Spatio-temporal Bamboo Forest Dynamics in the Lower Beles River Basin, North-western Ethiopia

Introduction: Bamboo forests, which are an integral part of the eco-system and an important source of socio-economic life for rural communities in the vast savannah lowlands of Ethiopia, are experiencing signicant changes. Therefore, examining bamboo forest cover changes and identifying responsible drivers for the changes are of the greatest importance for sustainable management of these useful resources. This study was intended to examine the spatiotemporal bamboo forest dynamics in the Lower Beles River Basin, north-western Ethiopia. Method: A combination of pixel-based hybrid classication techniques and Normalized Difference Vegetation Index (NDVI) was employed to analyze bamboo forest cover changes from 1985 – 2019. Focus group discussions, questionnaire, key informant interview and observation were used to identify the drivers for bamboo forest cover change. Results: The study ndings indicate that bamboo has experienced signicant spatio-temporal change over the study period (1985 - 2019) in the study District. In the base year (1985), bamboo covered 5.1% (5,277.1 ha) of the study area. Signicant decline in bamboo forest had occurred in 2001 when the bamboo forests suffered the greatest devastation and shirked to 1.6%. In 2019, the bamboo has been rehabilitated from degradation and increased signicantly. However, the net change over the study periods was negative where the bamboo forests declined by 0.8% (831.14 ha). Conclusion: In conclusion, the observed changes in bamboo forest cover was driven by an interplay of multiple factors. Agricultural land expansion, wildre, free grazing, lack of regulatory mechanisms and improper harvesting and expansion of settlement areas were the top ve drivers respectively while conict, mass owering and tenure contributed for the change. Therefore, the bamboo forests deserve great attention, and the results from this study imply the need for the concerted efforts of stakeholders for sustainable management, utilization and conservation of the bamboo resources. and 4.36%, respectively. In the second change analysis period (2001 - 2019), the bamboo forest increased from 1.6% (1,675.15 ha) to 4.25% (4,446.11 ha) at a rate of 2.65% (2,771 ha). However, over the study period (1985 – 2019), the bamboo forest declined by 0.8% (831.14 ha). Generally, results from this study revealed that bamboo has experienced signicant spatial and temporal changes over the last three and half decades (1985 - 2001) in the study District. concerted efforts of stakeholders for sustainable management, utilization and conservation of the bamboo resources. Improving bamboo forest protection, community awareness, introduction of controlled grazing system, and incentive and motivation for bamboo growers are among the possible solutions forwarded to conserve and sustain the bamboo resources.


