1. Introduction
The change of land use and land cover (LULC) has become a fundamental component of current strategies to manage natural resources and monitor environmental changes [
1]. LULC resource uses have resulted in major anthropogenic changes [
1,
2]. Increasing human activities have caused large-scale changes in the terrestrial surface, disturbing the productivity of global systems [
3]. Rapid LULC changes, especially in developing countries, have reduced essential resources, including vegetation, water, and soil [
1]. Furthermore, factors contributing to LULC change have indigenous causes, one of which is economic activities by the local community.
One such major economic activity responsible for LULC change is mining. The exploitation of mining resources significantly impacts the ecological environment [
4]. The study by [
5] stated that mining results in urban growth, leading to LULC changes in many areas worldwide, especially in developing countries such as Zambia [
6]. Open-pit, ore, and strip mining can lead to LULC change [
7] which have already contributed to severe environmental landscape degradation in mine-adjacent areas of the USA [
8].
Governments in many countries require the recovery of areas degraded by mining [
9]. Both sustainable and eco-friendly mining necessitate continuous LULC change monitoring to identify their long-term environmental impacts [
10]. Monitoring these changes provides fundamental security measures and data for planning ecological restoration and land reclamation strategies [
7]. Furthermore, the spatiotemporal change quantification in an area resulting from open-pit mining becomes crucial in understanding the impacts of mining activities and evaluating their socioeconomic, environmental, and ecological impacts [
11] through related legislation.
Copper mining plays an important role in Zambia’s economic development. Commercial copper mines in Zambia have been in operation since 1928, when the first mine opened [
12]. In Zambia, there are five major open-pit and eight underground mines that produce copper and other minerals, such as cobalt [
12]. Copper mining takes place in the Copperbelt Province, located in the northern region of the country, which is the most densely populated and urbanized in the country [
13]. Consequently, the development of urban centers around mines has posed spatial problems in reconciling the needs of a rapid population growth with the demands of the mining industry [
13].
For over seventy years, mineral resources have been mined near Kitwe, Copperbelt Province [
14]. The long mining history and the existence of other pollution sources complicate the assessment of environmental impacts in Kitwe, thereby necessitating an indicator of environmental change [
14]. Extensive quantities of mine residues, including broken rocks, fine particles, and slag, have been generated and deposited on the land [
15]. These residues indicate the mining extent and are indicators of its environmental impact [
14]. However, in Kitwe, the extent and dynamics of the changes have not been comprehensively studied, except for an old-time study before the privatization of mining companies in the 1990s [
16]. There is limited information about the spatiotemporal extent of the LULC changes in this district, and no evaluation on the information has been done after privatization to enhance land use planning.
Although Zambia has made headway toward incorporating mine closure-related laws and policies, either directly or indirectly, into its constitution, implementing such legislation has achieved little success [
17]. According to Clark and Clark (2005) [
18], Zambia’s current legal and policy frameworks for closure are too weak and fragmented to guarantee comprehensive mine closures because most governmental institutions involved in the management of closure are unable to fulfill their legal obligations owing to three factors: lack of political support, insufficient supply of human and financial resources, and hindrance by contractual agreements between mining companies and the government.
This study, therefore, aims to (1) understand the recent land use land cover (LULC) change dynamics resulting from mining activities in the Kitwe District of Copperbelt Province (1990–2020); (2) clarify the Zambian legal framework on mine closure; and (3) examine Zambia’s mine closure legislation to determine if it complies with the sustainable development principle and most recent international best practices for mine closure. This study selected the period from 1990 to 2020 because the privatization of mines in Zambia began in the 1990s. Therefore, this study focused on understanding the changes in LULC that occurred immediately before and after the mines were privatized.
2. Data and Methods
2.1. Study Area
2.1.1. Kitwe District
This study was conducted in Kitwe District (799.42 km
2), Copperbelt Province, Zambia (
Figure 1), located approximately 50 km northwest of Ndola City, the headquarters of the province. It is a mining district within the Neoproterozoic Katangan Supergroup basinal succession [
19]. Kitwe City (12.8024302° S and 28.2132301° E) lies on the west bank of the Kafue River.
