Consistent Changes in Land-Use/Land-Cover in Semi-Arid Areas: Implications on Ecosystem Service Delivery and Adaptation in the Limpopo Basin, Botswana

Abstract

Ecosystems in semi-arid areas remain essential to securing livelihoods and aiding climate change adaptation. However, land-use and land-cover change (LULCC) is the leading driver of biodiversity, ecosystem services, habitat, and ecosystem loss in most rural areas of developing countries. We evaluated LULCC in the Bobirwa sub-district of Botswana between 1995 and 2015. We employed the supervised classification’s maximum likelihood algorithm on the 1995, 2005, and 2016 Landsat images to establish the implications of LULCC on the delivery of provisioning ecosystem services (ES) and ecosystem-based adaptation in the Limpopo Basin part of Botswana. Five major LULC classes—vegetation, cropland, bare land, built-up areas, and water bodies—were identified in the sub-district. The decline in vegetation by 50.67 km²/year between 1995 and 2016 was characterized by an increase in croplands (34.02 km²/year). These changes were attributed to the growing human population that induced farming households to expand croplands. Government programs also encouraged agricultural expansions by offering free inputs and compensating smallholder farmers for land preparation. Higher agricultural yields remained critically low while the loss of vegetated areas to croplands threatened biodiversity, habitats, and the sustainability of provisioning ES through impaired ecosystem functions. There is an urgent need to arrest all unnecessary agricultural expansions and enhance agricultural productivity from current land parcels. The government and other relevant stakeholders also need to strengthen the ecosystem management capacities of local communities and support them to develop and implement biodiversity-based village action plans. Engaging communities through participatory, biodiversity-based action planning promotes biodiversity conservation and the sustainable use of ecosystem resources.

1. Introduction

Land-use and/or land-cover change has been identified as one of the most important drivers of change in ecosystems and their services. Changes in land use highlight the influence of human activities that have a direct impact on the delivery of ecosystem services from the landscape [1,2]. Economic growth and development often trigger such changes and result in several trade-offs with human dependence on the natural environment [3]. In recent years, humans have exerted considerable pressure on the natural environment. For instance, the growing human population has increased the expansion of human settlements [1,4,5]. This often involves the clearing of naturally vegetated areas, which fragments local ecosystems, causes loss of biodiversity, and interferes with several ecosystem functions—both underpin ecosystem service delivery—and often exceed their ability to regenerate, resulting in loss of some ecosystems and ecosystem services [6,7].

In some places, excessive human exploitation continues to threaten the sustainability of local ecosystems [2]. However, as shown elsewhere, effective policies and management regimes protect and conserve local ecosystems while also promoting sustainable utilization of their services. In Botswana, the protected areas, forest reserves, and wildlife management areas in the north, including the Chobe National Park, have not only conserved local ecosystems but have also ensured a constant delivery of ecosystem goods and services to the surrounding communities [8].

Several international agreements, such as the United Nations Sustainable Development Agenda and the Convention on Biological Diversity, have been critical in guiding and encouraging countries to strike a balance between development and environmental protection [9,10]. Moreso, the Intergovernmental Platform on Biodiversity and Ecosystem Services (IPBES) provides a platform for governments to improve the interface between science and policy on issues of biodiversity and ecosystem services [11]. Nonetheless, the diversity of landscapes, climate, ecosystem services, cultures, social values, legislation, and the state of development render different land-use and/or land-cover types to be of varying importance to local people [12].

Previous studies in the Limpopo Basin of Botswana have established the importance of provisioning ecosystem services (ES) such as fresh water, Mopane caterpillars, cultivated crops, natural pastures, firewood, medicinal plants, and wild fruits to rural livelihoods and climate change adaptation [13,14,15]. The Millennium Ecosystem Assessment defined and popularized provisioning ecosystem services as tangible products from ecosystems [16]. The exploitation of, and trade in, certain provisioning services form a common yet vital part of livelihood strategies the world over. Botswana’s National Adaptation Plan (NAP) Framework also acknowledges the importance of ecosystem-based adaptations in the NAP process [17]. Moreover, indigenous people living and depending on the natural environment in Botswana and elsewhere further reveal the importance of local ecosystems as a key livelihood and adaptation option given the susceptibility of rain-fed agriculture. However, the incremental climate change impacts as global average temperatures are set to warm beyond the 1.5 °C threshold of the Paris Agreement are expected to further limit the potential of ecosystems in semi-arid areas [18,19,20]. Without studies that quantify LULCC, especially in rural areas, understanding and redressing the simultaneous threats posed by climate change and anthropogenic pressure on provisioning ES becomes limited.

The nexus between land use, ecosystem services, and climate change adaptation has become more critical for researchers in recent years. Liao et al. found that land use/cover, climate (annual average rainfall, etc.), and human disturbance are among the key factors affecting ecosystem service provision in their study of the Northeast Forest Belt in China [12]. Another study by Biratu et al. evaluated the impact of landscape management scenarios on ecosystem service values in Central Ethiopia and found that cultivated land increased by 22% under the business-as-usual (BAU) scenario, whereas forest, water body, wetland, and shrubland were reduced [3]. The study, however, showed that forest cover increased, while bare land was reduced by 21% and 25% under the Ecosystems Protection and Agricultural Development (EPAD) and the Landscape Ecosystems Restoration and Conservation (LERC) scenarios. Chen et al. examined the impacts of land use changes on five typical ecosystem services (grain production, water yield, soil conservation, habitat quality, and carbon sequestration) in Guangdong province of China and concluded that land use and its changes had a significant impact on ecosystem services [7]. Mugari et al. note that climate change and anthropogenic pressure are increasingly modifying and/or interfering with ecosystem functions. This limits the delivery of ecosystem products, threatens livelihoods, and weakens the adaptive capacity of people in developing countries where direct dependence on the natural environment remains very high [15].

