The Bosomtwe Lake Basin stands in the Bosomtwe District of Ghana’s Ashanti Region, which is home to Kutunase as its capital (Figure 1). The district, which is situated at latitude 6˚ 30 Nand longitude 1˚ 25 W, was put to birth by dividing two separate districts, Atwima Kwanwoma and Bosomtwe Atwima Kwanwoma and covers an area of about 178 square km [36] . Residents in and around the basin, as

well as others in the surrounding region, earn a living from the lake resort, which is a well-known tourist destination [36] . Sand, clay, and gold that are extracted from the lake’s basin also add to the region’s economic growth. This basin, which lies in the equatorial zone, receives rainfall that is typical of the moist semi-deciduous forest zone, which is home to a variety of highly valuable tropical hardwood species [36] .

There are several tree species in the district, including Onyina (Ceiba pentandra), Mahogany (Khaya ivorensis), and Wawa (Triplochiton scleroxylon) [53] . The native forest cover in the district is being replaced by secondary forests and grasslands due to human activities such as unlawful small-scale gold mining, careless tree felling, illegal timber harvesting, and inappropriate farming practices [54] . The district’s expansion in settlements is evidenced by its closeness to Kumasi Metropolis. The district is also one of the most active in the Ashanti Region because of its favourable infrastructure, economic activity, and booming tourism sector [54] [55] .

We analyzed Landsat images from the following years: 2001, 2005, 2010, 2015, and 2018. To meet the need for a post-classification comparison, change detection technique, which involves grouping images obtained at different times, images of different dates were picked using the Landsat Enhanced Thematic Mapper (ETM+) satellite with a spatial resolution of 30 m × 30 m (Mas, 1999). The United States Geological Survey (USGS) website is where the above images were retrieved. The study’s foundation was provided by the data sources listed in Table 1. Image pre-processing was done to fix the geometric and radiometric distortions of the images due to changes in the earth’s rotation, atmospheric conditions, and sensor effects. An embedded haze correction module in the Erdas.

Imagine software was used to address haze in the 2001 and 2005 images, which is frequently caused by small floating particles in the atmosphere. The images from 2010 to 2018 were left unchanged because their reduced cloud distortions were not detrimental to the classification. To classify all types of land cover in the image, training samples from Google Earth were used, and to acquire a true representation of the various land cover classes, these samples were selected and looked at again.

The major software employed in this study included ERDAS Imagine 13.1, Environmental System Research Institute (ESRI) ArcGIS 10.6, and QGIS software version 3.4.2. The gathering of primary and secondary data is an essential part of all research endeavours [43] . The original data was acquired from ground-based coordinates and field observations; the secondary data was collected from the website of a private organization and Google Earth. Using the Landsat satellite, enhanced and geometrically adjusted images were gathered from the United States Geological Survey website for the years 2001, 2005, 2010, 2015, and 2018 [56] . The appropriate images were then downloaded and made available for further enhancement [43] .

Preparing images for classification is an essential step in land cover categorization as it affects the precision of the classification outcomes [43] [44] . The image preprocessing phase increases the accuracy of the classification findings by correcting the image’s geometric and radiometric distortions and establishing a clear connection between the secondary data and the primary data collected in the area under study [43] [44] [57] . Radiometric and geometric corrections were carried out to prevent obtaining false results, as photographs that bypass the rectification stage can result in unreliable data, due to the failure of some satellite scanlines and circumstances such as cloud cover [57] . Layer staking, picture improvement, radiometric correction, pan sharpening, study area extraction, data stacking, and edge masking were among the preliminary processing techniques employed in this study. The methods shown in Figure 2 were followed to process the satellite imagery.

The data acquired (Landsat 7) were susceptible to information gaps due to the failure of the scanline’s corrector linked with the ETM+ instrument. Hence, a process of data masking (gap filling) was carried out to improve each band using QGIS software. The data went through a stacking process after its masking. Since composite colour bands work best for grouping vegetation and highlighting the unique characteristics of different land cover classes, NIR, red, and green bands were combined to create a fake colour composite [43] [56] . The ERDAS Imagine computer application was used to complete the data stacking process.