Kitwe, which is the main commercial and industrial center of the province [
20], is the second largest city in Zambia and the largest city in the Copperbelt Province. Kitwe rapidly developed a copper mining industry especially after 1936 along with the establishment of secondary industries [
20]. It is famous for the Black Mountain, a copper slag dump located in the Wusakile Township. Kitwe has four mines, including the Mopani Copper Mine, a joint venture situated in Kitwe, where 95% of its operations take place. The ownership makeup includes Glencore International AG (73.1%), First Quantum Minerals Limited (16.9%), and Zambian Consolidated Copper Mines (ZCCM) Limited (10%), the national mining company [
14]. Prior to privatization, the Nkana Slag Dump (Black Mountain) was owned by Nkana Mine of ZCCM Limited. This licensed dump received slag from the Nkana Smelter until the designed limit was reached in the 1990s [
21]. The Konkola Copper Mine is the second mine in the study area. While the company is headquartered in Chingola, 15% of its operations, including the Nkana refinery, acid plants, and smelter, are situated in Kitwe [
21] with the Nkana smelter being Zambia’s largest main copper production plant. The third mine is Kagem Emerald Mine, owned by Kagem Mining Limited, the largest producer of emeralds, accounting for approximately 25% of the global emerald production. The fourth mine is the Mindola Underground Mine, once owned by Rokana Mine (now closed) and now owned by ZCCM. Extensive tailings are around this mine, and two small tailings dams are located in the city center.
2.1.2. Rokana Mine
The Rokana mine is one of the oldest copper–cobalt mines owned by ZCCM, located in the central part of Kitwe (12°49′59″ S, 28°12′6″ E) [
15]. It has been in continuous operation since 1928, and during mine nationalization (1970–1991), underground and open-pit sources were operated [
15]. Mining operations at Rokana were halted in the 1990s owing to unfavorable economic viability, resulting in ZCCM placing the mine under care and maintenance [
22].
The Rokana mine generated large amounts of mine waste in the form of tailings (tailings dams a-l in
Figure 2) and caused serious environmental problems [
15]. All tailings dams in the district and around the abandoned Rokana Mine are currently closed, which means that all mine waste currently produced in Kitwe is transported to TD 15A (located in Kalulushi District), the only operating tailing dam near Kitwe (
Figure 2).
2.1.3. Kitwe District Population
The 2010 population census shows that the Kitwe District population increased from 347,024 in 1990 to 517,543 (27% of the Copperbelt Province’s population) in 2010 (
Table 1) [
23]. The population reached 661,901 in 2022 [
24], with a population density of 814.7 people per km
2. Approximately 276,000 people in Kitwe District are older than 18 years [
25]. The average annual population growth rate of Kitwe is 2.1% [
24].
2.2. Remote Sensing Analysis (Landsat Imagery and Processing)
Open-access Landsat 5 top of atmosphere (TOA) and Landsat 8 TOA reflectance data available on the Google Earth Engine (GEE) were used to create the satellite images (
Table 2). Many Landsat images in this platform are processed with a relatively high level [
26]. The Landsat 8 and 5 TOA reflectance data from Collection 1 Tier 1 are the highest possible quality imagery available [
27]. In the Tier 1 collection, scenes were georegistered consistently, indicating that all images underwent correction for displacement using ground control points and digital elevation model data. Within this collection, a root-mean-square error ≤ 12 m was used to register all images [
27]. The geometric registration guarantees that pixel-to-pixel correspondence is essential for the multitemporal image integration [
28].
In addition to geometric registration, radiometric normalization is necessary for multitemporal imagery [
29]. This normalization ensures consistent spectral-radiometric properties throughout observations from different days or sensors [
30]. During the radiometric calibration, the unprocessed and raw digital numbers for each spectral band in a Landsat scene is converted into at-sensor radiance values that account for the specificities of the sensor acquiring the imagery, including mechanical failures, or deterioration in sensor quality and measurement changes [
30,
31]. For consistency with other scenes, the Tier 1 collection was elected in this study because all images in the collection have already been radiometrically calibrated with well-established methods [
28].
Following radiometric normalization, the at-sensor reflectance was converted into a planetary reflectance value [
32]. Images can be converted into either surface reflectance or TOA values [
33]. The TOA collection was selected instead of the surface reflectance data because initial tests indicated that a Kauth–Thomas linear transformation significantly enhanced classification accuracy [
34]. Presently, the Kauth–Thomas coefficients for Landsat 8 data are best established for TOA data [
33,
35]. TOA data have consistently been utilized to generate multitemporal image mosaics, resulting in high-accuracy land cover classifications. This is particularly notable when spectral indices or transformations are applied to enhance spectral signal [
11,
36,
37,
38,
39,
40]. Well-established calibration coefficients were used to compute the TOA reflectance values for the Tier 1 collection [
30].