Although changes in land use are mainly driven by policies and management regimes in the short term, long-term benefits from land use changes can in turn influence decision-makers and policies [7]. However, information on the consequences of land use change, for instance on ecosystem services and climate change adaptation, is largely lacking at local scales [1]. Thus, examining the consequences of changes in land use on the delivery of ecosystem services and adaptation can provide the necessary evidence to inform policies and decisions that contribute to sustainable adaptation and development.

Various techniques have been used to detect LULCC, including land cover classification for multiple periods, change vector analysis, and normalized index differencing techniques [21,22,23]. Freely available remotely sensed data allows similar analyses to be conducted inexpensively in data-poor areas to aid decision-making. Landsat data provides high spatial resolution images where variations can be observed at a scale of 30 m [24]. Supervised land classification techniques have been used extensively to classify land and detect changes in LULC over time [23,25,26].

A serious challenge to the sustainable management of local ecosystems, adaptation, and development in data-scarce regions, is how the required evidence is derived and utilized [1]. However, in order to proffer solutions to societal challenges, the process needs to engage local communities to co-produce the required knowledge. Cognisant of this, we have attempted to plug this knowledge gap, including how evidence for supporting decisions can be derived in data-scarce areas, by quantifying land-use and land-cover (LULC) change at a local scale, and engaging surrounding communities to identify bundles of ecosystem services derived from different LULC types. Similar to the studies mentioned above, we also apply the supervised classification technique to quantify LULC changes. We then discuss the consequences of recent changes in different LULC types on the delivery of provisioning ecosystem services and adaptation to climate change. The analyses in this study thus aim to provide the evidence needed to support local-level decisions and planning processes that can also contribute to the sustainable management of local ecosystems, adaptation, and development in Botswana.

The analyses in this study were guided by the following research question: What are the implications of recent changes in LULC on the delivery of key provisioning ES and ecosystem-based adaptations in the Limpopo Basin of Botswana? This work was part of the Adaptation at Scale in Semi-Arid Regions (ASSAR) and the South Africa/Flanders Climate Adaptation Research and Training Partnership (SAF-ADAPT) projects. Bobirwa sub-district was selected for this study since it exhibited desired attributes of rapid urbanization, increasing demand for agricultural land, high human dependence on the natural environment, proximity to protected areas, and sharing national boundaries with Zimbabwe and South Africa.

2. Materials and Methods

2.1. Study Area

The Limpopo River Basin part of Botswana is ecologically and economically significant to indigenous people and surrounding communities. Our case study, the Bobirwa sub-district is situated between 28°09′10″ E to 29°21′42″ E and 22°35′17″ S to 21°35′56″ S and lies entirely within the Limpopo River Basin part of Botswana. The sub-district has an altitude ranging between 590 and 886 m, making it the lowest part of Botswana, hence it has a network of channels that drain into the Limpopo River [14]. The sub-district boundary forms the national boundary with Zimbabwe to the northeast and South Africa to the southeast where the Limpopo River marks the boundary (Figure 1). According to the 2011 National Population and Housing Census report, the population of the Bobirwa sub-district was 71,936, comprising 34,247 males and 37,689 females from 19,213 households with an average household size of 3.74 and a population density of 5.05 people/km² [27].

The study area is highly susceptible to droughts and erratic rainfall which fluctuates well below 400 mm/year in most years [18]. The recent (2010–2016) average winter and summer temperatures were 15.8 °C and 24.5 °C, respectively. However, summer temperatures have often exceeded 38 °C in the last 5 years with the occurrence of heat waves [18]. Summers are hot and dry between August and November (pre-rainy season) and become hot and wet between December and February/March (rainy season). Thus, the period March–April is considered the post-rainy season. Winters are cold and dry and occur from late April to early August. Nonetheless, the study area experiences a less severe climate than other parts of the country, allowing crop and livestock production and supporting considerable biodiversity that underpins the delivery of several timber and non-timber forest products [14,28].

Natural vegetation in the study area is mostly represented by desert-like vegetation and a variety of species, such as grasses and grass-like plants, shrubs, and trees with patches of bare land [29]. Colophospermum mopane, commonly known as mopane, dominates the tree, shrubs, and other desert-like vegetation. The Limpopo Basin is richly endowed with resources such as wood fuel, medicinal plants, and Mopane caterpillars (larvae of Imbrasia belina moth), which thrive on Mopane trees (Colophospermum mopane) abundant in the area. The sub-district is well-known for Mopane caterpillars, which are an important source of protein and income, thus attracting harvesters and buyers from elsewhere in Botswana, Zimbabwe, and South Africa [14,30]. Village settlements are separated from “cattle posts” and crop fields, resulting in a ‘three-home’ system for many households whereby there is a temporary and seasonal movement across the three locations, i.e., village—crop fields—cattle posts.

