Impacts of farming and herding activities on land use and land cover changes in the north eastern corridor of Ghana: A comprehensive analysis

ABSTRACT

This study was conducted to investigate the effect of farming and pasture area extensions on land use and land cover in the North Eastern Corridor of Ghana. Landsat 5 TM⁺ image period of 2000 and Landsat 8 TOA satellite for the periods 2013 and 2022 were used. All images were captured at approximately the same period to ensure the selected images had the same reflectance values. A supervised machine learning technique using the algorithm of the random forest classifier was employed for the preparation of the classification of thematic maps. The Markov chain model was used to examine the dynamics of land use and land cover (LULC) changes in the study area. Visual appraisal of the images indicated some level of notable changes across the various classes from 2000 to 2022. The trend of the various changes in percentage terms also supports this observation. The results reveal that there was an improvement in the vegetation cover from 2000 to 2013 as reflected in the maximum and median NDVI values of the classified images within the period. However, the results show a considerable decline in vegetation health from 2013 to 22. Based on these results, we recommend that a more in-depth analysis to identify other possible anthropogenic activities and factors, that may serve as significant underlying causes of these vegetation cover changes in the region. The Ministry of Food and Agriculture (MoFA) should train farmers to incorporate tree planting into their farming whilst avoiding deforestation and bush burning within the area.

KEYWORDS:

• land use land cover change
• extension of farming
• markov chain model
• normalized difference vegetation index
• remote sensing data

1. Introduction

Understanding the changes in land use and land cover (LULC) is crucial across various domains that rely on Earth observations. These¹ fields include urban and regional planning (Pande et al., 2021; Rogan et al., 2008; Talukdar et al., 2020), assessments of environmental² vulnerability and impact (Barnetson et al., 2019; Shetty, 2019), monitoring of natural disasters and hazards (Barnetson et al., 2019; Kutlug Sahin & Colkesen, 2021), and the estimation of factors like soil erosion and salinity.³ This is especially important in agriculture due to land use disputes among different users of land.⁴ This land users include crop producers who are farmers on one hand and herders on the hand.

Over the years, West Africa, with a particular focus on Ghana, has witnessed an expansion of farming and pasture areas aimed at meeting the food, meat, and raw material demands of a growing population in the sub-region. This expansion is likely to adversely impact the ecosystem, vegetation, water, and soil quality, potentially leading to increased conflicts between crop farmers and animal herders (Aly et al., 2016; Bin, 2008). The terms Land Use and Land Cover (LULC) denote distinct concepts illustrating the interaction between humans and the natural land surface. Land use encompasses human activities on the land, while land cover represents the natural cover or envelope of the land surface (Burley, 1961; Chamling & Bera, 2020; Yiran et al., 2012). Farming and herding, as vital human activities, significantly influence the dimensions of land use and land cover, which constitute essential elements of the landscape. There exists a direct or indirect correlation between LULC and various socio-economic processes such as farming and herding, as well as the geophysical aspects of the earth (Chamling & Bera, 2020; Yiran et al., 2012).

Examining alterations in the boundaries of agricultural and pasture lands is most effectively accomplished by analyzing land cover changes through satellite remote sensing techniques. This approach enables the derivation of quantitative measures for such changes in land use and land cover, providing insights into how they have evolved and expanded over time (Masood et al., 2023; Yiran et al., 2012).

The practices of farming and excessive grazing carry the potential to diminish vegetation density, consequently leading to a decrease in crop yield and livestock productivity, as well as impacting the natural processes of the ecosystem (Barnetson et al., 2019; Masood et al., 2023; Rogan et al., 2008). This therefore makes the analysis of LULC changes, especially in areas of intensive agricultural production activities very paramount. This is crucial due to the fact that the examination of changes in land use and land cover allows for the acquisition of data concerning untouched lands, the conversion or intensification of agricultural areas, and deforestation (Aly et al., 2016; Fu et al., 2023; Yiran et al., 2012). This information holds particular significance because the rate of land cover change in Africa is surpassing expectations, driven by population growth and the imperative to expand and intensify farming and herding for increased food production to sustain the growing populace. This trend has heightened the extent and temporal-spatial scope of human-induced modifications to the land surface in the continent (Chamling & Bera, 2020; Pande et al., 2022).

In Ghana, conflicts between herders and farmers have been explicitly associated with the movement and damage caused to crops by the livestock of herders, along with intense competition for grazing land and water in the country’s savannah areas and forest fringes (Kyei-Poakwah, 2018). The ongoing competition for natural resources, including farming and grazing lands, as well as water resources (Govender et al., 2008; Pande et al., 2021), further exacerbates the tension and volatility in the farmer–herder relationship. This underscores the significance of comprehending the impact of farming and herding activities on land and vegetative cover, as it plays a crucial role in shaping effective policies to mitigate these conflicts and promote unity. Current and comprehensive information in this regard is of utmost importance for understanding the true consequences of agricultural land use, ecosystems, and other environmental processes in the study area, including the extent of expansion of farming and pasture lands (Awad, 2018; Forkuor, 2014; Masood et al., 2023).

However, conventional approaches to gathering demographic data and analyzing environmental samples over an extended period have proven to be ineffective. To comprehend the intricate and multifaceted changes in the environment and land cover, the adoption of innovative technologies such as satellite remote sensing and/or Geographical Information Systems (GISs) becomes imperative (Awad, 2018; Masood et al., 2023; Talukdar et al., 2020). Aerial spatial information data can be acquired to examine and chart alterations in natural resources and environmental conditions resulting from human activities like farming and herding (Masood et al., 2023; Onyango, 2015; Pande et al., 2021; Rogan et al., 2008). In this context, it is crucial to access enhanced and current Land Use/Land Cover (LULC) datasets for an agrarian nation like Ghana, particularly in the corridors of the Northern regions. This necessity arises from the prevalence of small-scale crop farming and herding activities among herdsmen, which are likely to impact the expansion of farming and pasture lands along the corridor. This holds significant importance, given the escalating competition for farming and grazing lands among rural farmers and herders in Ghana, leading to heightened conflicts (Alhassan, 2017; Arabameri et al., 2022; Baik et al., 2006). These conflicts pose national and sub-regional security and policy challenges.