The resolution of the launched photos was enhanced through a pan-sharpening process using the same ERDAS IMAGINE computer software, enabling small objects to be seen and seem finer than they were before [43] [58] . The study area was acquired by superimposing the Bosomtwe district shapefile over the acquired satellite picture. Subsequently, the research region was extracted making use of ERDAS IMAGINE computer software from the pan-sharpened image. Using the services of the Google Earth program, distinct locations with vegetation, bare terrain, water, and settlement were identified for every class. Fifty points were chosen for each of the four land cover classes, with the pre-cause that the points should be distributed equally and not geared to one point. This yielded a total of 200 training data points for all the different classified photos. Each class was assigned a value: water had a value of 1, settlement had a value of 2, bare ground had a value of 3, and vegetation had a value of 4.

Image Classification: Grouping all of the pixels in a photo into classifications related to land cover is the core objective of image classification. The satellite images were categorized into different classes with a pixel-based technique called the maximum likelihood classification algorithm. Experts in remote sensing assert that the maximum likelihood classification algorithm is a widely used technique because of its strong theoretical basis and adaptability to different kinds of land use and cover, satellite imagery, and data [50] . Supervised categorization groups pixels into classes based on a comparison between each pixel’s spectral characteristics and a set of typical pixels from the user’s training data [50] . Training datasets, classification, and accuracy evaluation were the three processes of supervised classification performed [43] . After a reconnaissance was conducted to identify land cover classes in the area under examination, four uniqueland cover classes were grouped: water body, vegetation, bare ground, and settlement (Table 2).

With the use of a common Microsoft Excel computing program, the surface areas of the various land cover classes were used in a trend analysis to determine the amount occupied by each class and create a graph using lines that illustrate the variations throughout the research years between the classes. This means that following classification, the ERDAS Imagine software provided us with the characteristics of the different land cover classes, from which their corresponding surface areas were derived. We then plugged those attributes into a written Excel program to create a chart that visually represented the trend of changes in land use and cover from the years 2001 to 2018. To calculate the producer accuracy, user accuracy, overall accuracy, and kappa coefficient, respectively, the following equations were used: Equation (1), Equation (2), Equation (3), and Equation (4) [43] .

P r o d u c e r ' s a c c u r a c y ( % ) = ( x k k x + k ) × 100 (1)

U s e r ' s a c c u r a c y ( % ) = ( x k k x k + ) × 100 (2)

O v e r a l l a c c u r a c y ( O A ) = 1 N ∑ K = 1 r n i (3)

K a p p a c o e f f i c i e n t ( k ) = N ∑ K = 1 r x k k − ∑ K = 1 r ( x k + ⋅ x + k ) N 2 − ∑ K = 1 r ( x k + x + k ) (4)

Where x_kk represents the total of the pixels in rows “k” and columns “k,” N denotes the total number of pixels, and r is the number of classes. Subscription x + k represents the total samples in column “k” of the error matrix, while x_k + represents the total samples in row “k” [43] . K > 80% denotes excellent accuracy and high agreement, 40% – 80% denotes moderate agreement, and < 40% denotes poor agreement [36] [59] .

After classifying images, determining accuracy is fundamental [43] . It is essential to adopt an evaluation procedure to guarantee the accuracy of the classification using dependable and effective means as LULC classification has the tendency to carry errors [60] [61] . We obtained the confusion matrix data from the ground truth data collected during the fieldwork, and we used this data with the kappa coefficient to assess accuracy. One well-liked instrument for assessing the precision of LULC classification methods is the confusion matrix [62] [63] [64] . The total accuracy (Equation 3), producer’s accuracy (Equation 1), user’s accuracy (Equation 2), and kappa coefficient (Equation 4) were computed from the generated error matrix. Table 3 and Table 4 present the accuracy level for the classified images.