Introduction
Bamboos are crucial vegetation resources mostly distributed in tropical and sub-tropical areas of Asia, Africa and Latin America (Ben-zhi et al. 2005;Zhang et al. 2014;Zhuang et al. 2015), covering 33 million ha area and accounting for about 1.0% of the forest area in the world (Zhou et al. 2011;Du et al. 2018). They are unique in their capacity to meet a wide range of socio-economic, sustainability, and conservation focused objectives (Sohel et al. 2015;Ling et al. 2016;Li et al. 2020), owing to their rapid growth, ease of propagation, and the range of ecosystem services they provide (Zhou et al. 2011;Nath et al. 2015;Li et al. 2016). Bamboos have very crucial ecological and environmental functions in soil and water conservation, land rehabilitation, and carbon sequestration due to their biological characteristics (Ben-zhi et al.s 2005;Yiping et al. 2010;Akinlabi et al. 2017). Therefore, mapping the spatio-temporal bamboo forest change is critical to devise strategies for sustainable use of bamboo resources (Li et al. 2020).
In Africa, Ethiopia stands in the rst position for its bamboo forest resources. The country possesses about 67and 7% of the total bamboo forest areas in Africa and the world, respectively (Endalamaw et al. 2013;Mekonnen et al. 2014;TsinghuaUniversity and INBAR 2018). It has two indigenous species of bamboo, namely Yushania alpina (K. Schum.) W.C. Lin (highland bamboo) and Oxytenanthera abyssinica (A. Rich) Munro (lowland bamboo) with area coverage of more than one million ha (Embaye 2000;Desalegn and Tadesse 2015;Zhao et al. 2018). The focus of the present study is on O. abyssinica, which grows in the vast western savannah lowlands along major river valleys bordering Sudan (Kelbessa et al. 2000), and constitutes about 85% of the total bamboo forest in the country (Sertse et al. 2011;Endalamaw et al. 2013;Desalegn and Tadesse 2015;Boissière et al. 2019).
However, reliable and consistent data on bamboo resources are limited. Different inventories conducted so far has been reporting different gures about bamboo resources. In 1960s, the total area of bamboo was estimated at 1.5 million ha (Lobovikov et al. 2007). However, a study conducted by GIZ-LUSO Consult Gmbh (1997) reported that there is 700,000 -850,000 ha of lowland bamboo and 129,626 ha of highland bamboo in Ethiopia. The International Bamboo and Rattan Network, currently, Organization (INBAR), production to consumption study (Kelbessa et al. 2000), reported the availability of 1.1 million ha of bamboo: 950,000 ha of lowland bamboo and 150,000 highland bamboo. Lobovikov et al. (2007) estimated that the bamboo forest area was about 849,000 ha in 2005. A recent remote sensing-based study conducted by Zhao et al. (2018) reported that Ethiopia possess 1.47 million ha of bamboo resources.
From the above reports, it is possible to infer that the bamboos have been undergone with spatial and temporal changes. Over the years, the bamboo forests have been under high pressure and suffering great depletion. The expansion of agricultural land and changing bamboo stands to other land uses, free grazing and re hazards are important anthropogenic factors causing depletion of bamboos (Kindu and Mulatu 2009;INBAR 2010;Desalegn and Tadesse 2015;Boissière et al. 2019), when gregarious owering is the major natural cause contributing for their degradation (Sertse et al. 2011). Nevertheless, little efforts were made so far to counterbalance the problem and reliable and consistent data on the status of the bamboo resource base as well as the policy options for sustainable bamboo resource use and development are scanty (Tsinghua University and INBAR 2018;Boissière et al. 2019;EFCC 2020).
The use of remote sensing data has been instrumental in monitoring the changing pattern of vegetation across diverse landscapes. Hence, several studies have been conducted to identify and map bamboo forest cover using remotely sensed data at global scale (Wang et al. 2009;Goswami et al. 2010;Xu et al. 2012;Carvalho et al. 2013;Han et al. 2014;Li et al. 2016;Fava and Colombo 2017;Du et al. 2018;Liu et al. 2018;Zhao et al. 2018;Zhang et al. 2019) However, studies conducted at the local level are of critical importance to better understand the cumulative impact at multiple scales (Sohl et al. 2004;Shah and Sharma 2015).
Hitherto, very little scientific studies have been conducted on spatio-temporal cover change of bamboo forests (Bessie et al. 2016;Zhao et al. 2018). Lack of reliable and consistent data on bamboo resources has impeded the proper management bamboo forests deserve and limited their potential in providing ecological and socio-economic services (Nat h et al. 2015; Li et al. 2016Li et al. , 2020Liu et al. 2018). In the milieu of sever bamboo resource degradation, on one hand, and the availability of abundant resource base, on the other hand, some authors deem that more research is critical in the lowlands of north-western Ethiopia, where the country's largest bamboo resource base is found. The rationale behind such scienti c studies is that empirical evidence on spatio-temporal bamboo forest cover change is a decisive instrument to devise strategies for sustainable bamboo resource management. Therefore, the objectives of this study were: (i) to examine the spatial and temporal changes of the O. abyssinica forest cover between 1985 -2019, and (ii) to identify drivers of the changes of O. abyssinica forest cover in the Lower Beles River Basin, north-western Ethiopia.

Materials And Methods
The study area The study area, Mandura District is one of the bamboo producing areas in Lower Beles River Basin, north-western Ethiopia. The District located between 10º 50' 55" to 11º 10' 10" N and 36º 02' 48" to 36º 32' 42" E, with a total area of 1,045 km 2 . It is part of the north-western lowlands where elevation ranges from 800 to 2186 m (Fig. 1). The basement complex rock covers the vast areas where, few areas, particularly, the Kar Mountain escarpments, situated in the northeastern part of the District are covered with plateau basalts (Addisu 2010). Nitosols, cambisols, luvisols and leptosols are the dominant soils in the sub-basin (Yilma and Awulachew 2009).
In the District, there are both perennial and seasonal rivers and streams that ow towards Gilgel Beles River, which is main tributary of River Beles (Fig. 1). The District shows a unimodal rainfall distribution pattern and ranges approximately, from 1,052 to 1,957 mm whereas temperature ranges between a maximum of 35 to 40 °C and a minimum of 18 to 20 °C (Nyssen et al., 2018). Generally, climate of the District is classi ed as wet tropical (wet kolla) except small areas around the Kar Mountain, which is wet sub-topical (wet woynadega).
The Gumuz, Amhara, Agew and Shinasha are the major ethnic groups in the area while the Oromo represent a minority. About 81.5% of the population is rural

