Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/taylor-francis/monitoring-of-forest-cover-dynamics-in-eastern-area-of-b-ni-mellal-JvouaZKKGj
References
References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.
- Publisher
- Taylor & Francis
- Copyright
- © 2018 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group
- ISSN
- 2474-9508
- DOI
- 10.1080/24749508.2018.1452478
- Publisher site
- See Article on Publisher Site
Abstract
GeoloGy, ecoloGy, and landscapes, 2018 Vol . 2, no . 3, 203–215 https://doi.org/10.1080/24749508.2018.1452478 INWASCON OPEN ACCESS Monitoring of forest cover dynamics in eastern area of Béni-Mellal Province using ASTER and Sentinel-2A multispectral data a a a b b Ahmed Barakat , Rida Khellouk , Aafaf El Jazouli , Fatima Touhami and Samir Nadem laboratory of Georesources and environment, Faculty of s ciences and Techniques, sultan My slimane University, Béni-Mellal, Morocco; polydisciplinary Faculty, sultan My slimane University, Béni-Mellal, Morocco ABSTRACT ARTICLE HISTORY Received 7 november 2017 This study aimed to monitor and analyze the spatial and temporal dynamic of forest cover in a ccepted 25 January 2018 Eastern area of Beni-Mellal Province (Morocco), using multispectral ASTER and Sentinel-2A MIS images acquired in 2001 and 2015, respectively. The supervised classification algorithm and KEYWORDS NDVI were combined within a GIS environment to quantify the extent and density change of Béni-Mellal province; forest forest cover stands, i.e., Holm oak, Aleppo pine, Thya, Zea oak, Crops & others and Bare ground. stand changes; asTeR and The classification overall accuracy was 97.76 and 95.80% in 2001 and 2015 images, respectively. s entinel-2a data; supervised The result revealed an overall forest cover change with an increase in forested area. All species classification stands showed expansion at the expense of the bare ground and crops & others classes. The density maps showed a net density change with an expansion of dense forest class. The observed forest cover expansion may be due to the favourable climate in the examined period, the protection, the reforestation programs and the regeneration through clandestine cutting. These results constituted the first attempt at mapping and monitoring of forest cover change in the study region that used a remote sensing-based product. They will help authorities and forest managers for the development of sustainable forest conservation and management decisions. 1. Introduction Even though this decline occurs in several arid areas in the world, and sometimes at a faster rate than to that Forest is a treasure trove of biodiversity and plays a of tropical forests (Seabrook, McAlpine, & Fensham, vital role in maintaining the ecological balance and 2006; Zak, Cabido, & Hodgson, 2004), the forest cover the health of our planet. The forest cover in arid and in arid and semi-arid regions has received less attention semi-arid ecosystems is important in support of global (Grainger, 1999). International agreements on forest biodiversity, carbon sequestration, and economic activ- monitoring have begun to include forest degradation ities. It provides such a key ecosystem goods and ser- and deforestation (Kissinger, Herold, & De Sy, 2012). vices to over one billion people living in the arid and According to statistics from the Food and Agriculture semi-arid lands (Safriel et al., 2005). It also conserves Organization of the United Nations, global forest area many animal and plant species (Food and Agriculture decreased by 3.1% (129 million hectares) between 1990 Organization of the United Nations [FAO], 2015), char- and 2015 (FAO, 2016). The excessive exploitation of acterized by many related functions such as timber pro- forest resources has led to depletion. Given this alarm- duction, climate regulation and recreation (Gamfeldt ing situation of over-exploitation and degradation of et al., 2013). Several scientific researches enumerated forest heritage, it needs good management and sus- ecological, economic and social prot o fi f forest for tainable development. humanity, and above all for local populations (i.e. FAO, Assessment forest areas and their dynamics at local 2010; Karjalainen, Sarjala, & Raitio, 2010; Rotenberg & to regional scales has therefore become an urgent neces- Yakir, 2010; Salehi & Karltun, 2010; Yang et al., 2012). sity given their vulnerability to anthropogenic and cli- The forest cover in arid and semi-arid regions (fragile matic pressures, importance for adaptive management environment) has undergone in the last few dec- and implementation of national directives and interna- ades negative changes (de Waroux & Lambin, 2012). tional treaties (United Nations Convention to Combat Its continued decline is due to rapid climate change Desertification of 1994 (UNCCD); Convention on the and human pressures such as overexploitation, over- International Trade in Endangered Species (CITES), pasturage, conversion to agriculture, and urbanization. Intergovernmental Forum on Forests (IFF), Evaluating CONTACT ahmed Barakat a.barakat@usms.ma The supplemental data for this article is available online at https://doi.org/10.1080/24749508.2018.1452478. © 2018 The a uthor(s). published by Informa UK limited, trading as Taylor & Francis Group. This is an open a ccess article distributed under the terms of the creative c ommons a ttribution license (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. 