In pursuit of these objectives, the mapping of the expansion of farming and pasture boundaries, along with subsequent identification of Land Use/Land Cover (LULC), holds particular significance. This is underscored by the profound impact these activities exert on the agricultural landscape, leading to rapid modifications that alter land cover patterns. Such changes have far-reaching consequences, influencing biogeochemical and hydrologic cycles, climate change, ecological processes, groundwater quality and quantity, and the overall economy (Daly et al., 2017; Onyango, 2015; Shetty, 2019; Talukdar et al., 2020). Having access to this information is crucial for policy planning, enabling the effective integration of farming and herding and the proper management of environmental resources in both the study area and the broader country. Recent studies have demonstrated the feasibility of utilizing remote sensing approaches and/or Geographic Information Systems (GIS) for mapping and detecting changes in Land Use/Land Cover (LULC) (for example; Casamitjana et al., 2020; Fahad et al., 2022; Macarringue et al., 2022; Mashala et al., 2023; Olorunfemi et al., 2020; Wangyel et al., 2021). The effectiveness of these methods lies in their capacity to collect data on a global scale regularly and provide geospatial series data, making them valuable tools for studies of this nature (Wiesner et al., 2012). Again, machine learning algorithms such as Random Forest (RF), Support Vector Machines (SVM), and Convolutional Neural Networks (CNN) have shown remarkable performance in LULC classification tasks (Barnetson et al., 2019; Karpatne et al., 2016; Shetty, 2019).

However, research conducted in Ghana concerning herders and farmers (Alhassan, 2017) and studies focused on land use, land cover, and other geospatial aspects (Akubia et al., 2020; Ampim et al., 2021; Boateng & Aduah, 2022; Kutir et al., 2022) have overlooked the mapping of the extension of farming and herding routes and the consequent effect on the changes in land use and land cover (LULC) classes. In the northern region of Ghana, herding and farming activities have been the predominant occupations for the majority of the population. Despite the significance of feed sources in livestock production, grazing practices have often been disorganized due to the absence of established rangeland management systems and cattle feed management (Mekasha et al., 2014). In the northeastern corridors of the country, livestock, particularly cattle, primarily graze on extensive uncultivated lands or crop residues, leading to intermittent conflicts between crop farmers and animal herders. Beyond resource conflicts, competition for natural resources, driven by irreversible activities such as land clearing, cultivation, overgrazing, settlements, industrialization, and urbanization, as well as various forms of land management forms (Bessah et al., 2019; Foley et al., 2005; Mekasha et al., 2014; Mengistu et al., 2005; Rahman et al., 2008; Tolera, 2012), severely impacts LULC dynamics in the northeastern part of the country. The limited existing literature on LULC dynamics in the study area results in a lack of information regarding the nature of farming and herding activities and their impact on the LULC dynamics of the region.

This research therefore aims to delineate the classes of land use and land cover (LULC) as well as the alterations resulting from the expansion of farming and pasture area boundaries in the North Eastern Corridor of Ghana. Specifically, this paper evaluates the precision of LULC classification, explores changes in land use classification and land cover, and employs the Markov Chain Model to analyze the detection of changes in land use and land cover. The study anticipates offering valuable information to assist policymakers, NGOs, traditional authorities, and landowners in addressing natural resource conflicts between herders and farmers in Ghana. Additionally, it aims to furnish crucial data and insights for future researchers engaged in remote sensing and geospatial studies, not only in the specified study area but also in Ghana as a whole.

2. Related literature

Extensive literature exists on the effects of agricultural activities, such as crop farming and animal grazing, on LULC (e.g. Köster et al., 2013; Mekasha et al., 2014; Bessah et al., 2019; Nébié et al., 2020, etc.). This body of literature is expanding and intensifying due to renewed calls for sustainable production systems to feed an ever-growing population. The following sections provide a review of some relevant literature on land use, land cover, and geospatial analysis.

2.1. Effect of herding activities on land cover and vegetative change

The activities of herding and farming have a higher tendency to cause changes to the environment in particular and the climate in general. The activities of herbivorous animals including cattle through the reduction of plant density and grasses that have an impact on land use land cover dynamics (Köster et al., 2013), cannot be underestimated. The effect of the activities of these animals on the environment and for that matter, LULC is further aggravated by the activities of humans through farming and herding (Köster et al., 2013). These activities are likely to intensify and further worsen the already fast-changing land use land cover (LULC) dynamics due to population surge and the rising demand for food. The potential land use intensity implies that more land cover is likely to be converted as a result of the expansion of farmlands and increasing herd population and demand for grazing space, water, settlement, and urban development (Aly et al., 2016; Bessah et al., 2019). These activities are the major anthropogenic causes of changes in land use land cover (LULC) which may have an impactful effect on climate change (Bessah et al., 2019; Mekasha et al., 2014).