Between 1990 and 2020, a cloud-screening algorithm was used to eliminate cloud-contaminated pixels from each Landsat image, utilizing quality assessment bands. Six-month composites were then generated by calculating the median value from the images of the target months (July to December) [
41]. For instance, for 1990, pixel values were calculated by taking the median of all cloud-free pixels from images between 1 July 1990, and 31 December 1990. A six-month window was used to ensure the availability of at least one cloud-free pixel for each composite and seasonality was considered. In this study, the dry season (July to December) was considered to clearly differentiate the spectral signatures of LULC types.
2.3. LULC Classification
Based on existing classifications of land cover (National Remote Sensing Centre) and field observations, the LULC for each year (1990, 2000, 2010, and 2020) was classified into five categories (
Table 3): bare land (including mining areas), built-up area, forest, grassland/pasture/agricultural land, and water. The Random Forest (RF) decision tree classification algorithm in the GEE was used to extract the five LULC classes.
Before selecting the training samples, empirical analyses of satellite imagery, Google Earth images, and topographic sheets of the district were carefully performed. For most of the classes, a minimum of 50 training samples were collected across the study area.
This study used RF, a tree-based classifier with K-decision trees, to perform supervised pixel-based classification [
42,
43]. The RF addresses the overfitting problem through building an ensemble of decision trees [
43,
44]. To classify the composite Landsat images into five LULC classes (
Table 3), this study trained the RF classifier on the GEE platform using 250 training samples.
2.4. Determination of the Mining Areas
LULC maps prepared by Landsat can identify bare land, including mining areas. However, the mining area must be distinguished from other bare lands; therefore, PlanetScope satellite images were used. The 2020 PlanetScope image was acquired from
https://www.planet.com/products/planet-imagery/ (accessed on 19 November 2020). The satellite image is a 4-band multispectral image (blue, green, red, near-infrared) with a 3 m spatial resolution (image ID planet/item_id:”2792051_3533219_2020-10-29) (accessed on 19 November 2020).
This study did not use a maximum likelihood classifier; instead, a Support Vector Machine was used to analyze the satellite images using ArcGIS Pro software 2.8.2 (ESRI, Redland, CA, USA). This is because statistical methods, such as the maximum likelihood classification method, possess certain limitations, particularly concerning distributional assumptions and constraints on data input [
45]. Many studies claim that machine learning algorithms, including Support Vector Machine, may frequently achieve higher accuracy in classifying a dataset than conventional classifiers [
46,
47,
48].
Before classification, training samples were created by the region of interest tool for the five classes (mining areas independently classified). Signature sets involve selecting a set of pixels with similar spectral values, specifically for one class. As a result, for each identified class in the image, a signature was assigned and a signature set was integrated. Finally, the PlanetScope image was classified, and the classified image of the mining area was merged with the Landsat image.
2.5. Validation
The composites from different years were separately trained and validated in the classification process. The classifier was trained using approximately 70% of the sample points, with the remaining 30% utilized to assess the accuracy and validate the RF classifier. The error matrix was used to calculate the RF classifier accuracy and kappa statistics. The final maps were compared with the high-resolution imagery on the Google Earth.
2.6. Class Smoothing Process
Class smoothing was performed during image processing to remove noisy pixels using ArcGIS Pro. This was done because the process of classification typically results in a tiny percentage of unclassified, poorly classified, or solitary pixels, which are frequently seen around the boundaries of two areas that are unambiguously assigned [
49]. This can create a “pointillist” or blurry appearance that may pose challenges for map production [
50,
51]. It is desirable to homogenize the classification by reassigning pixels to one or the other class [
52,
53]. To minimize unnecessary details and improve the classification accuracy, post-classification smoothing using a majority filter is fundamental [
54]. This was performed by eliminating pixels < 900 m
2 (less than the size of one pixel of the Landsat images). Filtering entails reassigning isolated pixels to the predominant class in which they are located [
55]. Classified images usually manifest a salt-and-pepper appearance, because of the inherent spectral inconsistency faced by a classifier when applied on a pixel-by-pixel basis [
56]. In the bare land class, for instance, scattered pixels throughout the mining area boundary may be labeled as built-up areas, or vice versa. To address such instances, it is desirable to smooth the classified output, highlighting only the main classification [
50,
57,
58].