204 A. BARAKAT ET AL. their changes and species diversity over time and space resources remain insufficient to reverse the observed is important to understand the effect of anthropogenic degradation trend, due mainly due to anthropogenic activities and climate change on forest cover. Despite and arid constraints. the importance of forest inventories in obtaining infor- To assess their status, more studies have been con- mation on changes in forest ecosystems, they are costly ducted, but they mainly concerned monitoring the and time-consuming (Mohajane, Essahlaoui, Oudija, threatened or vulnerable vegetal species, namely regres- El Hafyani, & Cláudia Teodoro, 2017). An attractive sion of Argan trees (Occidental High-Atlas and Souss and less-expensive method to carry out such moni- plain) (de Waroux & Lambin, 2012; Linares, Taïqui, toring is the remote sensing that has demonstrated its Sangüesa-Barreda, Seco, & Camarero, 2013), Atlas effectiveness for rapid assessment of forest dynamics cedar in the Middle Atlas forests (Bahri et al., 2007; and degradation over several decades and on a large- Haboudane & Bahri, 2007; Linares et al., 2013), cork scale (Eva et al., 2010; Hansen et al., 2009; Nagendra oak communities at Mâamora (Laakili et al., 2016; Saber, et al., 2013; Wani, Joshi, Singh, & Shafi, 2016). For Rhazi, Rhazi, & Ballais, 2008), and Juniperus thurifera mapping and estimating forest cover change, a num- in the High-Atlas of Marrakech (Montès, Bertaudière- ber of remote sensing techniques were used, such as Montes, Badri, Zaoui, & Gauquelin, 2002), Carob trees the multi-temporal vegetation indices (NDVI, SVI, in Azillal Province (Hajib, Lahlaoi, Alaoui, & Khattabi, EVI) (del Castillo, García-Martin, Aladrén, & de Luis, 2015) and green oak in High Atlas (Slimani, El Aboudi, 2015; Czerwinski, King, & Mitchell, 2014; Höpfner & Rahimi, & Khalil, 2017). Many of them used an approach Scherer, 2011; Mancino, Nolè, Ripullone, & Ferrara, based on field data and visual image interpretation from 2014), the spectral band analysis (Röder, Udelhoven, different sensor data to detect vulnerable tree species Hill, del Barrio, & Tsiourlis, 2008; Sexton, Urban, change. Donohue, & Song, 2013) and conventional classifica- e e Th astern part of the province of Béni-Mellal, a tion methods (Harrison, Rivard, & Sánchez-Azofeifa, study area, which is part of the Atlas mountain of Béni- 2018; Melville, Lucieer, & Aryal, 2018; Murray et al., Mellal, has a significant forest cover whose preserva- 2018). In this context, remote sensing data from several tion of biodiversity requires sustainable management. satellite sensors (i.e., Landsat, ASTER, Sentinel-2A, However, it remains highly vulnerable to various pres- MODIS), which is now available free and provides sures and aggression caused by multiple and complex temporal datasets for few decades, has become increas- interactions between natural and anthropogenic fac- ingly used to identify the forest cover types and their tors, and its monitoring is, therefore, necessary to check relative changes. and estimate the change in forest cover and to propose Morocco, a semi-arid/arid country, benefits from appropriate conservation plan. Thereby, monitoring of a considerable floristic diversity represented by for - this forest cover is, consequently, essential given their est ecosystems occupying 31,482 km of the total area vulnerability to anthropogenic pressures and those asso- (Lahlaoi, Rhinane, Hilali, Lahssini, & Moukrim, 2017). ciated with climatic fluctuations. While monitoring has Forest stands vary due to climatic, orographic, litho- oen r ft evolved around field data, remote sensing may logical and anthropogenic criteria, but they are often play a fundamental role in the diagnosis of sustaining linked to landforms. The forest provides important species and habitat diversity and the quantification of services to local and regional society such as global degradation or regeneration in the study region. The aim carbon storage, soil protection, hydrological and bio- of this work was to map, classify and evaluate the spatial chemical cycle, biodiversity, economy and especially and temporal distribution of tree species and densities local well-being. In the last five decades, these fragile in the Eastern area of Béni-Mellal province using high ecosystems have begun to suffer greatly from anthro- spatial resolution and free satellite images data. ASTER pogenic impacts and climate change that remain, and Sentinel-2A MIS images acquired in 2001 and 2015, nevertheless, under-treated. Deforestation due to the respectively, constituted a principal source of data based intense use of woody resources by local rural societies, on supervised classification and NDVI calculation. e Th over-pasturage, ageing of forests and hydric stress due field data, images of Google Earth and a local forest map to climate change, have an impact on the tree regen- were used to validate the accuracy of the derived forest eration and the forest production (Bahri, Haboudane, stand maps. Bannari, Bonn, & Chillasse, 2007; Haboudane & Bahri, 2007; Hammi et al., 2010; HCWFCD, 2017; Rejdali, 2. Materials and methods 2004). Despite efforts made by the Moroccan High 2.1. Study area Commissariat for Water, Forestry and Combating Desertification (HCWFCD), particularly through The study area corresponds to the Eastern mountain national forest programmes (PNF) promoting forest area of Béni-Mellal Province, and lies between 32°28′- conservation and development of the forest sector, 32°24′N and 5°54′-5°52′W with a total area of 91819 ha the reforestation and natural regeneration of forest (Figure 1). It is part of the Atlas chain of Béni-Mellal, GEOLOGY, ECOLOGY, AND LANDSCAPES 205 Figure 1. l ocation map of the study area. s ource: a uthor. and covers six mountain communes (El Ksiba, Aghbala, e Th altitude ranges from 869 to 5542 m, with an aver- Naour, Tzi Inisly, Tanougha and Boutferda) that are age of 1778 m. The forest ecosystem has much changed underdeveloped and faces many socio-economic chal- due to the land use change (such as conversion of for- lenges resulting from the high level of poverty and ests and grasslands into cropland) and climate change marginality, which manifests in a high dependency on in the past decades. Deforestation is prominent, and the forest resources leading to overexploitation. Agriculture