The main livelihood activities and economy revolve around rain-fed agriculture and exploitation of timber, firewood, and non-timber forest products such as Mopane caterpillars (Imbrasia belina), wild fruits, thatch, and palm leaves (Hyphaene petersiana) for basketry [14,30,31]. The main crops grown under rain-fed conditions are maize, sorghum, millet, cowpeas, groundnuts, round nuts, and watermelons. Subsistence livestock and poultry production, a mainstay of the local economy, is mainly characterized by rearing cattle, goats, and chickens under free-range [14]. Other land uses include game reserves and privately owned farms with abundant wildlife. In recent years, tourism development in the sub-district, for instance in the Tuli Block area, has been strengthened in part due to government support of local tourism and to a larger extent by the abundance of wildlife in the area.

2.2. Data Sources

Three sets of Landsat images with less than 0.1% cloud contamination were downloaded from the USGS GloVis for the post-rainy period of the years 1995, 2006, and 2016 (each year with 4 scenes) through the URL: http://glovis.usgs.gov/ (accessed on 16 August 2019) using 170-1/075-6 as path/row, respectively, as detailed in

Table 1. Landsat images used for this study.

Month	Year	Path/Row
		170/075		170/076		171/075		171/076
		Sensor	Date Acquired	Sensor	Date Acquired	Sensor	Date Acquired	Sensor	Date Acquired
March/April	1995	TM	17/03/95	TM	17/03/95	TM	09/04/95	TM	09/04/95
	2006	ETM	08/04/06	ETM	10/05/06	ETM	07/04/06	ETM	07/04/06
	2016	OLI	02/03/16	OLI	02/03/16	OLI	02/04/16	OLI	02/04/16

. Landsat offers the longest global record of the earth’s surface and is operated by USGS/NASA. The dates when the Landsat images were acquired (Table 1) are for the period after the rainy season; these were specifically chosen as vegetation, croplands and bare lands are more distinct after the rainy season and thus reduced chances of confusion. Since the study was limited by the unavailability of cloud-free imagery on desired dates, we appreciate the slightly lower classification accuracy when using Landsat imagery acquired on different dates for land-cover classification. Nonetheless, the multi-temporal analysis, use of imagery before the leaf senescence (greenness transition) period, and extensive ground-truthing minimized these errors [32].

As shown in Table 1, the satellite images used in this study were imagery from Landsat-5 Thematic Mapper (TM) Sensor (1995), Landsat-7 Enhanced Thematic Mapper (ETM) Sensor (2006), and Landsat-8 Operational Land Imager (OLI) Sensor (2016) sensors with a resolution of 30 m.

2.3. Image Pre-Processing

The pre-processing of the acquired Landsat images was executed in ENVI (v5.3) and included radiometric and atmospheric correction using the Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH). This is a first-principles atmospheric correction modeling tool for retrieving spectral reflectance from hyperspectral radiance images. FLAASH can accurately compensate for atmospheric effects and correct wavelengths in the visible through near-infrared and short-wave infrared range [33]. The Landsat images were corrected to remove atmospheric effects and then re-sampled to 30 m pixel size for all bands using the nearest neighbor method. The nearest neighbor algorithm is applied where knowledge of the underlying probabilities is limited.

2.4. Image Processing

2.4.1. Supervised Image Classification

The pre-processed satellite images for the post-rainy period (March/April) were classified using the maximum likelihood algorithm (MLA) using ENVI v5.3 software. Initially, regions of interest (ROIs) were carefully selected on each satellite image to create a training sample for supervised classification. Each ROI had between 80 and 150 polygons as training sites, and these were chosen randomly depending on the area covered by the class. Five ROIs were selected for this study, as described in

Table 2. Land cover classes used in the classification of Landsat images.

Land Cover Class	Description
Built-up areas Croplands	Village settlements, rock surfaces, paved roads, etc. Cultivated areas under rain-fed and irrigation, plantations
Water bodies	Dams, ponds, rivers, streams, etc.
Vegetation	Predominantly trees, shrubs, or grasslands
Bare land	Bare lands with exposed soil surfaces, with no vegetation all year round, or with very sparse vegetation

. The MLA uses the ROIs to classify a data point, i.e., classifies a data point based on how its neighbors are classified [34]. Despite a wide choice of algorithms for supervised classification, MLA was most suitable for classifying at least two classes with overlapping spectral characteristics, simple point clouds, and some correlation between the brightness in different ranges of the spectrum [34]. Several comparable studies elsewhere also used MLA [26,27,28].

The LULC classes in this study (Table 2) were distinguished clearly from each other with the aid of a feature space image, which plots a scatter of pixel values of two bands of the imagery and different color reflectance. The intensity of each portion of the feature space image shows all the pixels with the same reflectance values, which reduces the chances of mixing of any two classes Therefore, classification was performed after there was minimal mixing.

2.4.2. Accuracy Assessment

Accuracy assessments were performed to establish the relationship between the classified LULC images and the control reference data similar to previous studies [5,26,35]. This was based on ground data collected from the respective land-use and land-cover types in the study area. For the 2016 image, a minimum of 50 ground truth points was generated per class in 8 villages except for water bodies, which had about 25 ground control points as there were fewer open water bodies. For the 1995 and 2006 images, historical Google Earth was used for validation of the image accuracy assessments. Similarly, each class, except for water bodies, had a minimum of 80 ground control points taken for accuracy assessment.

