of 10,000 ha, the area is characterized by a slightly rugged flat terrain with relatively homogeneous relief at an altitude ranging between 1200 and 1400 m. The soils of the area are dominated by little evolved and shallow rendzine. The climate is semi-arid with cold and harsh winters (the average minimum temperature of the coldest month is 0.1 • C), and hot and dry summers (the average maximum temperature of the warmest month is 33.8 • C). There is, therefore, a sharp thermal contrast between seasons, as winters are cold and summers are warm. Mean rainfall is 300 mm [20] and, since 1972, a drying climate trend has been noted [19]. It is worth noting that this zone is considered a gateway to the Great Sahara, which makes it a region of major importance in fighting against desertification and protecting soil against wind erosion. Moudjbara was the first perimeter to be afforested using Aleppo pine (Pinus halepensis) in 1972 within the framework of the green dam project. The plantations were established Sustainability 2021, 13, 7953 3 of 15 near the existing natural Aleppo pine forests. In addition to natural and planted Aleppo pine forests, the area also includes urban zones (the town of Djelfa and its suburban areas), croplands used mainly for cereal production and degraded steppe pasture land dominated by Stipa tenacissima, Artemisia herba-alba and Lygeum spartum used for livestock grazing. importance in fighting against desertification and protecting soil against wind erosion. Moudjbara was the first perimeter to be afforested using Aleppo pine (Pinus halepensis) in 1972 within the framework of the green dam project. The plantations were established near the existing natural Aleppo pine forests. In addition to natural and planted Aleppo pine forests, the area also includes urban zones (the town of Djelfa and its suburban areas), croplands used mainly for cereal production and degraded steppe pasture land dominated by Stipa tenacissima, Artemisia herba-alba and Lygeum spartum used for livestock grazing.

Data Collection
For monitoring green dam evolution and changes in our study area during the following time intervals : 1972-1989, 1989-1999, 1999-2009, and 2009-2019, we used Landsat images with 30 m resolution. Five temporal datasets (see Table 1) were acquired from the US Geological Survey. We chose Landsat because it provides the longest archive of images [27], dating as far back as the early 1970s. Besides, the Landsat MSS data set has four spectral bands (4-7) with a spatial resolution of 60 m. Landsat TM data set has seven spectral bands with a spatial resolution of 30 m, and the Landsat 8 OLI data set has nine spectral bands (2-7) with a spatial resolution of 30 m. We used good quality and cloud-free scenes for the realization of LULC maps. Except for the 1972 images, which were taken in September because of the unavailability of cloud-free images in the same season, all Landsat images (1989,1999,2009, and 2019 images) were taken in July, which allows comparison of the resulting LULC maps.

Data Collection
For monitoring green dam evolution and changes in our study area during the following time intervals : 1972-1989, 1989-1999, 1999-2009, and 2009-2019, we used Landsat images with 30 m resolution. Five temporal datasets (see Table 1) were acquired from the US Geological Survey. We chose Landsat because it provides the longest archive of images [27], dating as far back as the early 1970s. Besides, the Landsat MSS data set has four spectral bands (4-7) with a spatial resolution of 60 m. Landsat TM data set has seven spectral bands with a spatial resolution of 30 m, and the Landsat 8 OLI data set has nine spectral bands (2-7) with a spatial resolution of 30 m. We used good quality and cloud-free scenes for the realization of LULC maps. Except for the 1972 images, which were taken in September because of the unavailability of cloud-free images in the same season, all Landsat images (1989,1999,2009, and 2019 images) were taken in July, which allows comparison of the resulting LULC maps.