demand for forest land for agriculture and pasturage is and livestock are the fundamental component of the so considerable. Therefore, a detailed spatial and tempo - rural economy throughout the region. In most cases, ral analysis of the forest resources is critical and urgent these are a traditional farming and extensive livestock for better forest resource management in the eastern herding. The total human population in 2014 was about areas of Béni-Mellal province. e Th study area experiences sub-humid climate that is 2.2. Materials governed by seasonal changes. e Th average annual pre- In this study, multispectral data from two different cipitation is between 22.7 and 1003.3 mm, with a mean satellite sensors, ASTER (Reflective Radiometer and of 438.7 mm. The wet season lasts from mid-October to Space Advanced er Th mal Emission) and Sentinel-2A May with the wettest month usually being November, MIS, were used to map species distribution and forest while the dry season takes up the rest of the year. The density in 2001 and 2015, respectively. All images are bedrock consists mostly of limestone and dolomites, acquired during sunny periods, without cloud cover. covered with soils that are generally fersialitic and brown ASTER images were obtained from the U.S. Geological forest types. There are other less well-represented soils Survey (USGS) (http://earthexplorer.usgs.gov//), and such as poorly developed soils occurred on steep slopes those of Sentinel-2A MIS were collected from the offi- and brown calcareous soils developed on soft limestones cial Copernicus Sentinel-2A MIS Data Hub (ESA 2016) or on marl-limestone. 206 A. BARAKAT ET AL. Table 1. s atellite images and bands used in the analysis. and a height over than 13 m. The cover land was also comprised of bare ground and crops & others (lower Image Date d’acquisition Resolution Bands (μm) plants and shrub). Crops & others correspond to areas asTeR 06 d ecember 2001 15 m Band 1: Green (0.52–0.60) cultivated with annual crops, vegetables, or fruit, lower 18 d ecember 2001 Band 2: Red wild plants and shrubs. Most of the cultivated area has (0.63–0.69) Band 3: near infra- been recently managed for modern fruit cultivation, like red (0.78–0.86) apple or cherry. Bare ground constitutes to the exposed s entinel-2a 16 d ecember 2015 10 m Band 3: Green (0.5425–0.5775) soils, urban and rural areas and roads. 26 d ecember 2015 Band 4: Red A field campaign was targeted to ascertain the accu- (0.65–0.68) racy of the classification by field observations, and the Band 8: near infrared accuracy of the land cover map generated. It should (0.7845–0.8995) be noted that two field missions were carried out, one of reconnaissance in April 2016 and the other of vali- dation of the results in May 2016. Forty-three control (https://scihub.copernicus.eu/s2/#/home). These images points showing complicated vegetation spatial patterns were downloaded with a preprocessing level (geometric or homogeneous plant stands were established using correction in UTM WGS84 projection), at the same time Global Positioning System (GPS) instrument (Figure 1). of the year to avoid problems of seasonal variation. Near- Validation and quantitative analysis of accuracy were infrared and visible bands have been used knowing that performed using confusion matrices, global precision, many studies carried out in the world show that spec- and Kappa coefficient. The forest map available from tral information from satellites can be used to identify HCWFCD served as a reference to complete the vali- vegetation cover (Guyot, Baret, & Major, 1988; Warner, dation by superimposing it and comparing it with the Skowronski, & Gallagher, 2017). The characteristics of satellite images. satellite images and bands used in this study are sum- marized in Table 1. All image processing, classification and GIS analyses 2.3. Methods were performed using QGIS 2.8.6, ArcGIS 10.2.2 and e m Th ethodological framework adopted in this study MATLAB R2013a softwares. The ASTER digital eleva- for monitoring changes in canopy extent and density of tion model (DEM) with a spatial resolution of 30 m was forest stands is presented in Figure 2. The spatial images used to extract topographic features of the study area, of ASTER (2001) and Sentinel-2A MIS (2015) were such as elevation and aspect. selected free archive availability. These images were pro - Dominant higher plant species are stratified into four cessed using QGIS 2.8.6 to correct atmospheric errors stands, with respect to classification map available from and resample the resolution of the ASTER image from 15 HCWFCD. This 1/20000 scale forest map was produced to 10 m. Atmospheric correction is made using the Semi- by the National Geographic Institute (IGN) (1962). The Automatic Classification Plugin extension (Congedo, four stands are Holm oak (Quercus rotundifolia), Aleppo 2016) based on the Dark Object Subtraction (DOS) pine (Pinus halepensis), Thya (Tetraclinis articulata ), model. Then the images mosaic method was applied Zea oak (Quercus canariensis). Some information on because of the study area that is located between two the height and diameter of the species is determined successive images for the two satellite products. from our field observations. Holm oak is the backdrop for all forest stands in the study area, occurring at an 2.3.1. Image processing elevation between 100 and 1819 m. Amongst the stands Before the application of the classification method on of oak, there are coppices of various sizes, and stands of processed images, the coloured compositions of each young to adults, with different levels of density. Aleppo product were tested to select the best combination pine occupies a small extent, and occurs in the form of of bands. e Th best band combination is the coloured adult stands, generally located on the south and south- compositions of 3-2-1 (RGB = NIR-Red-Green) for the east slopes. The