2.7. Field Survey and Accuracy Assessment
The classified images from the GEE were exported to ArcGIS Pro for post-classification, where an accuracy assessment was performed.
A widely employed tool for evaluating map accuracy is an error matrix [
59], which aligns and compares pixels in classified images with ground data [
60,
61]. The producer’s accuracy assesses errors of omission, measuring the effectiveness of classifying real-world land cover types [
62]. The user’s accuracy assesses commission errors, representing the likelihood of a classified pixel matching the land cover type of its corresponding real-world location [
49].
The Kappa statistic is a separate multivariate technique to assess accuracy [
49]. The reference data for the 2020 map were a combination of data collected during fieldwork and Google Earth Pro image archives. However, clear and updated Google Earth Pro images were lacking for 2010, 2000, and 1990. In this study, stratified random sampling was employed to collect a minimum of 40 reference points. The sampling was based on the sizes of the land use and land cover (LULC) classes for the classified image in 2020. The Kappa coefficient was calculated computed based on reference [
49]. A value > 0.80 shows excellent agreement, and that between 0.4 and 0.80 suggests moderate agreement between classification categories.
2.8. Collecting Legal Documents on Mine Closure
Legal documents on mine closures in Zambia were retrieved from Blackhall’s Laws of Zambia (
https://zambialaws.com (accessed on 20 July 2022)), which provides free access to Zambian laws containing primary and secondary legislation. All acts and subsidiary legislations were enacted in terms of principal legislation. This website was chosen because all acts are presented in fully revised and annotated forms, and their online database is amended in line with the publication of new and amended legislation. Blackhall’s Laws of Zambia are intended as tools for both the legal community and public. It is arranged such that any enactment can be easily searched for on the Internet in chronological and alphabetical order with annotated amendments.
The laws and policies regulating corporate environmental practices in Zambia, with an emphasis on mining, were examined. The analysis of the legal and regulatory framework involved an examination of the extent to which it met international best practices and standards of corporate conduct and to which self-regulatory mechanisms were accommodated under the framework. This was accomplished by reviewing the four mining-related acts, laws, and relevant statutes regarding corporate environmental practices in Zambia. These statutes include the Mines and Minerals (Environmental) Regulations of 1997, the Environmental Protection Fund Regulations of 1998, the Environmental Management Act of 2011, and the Mines and Minerals Development Act of 2015.
5. Discussion
The trends of general decrease in forest coupled with subsequent increments of grassland/pasture/agricultural land, mines/bare land, and built-up areas in the study area are attributed to population expansion, rapid urbanization of the district, influx of small-scale mine operators (usually unemployed youth), and expansion of commercial (private) mining activities. When the changes in LULC shown in
Table 10 are extrapolated, some potential threats are expected to arise in Kitwe District in the near future. One future concern is the continuation of urban expansion due to the constant increase in population from both natural growth and rural-to-urban migration [
91]. The United Nations predicts that Kitwe’s population will be 1,005,000 in 2030 [
92]. In general, urban expansion negatively impacts the environment because more land and building materials are required for the construction of houses and other buildings [
93]. This means that more forest resources are needed to construct houses and other buildings, which increases pressure on the remaining forests in the district [
94]. Second, as the population increases, the demand for food also increases. This implies that grassland/pasture/agricultural land is likely to increase in the future to meet the demand for food, with further pressure on forests, as land needs to be cleared when preparing for cultivation [
25]. Third, due to the current increase in the mining area and copper production, new bare land (mining area) will be created, which implies that there will be further loss in the area covered by forest and vegetation among the grassland/pasture/agricultural land classes (
Table 10).
When mineral resources are depleted, mines are closed. This implies that tailings dams and open pits should also be considered. According to mine closure legislation in Zambia, mines must be sustainably closed. However, the Zambian legal framework has inadequate provisions regarding mine closure planning and plans, financial assurance, incomprehensive relinquishment, and post-closure obligations [
75]. Furthermore, little is known about the socioeconomic dimensions from the perspective of mine closures because mine closures related to socioeconomic dimensions do not have legislation. While mine closure planning practices and the necessity to plan for closure are generally acknowledged within Zambia’s existing legal framework, reference [
75] evaluates that the framework may not be fully comprehensive and might not completely align with sustainable development principles. The evaluation, for instance, identified the absence of provisions mandating the submission of comprehensive mine closure plans. The extent of stakeholder participation in the current mine closure planning process appears somewhat insufficient, primarily dictated by the broader Environmental Impact Assessment (EIS) assessment process [
83]. To ensure the successful completion and rehabilitation of mine sites, certain aspects of the current framework must be improved. This study suggests the incorporation of detailed mine closure objectives/standards within the legal framework [
95]. Lastly, developers should be granted additional financial instrument options, enhancing provisions on financial assurances to accommodate the unique characteristics of each operation.