2.2. Mapping out herding areas

In the northern part of Ghana, there is no clear boundary between crop farming and grazing lands. This implies that there is no clear demarcation of lands for crop production and for other land management forms, including grazing. In some parts of the country, this lack of distinction has resulted in the restriction of pastoral accessibility, limiting the mobility of herders and subsequently leading to clashes between herders and crop farmers (Nébié et al., 2020). Mapping areas of herder activities along the corridor, therefore, needs to be undertaken to provide information on these herders in the area and to delineate farming and grazing area extensions. This information will help establish the basis for the settlement of land-use disputes and mitigate the ambiguity in land-use conflicts to promote peaceful coexistence (Nébié et al., 2020). It may also necessitate land-use monitoring, intending to provide crop farmers with information on herder locations and mobility guidelines for herders to reduce conflicts (Altmann et al., 2018).

To this effect, previous studies have addressed the mapping of farming—herding zones, cropping patterns, rangeland or land-use land cover management, and agricultural activities, particularly in relation to herder conflicts and environmental issues (Nébié et al., 2020). In this study, we build on these previous studies and adopt GIS and remote sensing data (Nébié et al., 2020; Pande et al., 2021) to investigate the impact of farming—herding activities on land use and land cover changes over time.

2.3. Remote sensing approach to the study of land and vegetative cover changes

The application of the remote sensing approach has been widely adopted for the study of land and vegetative cover changes that have resulted from the interaction between man and his environment (Al-Taisan, 2022; Aly et al., 2016; Barnetson et al., 2019). Information about land and vegetative cover changes may contribute to a large extent, land and forest resource management regimes, especially in relation to crop farming and herding. This is more important in terrestrial ecosystems where vegetation and grass cover play vital roles in the exchange of water and energy in the biogeochemical and climate cycle (Al-Taisan, 2022; Awad, 2018; Fu et al., 2023).

There are several remote sensing (RS) techniques that have been used to detect changes in land and vegetative covers. Landsat image series has been used to analyse the vegetation change in semi—arid areas as seen in the studies of Al-Taisan (2022), Pande et al. (2021), Fu et al. (2023), etc. Some studies have also relied on the combination of remote sensing data and GIS data including slope, topography, and land used. Remote sensing techniques involving multispectral and hyperspectral data have been widely adopted (Al-Taisan, 2022) for the monitoring and mapping of vegetation cover changes since 1972 especially when considering a vast area coverage (Govender et al., 2008; Pande et al., 2021). These two techniques which use a sensor to capture the wavelength dependence of light that has been reflected from the Earth’s surface, are regarded as the best methods for the detection and mapping of changes in land and vegetative cover (Al-Taisan, 2022). This study adopts a combination of data sources to perform mapping and change detection of the study area over time.

2.4. Geospatial analysis of land and vegetative cover changes

Various analytical methods are employed in the realm of remote sensing and GIS for the identification of changes in land and vegetative cover. Multispectral and hyperspectral remote sensing (Aly et al., 2016; Govender et al., 2008; Pande et al., 2022), aerial imagery (Álvares et al., 2018; Everaerts, 2014; Ural et al., 2011), ground truth data (Daly et al., 2017; Miyazaki et al., 2011) radar (Conyers, 2016), and LiDAR data (Csanyi & Toth, 2007; Rapinel et al., 2015) represent common types of data that can be obtained either individually or in fused multiple forms to investigate changes in land use and land cover. Typically sourced from satellite images or aircraft (Al-Taisan, 2022), these data are integral to the analysis process. Following the methodology outlined by Al-Taisan (2022), Figure 1 presents a schematic representation of the workflow for analyzing changes in land and vegetation cover, spanning from data acquisition to the generation of the final change detection map.

Figure 1. Approaches to the detection of land cover changes (modified, from Al-Taisan, 2022).

2.5. Machine learning algorithms in LULC classification

The application of Machine learning algorithms in LULC classification has brought about an improvement in the LULC literature. Machine learning algorithms have shown remarkable performance in LULC classification tasks (Barnetson et al., 2019; Karpatne et al., 2016). These algorithms are trained on labelled datasets, where each pixel is assigned a specific land cover class (Rogan et al., 2008; Shetty, 2019). From the LULC literature, some of the common machine learning algorithms in LULC classification include Random Forest (RF), Support Vector Machines (SVM), and Convolutional Neural Networks (CNN) (Shetty, 2019).

Support Vector Machines (SVM) is a supervised learning algorithm that effectively handles the combination of binary and multiclass classification (Amarappa & Sathyanarayana, 2011; Karpatne et al., 2016). It has demonstrated success in differentiating between different land cover classes in LULC classification. The Convolutional Neural Networks (CNN) can automatically learn hierarchical features from remote sensing images, capturing complex spatial patterns (Kattenborn et al., 2021; Maggiori et al., 2017). It has also provided magnificent performance in image classification tasks as a type of deep learning algorithm (Kattenborn et al., 2021; McIver & Friedl, 2001).

This study adopted the Random Forest (RF) algorithm due to its ability to handle noisy and high-dimensional datasets. The Random Forest (RF) approach is an ensemble learning algorithm that combines multiple decision trees to improve accuracy and robustness. Several studies including (Avci et al., 2023; Karpatne et al., 2016; Kutlug Sahin & Colkesen, 2021; Shetty, 2019), have adopted this approach to study LULC changes. These machine learning algorithms automate the classification process, reducing the need for manual interpretation. This leads to increased efficiency and the ability to process large datasets quickly (Arabameri et al., 2022; Karpatne et al., 2016; Rogan et al., 2008). They can also adapt to diverse and dynamic landscapes, making them suitable for classifying different types of land cover across various regions and time periods (Awad, 2018; Pande et al., 2021). Hence, this study adopted the Random Forest (RF) machine learning algorithm for the purpose of classifying the LULC classes.