height of the trees reaches 12 m, and image ASTER and Sentinel-2A MIS (Desclée, Bogaert, & their diameter oscillates between 15 and 35 cm. Thya Defourny, 2006). The choice of the bands in the infrared formation constitutes the main resinous species of the (band 3 for ASTER and band 8 for Sentinel-2A) and in studied forest. It is found at the mean elevations of the the red (band 2 for ASTER and band 4 for Sentinel-2A) southern and southeast mountainous slopes. The height allowed to increase the contrast and to discriminate the of the Thya varied between 1 (the youngest trees) and vegetation formations between them and to differentiate 5 m. Zea oak formation covers a very small area of the them from other land-use classes. Then we applied a clas - study area and occurs as a moderately dense adult forest sification by the maximum likelihood algorithm (Chutia, structure. It occupies the north and northwest moun- Bhattacharyya, Sarma, & Raju, 2017; Kavzoglu, 2017) on tains slopes with elevations between 1550 and 1770 m. the different images, which assumes a normal distribution Some Zea oak trees reach a diameter greater than 60 cm of spectral pattern for each class (Xie, Sha, & Yu, 2008). GEOLOGY, ECOLOGY, AND LANDSCAPES 207 Figure 2. Methodological flowchart of the procedure used in the study. This classification method requires the selection of Kappa values >0.80 represent a strong agreement and forest formations in the study area. For this reason, we good accuracy, between 0.40 and 0.80 indicate a middle used a regional forest map was used identify the forest accuracy, and <0.40 indicate a poor agreement between cover types, and the field data were used to support the classification and observation. interpretation process by assigning each pixel type to a 2.3.2. NDVI differencing specific land cover class. Six different land cover types, e m Th ethod of thresholding NDVI (Slimani et al., namely Holm oak, Aleppo pine, Thya, Zea oak, crops & 2017) was adopted to highlight the density of each for- others, and bare land, were used in the classification est stand. The NDVI considered as one of the common process. In this classification, the pixel categorization vegetation indices because of its known theoretical and process is done by specifying, a numerical description empirical relationships with vegetation abundance and of land cover classes previously defined. To check if the greenness (Goward, Markham, Dye, Dulaney, & Yang, training samples were different radiometric characteris - 1991; Pettorelli et al., 2005; Wang, Adiku, Tenhunen, & tics, the signature separability was used since the spectral Granier, 2005), was used to identify and quantify the signatures of similar features have similar shapes. vegetation. It was calculated from the red and near- Since the classified maps oen co ft ntain some error, infrared (NIR) band values, and its values range from -1 accuracy assessment is an important step in the classi- (unvegetated areas) to +1 (healthy vegetated areas). The fication process. Accordingly, an assessment to ascer - NDVI extracted from our study area has values that range tain the accuracy of our classification portrayed forest from −0.14 to +0.74. The two NDVI images generated in cover was conducted based on field data, local forest 2001 and 2015 were processed for extracting three forest map used in training step, and Google Earth image. cover density classes using image thresholding methods. e o Th verall accuracy, kappa statistics and user’s and According to the NDVI value, the image was classified producer’s accuracies (Congalton & Green, 2008) were into three distinct thematic classes: NDVI <0.1: with- determinate to evaluate the accuracy of the supervised out vegetation, 0.2 < NDVI < 0.4: slightly dense forest, classification used in this study for two dates (2001 and 0.4 < NDVI < 0.74: dense forest. The identification of the 2015) images. The Kappa values lie between 0 and 1, with 208 A. BARAKAT ET AL. best-fitting threshold value was based on the comparison Focusing on the class level (Table 2), the crops & between the Google Earth images and the NDVI results. others class also decreased by 0.51% (from 8135.91 ha in 2001 to 7666.65 ha in 2015), and the bare ground class retracted by 5.74% (from 32122.33 ha in 2001 3. Results to 26854.37 ha in 2015). Within the forest area, Holm 3.1. Analysis of forest cover changes oak is the most expanded species occupying the large extension of about 48.04% of the total area in 2001 and 3.1.1. Accuracy assessment 52.61% in 2015. Aleppo pine has established a small e o Th verall accuracy, kappa statistics and user’s and area of about 2537.31 ha and 2638.51 ha in 2001 and producer’s accuracies (Congalton & Green, 2008) have 2015, correspondingly. Thuya stand accounted for about been used to assess classification accuracy in the present 5.26% (4837 ha) and 17.1% (6302.2 ha) of the total study study. The confusion matrix for 2001 and 2015 image’s region, respectively, in 2001 and 2015. Zea oak species classification, including the results of accuracy and has the lowest area (67.73 ha) compared to other classes, kappa coefficient assessment are given, respectively, in which is about 0.07% in 2001 and 0.05% in 2015 of the Tables S1 and Table S2 in Supplementary material. study region. The bare ground area reduced from about e o Th verall accuracy of the supervised classification 35% of the study region in 2001 to 29.24% in 2015. method was about 97.76% in 2001 and 95.80% in 2015, From this analysis, an overall land cover change according to obtained overall Kappa values of 0.971 and between 2001 and 2015 was identified with an increase 0.946, respectively. Knowing that the critical accuracy in a forested area. All species stands showed expansion value of 75% beyond which a classification is deemed at the expense of the bare ground and crops & others acceptable (Girard & Girard, 2010), these results indi- lands. cated the best classification accuracy. In both dates, the bare ground, Aleppo pine, and Holm oak classes were 3.1.3. Density change of forest stands the most discriminated according to the scores on the e Th density assessment maps for 2001 and 2015 were diagonals of Table S1 and Table S2 (Supplementary mate- reclassified to three density classes separated into low rial). However, the great confusion occurred between (NDVI <0.1), moderate (0.2 < NDVI < 0.4) and high the Thya, Aleppo pine and crops & others stands. This density (0.4 < NDVI < 0.74) (Figures. 