Data sources
The study used both spatial and non-spatial (socio-economic) datasets. These datasets were collected from various sources, which include: Landsat imageries, Google Earth images, household survey questionnaire, focus group discussion, key informant interview and led observation. Other datasets were collected from government reports, published literature and policy documents.
The remote sensing data and techniques Satellite imageries of Landsat 5 (TM), Landsat 7 (ETM+), and Landsat 8 (OLI_TIRS) were taken for the year 1985, 2001 and 2019 respectively (Table 1). All Landsat images were taken during leaf-off season (Wang et al. 2009;Zhao et al. 2018). The reasons behind for choosing of years and seasons were: (1) the availability of cloud free (below 10% cloud cover) and good quality satellite images; and (2) the dry season is a proper time to spectrally distinguish bamboo from leaf-shedding vegetation and co-occurring grasses and shrubs. In addition, the Advanced Space Borne Radiometer (ASTER) Global Digital Elevation Model (GDEM), ASTER GDEM, acquired from USGS Earth explorer (http://gdex. cr.usgs.gov/gdex), was used to determine the elevation of the study area.
Image pre-processing In order to maximize classi cation accuracy, input images must have minimal contamination from clouds, haze, shadow, or other disturbances (Azzari and Lobell 2017). Thus, haze reduction and atmospheric correction was conducted on satellite imageries using Atmospheric and Topographic Correction (ATCOR) software which is an add-on module to ERDAS (Wang et al. 2009). Following the availability of panchromatic data, pan-sharpening or image fusion was 2015) was employed. Primarily, the unsupervised classi cation was carried out in Iterative Self Organizing (ISO) data analysis algorithm (Lillesand and Kiefer 2007) to determine the number of land use/land cover classes (LULCC) in the study area.
However, mapping bamboo forests using the spectral approach alone is challenging due to the similarity of spectra between bamboo and other forest types (Wang et al. 2009;Zhang et al. 2019). To offset the problem, phenological information plays a crucial role in bamboo extraction (Zhao et al. 2018;Zhang et al. 2019). Similarly, it was noted that taking into account phenology when performing land cover classi cation yield more accurate maps (Knight et al. 2006;Simonetti et al. 2014;Fan et al. 2015). In this regard, remote sensing phenology studies use data gathered by satellite sensors that measure wavelengths of light absorbed and re ected by green plants, and transform raw satellite data of these light waves into vegetation indices (Simonetti and Simonetti 2014).
In remote sensing scene, there are several vegetation indices highlighting vegetation bearing area (Gandhi et al. 2015). The most widely used indices in research on global environmental change, is the Normalized Difference Vegetation Index (NDVI) (Bhandari et al. 2012). It is calculated from the visible red and near-infrared (NIR) light re ected by vegetation as follows (Simonetti et al. 2014): Hence, leaf-off season Landsat 8 imageries spanning from 2015 -2019 were used to compute the NDVI map of the study area. Each feature has its own spectral re ectance, varying according to the wavelength (Simonetti et al. 2014). Therefore, as our aim was to discriminate bamboo form other vegetation, we took 100 points (60 for classi cation and 40 for validation) from bamboo forest prior to classifying the NDVI images. Then, NDVI images were classi ed in to ve land cover classes: non-vegetated land, grass, bamboo, shrub/ bushland and woodland (Fig. 2).
Values in NDVI range between -1 to 1. According to Bhandari et al. (2012), very low value of NDVI (0.1 and below) correspond to barren areas of rock, sand or snow. Moderate values (0.2 to 0.3) represent shrub and grassland, while high value (0.6 to 0.8) indicates temperate and tropical rainforests. Bare soil is represented with NDVI values, which are closer to 0, and water bodies are represented with negative NDVI values. In our study, the average NDVI value of bamboo was found between 0.14 to 0.165, which was less than shrubs, but greater than grasses. When the maximum value of NDVI was low (0.4 -0.58), the NDVI of bamboo was 0.14 -0.162. Whereas, as the value of NDVI reach as high as 0.6 and above, the bamboo had a value of 0.148 -0.165. Given this threshold value, the NDVI maps were produced for the year 1985, 2001 and 2019 (Fig 2). Finally, the NDVI maps of each year was used to extract features to aid and improve the supervised classi cation.
The supervised image classi cation, with maximum likelihood classi cation algorithm (Rogan and Chen 2004), was conducted using training samples collected from different sources. The highest number of samples were taken from bamboo land cover class in order to increase the classi cation accuracy. For 1985 and 2001 imageries, classi cation was undertaken with the help of high resolution Google Earth images, knowledge of elders, NDVI maps and color visualization and interpretation of the raw images (Lossou et al. 2019). For classifying 2019 Landsat imagery, eld survey, NDVI maps and Google Earth datasets were used to collect training samples (Fig 3).