The user’s accuracy was calculated as the number of correctly classified pixels as a proportion of the total number of pixels assigned to a specific category and accounts for errors of commission. The producer’s accuracy was computed to show the number of pixels correctly classified in a specific category as a proportion of the total number of pixels actually belonging to that category and accounts for errors of omission. The overall classification accuracy was also computed to show the total of all the correctly classified sample pixels as a proportion of the total number of reference pixels.

2.4.3. Error Matrix

An error matrix showing errors of omission (producer’s accuracy) and commission (user’s accuracy), overall classification accuracy, and Kappa coefficient was used to determine the accuracy of the LULC maps [26]. Error matrices are commonly used to report classification accuracies, as they compare the relationship between known reference data (ground truth) and the corresponding results of an automated classification on a category-by-category basis [36]. The error matrices used in this study to compare the reference class values with the classified class values are a 5 × 5 matrix, where 5 is the total number of classes used for this study. The overall accuracy and the Kappa coefficient values were calculated using the producer’s and user’s accuracy for all the LULC classes similar to several studies elsewhere [22,24,37]. The Kappa coefficient measures the deviation of chance agreement from the actual agreement. A minimum overall classification accuracy of 80% is considered efficient.

2.5. Household Survey

The study applied the sampling error formula [1] for a target population exceeding 10,000 to determine the sample size. The sampling unit for this study was the household, and the target population was the 10,896 households in the study area [2]. Stratified simple random sampling was then employed to select 317 study participants in the eight villages using numbered village lists from the local Sub-Land Board and random numbers generated in R studio software. Commands for random number generation required the first and last positions of households on each village list, including the total number required from that range without repeating. All participants were informed prior to the interview with the assistance of Village Development Committees in each village, and written consent was sought before each interview.

A semi-structured questionnaire prepared in collaboration and with input from members of the Adaptation at Scale in Semi-Arid Regions (ASSAR) project Southern Africa regional team was administered to household heads or the most senior member available. This study only focused on questions that sought to establish the provisioning of ecosystem services in the Limpopo Basin, their relative importance, land-use and/or land-cover types providing them, and their recent trends. However, only 310 questionnaires were considered complete and usable for analyses.

2.6. Data Analysis

Maps were used to show the spatial distribution of the different LULC types, and band math were then computed on the classified LULC images. The absolute area (in km²) and the proportion covered (in %) by different LULC class categories were calculated for 1995, 2006, and 2016 using the number of pixels covered by each LULC class. The overall classification accuracy and the Kappa coefficient were used to analyze the efficiency of the classified LULC classes similar to other studies [22,24]

Post-classification change detection was performed by computing the changes in the area covered by each LULC class between 1995 and 2006, 2006 and 2016, and 1995 and 2016. Land cover classification images for the different years were compared on a class-by-class basis, and the differences in terms of the number of pixels covered by each category in the different time periods were used to compute the area statistics for these changes. The absolute changes (in km² and %) and average rate of changes (in km²/year and %/year) in the area covered were computed to detect changes in each of the three time periods. Splitting the analyses into decadal periods was key to showing the period with significant changes and unmasking the most recent changes. The sections below present the results of the change matrix for the periods 1995–2006, 2006–2016, and 1995–2016. LULCC change matrices were used to examine how a decline in one LULC was distributed among other LULC classes. Similarly, they also revealed how an increase in one LULC occurred.

Data from the household survey were analyzed using STATA software (version 14.2). Ranking of provisioning ecosystem services by households considered their relative importance, i.e., the contribution of each provisioning ecosystem service towards household food security, income, employment, and provision of raw materials. It also considered the possible impacts of a decline in the ecosystem service on the same. Ranking was achieved by averaging all the individual responses for each of the provisioning ecosystem services identified into a single metric ranging between 1 and 100 points. Thus, those provisioning ecosystem services within the 66–100 points range were highly ranked, while those in the range of 33–66 were ranked medium and those lowly ranked (0–33) provisioning ecosystem services. Highly ranked provisioning ecosystem services contributed more to livelihoods, while the lowly ranked contributed the least.

3. Results

3.1. Land-Use and/or Land-Cover Change in Bobirwa Sub-District

The spatial distribution of LULC types in Bobirwa for 1995, 2006, and 2016 is shown in Figure 2 below.

Figure 3 summarizes the absolute areas (km²) and proportions (%) covered by the different LULC types in 1995 and 2016. In 1995, approximately 5130 km² (68%) of the total surface area in the Bobirwa sub-district was covered by natural vegetation such as woodlands, shrublands, and grasslands. Croplands covered only 395 km² (5.24%). The overall classification accuracy and the Kappa coefficient for the 1995 LULC were 79.82% and 0.7035, respectively. In 2016, the area covered by natural vegetation was about 4066 km² (53%) of the total surface area of the Bobirwa sub-district. Croplands spanned nearly 1110 km² (14.71%), while bare lands were distributed over 1798.33 km² (23.84%). The overall accuracy and the Kappa coefficient of the 2016 classified image were 78.51% and 0.7249, respectively. The area covered by natural vegetation was significantly different from the area covered by built-up areas (p = 0.0080) and water bodies (p = 0.0068). The area covered by bare lands was also significantly different from the area covered by built-up areas (p = 0.0040) and water bodies (p = 0.0006).

Table 3. Comparison of LULCCs in Bobirwa sub-district between 1995 and 2016.