5 and 6), using confusion results in close spectral responses. Regarding ArcGIS 10.2.2 software. The results of the spatial and producer’s accuracy, all classes were above 85% in both temporal evolution of the density of the canopy in years. The highest values are for the Aleppo pine stand the study area during the 2001 and 2015 periods are followed by the crops & others class, and then by Holm reported in Table 3(a). The spatial pattern of classified oak in both years. The user’s accuracy was above 90% forest cover density zones indicates that the areas with except for the Aleppo pine stand showing user’s accuracy high canopy density are located mainly in the west of the of 83.33% in 2001. The strongest scores are also for the study region, while the areas with low canopy density classes of Holm oak, bare ground, and Thya. are located in the northwest, central east and southeast A spectral signature analysis between the identified of the study region (Figures 5 and 6). Overall, the forest land cover classes performed during the classification experienced a somewhat slight density change in the process to discriminate these classes, showed that it is period analyzed (Table 3(a)). The dense forest increased impossible to separate between all species, namely the from 24814.50 to 27160.87 ha between 2001 and 2015, forest trees rarely represented and the shrub. For this, signifying a rate of 0.02%. This increase in high-density these species, shrubs, and trees seldom represented, were areas occurred at the expense of low-density areas, and merged with crops into a single class termed “crops & could likely to be related to tree development in these others.” areas especially the Holm oak and Aleppo pine stands in 2015. 3.1.2. Spatial distribution of forest stands e fin Th al map with the density of each forest stand e r Th esults of the digital classification deemed reliable was also derived using the geometric intersection of were used to assess the vegetation stands being changed the land cover density map and the spatial distribution over this period from 2001 to 2015. At the class level, the map of forest stands by applying intersect geoprocessing area per class of land cover and its expansion or retrac- function in ArcGIS. e Th overlay of the stand type map tion quantified for each map are represented in Figures 3 with the density map concerned only the Holm oak, and 4, Table 2. ya, A Th leppo pine and Zea oak stands (Figures 7 and 8, Forest was the dominant land cover in 2001 with Table 3(b)). The results indicate that the Holm oak 56.15% of the entire study region of the eastern area of species, moderately dense in 2001, became very dense Beni-Mellal province, followed by the bare ground class in 2015, especially in the northwest of the study area, (34.98%) and the crop & others class (8.86%). The forest the north of Koumch Village and south of Boutferda area increased by about 6.25% of the whole area in 2015 Village. The Aleppo pine density increased south of the study region (Tizi n’Ait Ouirra) and west of the rural (Table 2) or by about 5728.5 ha. GEOLOGY, ECOLOGY, AND LANDSCAPES 209 Figure 3. Forest stand map for 2001. s ource: a uthor. commune of Tizi-Nisly. The Zea oak formation density this increase in surface and density of the forest tree remained stable and dense. The Thya formation observed stands remains relative as the rates of change remain throughout the forest area except for southern part has low to moderate, and as the resolution of the ASTER undergone a significant density evolution in the period and Sentinel-2A MIS products used could influence 2001–2015. the accuracy of the spatial analysis. Such weak changes would also be linked to the high anthropogenic pres- sure that impedes the regeneration and dynamics of 4. Discussion these ecosystems. The high anthropogenic pressure Analysis of the dynamics of change revealed a slight could also explain such weak changes in the region, increase in forest stands between 2001 and 2015. This which is hampering the regeneration and dynamics of increase in surface could mainly be explained by the these ecosystems. effect of the enhanced protection by forestry authori- The results of this study will be certainly a refer - ties, the reforestation programs initiated by the High ence to the state of the forests and the evolution of Commission of Water and Forests for the reconstruc- the forest formations in the study area, which remain tion of certain vegetation facies, namely pine, the cli- very vulnerable to anthropogenic activities as well as matic conditions that were more favourable during climate change. The dominant anthropogenic drivers the period 2001-2015, and also the geographical and impacting the forest cover in the study region, lie in topographical position characterizing by a microcli- the clandestine deforestation, overgrazing, expansion mate attenuating the effects of drought in dry seasons of agricultural land and urbanization. The causes of on the vegetation. The transformation of forest areas deforestation are mainly due to the habits of local may be partly associated with the impact of a tree cut- populations and their extreme poverty, which leaves ting promoting the stand regeneration. Nevertheless, them too dependent on forest resources, and despite 210 A. BARAKAT ET AL. Figure 4. Forest