2.6. Study area

The present study was conducted along the Northeast corridor of the Northern regions of Ghana. This part of Ghana lies eastward to the southern part of Burkina Faso with most communities in the area lying close to the Volta Rivers (White and Black) (Yenibehit et al., 2022). The area is suitable for crop farming and its nearness to rivers and streams makes it preferred grazing land. It is noted as the hub of crop and animal production due to the heavy dark coloured savannah ochrosols soils, Savannah Glysols and ground water laterite (Cudjoe et al., 2013). The entire area of the corridor is said to start from the Tema municipal in the Greater Accra region and stretches through the regions of Volta, part of Eastern and Ashanti regions, Oti, Northern, Savanna, Northeast and ends at Kulungugu in the Upper East region (Zakaria et al., 2014). Apart from Burkina Faso, the corridor also borders externally to Togo to the east and several districts to the west of the country (Cudjoe et al., 2013). We used geospatial data of the northern part of the corridor where the four districts that constituted the study are located. The map of these districts as located in this part of the eastern corridor of the country, is presented in Figure 2 with the selected districts indicated in the map. The four districts that data were collected for this study include the Karaga District, Gushegu District, Savelugu Municipal, and the West Mamprusi Municipal. The map of the part of the corridor considered for this study is illustrated in Figure 2.

Figure 2. Map of the study districts marked out in a map of the eastern corridor.

Source: Adapted from Yenibehit et al. (2022)

2.7. Data and image pre-processing

To perform analysis on mapping and detection of LULC in the study area, we utilize Geographic Information System (GIS) data obtained from Landsat satellite imagery to spatially locate areas of farming and grazing and detect LULC in the area. Landsat 5 TM⁺ image period of 2000 and Landsat 8 TOA satellite for the period 2013–2022 were used for the analysis. All images were captured at approximately the same period to ensure the selected image had the same values of reflectance and were derived through the Google Earth engine. To eliminate possible noises and haze in the selected images, radiometric correction was performed. There was also a geometric projection of the image to the Universal Transverse Mercator (UTM) zone 30 North (Verpoorter et al., 2012).

The satellite image data utilized for further analysis are further explored in section four of this paper. From the assertion of Yu et al. (2019), band 4, 5, 6 is the best three-band combination and band 1, 2, 5, 7 is the best four-band combination which produces almost identical performance for LULC classification. Hence bands 1,2,5,7 of the Landsat 8 image were stacked with the application of R software, (RF) algorithms and QGIS (version 3.28.13) for the classification in this study.

2.8. Data analysis

In this study, a supervised machine learning process using the random forest (RF) classifier algorithm was adopted for the classification of the selected imagery. RF is a robust algorithm for the classification and regression of data of high dimensionality with the potentiality of discrimination and grouping of digital levels of image classification which produces good results suitable for the classification of multisource remote sensing and geographic data (Avci et al., 2023; Barnetson et al., 2019; Fu et al., 2022; McIver & Friedl, 2001). This approach was followed due to its ability to minimize the correlation between the classifiers in the ensemble by bootstrapping and aggregating the ensemble to form a classification and regression tree (CART) ((Karpatne et al., 2016; Rogan et al., 2008; Shetty, 2019). The Random Forest classifier is not sensitive to noise or overtraining as the resampling is not based on weighting and is computationally much lighter based on boosting as well as Out of the Bag (OOB) error estimate and evaluation of variable performance (Talukdar et al., 2020).

For the purpose of environmental monitoring and management, researchers often extract vegetation indices from Landsat imagery to gain insights into ecosystem dynamics. These indices, such as the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Soil Adjusted Vegetation Index (SAVI), and Normalized Difference Water Index (NDWI) are calculated using different bands of the electromagnetic spectrum to assess various ecological parameters, including vegetation health and density (Alvino et al., 2020; Sousa et al., 2020; Xue & Su, 2017).

In the context of land use and land cover (LULC) literature, NDVI EVI, SAVI, and NDWI are commonly utilized vegetation indices. However, in the present study, the normalized difference vegetation index (NDVI) was specifically employed due to the sparse nature of the forest in the study area. The NDVI values were utilized for preliminary analysis to establish the baseline for changes in vegetation growth and health, potentially influenced by extensive farming and grazing (Dent & Tucker, 2015). This rudimentary analysis serves as a foundation for identifying areas of stress or damage, paving the way for further in-depth analysis. The normalized difference vegetation index (NDVI) in this study following Mihi et al. (2019), is mathematically expressed as:

(1) $\mathit{NDVI}=\frac{\mathit{NIR}\mathit{\hspace{0.17em}}-RED}{\mathit{NIR}+\mathit{RED}}\mathit{\hspace{0.17em}}$(1)

where NIR and RED are the spectral information in the Near Infra-Red and Red bands respectively of the satellite imagery. By design, the NDVI varies between −1.0 and + 1.0. NDVI of higher and positive values indicate higher photosynthetic activity or vegetation greenness. Values near zero (0) are usually non-vegetated areas and negative values indicate water bodies (Mihi et al., 2019).

2.9. Assessment of accuracy

Assessing the accuracy of the land use and land cover thematic map enables us to evaluate the errors that could occur during the preparation of the final thematic LULC maps (Rossiter, 2004). Therefore, land use and land cover (LULC) maps prepared were subjected to the overall accuracy and Kappa coefficient statistical tests. The Kappa coefficient was determined by using the following equations (Rossiter, 2004):

(2) $\hat{K}=\frac{\mathit{n}{\sum }_{\mathit{i}=1}^r{x}_{\mathit{ii}}-{\sum }_{\mathit{i}=1}^r{x}_i+\mathit{x}+\mathit{i}}{n^2-{\sum }_{\mathit{i}=1}^r{x}_i+\mathit{x}+\mathit{i}}$(2)

The overall accuracy of image classification was also determined by the given equation below:

(3) $\mathit{OverallAccuracy}=\frac{{\sum }_{\mathit{i}=1}^r{x}_{\mathit{ii}}}{\mathit{x}}$(3)

where n stands for a total number of observations (pixels), while r is the number of rows in the matrix. $x_{\mathit{ii}}$represents a number of observations in row i and column i. On the other hand, the marginal sum of row and column is denoted by $x_i\mathit{andx}+\mathit{i},$respectively (Islam et al., 2018).