LULC Class	Changes from 1995–2006		Changes from 2006–2016		Changes from 1995–2016
	km2 (km2/Year)	% (%/Year)	km2 (km2/Year)	% (%/Year)	km2 (km2/Year)	% (%/Year)
Bare land	727.94 (66.18)	43.25 (3.93)	−612.83 (−61.28)	−25.42 (−2.54)	115.11 (5.48)	6.84 (0.33)
Built-up areas	165.18 (15.02)	65.82 (5.98)	64.87 (6.49)	15.59 (1.56)	230.05 (10.95)	91.67 (4.37)
Croplands	168.66 (15.33)	42.65 (3.88)	545.72 (54.57)	96.74 (9.67)	714.39 (34.02)	180.65 (8.60)
Water bodies	−14.01 (−1.27)	−16.77 (−1.52)	18.61 (1.86)	26.77 (2.68)	4.60 (0.22)	5.51 (0.26)
Vegetation	−1047.78 (−95.25)	−20.42 (−1.86)	−16.37 (−1.64)	−0.40 (−0.04)	−1064.15 (−50.67)	−20.74 (−0.99)

indicates the dynamics of LULCC in the Bobirwa sub-district during study periods. Between 1995 and 2006, croplands increased by approximately 169 km² (42.65%), which translates to an increase of 15.33 km²/year. Natural vegetation declined by about 1048 km² (20.42%), an average decline of 95.25 km²/year during the same decade. Between 2006 and 2016, croplands increased by approximately 546 km² (96.74%), at a rate of 54.57 km²/year. Natural vegetation and bare land decreased by about 16.37 km² (0.40%) and 612.83 km² (25.42%) at an average of 1.64 km²/year and 61.28 km²/year, respectively. Overall, between 1995 and 2016, croplands and bare land increased by 714.39 km² (180.65%) and 115.11 km² (6.84%), respectively (Table 3). This was an average increase of 34.02 km²/year and 5.48 km²/year in croplands and bare lands. Natural vegetation declined by 1064.15 km² (20.74%) at an average rate of 50.67 km²/year during the 21 years. None of these changes were significant (p = 0.1906).

Figure 4 shows how gains or losses in each of the five LULC classes came about during the period under study. Figure 4a shows that from 1995 to 2006, 404.31 km² (7.88%) of vegetation was converted into croplands, and a further 1123.23 km² (21.89%) was converted into bare lands. A total of 3457.67 km² (67.39%) of natural vegetation remained unchanged. During the same decade, 262.71 km² (15.61%) of bare land and 238.27 km² (60.25%) of croplands were converted into naturally vegetated areas. A total of 108.6 km² (27.46%) remained as croplands, while the rest was converted to other LULC types.

From 2006 to 2016, Figure 4b shows that 627.83 km² (15.38%) of vegetation was converted into croplands, while 3079.28 km² (75.42%) remained unchanged. Only 243.40 km² (5.96%) of vegetation was converted into bare land. A total of 221.42 km² (39.25%) of croplands and 726.81 km² (30.14%) of bare land were revegetated. Approximately 306.44 km² (54.32%) of croplands remained unchanged between 2006 and 2016. Overall, considering the entire 21 years from 1995 to 2016, Figure 4c shows that 785.62 km² (15.31%) of vegetation was converted into croplands, and 639.20 km² (12.46%) was converted into bare land. About 3509.01 km² (68.40%) remained as natural vegetation between 1995 and 2016. Only 172.24 km² (3.36%) of vegetation was converted into built-up areas. A total of 159.93 km² (40.44%) of croplands and 308.17 km² (18.31%) of bare land were revegetated during this period. Approximately 190.58 km² (48.19%) of croplands in 1995 remained as croplands in 2016.

3.2. Household Socio-Economic Characteristics

Table 4. Socio-economic characteristics of survey participants.

Socio-Economic Characteristics		N/Mean	% of Cases	χ2	p-Value
Age of household head	Young (20–40)	57	18.4
	Adult (40–60)	128	41.3	9.332	0.009 **
	Elderly (>60)	125	40.3
Gender of household head	Female	247	79.7
	Male	63	20.3	7.688	0.006 **
Education level of household head	None	84	27.1
	Primary	122	39.4
	Secondary	71	22.9		0.014 *
	Tertiary	33	10.7
Employment status of household head	Full-time farmer	100	32.3	0.160	0.689
	Part-time farmer	210	67.7
Household size		5.88	-
Arable farm size (ha)	Total land owned	8.85	-
	Cultivated area	2.32	26.2
Annual income of household (BWP) (USD 1 ≈ BWP 13.4)	<5000	193	62.3
	5000–10,000	39	12.6
	10,001–15,000	19	6.1	4.160	0.385
	15,001–20,000	15	4.8
	>20,000	44	14.2
Household sources of income	Crop production	164	52.9
	Casual labor (Ipelegeng)	115	37.1
	Mopane caterpillars	91	29.4
	Formal employment	86	27.7
	Small livestock	76	24.5
	Poultry products	76	24.5
	Informal trade	64	20.6
	Cattle	48	15.5
	Timber/firewood	48	15.5
	NTFPs	10	3.2
Livestock owned	Cattle	6.27	37.4	0.201	0.654
	Donkeys	1.42	26.1	5.117	0.077
	Goats	7.77	55.5	9.633	0.008 **
	Sheep	1.02	10.3	0.649	0.421
Government (ISPADD) input support received	Full subsidy	145	46.8
	Partial subsidy	17	5.5	1.145	0.887
	None	148	47.7
Major farm implements used	Hand-held	204	65.8	3.514	0.037 *
	Animal-drawn	137	44.2	9.257	0.026 *
	Tractor-drawn	11	3.5	0.004	0.950
Access to climate and agricultural information	Yes	282	91.0	0.071	0.790
	No	28	9.0
Average grain yield (kg/ha) [last 3 seasons]	Sorghum	168.9
	Millet	167.5
	Maize	164.3
	Cowpeas	122.3