stand map for 2015. s ource: a uthor. Table 2. Total area per class and temporal variation. aggression caused by interactions between natural and anthropogenic factors, the results showed that Land Area in 2001 Area in 2015 Variation cover the study forest is doing quite well. This is, unfortu- classes ha % ha % ha % nately, the case for many Moroccan regions suffering Holm 44119.11 48.04 48311 52.61 4191.89 +4.57 from intensive overgrazing; criminal harvesting of oak aleppo 2537.31 2.76 2638.51 2.87 101.2 +0.11 wood and parasitic diseases that are all factors that, pine combined with climatic uncertainties, cause abnormal Thya 4837.00 5.26 6302.16 6.80 1465.16 +1.54 Zea oak 67.73 0.07 46.73 0.05 −21 −0.02 evolution and regeneration in some parts of these for- crops & 8135.91 8.86 7666.65 8.35 −469.26 −0.51 ests (i.e., Bahri et al., 2007; de Waroux & Lambin, 2012; oth- Haboudane & Bahri, 2007; Hajib et al., 2015; Hammi ers Bare 32122.33 34.98 26854.37 29.24 −5267.96 −5.74 et al., 2010; Laakili et al., 2016; Linares et al., 2013; land Rejdali, 2004; Saber et al., 2008). This forest state of Total 91819.42 100 91819.42 100 the study area that appears to be normal would be due to favourable climatic conditions during the period attempts to implement a management transfer sys- 2001–2015, above all, to the ambitious approach fol- tem, the illegal exploitation remained considerable. lowed by High Commission for Water, Forests and Furthermore, the local population more accustomed the Fight against Desertification – HCWFFD, where, to traditional methods in agriculture encroached the in addition to protection, reforestation is the central forest areas by eliminating stubble and shrubs to con- action (HCWFCD, 2017). quer the new cultivable land. One underlying cause of The results also show that using and combining forest cover change is related to the excessive expand- spectral signatures and NDVI index lead to interesting ing pasturage. Despite these multiple pressures and results on mapping land cover stands and estimating GEOLOGY, ECOLOGY, AND LANDSCAPES 211 Figure 5. Forest density map for 2001. s ource: a uthor. Figure 6. Forest density map for 2015. s ource: a uthor. 212 A. BARAKAT ET AL. Table 3. Temporal variation in forest cover density (a) and density variation per forest stand (b). Area in 2001 Area in 2015 Variation Density ha % ha % ha % (a) Forest cover density l ow dense 29818.70 0.33 27002.25 0.30 −2816.45 −4.96 d ense 24814.50 0.27 27160.87 0.29 +2346.37 +4.51 (b) Forest stand density Holm oak l ow dense 22271.33 24.25 21413.20 23.32 –858.13 +1.66 d ense 21847.78 23.79 26897.80 29.29 +5050.02 +9.79 aleppo pine l ow dense 258.24 0.28 102.93 0.11 −155.31 −0.30 d ense 2279.07 2.48 2535.58 2.76 +256.51 +0.50 Thya l ow dense 1086.90 1.18 161.75 0.17 −925.15 −1.79 d ense 3750.10 4.08 6140.41 6.68 +2390.31 +4.64 Zea oak l ow dense 12.15 0.013 13.09 0.01 +0.94 +0.01 d ense 55.18 0.06 33.64 0.03 −21.54 −0.04 Figure 7. Forest stand density map for 2001. s ource: a uthor. their densities. The results obtained constitute the first can contribute to improved management of forest remote sensing-based product for mapping forest cover reserves, and highlights initiatives and challenges in the region. The present study demonstrates how sat - in this domain. Particularly since access to remote ellite-based detection of vegetation change can pro- sensing data with better quality becomes free such as vide reliable results in the assessment of natural forest the Sentinel 1A and 2A satellites that provide high- dynamics. The study confirms thus how remote sensing resolution radar and optical imagery on the Internet. GEOLOGY, ECOLOGY, AND LANDSCAPES 213 Figure 8. Forest stand density map for 2015. s ource: a uthor. 5. Conclusion respectively, in the examined period, and their densities showed significant growth in 2015. Zea oak, Crops & This study explored the potential of remote sensing for others and Bare ground classes displayed a net retraction. forest stands mapping and density in the eastern part e r Th esults of this study showed the potential of the of Beni-Mellal province, using ASTER and Sentinel-2A remote sensing products to assess the large forest area. MIS images acquired in 2001 and 2015 respectively. The Long-term monitoring of forest cover becomes possible principle of the methodology was the use of supervised to forest managers using free multi-date satellite images classification to map the different types of forest stands. due to their availability and the relative simplicity of Also, the NDVI vegetation index was calculated to esti- the methodology. Nevertheless, to discriminate vegeta- mate the density of the forest. The field data, Google ble species using multispectral images alone remains a Earth images, and a local forest map were used to vali- challenge but is possible with additional data and expert date the accuracy of the derived forest stand maps. This information. assessment constitutes the primary study of the forest This study allowed highlighting more information in the region using remote sensing tools and provides about forest stands and their dynamics that can be con- the first forest maps oer ff ing more information about a sidered a useful starting point to further spatial and tem- spatial distribution of forest stands in the study area. The poral analyses of forest cover change in other regional resulting maps showed expansion in extent, and den- and national areas. sity of forest stands on bare ground class resulting from natural processes and management actions adopted by Acknowledgements HCWFCD. Forest area and density increased to about 6.25% and 4.51% in 2015, respectively. Holm oak stand This work was conducted in collaboration with the Provincial dynamic appeared to be most important with an expan- Direction of Water, Forests and Desertification Control, Béni-Mellal, Morocco. We are grateful for their