Accuracy assessment
Before proceeding to other activities, it was essential to evaluate the accuracy of the classi cation results (Liu and Mason 2016). Hence, accuracy assessment was conducted with independent samples collected through eld survey, knowledge of elders and ancillary data, namely Google Earth images and NDVI maps, using strati ed random sampling approach (FAO 2016). Then, confusion matrix or error matrix, which is one of most common means of expressing or assessing classi cation accuracy (Kindu et al. 2015;Liu and Mason 2016), was used for accuracy assessment purpose.
An error matrix compares information from reference data to information on the classi ed map for a number of sample area (Congalton 1991). The overall, producer's and user's accuracies were computed from error matrix. In addition, the omission and commission errors of bamboo and non-bamboo classes were calculated to evaluate the performance of the bamboo classi cation. Commission error is the percentage of pixels, classi ed as a certain class, which do not belong to that class according to the reference data. Omission error is the percentage of sample points, belonging to a certain class in the reference data, which were classi ed as other classes (Congalton 1991;Li et al. 2016). In addition, the kappa coe cient (K), the most commonly used statistical measure of classi cation accuracy and quality (Lillesand et al. 2015), was employed. The result of performing a kappa analysis is a KHAT statistic (an estimate of Kappa), which is agreement of accuracy (Congalton 1991).
The KHAT statistic (kappa coe cient) calculated as follows:

Change analysis
Change analysis was carried out using post-classi cation image comparison technique. First, images of each reference years (1985,2001 and 2019) were classi ed separately. Then, change detection tasks were conducted. In order to examine the bamboo forest cover change over the study periods, we carried out change analysis on two intervals: 1985 -2001 ( rst) and 2001 -2019 (second). The percentage of land cover change detection was conducted using the following formula (Kindu et al. 2013): ArcGIS ® 10.5 and Earth Resource Data Analysis System (ERDAS) Imagine ® 2014 were used throughout the image pre-processing, processing, classi cation, post-classi cation as well as production of the nal land cover maps of the study period.

Socio-economic data and methods
A total of 160 respondents were selected for questionnaire using systematic (random) sampling from resident lists found in the sample kebeles (the smallest administrative unit in Ethiopia), and the household head (HH) was interviewed (Creswell 2009). Prior to selecting sample respondents, sample villages or kebeles were selected based on information collected through reconnaissance survey, literature review and expert opinion. Accordingly, four kebeles: Kutirhulet, Deha-anzabuguna, Bahus and Deha-nubeshe were selected based of their bamboo forest coverage, representativeness and accessibility to transportation service.
Close-ended questions were prepared and written in English rst and, later, translated to the Amharic Language. Amharic and Gumzigna (mother tongue language of the Gumuz community) languages were used (as required) to conduct the household surveys, while the data were recorded in Amharic and, then, translated into English. Four language translators were used to translate the Gumzigna language to Amharic and vice-versa. The main theme of the survey was to understand the perception of respondents about bamboo forest cover change and responsible drivers for the change.
Focus group discussions (FGDs) were held with 32 individuals selected from bamboo cultivators, kebele administrators and village level experts of the sample kebeles. The members discussed on issues regarding their views and experience on bamboo forest cover change and responsible drives for the change. During the entire discussion process, the investigator served as a facilitator while insiders fully participated in the dialogue (Lune and Berg 2017). Moreover, 12 elders, whose age was above 60 years old, and lived for long period (more than 35 years) in the study area, were selected as key informants. They were selected based on recommendations of village-level experts and o cials.