Significance level; * p < 0.05, ** p < 0.01. Source: Household Survey Data, 2017.

summarizes the socio-economic characteristics of survey participants and presents their frequencies (or means) and proportions (%), including Chi-square tests (and p-values) of differences among participants. There were 79.7% females and 20.3% males with an average age of 56.3 years. Only 27.1% of the participants had not gone past primary school, while 32.3% reported being full-time farmers and 67.7% as part-time farmers. The average size of farmland owned by a household was 8.85 ha, but the cultivated area was 2.32 ha. Average household size was 5.9 (3.3 females and 2.6 males). A proportion of 37.4%, 26.1%, 55.5%, and 10.3% of the households owned cattle, donkeys, goats, and sheep, respectively.

More than 62% of the households had income below BWP 5000 (USD 374). Sources of income for households included crop production (52.9%), casual labor (37.1%), selling Mopane caterpillars (29.4%), formal employment (27.7%), selling small livestock and poultry (24.5%), selling timber and firewood (15.5%), and selling other ecosystem products such as palm wine and wild fruits (3.2%). At least half of the participants had received free inputs from the government through the Integrated Support Programme for Arable Agricultural Development (ISPAAD). Households used hand-held (65.8%), animal-drawn (44.2%), and tractor-drawn (3.5) farm implements. The average yield for sorghum, millet, maize, and cowpeas under rain-fed conditions for the past 3 seasons ranged between 122 and 170 kg/ha.

3.3. Key Provisioning Ecosystem Services in Bobirwa Sub-District

Figure 5 shows that a total of 15 provisioning ecosystem services were identified as critical to livelihoods from the survey (see Supplementary Table S1 for a more detailed list); seven contributed directly towards household food requirements, nine contributed towards household income and four (4) provided raw materials, e.g., palm leaves for weaving. Reserved areas delivered almost all the identified provisioning ES (14), followed by woodlands (13), croplands (9), areas within village settlements (9), water bodies (3), and bare lands (1).

Figure 5 also shows that the highly ranked provisioning ES in terms of relative importance and contribution to household food requirements, income, raw materials, and employment opportunities were natural pastures, livestock production (cattle, donkeys, goats, sheep, and poultry), freshwater, Mopane caterpillars, and firewood. The medium ranked were timber and poles, palm plants, wild fruits, and thatch. The lowly ranked included natural medicines and medicinal plants, natural dyes (for decorating palm leaves used for weaving), wild animals, fisheries and sand, and precious stones. These rankings also considered the effects of marginal changes in the delivery of each ES on household food security, income, provision of raw materials, and employment opportunities.

A general decline in accessibility to several timber and non-timber forest products was reported by at least 78% of participants in the study area (Figure 6). Access to palm plants and natural dyes was reported to be largely unchanged though concerns were raised over the destructive harvesting and/or extraction of palm wine. Access to water for livestock consumption at the communal grazing areas (cattle posts) was also reported to have declined by 34% of the participants. Almost 23% of the participants reported difficulties in accessing fertile land for cultivating crops, while about 60% had the same access. Only 7% reported more access to fertile land for cultivation.

3.4. Perceived Drivers of Change in Provisioning Ecosystem Services

Figure 7 shows the perceived drivers of change in (a) total area (size) of ES sites and (b) walking distance to sites providing ES. The perceived drivers of change included primary (e.g., adverse climate) or secondary, i.e., those driven by one or more of the stated (primary) drivers (e.g., land degradation).

Fluctuations in the total area or size of historical sites delivering provisioning ES (Figure 7a) were perceived to be driven by adverse climate and/or weather (83.1%, p = 0.000), expansion of human villages (80.3%, p = 0.000) and changes in land-use (71.8%, p = 0.002). Land degradation (43.7%, p = 0.132), bush encroachment (40.8%, p = 0.217), dam construction (32.4, p = 0.000), and veld fires (12.7%, p = 0.000) were perceived by fewer participants. Similarly, causes of changes in distance to sites delivering provisioning ES were perceived as overexploitation and/or overgrazing (98.6%, p = 0.000), degradation of nearby sites (85.9%, p = 0.000), village and/or settlement expansion (81.7%, p = 0.000) and land-use change (70.4%, p = 0.001). Fewer participants perceived changes in regulations (25.4%, p = 0.000) and restricted access to privately owned farms (19.7%, p = 0.000) as limiting access to nearby sites.

4. Discussion

4.1. Dynamics of Change in Bobirwa Sub-District

The majority of provisioning ES in the Limpopo Basin are derived from, or depend on the extent and richness of, natural vegetation. Agriculture is also an important livelihood, thus our discussion mainly focuses on changes in natural vegetation and croplands. Although this study did not quantify the changes in individual provisioning ES, our analyses provide a suitable quantitative measure to understand the implications on bundles of provisioning ES delivered by natural vegetation and croplands.