support, sion of 4.57% and a density of 9.79%. Aleppo pine and especially Mr Abdelfettah Naouri. ya s Th tands also expanded by about 0.11% and 1.54%, 214 A. BARAKAT ET AL. measurements from the advanced very high resolution Disclosure statement radiometer. Remote Sensing of Environment, 35, 257–277. No potential conflict of interest was reported by the authors. Grainger, A. (1999). Constraints on modelling the deforestation and degradation of tropical open woodlands. Global Ecology and Biogeography, 8, 179–190. Funding Guyot, G., Baret, F., & Major, D. (1988). High spectral resolution: Determination of spectral shifts between the This work was supported by Sultan My Slimane University. red and near infrared. e I Th nternational Archives of the Photogrammetry, Remote Sensing, 8, 1307–1317. References Haboudane, D., & Bahri, E. (2007). Deforestation detection and monitoring in cedar forests of the moroccan Middle- Bahri, E., Haboudane, D., Bannari, A., Bonn, F., & Chillasse, Atlas mountains. In Geoscience and Remote Sensing L. (2007). Essai de cartographie des espèces forestières Symposium. IGARSS 2007. IEEE International, 2007 (pp. dominantes dans le Moyen Atlas (Maroc) à l’aide des 4327–4330). IEEE. données [Aster mapping major forest species of the Hajib, S., Lahlaoi, H., Alaoui, H.M., & Khattabi, A. (2015). Middle Atlas Mountains (Morocco) using Aster data]. Modelling spatial distribution of the carob tree (Ceratonia Revue Télédétection, 7, 283–301. siliqua L.) in Azilal Province, Morocco. Journal of Chutia, D., Bhattacharyya, D.K., Sarma, J., & Raju, P.N.L. Geography and Geology, 7(4), 33. (2017). An effective ensemble classification framework Hammi, S., Simonneaux, V., Cordier, J.B., Genin, D., using random forests and a correlation based feature Alifriqui, M., Montes, N., & Auclair, L. (2010). Can selection technique. Transactions in GIS, 21, 1165–1178. traditional forest management buffer forest depletion? Congalton, R.G., & Green, K. (2008). Assessing the accuracy Dynamics of Moroccan High Atlas Mountain forests using of remotely sensed data: Principles and practices (p. 200). remote sensing and vegetation analysis. Forest Ecology and London, UK: CRC Press. Management, 260, 1861–1872. Congedo, L. (2016). Semi-automatic classification plugin Hansen, M.C., Stehman, S.V., Potapov, P.V., Arunarwati, B., documentation. Release, 4(29), 1–266. doi:10.13140/ Stolle, F., & Pittman, K. (2009). Quantifying changes in RG.2.2.29474.02242/1 the rates of forest clearing in Indonesia from 1990 to 2005 Czerwinski, C.J., King, D.J., & Mitchell, S.W. (2014). Mapping using remotely sensed data sets. Environmental Research forest growth and decline in a temperate mixed forest Letters, 4, 034001. using temporal trend analysis of Landsat imagery, 1987– Harrison, D., Rivard, B., & Sánchez-Azofeifa, A. (2018). 2010. Remote Sensing of Environment, 141, 188–200. Classification of tree species based on longwave Desclée, B., Bogaert, P., & Defourny, P. (2006). Forest change hyperspectral data from leaves, a case study for a detection by statistical object-based method. Remote tropical dry forest. International Journal of Applied Earth Sensing of Environment, 102, 1–11. Observation and Geoinformation, 66, 93–105. de Waroux, Y. L. P., & Lambin, E. F. (2012). Monitoring HCWFCD. (2017). Développement forestier: Reboisement degradation in arid and semi-arid forests and woodlands: [Forest development: Reforestation]. Retrieved from http:// e c Th ase of the argan woodlands (Morocco). Applied www.eauxetforets.gov.ma/fr/text.aspx?id=1038&uid=54 Geography, 32, 777–786. Höpfner, C., & Scherer, D. (2011). Analysis of vegetation and del Castillo, E.M., García-Martin, A., Aladrén, L.A.L., & de Luis, land cover dynamics in north-western Morocco during M. (2015). Evaluation of forest cover change using remote the last decade using MODIS NDVI time series data. sensing techniques and landscape metrics in Moncayo Biogeosciences, 8, 3359–3373. Natural Park (Spain). Applied Geography, 62, 247–255. Karjalainen, E., Sarjala, T., & Raitio, H. (2010). Promoting Eva, H., Carboni, S., Achard, F., Stach, N., Durieux, L., Faure, human health through forests: Overview and major J.-F., & Mollicone, D. (2010). Monitoring forest areas from challenges. Environmental Health and Preventive Medicine, continental to territorial levels using a sample of medium 15(1), 1–8. spatial resolution satellite imagery. ISPRS Journal of Kavzoglu, T. (2017). Object-oriented random forest for Photogrammetry and Remote Sensing, 65, 191–197. high resolution land Cover mapping using quickbird-2 FAO. (2010). Global Forest Resources Assessment 2010. imagery. Handbook of neural computation (pp. 607–619). Retrieved from http://www.fao.org/docrep/013/i1757e/ Amsterdam: Elsevier. i1757e00.htm Kissinger, G., Herold, M., & De Sy, V. (2012). Drivers of FAO. (2015). Global Forest Resources Assessment 2015 – How deforestation and forest degradation: A synthesis report are the world’s forests changing? (Food and Agriculture for REDD + policymakers. Vancouver Lexeme Consulting. Organization of the United Nations). Retrieved from Laakili, A., Belkadi, B., Gaboun, F., Yatrib, C., Makhloufi, http://www.fao.org/resources/infographics/infographics- M., EL ANTRY, S.A.L.W.A., Medraoui, L., Laamarti, A., details/en/c/325836/ & Filali-Maltouf, A. (2016). Analysis of dendrometric FAO. (2016). State of the World’s Forests (2016). State of the diversity among natural populations of cork oak (Quercus World’s Forests. Forests and agriculture: Land-use challenges suber L.) from Morocco. Turkish Journal of Agriculture and and opportunities. Rome, Italy: FAO. Forestry, 40, 127–135. Gamfeldt, L., Snäll, T., Bagchi, R., Jonsson, M., Gustafsson, L., Lahlaoi, H., Rhinane, H., Hilali, A., Lahssini, S., & Moukrim, & Kjellander, P. (2013). Higher levels of multiple ecosystem S. (2017). Desertification assessment using MEDALUS services are found in forests with more tree species. Nature model in Watershed Oued El Maleh, Morocco. Geosciences, Communications, 4, 1340. 