Image Preprocessing and Classification
The acquired images were initially subject to radiometric calibration to convert digital numbers to the top of atmosphere radiance (i.e., converting pixels values to spectral radiance and reflectance). By using the FLAASH tool in the ENVI 5.3 (Exelis Visual Information Solutions, Boulder, CO, USA) we then performed atmospheric correction, which consists of the conversion of the top of atmospheric values that were generated from radiometric calibration to the bottom of the atmosphere reflectance (surface reflectance) and to remove the atmospheric effects to obtain the reflectance of ground images. Finally, all images were clipped out according to the vector of the study area and projected in ArcGIS 10.8.1 (Environmental Systems Research Institute, Redlands, CA, USA). There are various techniques for classifying satellite imagery to generate LULC maps. Among the existing techniques, supervised classification is an image-processing method regularly used to detect changes in land surface features in different multi-spectral global datasets [28,29]. In this study, we applied the supervised classification method using a maximum likelihood classifier (MLC) in ArcGIS 10.8.1 (Environmental Systems Research Institute, Redlands, CA, USA) to classify the acquired Landsat images. The MLC technique depends on the likelihood that each pixel fits in a particular class based on a normal distribution of each band's data in the classification [30][31][32]. The first step in the classification process is to create training samples. In this study, training samples were created from high-resolution Google Earth aerial photographs by digitalizing at least 20 polygons for each one of the five LULC classes considered in this study, which included degraded steppes (degraded steppe pasture land dominated by Stipa tenacissima, Artemisia herba-alba and Lygeum spartum used for livestock grazing), urban areas (the town of Djelfa and its suburban areas), crops (croplands used mainly for cereal production), dense plantations and natural forests (natural forests and dense plantations of Aleppo pine), and open plantations (Aleppo pine plantations planted within the framework of the green dam project). It is worth stating that the separation of natural forests and dense plantations of Aleppo pine was impossible, so these two classes were grouped into one LULC class. The low number of LULC classes included in this study made the classification procedure easy for the green dam monitoring study.

Accuracy Assessment
Accuracy assessment is very useful for individual classification when the resulting data are used for change detection analysis [27]. To assess the classification accuracy of the resulting maps, over 150 random points (validation samples) were created on the study area on each classified map for the five LULC classes using ArcGIS 10.8.1 (Environmental Systems Research Institute, Redlands, CA, USA). The validation data were derived through the interpretation of high-resolution Google Earth aerial photographs close to the date of the acquisition of the images. Kappa coefficient and overall accuracy were computed for the resulting LULC maps based on confusion matrices.

LULC Changes Analysis and Prediction of the Future Scenario for the Year 2029
Chronological series of LULC maps are required to understand the range, types, and spatial configuration of changes that have occurred to a specific site, and to understand what drives those changes as well as to predict future changes. To identify spatially explicit LULC change in the Moudjbara region, we used the Land Change Modeler (LCM) tool and the OpenLand package in Idrisi TerrSet 18.7 (Clark Labs, Worcester, MA, USA) and R 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria), respectively. The LCM is the most widely used spatial model for LULC change analysis and for predicting future scenarios [33], and OpenLand implements a complete intensity analysis including tools for the generation of standardized multilevel output graphics [34]. Change analysis was performed for the following time intervals : 1972-1989, 1989-1999, 1999-2009, and 2009-2019, with the aim of (1) quantifying LULC changes in each time interval, (2) assessing whether the size and annual rate of change differ over periods by comparing the observed rates to the uniform rate (i.e., the rate that would occur if the annual changes were spread evenly over the entire time extent), and (3) scrutinizing the spatiotemporal evolution of the green dam project in the study area (i.e., quantifying gains, losses, and net changes in pine plantations in each time interval).
The future evolution of pine plantations in the Moudjbara region for the year 2029 was also forecast using LCM, based on the following steps: (1) we grouped all the transitions from pine plantations (i.e., dense-cover or low-cover pine plantations) to other LULC classes (i.e., degraded steppes, urban areas, and croplands) in one sub-model, assuming that the underlying drivers of these changes are the same (i.e., the anthropogenic pressure); (2) based on their potential in explaining the changes, i.e., pine plantation destruction, the following explanatory variables were used: digital elevation model (DEM), distance from croplands, distance from steppes, distance from roads, and distance from urban areas (we used Cramer's V coefficient to test the significance of each variable); (3) after selecting the suitable set of explanatory variables, transitions were modeled in LCM using an Artificial Neural Network (ANN) approach based on multi-layer perceptron (MLP); (4) change prediction was executed based on the Markov chain, and the calibration periods from 1999 and 2009 were used to perform the modeling for 2019; (5) after evaluating the accuracy and skill measure of the modeling, the developed model and the maps for 2009 and 2019 were used to estimate the changes in pine plantations for the year 2029.