Starting with the categorization of the 2001 satellite images, the researcher produced the following accuracy results (Table 3). The overall accuracy of the classification was 98.80% (Table 3). The producer accuracy for the classes of water and settlement was 100%, meaning that there was a perfect chance that the water and settlements on the ground would be classified and depicted as water and settlement on the classified map. The percentage of bare land was 60%, indicating that it was not categorized precisely when compared to water and settlement. In contrast, vegetation achieved 100% accuracy, indicating a good approximation of the actual ground feature. The percentage of sites that were absent (omitted) from the appropriate class on the classified map referred to as the error of omission was 6.4% for water, 0% for settlement, 0% for bear land, and

4.35% for vegetation. The user accuracy for each class on the map, which shows the likelihood of the class appearing on the ground, was 94.6% for water, 100% for settlement, 90.2% for bare land, and 95.83 for vegetation (Table 3). This means that, in the majority of cases, a user using the classified map has a chance to recognize the classified classes on the ground where they are located. In terms of water, bare land, settlement, and vegetation, the error of commission was 6%, 0%, 0%, and 4.17%, respectively.

A satisfactory total accuracy of 88.96% was achieved in the 2005 classification (Table 3), indicating that the outcomes were precise and, therefore, reliable. Water and settlement had a producer accuracy of 100% (Table 3), which means that what was collected on the ground precisely matched the data on the categorization map. Vegetation had a better depiction of the actual ground feature with 76% accuracy, while bare land with an accuracy of 63% was not properly identified. The locations that were absent (omitted) from the appropriate class, were 0.03% for settlement, 9.09% for bear land, 8% for vegetation, and 0% for water. User accuracy, which measures the likelihood that a class on the map will exist on the ground was 100% for water, 96.67% for settlement, 87.50% for bare land, and 90.48 for vegetation (Table 3). This means that a user using the classified map is almost certain to see the classified classes on the ground. For water, the error of commission was 0%, for habitation, it was 3.33%, for bare ground, it was 12.5%, and for vegetation, it was 9.52%.

2010 recorded a 96.86% total precision rate (Table 3). On the classification map, water on the ground was precisely categorized as water because the producer accuracy for the water class was 100% (Table 3). The percentages of settlement, barren land, and vegetation were 94.34%, 96.20%, and 95.45%, respectively (Table 3), implying that all three had a good depiction of the actual ground feature. The percentage of the classified map that was not included in the correct class was 0% for water, 0.94% for settlement, 6.33% for bear land, and 4.55% for vegetation. 99.01% for settlement, 93.83% for barren terrain, 95.45% for vegetation, and 100% for water were the user accuracy rates on the field (Table 3). Generally speaking, those who utilize the classified map are likely to see the actual classified classes in reality. Concerning the classes of water, settlement, barren ground, and vegetation, the errors of commission were 0%, 0.99%, 6.17%, and 4.55%, respectively.

In 2015, the overall accuracy rate was 95.24% (Table 4). The producer achieved 100% accuracy for both water and bare land classes (Table 4). This implies that naked land and water on the ground were precisely categorized as water and bare land on the classification map. Both the vegetation and the settlement exhibited a good depiction of the actual ground features, with respective percentages of 90.48% and 94.44% (Table 4). The percentage of omitted class was 0% for water, 9.52% for settlement, and 14.28% and 5.55% for vegetation and bare land, respectively. The user accuracy was 87.50% for bare terrain and 100% for water, settlement, and vegetation, suggesting that the classified classes are most certain to be observed as they are in reality (Table 4). The percentage of error for water, settlement, bare land, and vegetation were respectively 0%, 0%, 12.50%, and 0%.

A total of 98.95% accuracy was achieved in 2018 (Table 4), with a producer accuracy of 100% for water and vegetation (Table 4), these ground features were precisely categorized as water and vegetation on the classification map. The accuracy of bare land was 96.0%, whereas that of settlement was 99.32% (Table 4). Water had an error of 0%, settlement had 0.68%, while bare ground and vegetation had 4.08% and 0% error, respectively. It is evident from the user accuracy of 100% for water and vegetation, 99.32% for settlement, and 95.92% for bare land that users using the classified map are likely to see the categorized classes on the ground. For bare land, the error of commission was 4.0%, 0.68% for settlement, and 0% for water and vegetation, respectively.