Data analysis
Both qualitative and quantitative methods of data analyses were employed in the study (Creswell 2009). Qualitative technique of content analysis (Lune and Berg 2017) was used to analyze data generated through key informant interview, eld observation and FGDs. In this method, the data were categorized in different themes and coded, organized categorically and chronologically, reviewed repeatedly and continually coded. Then, a list of major ideas was chronicled and transcribed to verbatim. Moreover, descriptive statistics of percentages and averages were used for describing sample respondents' responses and to summarize results generated through GIS and remote sensing techniques.

Accuracy assessment results
According to accuracy assessment result (Table 3) More speci cally, as shown on the confusion matrix report (Table 4), 17.1% of reference points corresponding to bamboo were misclassi ed as grassland (11%) and shrub/ bushland (6.1%) in the 1985 map. Correspondingly, bamboo had the lowest producer's and user's accuracy values accounted for 82.9% and 81.9% respectively. This is due to the fact that major data used in this study was the medium resolution (30 m) Landsat images, which is too coarse for mapping bamboo forest areas as they are often and confused and scattered with other forests (Zhao et al. 2018;Huy and Long 2019). Therefore, to offset this problem, image fusion was applied to merge the high resolution (15m) Table 3). The user's and producers' accuracies were 85.5 and 86.6% in 2001, and 87.1 and 88.1% for 2019 maps, respectively. The EC of bamboo were 12.2 and 12.9%, and the EO accounted for 11.1 and 11.9% in 2001 and 2019 maps, respectively. Overall accuracies of maps were between 85.5 and 89.1% with kappa coe cient of 0.83 and 0.87. According to Landis and Koch (1977), the value of kappa coe cient is categorized into three, a value between 0.40, representing poor agreement, between 0.40 and 0.80, representing moderate agreement and above 0.80 representing strong agreement. Therefore, the kappa coe cient results of this study indicated strong agreement for each of the produced maps, and the overall accuracies were in acceptable range to monitor the bamboo forest dynamics over the last 34 years.

Spatio-temporal Bamboo forest cover changes in the study area
In 1985, bamboo covered 5.1% (5,277.1 ha) when non-bamboo vegetation types (woodland, shrub/bushland and grassland) comprised 33% (34,464,7 ha) and non-vegetated land (agricultural land and barren land) covered the largest area (62%) ( Table 5; Fig. 5). The spatial distribution of bamboo was scattered along with grasses, shrubs and forests, where thick bamboo forests were found in the peripheries of north-western, north-eastern and western parts of the District (Fig. 5).
Signi cant decline in bamboo forest had occurred in 2001 when the bamboo forests suffered the greatest devastation and shirked by 3.5% (Table 5). Compared with the bamboo, non-bamboo vegetation exhibited a slight reduction by 1.86%. Conversely, non-vegetated land had increased by 5.4%. The 2001 map clearly illustrates that concentrated bamboo forest was hardly found except at few places in the eastern part of the District. It was rather possible to observe scattered bamboo patches, mostly, at the north-eastern and eastern parts of the study District (Fig. 5). Generally, in the rst change analysis period (1985 -2001), the bamboo forests had declined at a rate of 3.5% (3,602 ha).
The 2019 map (Fig. 5) illustrates a good bamboo forest cover, implying that the bamboo has been rehabilitated from degradation and increased signi cantly. In this period, dispersed bamboo forests were common in most areas of southern, north-western and north-eastern parts, while some dense bamboo forests were found in the peripheries of north-western, central, eastern and north-eastern portions of the District. The bamboo has own 4.25 % of the District, i.e. covering 4,446.11 ha. However, except the grassland, non-bamboo vegetation and non-vegetated land declined by 1.26% and 4.36%, respectively.
In the second change analysis period (2001 -2019), the bamboo forest increased from 1.6% (1,675.15 ha) to 4.25% (4,446.11 ha) at a rate of 2.65% (2,771 ha). However, over the study period (1985 -2019), the bamboo forest declined by 0.8% (831.14 ha). Generally, results from this study revealed that bamboo has experienced signi cant spatial and temporal changes over the last three and half decades (1985 -2001) in the study District.