Temporal Changes in Human Population and Livestock Numbers
In this study, the temporal changes in human population and livestock numbers, mainly sheep and cattle, within Djelfa Province were used as indicators for reconstructing past changes in land use and for indirectly assessing changes in human pressure on the green dam. Demographic and livestock data were obtained from the Office for National

Accuracy Assessment and LULC Changes
The accuracy assessment of the LULC maps of the five temporal periods is shown in Table 2. The overall accuracy ranged from 75.5% to 94%, and the Kappa from 64% to 89.9%. The 2019 LULC map had the highest overall accuracy (94%) and Kappa (89.9%), whereas the 1989 LULC map had the lowest overall accuracy (75.5%) and Kappa (64.1%). The LULC change analysis results clarify the historical change and give us a clear statement about the land use practice in different years and the dynamic change in each cover class. The spatial distribution of the five LULC classes (degraded steppes, crops, urban, dense plantations and natural forests, and open plantations) during the five-time periods 1972,1989,1999,2009, and 2019 are presented in Figure 2, and the statistics (including the changes from one period to another) are shown in Table 3 Figure 4 shows that the rate of LULC change is not perfectly stationary during the studied time intervals, because the bars on the right do not all equal the uniform line. The left side of Figure 4 shows that the size of the change during the second time interval (1989-1999) is the smallest. However, the size of the change during the first time interval (1972)(1973)(1974)(1975)(1976)(1977)(1978)(1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989) is the greatest, and its duration is the longest (17 years). The right side of Figure 4 shows that the annual rate of LULC change was slower in the second time interval (1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999) compared to the other time intervals, and the rate of LULC change became fast in 1999.     (1972-1989, 1989-1999, 1999-2009, and 2009-2019) showing how the size and speed of change vary across time intervals. Bars that extend to the left of zero show the area of overall change in each interval, while bars that extend to the right of zero show the intensity of the annual area of change within each time interval. Uniform rate (U) refers to the rate that would occur if the annual changes were spread evenly over the entire time extent.

Green Dam Evolution during the Last 47 Years and Its Future Projection by the Year 2029
Firstly, we should note that the green dam in the Moudjbara region includes all open plantations and a great portion of dense plantations and natural forests. The results clearly show both losses and gains in these two LULC classes during the studied periods ( Figure  5). During the first period, 1972-1989, the open plantations gained approximately 10,000 ha, which had been occupied by degraded steppes; during the same period, the transition of dense plantations and natural forests gained 600 ha on the degraded steppes and lost  (1972-1989, 1989-1999, 1999-2009, and 2009-2019) showing how the size and speed of change vary across time intervals. Bars that extend to the left of zero show the area of overall change in each interval, while bars that extend to the right of zero show the intensity of the annual area of change within each time interval. Uniform rate (U) refers to the rate that would occur if the annual changes were spread evenly over the entire time extent.

Evolution of Human Population and Livestock Numbers
The human population of Djelfa Province has increased exponentially between 1996 and 2018 and will continue increasing until it reaches a value of more than two million inhabitants in 2030 (Figure 8). Similarly, the livestock numbers have increased markedly

Evolution of Human Population and Livestock Numbers
The human population of Djelfa Province has increased exponentially between 1996 and 2018 and will continue increasing until it reaches a value of more than two million

Evolution of Human Population and Livestock Numbers
The human population of Djelfa Province has increased exponentially between 1996 and 2018 and will continue increasing until it reaches a value of more than two million inhabitants in 2030 (Figure 8). Similarly, the livestock numbers have increased markedly between 1998 and 2020 ( Figure 8). Between 2020 and 2030, the increase in livestock numbers will continue with the same rate, and the power function predicted a value of more than five million heads in 2030 (Figure 8).
x FOR PEER REVIEW 11 of 15 between 1998 and 2020 ( Figure 8). Between 2020 and 2030, the increase in livestock numbers will continue with the same rate, and the power function predicted a value of more than five million heads in 2030 ( Figure 8).

Discussion
The spatiotemporal dynamics of LULC in the Moudjbara region experienced both negative and positive changes during the five-time intervals studied. The rate of the LULC