In the classification process, the land classes under investigation were formulated and utilized as training sites for the Random Forest (RF) algorithm, aiming to classify geomorphological and land use features within the study area. These training sites were established by combining information from Google Earth maps with expert knowledge of the spectral signatures of prominent landmarks, derived from an understanding of the geological composition of the region.

2.10. Dynamics of land use and land changes

For investigation into the dynamics of land use and land cover changes in the study area we employed Markov chain analysis. This approach relies on probability assumptions, offering more comprehensive information compared to the regression approach (Baik et al., 2006; Use et al., 2023; Wangyel et al., 2021). Consequently, the probability distribution of a specific land change class in the current study is contingent upon its distribution in the preceding period, and these dependencies exhibit consistency across all periods. This implies that the structural change process can be effectively represented by a stationary first-order Markov chain (Baik et al., 2006; Wangyel et al., 2021). A stochastic matrix with the probability, A_ij of movement of the area under a given land class i in period t to another land use class group j in period t +1 is called transitional probability matrix (TPM) expressed as;

(4) $\text{P}={\text{ A}}_{\text{ij}}\cdots {\text{ A}}_{\text{is}}⋮\ddots ⋮{\text{A}}_{\text{ij}}\cdots {\text{ A}}_{\text{ss}}=\frac{\text{Q}}{0}\frac{\text{R}}{\text{I}}$(4)

where $A_{\mathit{ij}}\ge 0$ and $\sum _j{A}_{\mathit{ij}}=1$, P is a square matrix (n × n), with n being the total number of land classes under study. Each element (A_ij) represents the probability of moving from one land use class i to other j (Baik et al., 2006).

3. Results and discussion

The results obtained from the remote sensing and geospatial analysis are presented and discussed in the preceding sub-sections, focusing on the accuracy of LULC Classification, examining the Land Use Classification and Land Cover Changes, and analysing land use and land cover change detection using the Markov Chain Model.

3.1. Accuracy assessment of LULC classification

Since the quality of a thematic map from a satellite image is determined by its accuracy, we present the overall accuracy and Kappa statistics of the classified images of the years 2000, 2013, and 2022 in Table 1. The accuracy for each year was computed by analyzing the confusion matrix, which delineates the land classes in focus (Abdu, 2018). However, inclusion of the entire matrix in the paper is deemed practically impossible due to space. Hence, in Table 1, we provide a summary of the overall accuracy values and Kappa statistics, offering a concise representation of the assessment results.

Table 1. Accuracy of land use land cover classification

Year	Overall Accuracy	Kappa Statistics
2000	0.86	0.82
2013	0.84	0.83
2022	0.81	0.78

Source: Analysed from Satellite Images of 2000, 2013 and 2022.

The overall accuracy in Table 1 represents the proportion of correctly classified land cover pixels to the total number of pixels in the study area. In the given table, the overall accuracy values for the years 2000, 2013, and 2022 are 0.86, 0.84, and 0.81, respectively. These values indicate that the land classification models achieved a high level of accuracy in identifying land use and land cover classes in the North Eastern part of Ghana.

The Kappa statistic measures the agreement between the predicted classifications and the observed classifications, accounting for the possibility of agreement occurring by chance. Kappa values range from 0 to 1, where 0 represents no agreement beyond chance and 1 represents perfect agreement (Tang et al., 2015). From the table, the Kappa values for the years 2000, 2013, and 2022 are 0.82, 0.83, and 0.78, respectively. These values indicate substantial to almost perfect agreement between the predicted and observed land cover classifications.

3.2. Vegetation cover change

The study of the nature of changes in land use land cover classes due to anthropogenic effects by exploring the dynamics of vegetation cover changes was examined by calculating NDVIvalues obtained from the images (Mihi et al., 2019). The results of the NDVI map in Figure 3 show deteriorated vegetation as the NDVI map of the year 2000 was greener than the NDVI map of 2013 which only got worse off in the year 2022 indicating a decrease in the vegetation cover of the area probably as a result of loss of forest and grasslands to other land use classes. To understand the extent of degradation in this study, the summary statistics of the NDVI values from the hyper-temporal NDVI data obtained from the satellite images were utilized.

Figure 3. Vegetation cover change of ndvi maps for north eastern corridor of Ghana, for the years 2000, 2013 and 2022.

The summary statistics of the NDVI values for the Northeastern corridor of Ghana for the years 2000, 2013, and 2022 are presented in Table 2. From the results, in 2000, the minimum NDVI was −0.12, which reduced to −0.28 in 2013 and then slightly increased to −0.22 in 2022. These negative values suggest areas with very limited vegetation or areas dominated by non-vegetated land covers such as bare soil. The results could imply bush clearing and farming by crop farmers and bush fires which are always common with animal herding in the area. The maximum NDVI values indicate areas with the densest vegetation cover.

Table 2. Descriptive statistics of NDVI values for the sample years

	Statistics of NDVI Values for the Years Sampled
Statistic	2000	2013	2022
Minimum	−0.12	−0.28	−0.22
Maximum	0.53	0.78	0.68
Median	0.27	0.37	0.24
Mean	0.27	0.37	0.25
Standard Deviation	0.04	0.06	0.05

Source: Analysed from Satellite Images of 2000, 2013 and 2022.