The dominant land-cover type in Bobirwa during the study period was natural vegetation followed by bare lands (with sparse vegetation) and croplands. The decline in natural vegetation between 1995 and 2016 at a rate of 50.67 km²/year, was characterized by a significant increase in croplands of 34.02 km²/year during the same period. This indicates an increased demand for agricultural land, particularly for crop production similar to studies in Central Ethiopia [3].

Besides the expansion of croplands, the decline in natural vegetation is also attributed to the expansion of village settlements caused by a growing human population. According to the 2011 Botswana National Census, the total human population of the sub-district grew by 8.2% between 2001 and 2011, which translates to an annual growth rate of 0.8% during that decade [27]. For instance, the expansion of villages such as Bobonong, Tsetsebjwe, Mathathane, Motlhabaneng, and Molalatau to establish built-up areas involved clearing vegetated areas, particularly the Colophospermum Mopane woodlands that are prevalent in the area as noted by earlier studies [30,38]. With such growth and urbanization of some villages, pressure on surrounding land for settlement and croplands could further reduce natural vegetation in the Bobirwa sub-district.

The three-tier household system among many of the rural households in Botswana, i.e., spread across the village settlement, communal grazing areas, and crop fields also has implications on LULCC in the study area. The increase in built-up areas and bare lands around the villages in Bobirwa is consistent with the effects of a growing human population [35]. Expansion of settlements that involve the clearing of woodlands, shrublands, and grasslands has also been reported in the Likangala catchment in Malawi [39] and Mukuvisi and Marimba catchments in the outskirts of Harare [40]. The Mopane woodlands are also sources of firewood, some of which are harvested illegally for commercial purposes. Thus, an increase in human population in the Bobirwa sub-district, coupled with limited livelihood opportunities, and a unique dependency on the natural environment, particularly agriculture, exerts pressure on land [5].

The increase in croplands in Bobirwa during the assessment period is consistent with most rural communities in Africa where subsistence, rain-fed crop, and livestock production are critical sources of livelihoods, as shown elsewhere in Botswana [2]. Given that only 25% of the arable farmland owned by households was under cultivation, expansions in croplands are likely to perpetuate. However, increases in croplands often fragment natural ecosystems and could significantly impair some ecosystem functions [41]. For instance, some households in the study area resorted to establishing croplands at the cattle posts, areas only reserved for communal grazing, thereby reducing natural pastures among other natural products obtained therefrom. Such actions were mostly driven by delays in obtaining land officially. In addition, more frequent droughts experienced in recent years were cited by some farmers as forcing them to clear more land for growing crops. The critically low grain crop yields (sorghum, millet, maize, and cowpeas), i.e., <170 kg/ha, obtained by farmers in the study area highlight the precarity and susceptibility of rain-fed agriculture. An earlier study in the area shows a decline in rainfall events and increasing deviations below average rainfall [38]. The same study suggests an increased potency of droughts in the Limpopo Basin of Botswana and highlights the adverse effects on rain-fed crop production. Thus, some farmers viewed the expansion of area under cultivation as a coping strategy and insurance against drought-induced total crop failure. Kosonei et al. [42] also found increased demand for agricultural land in the Marigat sub-county of Kenya driven by the need to salvage yields in case of severe droughts. This is in contrast to studies elsewhere in Botswana that found increased abandonment of croplands due to the increasing severity of droughts [2].

Another driver of expansions of croplands in Botswana is the Integrated Support Programme for Arable Agricultural Development (ISPAAD) by the Ministry of Agriculture. The program supports smallholder farmers with free inputs such as fertilizer, seeds, herbicides, tillage, and access to credit [43]. While the study shows that at least half of the surveyed households were beneficiaries of the ISPAAD program, the effect of such programs is that they attract new farmers and influence current farmers to increase the area under cropland. Considering that the average area under cultivation by the surveyed households was about 2 ha, a quarter of total arable landholdings, and that the ISPAAD program gives free inputs for up to 5 ha, farmers were likely to expand their croplands. Our findings, similar to others before, show that the program has not enhanced crop productivity over the years [44]. Although current efforts to deviate from the traditional system where grazing and croplands are separate, to integrated farms of up to 15 ha where farmers can simultaneously practice crop and livestock production, more efforts need to go towards enhancing agricultural productivity. This could arrest unnecessary agricultural expansions since land fragmentation interferes with ecosystem functions and threatens biodiversity and the delivery of essential ES [2].

4.2. Impacts of LULCC on Ecosystem Service Delivery

The findings show that the increase in croplands, built-up areas, bare lands, and water bodies was mainly at the expense of natural vegetation. While increasing area under croplands could increase food production, the average grain yield obtained in the study area (Table 4) paints a gloomy picture. Instead, the observed decline in natural vegetation threatens biodiversity and reduces primary production [45]. Consequently, the delivery of key provisioning services such as timber, firewood, Mopane caterpillars, natural medicines, and natural dyes by vegetated areas becomes threatened. This further threatens household sources of income since some of these services feature among the reported household sources of income, e.g., Mopane caterpillars and firewood. With at least 60% of the households earning below BWP5000 (USD 373) per year, the loss of these sources of livelihoods could push them further into poverty. There is widespread evidence of the effectiveness of ecosystem services in safeguarding livelihoods, as well as providing a sustainable adaptation pathway, as also highlighted in Botswana’s National Adaptation Framework and National Biodiversity Strategy and Action Plan [17,46]. Ecosystem services from natural systems are particularly important in the study area given the precarity of rain-fed agriculture. Thus, a further loss of vegetated areas in the Limpopo Basin of Botswana could have cascading negative effects, especially on low-income households, which are the majority of the surveyed households.