Discussion
The spatiotemporal dynamics of LULC in the Moudjbara region experienced both negative and positive changes during the five-time intervals studied. The rate of the LULC change was not stationary during the studied time intervals. The intensity of the LULC change was slow in the first and second time intervals and became relatively fast from 1999. Likewise, our results revealed a nonlinear evolution of the area that was occupied by pine plantations between 1972 and 2019 and predicted a huge degradation of the green dam by the year 2029.
Our findings showed that during the first period of study (1972)(1973)(1974)(1975)(1976)(1977)(1978)(1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989), the area of the Moudjbara gained a huge extent of pine plantations (the natural forest was the central point for plantations), which replaced the steppe pasture lands. Although the increase in pine forests (mainly the planted ones) is a positive change, it is worth highlighting some flaws of this anthropogenic revegetation. The first error was the conversion of the steppe pasture lands, where the natural vegetation of Artemisia herba-alba, Lygeum spartum, Stipa tenacissima, and many other plants have adapted to the harsh climatic conditions, and that constitutes, in itself, a natural barrier against LDD. Another issue was the lack of integration of the local populations; the project did not allow the local populations to adopt it as a complementary initiative to their socio-economic way of life, which is based mainly on livestock farming [12,28]. Many pastoralists opposed the establishment of these plantations, which led to the emergence of permanent conflicts that lasted for a long time [7].
Although many efforts were made, during the second period (1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999), to improve the quality of reforestation and the application of other operations, such as dune fixation and track openings [29], the total area of the pine plantations decreased sharply. This degradation was mainly due to the expansion of croplands (pine plantations were harvested at some locations to increase the amount of arable land, especially for the cultivation of cereals), livestock overgrazing, and phytophagous attacks [35]. It should be noted that the use of monospecific stands of Aleppo pine has resulted in the proliferation of pine caterpillar processionary, which has destroyed most of the reforestations [6,36,37]. Furthermore, in many sites, the trees were planted on encrusted soils that impeded the penetrability of their roots, and, as a consequence, led to the development of dwarf trees.
The third period, between 1999 and 2009, was characterized by a notable increase in the area occupied by open plantations, but also by a decrease in dense plantations, due to human pressure on the territory (i.e., tree cutting for timber and fuelwood provision), or due to their direct use as a sylvo-pastoral system. The expansion of plantations during that period was mainly due to the reforestation program that was carried out by the General Directorate of Forests (DGF) in 2000, within the framework of the National Reforestation Plan (PNR). The PNR was characterized by the diversification of actions, ranging from reforestation to improving steppic rangelands by regulating livestock farming.
Our findings revealed a massive loss in pine plantations between 2009 and 2019, and most of the lost areas were converted to degraded steppes. To satisfy the need for increasing the number of livestock, pine plantations were completely harvested in some locations, to increase the area of pastures. In other locations, and as cited above, pine plantations were used as a sylvo-pastoral system, with a high livestock grazing pressure. Long-term overgrazing, coupled with prolonged cyclical droughts [38], has hampered the natural regeneration of these plantations. It is also worth noting that this period, in particular, was characterized by a widespread occurrence of fires and a huge proliferation of the pine caterpillar processionary, which led to the disappearance of large areas of the green dam.
Based on the anthropogenic scenario, i.e., where anthropogenic pressure is the main driver of the green dam destruction, our model predicted a loss of a huge amount of pine plantations in the Moudjbara region during the next decade. This finding is in line with the study by [28], who found only traces of the green dam, formed by a few bands of puny Aleppo pines that were scattered over a desertified area.

Conclusions
This paper has presented a study of LULC changes and their effect on the Moudjbara plantations during the last 47 years (from 1972 to 2019), using freely available data (Landsat imagery) and geographic information systems. The findings revealed that the green dam project was effective for a few years and, after this period, pine plantations underwent a considerable deterioration due to forest clearing, livestock overgrazing, and the proliferation of the pine caterpillar processionary, which has destroyed much of the reforestation. Our analysis also predicted that if the degradation continues at the same rate, the green dam project will disappear during the next few decades, in the analyzed region. Being aware of the threat of the destruction of the green dam, the General Directorate of Forests (DGF) is currently planning to reforest more than 1.2 million ha in the region, under the latest rural renewal policy, by introducing new principles related to sustainable development, fighting desertification, and climate change adaptation. Today, the DGF has learnt lessons from former programs, and plantations with monospecific stands are barred.
In our opinion, for effective control of LDD in Algeria, the institutional approach has to be reviewed and science has to play a more important role in guiding policies. The authorities should care about what people want and what nature needs. The focus only on tree planting often does not serve the livelihood demands of the local people, and the use of this practice as a key approach to stop the encroachment of sand on dryland where trees do not naturally grow, and where shrubs or grass are the native and natural land, needs to be reviewed since woody vegetation encroachment could affect the functioning of this fragile ecosystems. Scientific research has demonstrated that natural recovery is much more effective in restoring degraded arid steppes [38,39]. Hence, we strongly recommend moving away from the singular tree planting focus to diversifying desertification control methods. In this context, the application of dune stabilization methods (including mechanical, chemical and biological methods) is effective in reducing sand and dust storms [40][41][42]. Likewise, the implementation of sustainable grazing management practices, such as "grazing exclusion", is necessary to enable the natural recovery of the degraded steppes [38,39].