In 2000, the maximum NDVIwas 0.53, which increased to 0.78 in 2013 and then slightly decreased to 0.68 in 2022. Again, the mean NDVI indicateis a similar trend as seen in Table 2 with relatively close average NDVI values to their respective median values across the three periods (2000, 2013, and 2022). This suggests that the distribution of vegetation cover is relatively symmetrical without significant outliers in the sample images under this study. This is also supported by the moderately lower standard deviation values which range from 0.04 to 0.06, indicating relatively low variability in vegetation cover.

The upward trend from 2000 to 2013 suggests a general improvement in vegetation cover, which probably emanates from the implementation of tree planting exercises across the country in the early 2000s. However, the decrease in the maximum NDVI value from 2013–2022 May indicate the extension of farming/herding activities that have likely reduced the forest cover of the corridor.

In general, based on the NDVI values, it can be observed that there have been changes in vegetation cover in the North Eastern Corridor of Ghana over time. The maximum and median NDVI values show an overall improvement in vegetation cover from 2000 to 2013, followed by a slight decline in 2022. The trend agrees with the other land use changes and is in line with similar studies such as Forkuo and Frimpong (2012) Simonetti and Simonetti (2014); Ayele et al. (2018); Köster et al. (2013) who all recorded significant vegetation cover change using NDVI maps and values and attributed the changes to human activities such as deforestation, bush burning, overgrazing, etc. A summary of the descriptive statistics of the NDVIvalues is presented in Table 2.

3.3. Land use classification and land cover changes

We employed image enquiry with colour as a guide to assign land use classes and determine the class interval of each of the land classes. This was done after the NDVI results, to further comprehend the dynamics in land cover changes and distinguish between vegetation greenness covers of crops, dense forest and relatively light dense forest. Based on the classification, the following classes of land use were mapped in this study: water, urban, grassland, forest, deciduous, bare land, and agriculture (farming/herding). The results are presented in Figures 4, and Table 3. Visually appraising and comparing the images in the figures indicate some level of notable changes across the various classes from 2000 to 2022. The trend of the various changes and the percentage changes have been included in Table 3.

Figure 4. Land-cover land-use classification map for north eastern corridor of Ghana, for the years 2000, 2013 and 2022.

Table 3. Proportion and percentage changes in land use land cover classes

				% Change
Land Use Class	2000	2013	2022	2000 – 2013	2013–2022
Urban	2169.52	2131.16	2784.58	−1.76813	30.6603
Forest	1181.75	1301.76	337.23	10.15528	−74.0943
Agriculture	1224.8	1547.16	829.86	26.3194	−46.3624
Grassland	926.5	916.8	633.5	−1.04695	−30.901
Water	1973.5	1789.61	2775.97	−9.31796	55.11592
Deciuous	1909.37	1669.16	1556.16	−12.5806	−6.76987
Bareland	799.33	829.13	1267.46	3.728122	52.86626
Total	10184.8	10184.8	10184.8

Source: Analysed from Satellite Images of 2000, 2013 and 2022.

With reference to Table 3, in 2000, the urban land use class covered an area of 2131.16 km² which increased slightly to 2169.52 km² in 2013 but significantly increased to 2784.58 km² in 2022. The percentage change from 2000 to 2013 shows a slight increase of 1.77%, while the change from 2013 to 2022 indicates a substantial increase of 30.66%. The growth of urban areas suggests an expanding population and increasing urbanization in the North Eastern part of Ghana. This can have several implications, including increased demand for infrastructure, services, and housing, as well as potential environmental challenges such as habitat fragmentation and loss of natural ecosystems accompanied with the high demand for food both from crops and animals that have the potential to change or extend the farming and herding routes.

The results also reveal that the forest land use class had an area of 1181.75 km² in 2000, which slightly increased to 1301.76 km² in 2013 but experienced a significant decrease to 337.23 km² in 2022. The percentage change from 2000 to 2013 shows a positive trend of 10.16%, indicating some expansion of forests, while the change from 2013 to 2022 indicates a considerable decline of −74.09%. The substantial decline in forest cover raises concerns about deforestation and its potential impacts on biodiversity, ecosystem services, and climate change mitigation. The recent increase in cutting of rosewood which is the typical composition of the forest in the study area, bush fires, slash and burn system of farming, and the activities of herders may also contribute to a substantial change in the forest land class within this period.

In terms of agriculture, the results show that the agricultural land use class (crop production and animal herding) covered an area of 1224.8 km² in 2000, increased to 1547.16 km² in 2013 and decreased to 829.86 km² in 2022. The percentage change from 2000 to 2013 indicates a significant growth of 26.3194% in agricultural land, while the change from 2013 to 2022 shows a notable decline of −46.3624%. The decrease in agricultural land suggests potential shifts in land use practices, possibly due to factors such as changing economic conditions, urbanization, or agricultural intensification and destructive activities in herding which might force some farmers to abandon their farm lands that are dominated by herding. Understanding the drivers of this change is crucial for ensuring sustainable land use practices and food security in the region. It is however surprising to see such a significant decrease in agricultural land use class since most of the areas within this region are rural and agrarian in nature. The increase in urban land class could however support the diversion to other occupations and land use practices and migration to urban cities especially the national capital in search of better employment opportunities.

From Table 3, the results provide evidence that grasslands have declined across the periods under study. The grassland area in 2000 was 926.5 km², which remained relatively stable at 916.8 square kilometres in 2013. However, there was a decrease in grassland cover from 2013 to 2022, with the area reducing to 633.5 km². Also, the percentage change from 2000 to 2013 shows a slight decline (−1.05%) in grassland cover, while the change from 2013 to 2022 indicates a notable decrease (−30.90%).