Drivers of bamboo forest dynamics in the study area
In the present study area, there are many intertwined and connected factors or drivers for bamboo dynamics. Efforts were made to identify drivers of bamboo dynamics and a total of 13 factors were perceived by respondents as main drivers of bamboo dynamics in the study area (Table 7). However, there were variations about each of the drivers to which the local people viewed as important drivers of bamboo forest dynamics. Agricultural land expansion, wild re, free grazing, lack of regulatory mechanisms and expansion of settlement areas were perceived as the top ve drivers respectively.
Agricultural land expansion was viewed as the most important drivers by 85.6% of the respondents. According to the FGD participants and KII, bamboo forest degradation was associated with population pressure. According the CSA, the population in Mandura District has escalated. The population of the study area was 18,017 in 1984, 22,593 in 1994, and 40,746 in 2007. Over the last 23 years, the population was doubled. This implies that as the population is growing over time, demand for agricultural land, bamboo products and settlement increases. agricultural land expansion is a key driver for bamboo degradation in the study area. The FDG participants added that people deforest bamboo forests to expand agricultural land as they prefer the immediate bene t and ignore longterm bene t they could get from bamboo forests.
Induced by population pressure, expansion of settlement areas (68.8%) and increasing demand for bamboo products (63.8%), were also amongst the main drivers mentioned by respondents in the study area.
Wild re was the second most important driver mentioned by 76.9% of respondents in the study area. It was also understood from the account of FGDs participants and eld observation protocol that wild re was one of the major driver for bamboo dynamics. In the study area, very commonly, the community induce re on natural forests including bamboo grooves to stimulate the growth of grass for their livestock and or to open up the forest for hunting. Hence, the re burns bamboo culms and the emerging bamboo shoot which could let the clump ceased.
Moreover, Cattle keeping and grazing system was perceived by 76.3% of the respondents as the third important driver for bamboo dynamics in the study District. The cattle are allowed free to graze on any land use system including bamboo forest. When the cattle go through bamboo stands, they graze the bamboo shoot and leaves which are critical for development of the bamboo plant. Weak regulatory and enforcement was ranked as the fourth driver by 69.4% of the respondents. Similarly, improper harvesting was mentioned by 63.1% participants as driver bamboo forest cover change. In this regard, FGD participants justi ed that there is clear gap on use and management of bamboo forest. They added that most of the bamboo resources of the district are natural or stateowned bamboo forests. However, still now, there is no rule to govern the use of these natural or state-owned bamboo forests, which lead to over exploitation and degradation of bamboo resources.
Con ict 61.9%, mass owering (52.5%) and tenure (43.8%) were viewed as other important drivers of bamboo forest dynamics in the study area. However, fuelwood (7.06%) and charcoal production (5.14%) were perceived by respondents as the least preferred drivers of bamboo dynamics.
As we understood from eld observation protocol and participants re ection during FGDs, con ict and insecurity problem has prohibited bamboo cultivators to control the wild re and keep cattle away from the bamboo stands. Recently, insecurity and con ict are becoming common social problems in the district. Hence, when problem happens, people improperly harvest and loot bamboo culms, bamboo shoot, send their cattle for grazing and sometimes re the bamboo stands. Moreover, they viewed mass owering as important natural factor for bamboo degradation. They added that bamboo owers between 30-35 years, at this time bamboo could regenerate if it is free from animal and human interference though it does not work in the study area.