A statistical test was used to determine the kappa coefficient, which indicated how accurate the classification was. In essence, Kappa assesses how successfully the classification was carried out in relation to values that were assigned at random. Thus, did the categorization do better than values that were chosen at random? The range of the kappa coefficient was −1 to 1. Whereas a kappa of 1 indicates that the classification is significantly better than randomly assigning values, a value of 0 indicates that the classification is no better than randomly assigning values. A negative kappa coefficient also indicates that the classification is significantly worse than randomly assigning values. Equation 4 gives the formula for the kappa coefficient. As stated in Table 3, the 2001 total kappa value was 0.9659. For both bare ground and vegetation, the conditional kappa was 1, indicating a considerable improvement over values that were chosen at random. The scores obtained for settlement and water were 0.9492 and 0.9265 respectively. This means that all of the coefficients are near 1, meaning that classification is much stronger than choosing values at random. The total kappa value for 2005 was 0.8568 (Table 3). The conditional kappa for water was 1, while the conditional kappa for bare ground, vegetation, and settlement were 0.8654, 0.8590, and 0.9589, respectively. These values were all near 1, suggesting that the classification of these variables was superior to assigning values at random. In 2010, the overall kappa value was 0.9588 (Table 3).The conditional kappa values were as follows: for water, it was 1; for bare land, it was 0.9851; for vegetation, it was 0.9512; and for settlement, it was 0.9179.This further demonstrates that all of their coefficients are near 1, which considerably increases their classification over value allocation at random. The overall kappa value for 2015 was 0.9359 (Table 4). Whereas the conditional kappa for bare land was 0.859, it was 1 for water, vegetation, and settlements. It is evident that all the coefficients are equal to or nearly equal to 1, showing that the classification is noticeably superior to choosing values at random. The 2018 total Kappa value was 0.9842 (Table 4). As opposed to 0.9518 and 0.981 for bare ground and settlement, the conditional kappa for water and vegetation was 1. Since all of these values were relatively near to 1, it was clear that their classification was more excellent than choosing values at random.

A map showing classified land cover in the Bosomtwe Lake Basin from 2001 to 2018 is presented in Figure 3. The figure shows that, despite some areas having already experienced depletion, there was still a sizable amount of natural vegetation cover in 2001.After water, which makes up the majority of the basin, the histogram in Figure 4 shows that natural vegetation is the next land cover with a significant surface area, indicating that in 2001, the basin had a good proportion of natural vegetation. The histogram also indicates that the percentage of settlement was 11.92% of the Lake Bosomtwe Basin’s total surface area. This indicates that the basin did not experience significant development during the reporting year, which helps to explain why the Bosomtwe Basin had a high percentage of natural vegetation in 2001.

Figure 3 reveals that in 2005, water made up the majority of the basin’s cover, followed by vegetation, settlements, and bare ground. The various land cover classes’ percentages are shown in Figure 4’s histogram. The majority of the basin (49.32%) is made up of water, with the remaining portions being made up of barren land (8.05%), vegetation (32.19%), and settlement (14.44%). The histogram shows that while vegetation declined in the Bosomtwe Basin area in 2005, the number of areas without vegetation and settlements increased.

Figure 3 illustrates how, in 2010, water remained the predominant type of cover in the basin, with vegetation accounting for the second-highest percentage, behind bare ground and settlement, respectively. The various land cover classes’ percentages are also displayed on the histogram. The basin’s major component is made up of water (49.76%), which is followed by vegetation (22.40%), settlement (18.83%), and bare land (10.01%). Judging from the histogram, the Bosomtwe Basin area saw an increase in bare land, a decrease in vegetation, and an increase in settlements in 2010. The basin’s cover was still largely made up of water in 2015, as seen in Figure 3. Vegetation, habitation, and bare ground were the next categories of cover. In line with the histogram, water makes up the majority of the basin (47.09%), with vegetation (24.97%), settlement (17.49%), and bare land (8.75%) coming in second, third and fourth respectively. Vegetation, habitation, and bare ground were the next categories of cover. In line with the histogram, water makes up the majority of the basin (47.09%), with vegetation (24.97%), settlement (17.49%), and bare land (8.75%) coming in second, third and fourth respectively. From the findings, it can be deduced that in the Bosomtwe Lake area in 2015, there were fewer settlements, more vegetation, and fewer bare areas. Figure 3 shows that in 2018, water accounted for the largest portion of the basin’s cover, followed by vegetation, habitation, and bare ground. The histogram reveals that the majority of the basin (47.99%) consisted of water, with the other portions being made up of vegetation (20.78%), bare land (12.45%), and settlement (18.46%). During the reporting year 2018, the histogram explains why there was a rise in habitation, a corresponding increase in bare land, and a loss in vegetation in the Bosomtwe Lake area.