Adequate policies and intervention should be put in place [
96] to minimize the above mentioned negative impacts on LULC changes. One of the intervention measures to put in place to minimize the impacts of LULC is putting stringent and rigorous efforts into the re-afforestation of affected areas, such as old tailings dams and overburdens. Resettlement and other measures aimed at restoring degraded land to its original state after mining activities should be intensified by mining companies [
81].
For sustainable mine closures, revisions are required in the provisions of mine closure regulations. According to Section 119 of the Mines Act [
84], the Minister of Mines and Minerals Development should be empowered to regulate mine decommissioning and closure. This study proposes amending the current Mines Environmental Regulations, the primary legislation on mine closures, to remove the closure provisions. This study also recommends incorporating mine closure provisions in the well-established Environmental Impact Assessment (EIA) process, despite the requirement for specific regulations under Section 119 by the Minister of Mines and Minerals Development. This integration ensures greater visibility for mine closures throughout the process.
Collaboration and coordination are necessary among governmental agencies, such as the ZEMA and the Ministry of Mines and Minerals Development. To improve the sustainability of ecosystems and smart land use, governmental agencies and organizations must refine their planning. These collaborative efforts should encompass social, environmental, and economic factors to ensure the sustainability of services affected by rapid urban expansion. Enhancements are needed in afforestation of mine dumping sites, forest harvesting regulations, and food production. Thus, the emphasis in planning and decision-making should be on safeguarding remaining forest reserves such as Mwekera Forest, which is currently facing encroachment. Incorporating LULC analysis in a mining area is crucial for policymaking as it provides insights into the environmental impact of mining activities, aiding in the formulation of effective regulations. By utilizing LULC data, policymakers can identify areas of concern, such as deforestation or habitat loss, prompting necessary amendments to mine closure legislation to ensure sustainable resource extraction practices and environmental protection.
Finally, the limitations of this study include the lack of high-resolution images from 1990 to 2010, the unavailability of copper production data at the district level, hindering a precise examination of the relationship between mining activities and environmental impacts, and the study’s dependence on linear extrapolation for predicting future threats that may generalize the complex socioeconomic and environmental dynamics, potentially resulting in forecasting inaccuracies. Therefore, by integrating advanced remote sensing technologies, acquiring critical data through collaborations (with government agencies), adopting sophisticated modeling techniques, and incorporating qualitative methods, future studies can bridge the identified gaps and provide a more robust foundation for understanding the complex dynamics of LULC changes in Kitwe District, thereby contributing to improved environmental management and policy formulation.
6. Conclusions
Remote sensing and GIS tools were adopted to analyze LULC changes in Kitwe District, Zambia for the first time. This study identified LULC change patterns for 1990–2020. Significant changes in bare land (increased from 20.90 km2 in 1990 to 35.58 km2 in 2020, with 64.5% (22.95 km2 in 2020) being mining area), built-up area (increased from 37.06 km2 to 72.90 km2), and grassland/pasture/agricultural area (increased from 369.43 km2 to 412.96 km2) were observed. Forest area decreased from 366.34 km2 in 1990 to 271.04 km2 in 2020. Population and mining are the main drivers of the overall increase in built-up areas and bare land in the study area. To mitigate the negative impacts of LULC change and sustain community livelihoods, the interactions between the geoecological and socioeconomic processes leading to LULC change and associated land degradation must be understood. This study suggests that ongoing programs for government-initiated sustainable land management should be strengthened. Effective collaboration and coordination among governmental agencies are essential to address the multifaceted challenges posed by rapid urban expansion and mining activities. By integrating LULC analysis into policymaking, stakeholders can proactively identify and mitigate the environmental risks associated with mining, promote sustainable land use practices, and safeguard critical ecosystems for future generations.
Zambia’s legislation on mine closures has been problematic with no significant amendments made since 1995, when the country enacted its first legislation on mine closures. Despite the existence of a mine closure framework, the legislation has yet to be effectively implemented, particularly by the state. This knowledge gap should be filled by examining the impact of mines on the environment and the Zambian legal framework for sustainable mine closures.