Discussions
This study revealed that bamboo had experienced signi cant spatio-temporal changes over the last three and half decades (1985 -2019). In the base year (1985), bamboo covered 5.1% (5,277.1 ha) of the study area. Signi cant decline in bamboo forest had occurred in 2001 when the bamboo forests suffered the greatest devastation and shirked to 1.6%. In 2019, the bamboo has been rehabilitated from degradation and increased signi cantly. However, the net change over the study periods was negative where the bamboo forests declined by 0.8% (831.14 ha). This nding is in agreement with previous studies (Embaye 2000;Lobovikov et al. 2007;Emiru and Taye 2012;Mekonnen et al. 2014;Desalegn and Tadesse 2015;Bessie et al. 2016) which reported that bamboo forests in Ethiopia are declining from time to time.
In the study District, a range of drivers were identi ed for bamboo forest cover change occurred over the last three and half decades. The local people had perceived responsible drivers of the change though their level is different. These drivers are broadly classi ed as direct or proximate and indirect or underlying factors. In this study, it was fund that emanated from population pressure, agricultural land expansion, was the most important driver of bamboo forest cover change. In line with this, similar studies conducted so far (Embaye 2000;INBAR 2010;Bessie et al. 2016;Zhao et al. 2018;Boissière et al. 2019) reported that expansion of agricultural land was the man driver for bamboo forest cover change in lowlands of North-western Ethiopia.
Wild re was identi ed as the second most important driver of bamboo forest change. In the study district, people induce re to stimulate the growth of grasses for their livestock and or to open up the forest for hunting. At that time, the re burns bamboo culms and the emerging bamboo shoots, which could let the clump ceased. Fire has devastating effect if it occurs after owering of bamboo as it also burns bamboo seeds. According to Mulatu et al. (2016) seedling establishment is unlikely on burnt sites since the re affects both the soil seedbank and the regenerated seedlings. In this regard, Embaye (1998); Kindu and Mulatu (2009);Sertse et al. (2011);Boissière et al. (2019) also documented that re is one of the principal factor for bamboo forest cover change in Ethiopia.
Cattle keeping and grazing system was one of the most signi cant drivers for bamboo forest cover change in the study area. The cattle are allowed to graze freely on any of land uses including bamboo stands. When the cattle go through bamboo stands, they graze bamboo shoots and lives, which are critical for the development of the plant. Thus, free grazing has continued as a main driver for bamboo forest degradation in the study area. This nding is in agreement with the ndings of Kindu and Mulatu (2009);Bessie et al. (2016); Durai et al. (2018) that found open grazing system as in uential factor for bamboo forest degradation.
Increasing demand for bamboo products and expansion of settlement area driven by population growth were amongst the main drivers for bamboo forest cover change. According to the CSA, the population of the study area was 18,017 in 1984, 22,593 in 1994, and 40,746 in 2007. Over the last 23 years, the population was doubled. As the population grew over time, demand for bamboo products and settlement plot increased, which subsequently led people to clear bamboo forests for construction material and more space for resident plots. In connection with this, Zhao et al. (2018) noted the natural bamboo forests are under threat from deforestation and degradation for people are clearing them away to have more space for agricultural activities and residential plots.
Con ict and insecurity problem was a major driver that contributed for bamboo forest cover change. It had triggered other drivers like re and free grazing to cause a change in bamboo forest cover. According to the FGD participants, the prevalence of con ict and insecurity risk had prohibited bamboo cultivators to control re and keep cattle away from the bamboo stands. Recently, insecurity and con ict are becoming common social problems in the study district. Hence, when problem happens, people improperly harvest and loot bamboo stands, bamboo shoots; send their cattle for grazing and sometimes re bamboo stands.
Policy related factors like weak forest protection and tenure were among the underlying driving forces for bamboo forest cover change. In the study district, there is no strict rule to govern the use and management of bamboo forests, mainly, the natural bamboo forests. Hence, natural bamboo forests are considered as common pool resources which lead to overexploitation and degradation of bamboo. However, compared to the natural bamboo forests, bamboo grown on private land are protected. In this regard, Durai et al. (2018); Zhao et al. (2018); Boissière et al. (2019), reported that coupled with other anthropogenic factors, the absence of regulation and enforcement mechanisms triggered the degradation of bamboo in north-western Ethiopia.
Improper harvesting was the other cause for bamboo cover change. During ied observation, we found that bamboo management in the study area was improper and unsustainable. There was an instance of harvesting very young bamboo culm while many old bamboo culms lie down on bamboo clumps. During FGDs, participants justi ed that they have knowledge gap on bamboo resource management. They relied on the knowledge they gained from their fathers and forefathers and personal experiences. In this regard, Embaye (2000); Durai et al. (2018) reported that lack of awareness and dearth of scienti c knowledge on its growth, management, harvesting, processing and utilization are major causes for bamboo degradation.
Mass owering was an important natural phenomena causing bamboo forest cover change in the study landscape. According FGD participants, lowland bamboo owers between 30 to 35 years. At this time, it has to be kept free from animal and human interference to get regenerated. However, people tend to send their cattle for grazing, and otherwise make a re on it. For this reason, bamboo regeneration after mass owering was limited in the study area. This nding is in agreement with Embaye (2000); Sertse et al. (2011);Durai et al. (2018); Dereso (2019) which reported that the recent gregarious owering and die back of bamboo, and lack of efforts to assist regeneration, coupled with anthropogenic forest re and free grazing has resulted in poor regeneration and degradation and loss of bamboo forests.
In the study District, fuelwood and charcoal production were the least signi cant drivers for bamboo dynamics. As we understood from portfolio of FGDs, the local people did not prefer bamboo for fuelwood and charcoal. Although they are being depleted, there are natural forests that have higher biomass accumulation to use as sources of charcoal and fuelwood. For this reason, fuelwood and charcoal production are insigni cant drivers for bamboo forest cover change in the study area. This nding disagrees with the ndings of (Bessie et al. 2016) which reported that rewood and charcoal production are the main drivers for bamboo forest degradation in lowlands of north-western Ethiopia.