The land cover patterns and transformations in the Bosomtwe Lake Basin from 2001 to 2018 are presented in the chart in Figure 5. With relatively little development taking place in the basin at the time, the land cover map in 2001 indicated a high proportion of vegetation cover. Even though the basin had already been drained, there was still a sizable amount of vegetative cover because the settlement made up about 11% of the basin’s overall proportion of land cover classes.

The findings indicate that the low depletion can be linked to the area’s low population growth or development. Proceeding to 2005, it was observed that there had been a decline in vegetative cover as compared to 2001, as well as an increase in settlements in the area. When comparing 2005 to 2001, the percentage of bare land increased. According to [32] , one important element that causes degradation and deforestation is the increase in population and settlement of an area, which could have contributed to the decline in vegetation during the time. As said by [65] this could also imply that formerly forested areas have changed into or are already being transformed into barren ground or farming land. As the population of the Bosomtwe area increased due to immigration, there may have been widespread tree-cutting for residential, agricultural, or commercial purposes, which contributed to the reduction in vegetation. The loss of vegetation may also be due to the absence of environmental policies and practices aimed at protecting the forest. The Bosomtwe Lake Basin may have lost vegetation cover in 2005 as a result of all of these aforementioned activities.

The vegetative cover of the basin was found to have significantly decreased in 2010, as Figure 5 shows. The area’s population growth can be attributed to the significant amount of development that occurred in the basin in 2010 as a result of various human activities. Given that the basin’s previous forest may now have been replaced by farms, settlements, or even barren ground, the decline can also

be linked to the region’s growing population, the pace of migration, and urbanisation [8] , [65] . Environmental laws and regulations that safeguard forests were absent in the basin, which could also be another explanation for the decline in vegetation.

2015 saw a rise in the amount of vegetative cover in the Bosomtwe Lake basin, defying the overall trend of declining cover. As the total area of barren ground and settlements reduced, the percentage of natural vegetation increased. Subsequent investigation showed that practices and efforts for environmental conservation that the basin’s residents and non-residents observed contributed to the increase in vegetative cover in 2015. It became known that, as part of the World Environmental Day celebration, students from the Kumasi Institute of Tropical Agriculture (KITA) had planted trees in the Bosomtwe basin. This action had been acknowledged by the district chief executive of the Bosomtwe Atwima Kwanwoma district in 2015.250 trees or more were planted by them according to (business Ghana news.com). Additionally, farming was prohibited within 60 meters of the Bosomtwe basin, with penalties for those found farming there (graphic online.com). It is reasonable to conclude that the policies and practices implemented in 2015 to safeguard the environment contributed to the rise in vegetative cover in the basin.

Nevertheless, in 2018 everything had returned to normal and the general trend of declining vegetative cover persisted. The pace of urbanization in the region accelerated, leading to an increase in the number of settlements. There was more bare ground than there was in 2015. The abandonment of the environmental protection policies and practices that were implemented and carried out in 2015 may be the cause of this. As Figure 5 illustrates, immigration, population expansion, and human activity can be cited for the Lake Bosomtwe Basin’s decline in vegetation and rise in settlements and barren areas between 2001 and 2018.