Globally, the impairment of the delivery of provisioning ES such as timber, medicinal plants, wild fruits, resins, and natural dyes due to a decline in biodiversity is well-documented [47,48,49]. LULCC has been reported as the most severe direct driver of biodiversity loss by several studies [50,51,52]. The loss of vegetated areas in the study area, particularly at the communal grazing areas (cattle posts), also threatens livestock production, which, for low-income households, largely depends on natural pastures [53]. Elsewhere, land fragmentations have also caused habitat changes and significant biodiversity losses often through sudden disruptions of the balance in the ecosystem [41]. As natural habitats decline further in the Limpopo Basin of Botswana, conflicts between humans and wildlife could become rife, particularly for villages such as Mabolwe, Tsetsebjwe, Gobojango, and Motlhabaneng, which are close to protected areas. All these interactions and trade-offs negatively affect the well-being of communities, more so the poor.

The increasing severity of droughts in the Bobirwa sub-district, coupled with a lack of irrigation, presents immediate challenges to smallholder agriculture in the study area. Despite government support of smallholder farmers with free productive inputs and access to cheap credit under ISPAAD, without addressing the drought-induced moisture deficits during the growing season, cultivated crop yields are likely to remain low. Studies elsewhere in Botswana have highlighted the precarity of rain-fed crop production [2]. Thus, without addressing productivity issues that limit smallholder agriculture, trade-offs between forest ES and agricultural ES will leave local communities worse off [54]. There is an urgent need to ensure that the ISPAAD program is climate-proofed, for instance by promoting climate-smart agricultural practices and crop varieties and types that suit local agroecological conditions.

Besides provisioning ES, the loss of natural vegetation also has negative implications on several supporting and regulating ES. For instance, the important functions of regulating the micro-climate [55], water regulation [56], and water purification [57] performed by vegetation are also lost. Without vegetation to regulate the micro-climate, human and animal health will be highly exposed, e.g., to heat stress [58]. The loss of vegetated areas could be reducing infiltration, hydrological regulation, and accelerating run-off, soil erosion, and siltation of water bodies such as the Thune dam. The consequences include limited recharging of the groundwater system. Studies elsewhere have associated declining woodlands with land degradation, soil erosion, and the formation of gullies [59].

The expansion of croplands and the increased use of agrochemicals such as fertilizers, pesticides, and herbicides issued under the ISPAAD program may be contaminating surface water and groundwater sources. An increase in nitrates (from nitrogenous fertilizers) in water bodies such as the Thune Dam would cause algal blooms and increased biological oxygen demand. These threaten fish and other aquatic life if the dissolved oxygen in water bodies becomes insufficient to sustain these organisms. This would be a threat to an emerging fish industry in the study area. The contamination of water bodies and loss of aquatic organisms such as fish due to increased pollution from agricultural lands has been a huge problem elsewhere [60].

5. Conclusions and Recommendations

LULCC in the Bobirwa sub-district was characterized by declining vegetation and expansion of croplands. These LULCC regimes, which also fragment land, threaten the sustainability of many provisioning ES through impairing ecosystem functions and loss of biodiversity that underpins ES delivery. If unchecked, LULCC in the study area could irreversibly destroy biodiversity and natural habitats. Despite the support with free inputs from ISPAAD, the incessantly low crop yields among smallholder farmers highlight the ineffectiveness of the program and/or the precarity of rain-fed agriculture. The observed patterns of LULCC in the study area are expected to perpetuate, driven by the growing human population and more intense climate change impacts, and more efforts are needed to enhance crop productivity. However, the current conversion of natural vegetation to croplands needs a more balanced approach to reduce the fragmentation of ecosystems. This is required to ensure the sustainability of agricultural and natural systems and is necessary for reducing pressure on biodiversity, which underpins ecosystem service delivery.

A growing population and increasing demand for food from agriculture is a unique challenge to ecosystem management. Such actions as providing drought-tolerant, high-yielding crop varieties and fostering climate-smart agricultural practices could increase crop productivity, reduce agricultural expansions, and relieve pressure on natural ecosystems. Currently, natural vegetation delivers a wide range of provisioning ecosystem services that are critical for adaptation. Therefore, any unnecessary agricultural expansions, which severely threaten ecosystem-based livelihoods and adaptations need to be curtailed. In addition, there is a need to strengthen and capacitate traditional systems to take a leading and active role in managing local resources. Engaging communities through participatory, biodiversity-based action planning may play a significant role in promoting biodiversity conservation and the implementation of associated measures. Thus, the government and other relevant stakeholders need to enhance the ecosystem management capacities of local communities and support them in developing and implementing village action plans. However, village action plans need to mainstream biodiversity actions and be smoothly integrated with development and adaptation planning processes at the district and national levels. Moreover, such community- and ecosystem-based initiatives are emphasized in the country’s National Adaptation Plan (NAP) Framework and also contribute to achieving international biodiversity targets enshrined in the National Biodiversity Strategy and Action Plan [17,46].