Conclusions
Understanding the bamboo forest cover change has enormous potential to facilitate sustainable management of the resource. This study examined the spatio-temporal bamboo forest cover change in Lower Beles River Basin, North-western Ethiopia for the last 34 years (1985 -2019). Under data scarce environment, we employed simple, but critical techniques to offset the limitations of the coarse resolution Landsat image, and improve the accuracy of mapping of bamboo. First, we integrated vegetation indices, Normalized Deference Vegetation Index (NDVI) technique with supervised (pixel-based) classi cation scheme. The NDVI technique, with different threshold values, was employed to extract features. Secondly, image-fusion was applied to merge the high-resolution panchromatic image (band 8) of Landsat 7 and 8 with the multi-spectral images to have an image with better quality. Moreover, as bamboo forests are, often, scattered with other vegetation types, we took a large number of samples from bamboo land cover class to improve the classi cation accuracy.
According to the resultant maps, bamboo has experienced signi cant spatio-temporal change over the study period (1985 -2019) in the study District. The bamboo forests dwindled substantially between 1985 and 2001, speci cally during 2001. During this period, the bamboo forests were devastated by 3.4% and the District lost 3,602 ha of bamboo resource. In 2019, the bamboo has been rehabilitated from degradation and increased signi cantly. However, the net change over the study periods was negative where the bamboo forests declined by 0.8% (831.14 ha). The observed changes in bamboo forest cover was driven by an interplay of multiple factors. Agricultural land expansion, wild re, free grazing, lack of regulatory mechanisms and improper harvesting and expansion of settlement areas were the top ve drivers respectively while con ict, mass owering and tenure contributed for the change.
Generally, the bamboo forests deserve great attention, and the results from this study imply the need for the concerted efforts of stakeholders for sustainable management, utilization and conservation of the bamboo resources. Improving bamboo forest protection, community awareness, introduction of controlled grazing system, and incentive and motivation for bamboo growers are among the possible solutions forwarded to conserve and sustain the bamboo resources.

Declarations
Authors contribution: All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Shiferaw Abebe, Amare Sewnet Minale and Demel Teketay . The rst draft of the manuscript was written by Shiferaw Abebe and all authors commented on the previous version of the manuscript. All authors read and approved the nal manuscript.
Funding: This work was nancially supported by the International Foundation for Science (IFS), Stockholm, Sweden (grant number D/6296-1). Land predominantly covered with bamboo forests. It includes bamboo patches that are waiting for restoration or restocking. In some cases, bamboos are mixed with grasses when they are in the rst stages of restoration.

Shrub/ Bush land
It is a land cover less dense than woodland and covered by small trees, bushes, and shrubs two up to ve meters tall and a canopy cover of more than 20%. In some cases, such land covers are mixed with grasses.

Woodland
Land covered by an open stand of trees taller than ve up to 20 m with a canopy coverage of more than 20%.

Grassland
Small grasses are the predominant natural vegetation. It also includes land with scattered or patches of trees and a land cover used for grazing and browsing.
Barren land Area that has little or no vegetation cover, which is, mainly, covered by bare soil, rock out crops and land covered by structures, including roads, towns and rural villages.

Agricultural land
Contiguous areas allotted to extended rain-fed crop production, and where, mostly, oil seed, cereals and pulses are managed.
Source: Adapted from Emiru and Taye (2012)    Photos showing bamboo forests: Field photos (a, b and c) taken on bamboo forest; and Photos taken (d, e, f and g) from high resolution Google Earth images.

Figure 3
Photos showing bamboo forests: Field photos (a, b and c) taken on bamboo forest; and Photos taken (d, e, f and g) from high resolution Google Earth images.

Figure 4
Flowchart showing the procedures employed to prepare the nal LULC map Figure 4 Flowchart showing the procedures employed to prepare the nal LULC map Land cover map of the study area (1985, 2001 and 2019)