The reduction in grassland areas could be due to various factors, such as land conversion for other purposes or changes in land management practices. Farming and herding as well as the increase in other land use classes may also explain the reason for the decline in grasslands. This has a negative impact on grazing lands and may alter the grazing or herding/farming routes. This trend increases competition for natural resources that may intensify conflicts between herders and crop farmers. The water land use class covered an area of 1973.5 km² in 2000, decreased to 1789.61 km² in 2013 but significantly increased to 2775.97 km² in 2022 as presented in Table 2. This was translated into the percentage change in water cover between 2000 and 2013 as −9.32%, indicating a reduction, while the change between 2013 and 2022 was 55.12%, representing substantial growth in water as a land unit.

The initial decrease in water could probably be attributed to drying up of water bodies between 2000 and 2013 as a result of anthropogenic factors such indiscriminate illegal mining and adverse climatic conditions within the period. The increase in water cover as seen in the 2022 could suggests potential changes in hydrological patterns, such as the formation of new water bodies or the expansion of existing ones. Notable among this may include the government ‘’one village one dam’’ project that has led to the spring up in the number of water bodies in the north eastern corridor. This again provide positive indication of livestock and animal herding activities along the corridor.

Similar to the grassland units, the proportion of deciduous land cover was 1909.37 km² in 2000, decreased to 1669.16 km², resulting in a percentage change of −12.58% in 2013 and a further decreased in proportion to 1556.16 km², resulting in a percentage change of −6.77% by 2022. The trend observed for deciduous land cover in the North Eastern part of Ghana which indicates a decline over time. The percentage changes suggest that there has been a consistent reduction in the proportion of deciduous land cover, with a more significant decrease observed between the first and second time periods (−12.58%) compared to the decrease between the second and third time periods (−6.77%). This again points to the deleterious effects of human activities including illegal mining, slash and burn system of agriculture, the increase in built up areas due to urbanization among others.

Contrary to the other land use classes, the proportion of bareland was 799.33 km² in 2000 but during the second time period (2013), the proportion slightly increased to 829.13 km², representing a percentage change of 3.73%. In the third time period of 2022, the proportion significantly increased to 1267.46 km², resulting in a substantial percentage change of 52.87%.

This trend observed for bareland in the region indicates an increase over time. The percentage changes suggest that there has been a notable rise in the proportion of bareland. This implies that agricultural and grazing or herding activities have been extended over time without any intervention in the form of tree planting. The substantial increase in barelands between periods of 2013 to 2022 coincides with the era of a tough fight against illegal mining in the country coupled with indiscriminate bush burning and these issues could explain the trend of barelands in the area.

In conclusion, the remote sensing data reflects changes in land use and land cover in the North Eastern part of Ghana. The decrease in deciduous land cover and the increase in bareland suggest potential deforestation or changes in vegetation patterns in the region. The dynamics of land use classes obtained in this study is in consonance with several other studies in the LULC literature. Studies such as (Bessah et al., 2019; Mekasha et al., 2014; Onyango, 2015; Tolera et al., 2012; Wiesner et al., 2012) have all reported significant changes in land use land cover classes in their studies.

Although, we may conclude that changes in the other land classes under this study resulted in this trend, it is still important to conduct further analysis and investigate the underlying causes of these land use changes in the area.

3.4. Land use and land cover change detection analysis (Markov chain model)

To determine the transitional change from one land use class to the other, we employed the Markov Chain Model. It is a mathematical model used to analyse and predict the transition of states in a system over time (Baik et al., 2006). Table 4 provides the transitional probabilities for changing land use and land cover classes in the North Eastern Corridor using the area under each class of land cover as evaluated from satellite data of 2000, 2013 and 2022.

Table 4. Transitional probabilities to change in land use land cover classes (Markov chain model)

	urban	Forest	Agric	Grassland	Water	Deciduous	Bareland	Total
Urban	0.19	0.00	0.00	0.00	0.10	0.72	0.00	1.00
Forest	0.00	0.00	0.00	0.00	1.00	0.00	0.00	1.00
Agric	0.00	0.00	0.00	0.00	0.32	0.00	0.68	1.00
Grassland	1.00	0.00	0.00	0.00	0.00	0.00	0.00	1.00
Water	0.00	0.66	0.28	0.00	0.00	0.06	0.00	1.00
Deciduous	0.00	0.00	0.52	0.48	0.00	0.00	0.00	1.00
Bareland	1.00	0.00	0.00	0.00	0.00	0.00	0.00	1.00

Source: Analysed from Satellite Images of 2000, 2013 and 2022.

In the table, the different land use and land cover classes are represented by rows and columns. Each cell in the table represents the probability of transitioning from one land use/cover class (represented by the row) to another class (represented by the column) in a given time period.

From the results in Table 4, the probability of transitioning from an urban area to another urban area is 0.19. This indicates that there is a 19% chance that an urban area will remain unchanged in the next time period. In recent times the creation of new regions and districts couple with the fast pace of urbanization in many parts of the country including the North Eastern part of the country, could be the result of this observation. Hence, one of the reasons why urban areas may remain unchanged and may however increase in size.

Again, the results in Table 4 show unexpected trend of a high likelihood (72%) of urban areas converting into deciduous forests. This suggestion is unlikely but it points to the fact that most of the settlers are herders with high tendency to abandon their structures that are likely to be converted into deciduous forests.