Growing populations both locally and globally have put excess pressure on the environment’s scarce resources, modifying land cover and use in the process [66] . The growing number of individuals relocating to towns surrounding the Bosomtwe Lake Basin subsequently led to the growth in settlements, barren land, and a decline in vegetation. This may be linked to the emergence of resorts and other tourism-related amenities around the basin. The built-up region surrounding Chaohu Lake expanded due to an upward trend in the province of Anhui’s immigrant population [67] . [44] also acknowledged that urbanization and its subsequent disturbances to the natural environment were primarily caused by population growth and economic development.

The inactive or absence of environmental protection policies and management practices contributed significantly to land cover change in the lake basin. The majority of available laws and initiatives prioritised supplying wood for the country’s wood industry over protecting or conserving the natural vegetation, which encouraged excessive use of the forest and led to the degradation of forest reserves. Reforestation and other restoration initiatives intended to protect forest reserves and promote sustainable forestry also failed because of financial limitations and a focus on the economic value of forests rather than their sustainability, highlighting the differences between the policies’ intended and actual execution [68] .

The rising need for resources is a further issue to speculate. This finding agrees with the outcome of [69] who noted in their study that rapid population increase and the demand for forest products deplete forest cover. The demand for goods and services, fuel, housing, household appliances, and land for agricultural operations rises as more people relocate to the Bosomtwe Lake region. Pressure from rising human populations and increased demand for environmental resources deplete the environmental resources that are already accessible [32] [37] . Thus, every environmental management strategy should consider how population growth is affecting the physical makeup of forest cover [8] .

Changes in the environment also occur as a result of human interference. Land used for agricultural purposes, which is an economic activity carried out by humans, has replaced majority of the original natural vegetation [29] . Human activities such as population growth, poverty levels, food insecurity, livestock husbandry, and infrastructure relocation are the key drivers of environmental change [70] . Likewise, [71] noted decreasing output, rising incomes, and increasing pressure from populations. In addition to the modifications induced by human activity, other factors that have been linked to land cover change include population pressure, poverty, farming operations, habitat development, infrastructure growth, lack of alternative livelihood, unemployment, and drought [37] . Our results support that of other scholars who discovered that forest lands are being converted to farmlands to meet the current demand for land for agriculture [72] . In addition to agriculture, the Bosomtwe Basin’s territory is exploited for human settlements, feeder roads, local track construction, and small-scale illicit mining [37] . The basin’s natural cover changed as a result of all these human activities.

The incorrect management of both liquid and solid waste in the basin is not overlooked. People dispose of their waste in the lake or the forest since some communities lack waste bins, the available ones are also inaccessible owing to the rising numbers of people in the area. Some communities do not have access to public restrooms, thus those who live in areas without toilets or with limited access often use the lake or the forest as their lavatory [37] .

Ghana began administering its natural environment around 1906. However, following the United Nations Environment Conference in 1972, which resulted in the establishment of the Environmental Protection Council (EPC) and the implementation of national, regional, and local environmental protection policies in 1974, concerns about environmental degradation started to receive attention [68] . Significant forest management initiatives, including the creation of permanent forest estates, forest reserves, and legislation protecting forests, were implemented to guarantee the sustainability of forest resources, which slowed down the nation’s forest resource depletion [73] . Surprisingly, the majority of the initiatives focused on providing wood for the nation’s wood industry, which promoted overuse of the forest and caused forest reserves to deteriorate. Due to funding constraints and an emphasis on the financial worth of forests over their sustainability, afforestation and other restoration projects meant to preserve forest reserves and advance sustainable forestry failed. There was a clear difference between the intended and actual implementation of the forest policy [68] .

To guarantee sustainable development and safeguard the environment, natural resource management, biodiversity management, protected forest management, land management, and forest restoration all require statutory legislation [28] . This study will assist the appropriate regulatory bodies in putting rules in place that will ensure that the sustainable use of natural resources is monitored. The implementation of community-based forestry, a shift from conservation for timber supply to conservation for sustainable development, and the enforcement of other forest policies are further recommendations made by this paper. For sustainable forest management to be achieved while safeguarding resources for future generations, local people and communities must be engaged and empowered [68] .