7(3), 50. Girard, M.C., & Girard, C.M. (2010). Traitement des Linares, J.C., Taïqui, L., Sangüesa-Barreda, G., Seco, J.I., & données de télédétection (2è édition) [Remote sensing data Camarero, J.J. (2013). Age-related drought sensitivity of processing (2nd edition)] (p. 553). Paris, France: Dunod. Atlas cedar (Cedrus atlantica) in the Moroccan Middle Goward, S.N., Markham, B., Dye, D.G., Dulaney, W., & Atlas forests. Dendrochronologia, 31, 88–96. Yang, J. (1991). Normalized difference vegetation index GEOLOGY, ECOLOGY, AND LANDSCAPES 215 Mancino, G., Nolè, A., Ripullone, F., & Ferrara, A. (2014). Landsat remote sensing and geographic information systems to TM imagery and NDVI differencing to detect vegetation assess landscape dynamics – Application to a wooded area change: Assessing natural forest expansion in Basilicata, in the Benslimane Province (Western Morocco)]. Revue southern Italy. iForest - Biogeosciences and Forestry, 7, 75. forestière française, 60(5), 657–666. Melville, B., Lucieer, A., & Aryal, J. (2018). Object-based Safriel, U., Adeel, Z., Niemeijer, D., Puigdefabregas, J., White, random forest classification of Landsat ETM+ and R., Lal, R., … McNab, D. (2005). Dryland systems. Current WorldView-2 satellite imagery for mapping lowland state and trends (Vol. 1, pp. 623–662). Washington DC: native grassland communities in Tasmania, Australia. Island Press, Millennium Ecosystem Assessment. International Journal of Applied Earth Observation and Salehi, A., & Karltun, L.C. (2010). Livelihood dependency Geoinformation, 66, 46–55. on woodland resources in southern Zagros, Iran Caspian. Mohajane, M., Essahlaoui, A., Oudija, F., El Hafyani, M., & Journal of Environmental Sciences, 8, 181–194. Cláudia Teodoro, A. (2017). Mapping forest species in the Seabrook, L., McAlpine, C., & Fensham, R. (2006). Cattle, central middle atlas of Morocco (Azrou Forest) through crops and clearing: Regional drivers of landscape change remote sensing techniques. ISPRS International Journal of in the Brigalow Belt, Queensland, Australia, 1840–2004. Geo-Information, 6(12), 275. Landscape and Urban Planning, 78, 373–385. Montès, N., Bertaudière-Montes, V., Badri, W., Zaoui, E.H., Sexton, J.O., Urban, D.L., Donohue, M.J., & Song, C. (2013). & Gauquelin, T. (2002). Biomass and nutrient content of Long-term land cover dynamics by multi-temporal a semi-arid mountain ecosystem: The Juniperus thurifera classification across the Landsat-5 record. Remote Sensing L. woodland of Azzaden Valley (Morocco). Forest Ecology of Environment, 128, 246–258. and Management, 166, 35–43. Slimani, M.A., El Aboudi, A., Rahimi, A., & Khalil, Z. (2017). Murray, N.J., Keith, D.A., Bland, L.M., Ferrari, R., Lyons, Use of GIS and Satellite Imagery in the Study of the Spatial M.B., Lucas, R., & Nicholson, E. (2018). The role of satellite Distribution of Vegetation in the Entifa Forest (High Atlas remote sensing in structured ecosystem risk assessments. Central, Morocco). Euro-Mediterranean Conference for Science of The Total Environment, 619–620, 249–257. Environmental Integration. Nagendra, H., Lucas, R., Honrado, J.P., Jongman, R.H.G., Wang, Q., Adiku, S., Tenhunen, J., & Granier, A. (2005). Tarantino, C., Adamo, M., & Mairota, P. (2013). Remote On the relationship of NDVI with leaf area index in a sensing for conservation monitoring: Assessing protected deciduous forest site. Remote Sensing of Environment, 94, areas, habitat extent, habitat condition, species diversity, 244–255. and threats. Ecological Indicators, 33, 45–59. Wani, A.A., Joshi, P.K., Singh, O., & Sha, fi S. (2016). Multi- Pettorelli, N., Vik, J.O., Mysterud, A., Gaillard, J.M., Tucker, temporal forest cover dynamics in Kashmir Himalayan C.J., & Stenseth, N.C. (2005). Using the satellite-derived region for assessing deforestation and forest degradation Normalized Difference Vegetation Index (NDVI) to assess in the context of REDD+ policy. Journal of Mountain ecological effects of environmental change. Trends in Science, 13, 1431–1441. Ecology & Evolution, 20, 503–510. Warner, T.A., Skowronski, N.S., & Gallagher, M.R. (2017). Rejdali, M. (2004). Forest cover changes in the Maghreb High spatial resolution burn severity mapping of the countries with special reference to Morocco. Recent dynamics New Jersey Pine Barrens with WorldView-3 near-infrared of the Mediterranean vegetation and landscape, 21–31. and shortwave infrared imagery. International Journal of Röder, A., Udelhoven, T., Hill, J., del Barrio, G., & Tsiourlis, G. Remote Sensing, 38(2), 598–616. (2008). Trend analysis of Landsat-TM and-ETM+ imagery Xie, Y., Sha, Z., & Yu, M. (2008). Remote sensing imagery in to monitor grazing impact in a rangeland ecosystem in vegetation mapping: A review. Journal of Plant Ecology, 1, Northern Greece. Remote Sensing of Environment, 112, 9–23. 2863–2875. Yang, Q., Cogswell, M.E., Flanders, W.D, Hong, Y., Zhang, Z., Rotenberg, E., & Yakir, D. (2010). Contribution of semi-arid Loustalot, F., Gillespie, C., Merritt, R., & Hu, F.B. (2012). forests to the climate system. Science, 327, 451–454. Trends in cardiovascular health metrics and associations Saber, E.-R., Rhazi, M., Rhazi, L., & Ballais, J.-L. with all-cause and CVD mortality among US adults. (2008). Utilisation de la télédétection et des systèmes JAMA, 307, 1273–1283. d’informations géographiques pour l’évaluation de la Zak, M.R., Cabido, M., & Hodgson, J.G. (2004). Do dynamique du paysage: Cas d’une zone boisée de la subtropical seasonal forests in the Gran Chaco, Argentina, province de Benslimane (Maroc occidental) [Using have a future? Biological Conservation, 120, 589–598.
Journal
Geology Ecology and Landscapes – Taylor & Francis
Published: Jul 3, 2018
Keywords: Béni-Mellal Province; forest stand changes; ASTER and Sentinel-2A data; supervised classification