In line with aprior assumption, the results reveal that, the probability of transitioning from a grassland to an urban area is 1.00, indicating that grasslands have a high chance (100%) of being converted to urban areas. This is probably providing indications that areas that are currently left with grasses for grazing are likely to be converted into settlements for settlers who may be nomadic in nature in the case of herders or permanent residents as population growth is inflicting pressure on land use classes including grasslands.

Again, the results further indicate that, the probability of transitioning from water bodies to grasslands is 0.66, suggesting that there is a relatively high chance (66%) of water bodies transforming into grasslands. There has been a dramatic increase in water bodies around the area due to probably, dams constructed under the ‘’one village one dam’’ project of the government of Ghana. However, the nature of their contraction and the fact that some of them are already lost through runoff may imply that these water bodies are more likely to be converted into grasslands.

Finally, the results support the current trend of land use classes as most of them are been converted to agriculture including crop production and grazing/herding. The results in Table 4 posit that, the probability of transitioning from deciduous forests to agricultural areas is 0.48, implying that there is a moderate chance of deciduous forests being converted to agricultural land. This points to the fact that population growth and the fast pace of urban development in the corridor may lead to high demand for food and meat which should be compensated with high levels of agricultural production hence the likelihood of conversion of deciduous land classes to agriculture.

This study has gathered evidence of land use land cover transitions which imply that there are probabilities that one land use class could transition into another land use as reported in other studies such as (Abass et al., 2018; Addae & Oppelt, 2019; Aniah et al., 2023; Braimoh & Vlek, 2004; Koranteng et al., 2023; Viana & Rocha, 2020). The results of the transitional probabilities to change in land use land cover classes fitted with the Markov Chain Model is presented in Table 4.

4. Conclusion and recommendation

The growth in population and the development of urban built-up cities pose potential high demand for food and meat products which inflicts high transition for land use land cover (LULC) classes in Ghana and the North Eastern corridor in particular. Over the years, West Africa and Ghana have experienced an extended farming and pasture area in a bit to meet the food, meat and raw materials needs of a rising population of the sub-region. This has resulted into competition for farmlands and grasslands for grazing farm animals. This study was therefore designed to investigate the effect of farming and pasture areas extensions on land use land cover classes in the North Eastern Corridor of Ghana.

Landsat 5 TM⁺ image period of 2000 and Landsat 8 TOA satellite for the period 2013–2022 were used for the analysis. All images were captured at approximately the same period to ensure the selected images had the same values of reflectance. Markov chain analysis was used to examine the dynamics of land use and land changes in the study area. Data analysis was carried out with R software algorithms and QGIS for mapping the land use land cover change (LULC) classification using the satellite images (maps).

Based on the supervised classification and theNDVIvalues, it was observed that there have been changes in vegetation cover in the North Eastern Corridor of Ghana over time. Based on the analysis, the following classes of land use were mapped in this study: water, urban, grassland, forest, deciduous, bare land, and agriculture.

Visual appraisal of the images indicated some level of notable changes across the various classes from 2000 to 2022. The trend of the various changes in percentage terms also supports this observation. The maximum and median NDVI values show an overall improvement in vegetation cover from 2000 to 2013, followed by a decline in 2022.

The study can therefore conclude that, the remote sensing data reflects changes in land use and land cover in the North Eastern part of Ghana. The decrease in deciduous land cover and the increase in bare lands suggest potential deforestation, slash and burn forms of farming or changes in vegetation patterns in the region.

The weakness of this current study is the use of Landsat image data alone for the analysis. Therefore, for an improved classification and higher accuracy of the results, we recommend the use of multi sensor data classification using neural networks with the combination of ancillary data (i.e. elevation and aspect) with the Landsat image data. We also recommend that a more in-depth analysis be conducted to identify other anthropogenic activities and factors, that serve as significant underlying causes of these vegetation cover changes in the region.

Until then, the following policy recommendations are made, based on the findings of this study:

• The Ministry of Food and Agriculture (MoFA) should train farmers to incorporate tree planting into their farming whilst avoiding deforestation and bush burning. This will promote vegetation growth and minimize the impact of global warming.

• The government should establish cattle ranges in the area in order to avoid sporadic grazing of cattle and the indiscriminate burning of bush by herders in the region.

• Urban settlers should also be given training and education to plant trees around their homes whilst efforts are made to grass and plant trees on bare lands if possible.

Author contributions

NY and JKB conceived, designed the research, analysed the data and wrote the first draft of the manuscript. JK and AA made critical revisions and approved the final version. All the authors contributed to the writing of the manuscript, revisions, and agree with the manuscript results and conclusions. All authors reviewed and approved the final manuscript.

Acknowledgements

Data collection for this work was sponsored by the Access and Authority Nexus in Farmer-Herder Conflicts (AAN Project), a Danida-funded project.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Data used for this work is a public data that was downloaded from Google Earth Engine. Landsat 5 TM⁺ image period of 2000 and Landsat 8 TOA satellite image for the periods 2013 and 2022. Code for data extraction was obtained through the link below https://code.earthengine.google.com/

R-codes for data processing and analysis are available on request

Notes

1. Land Use and Land Cover (LULC) are two different terms that represent the interaction between humans and the natural land surface. Land use represents the activities of humans on land which in this study focuses on farming and herding activities; whereas land cover is a representation of the natural cover or envelope of the land surface (undisturbed land).

2. The north eastern part of the country is a hot spot for violent farmer—herder clashes due to competition for farming and grazing lands which also affect land use land cover changes.

3. Remote sensing refers to the process whereby the physical characteristics or features of an area (northern eastern Ghana) are been measured using its reflected and emitted radiation from a distance in order to monitor and detect any change.

4. Geospatial analysis is the activity of collecting, combining, and visualizing various types of geospatial data such as satellite imagery as adopted for this study.