Dynamics of Land Cover Change in the Anambra River Basin of Nigeria and Implications for Sustainable Land Management

The Anambra River Basin is predominantly situated in the southeastern geopolitical region of Nigeria (Fig. 1) and encompasses the Anambra River and its associated drainage systems, constituting a significant component of the region’s inland waters (Nwani, Ude 2005). Given Nigeria’s projected population increase from 208 million in 2020 to 419 million by 2058 (Chamie 2022), the Anambra River Basin is expected to experience increased human pressures on land cover and natural resources. The basin, covering approximately 11,000 km² (Fig. 1), extends from the Ankpa hills and merges with the River Niger at Onitsha. It exhibits a diverse range of biophysical characteristics, with elevations ranging from a maximum of approximately 594 m a.s.l. to a minimum of −2 m b.s.l. Intersecting eight states, namely Abia, Anambra, Benue, Delta, Enugu, Imo, Edo, and Kogi, the Anambra River Basin hosts abundant natural resources, including substantial deposits of coal and lignite, along with significant hydrocarbon potential. Notably, the basin is estimated to possess around 10 trillion cubic feet of gas reserves. It experiences two distinct seasons: a dry season from October/ November to March and a rainy season from April to September/October. These seasonal variations correspond to the hydrological regimes, with dry and flood phases observed in the region. Its climatic conditions further complement the basin’s biophysical attributes. It maintains an annual mean maximum temperature of 32°C and an annual mean minimum temperature of 24.5°C. The area receives an accumulative annual rainfall ranging from 1900 mm to 2707 mm (Akafor, Obiezue 2004).

Fig. 1.

In addition to its natural characteristics, the Anambra River Basin supports a growing population and diverse land uses. Agriculture employs over 60% of the population of South-eastern Nigeria and is the backbone of the economy in the region (Ogoke 2023). Within the Anambra River Basin, agriculture serves as the primary land use, encompassing activities such as forestry, fishing, farming, field crop production, and animal pasture. The region’s agricultural and fishery projects, including the World Bank’s rice initiatives, exemplify the immense potential within the basin. The Anambra-Imo River Basin and Rural Development Authority was established for sustained development and resource management.

Given the dynamic interplay of biophysical features, climatic patterns, and human activities, performing a spatiotemporal characterisation of the Anambra River Basin is imperative. Understanding the basin’s evolving dynamics and changes over time is crucial for effective planning, management, and sustainable development of the area. By studying the basin’s spatiotemporal aspects, valuable insights can be gained regarding its ecological integrity, water resources, land use patterns, and potential impacts on the surrounding communities.

The dynamics of land cover in the study area were examined using Landsat satellite imagery from three approximate periods: 1987, 2000, and 2018. Six Landsat scenes covering the study area were downloaded from the United States Geological Survey website (USGS 2023) in GeoTIFF format. Table 1 provides an overview of the characteristics of the Landsat imageries employed in the analysis.

The population data for the states within the river basin, spanning the years 2006–2018, were sourced from the National Bureau of Statistics website (NBS 2023). These population datasets are instrumental in understanding demographic changes and their potential influence on regional land cover dynamics. By utilising these resources, we aim to gain a comprehensive understanding of the changes in land cover over time and explore any potential correlations with population dynamics in the study area.

During image interpretation, the following five classes were detected, following the Level I classification scheme of Anderson et al. (1976): vegetation, wetland, waterbody, built-up area, and bare land (described in Table 2).

In recent years, the field of land cover classification has witnessed the development of various techniques, with neural networks emerging as a particularly accurate and effective method (Jansen et al. 2009). The application of neural networks for extracting land cover information from remotely sensed data has been widely demonstrated by researchers (Frizzelle, Moody 2001, Muchoney, Williamson 2001, Tatem et al. 2001, Benediktsson, Sveinsson 2003, Jensen, Binford 2004, Al-Hameedi et al. 2021, 2022,).

The utilisation of neural networks as a classification method offers compelling justifications. Firstly, neural networks have demonstrated a remarkable ability to handle complex patterns and relationships within data. They can capture intricate non-linearities and subtle interactions, enabling them to effectively discriminate between different land cover classes based on their unique spectral characteristics and spatial patterns. This capability makes neural networks well-suited for accurately categorising land cover types from remotely sensed imagery. Additionally, neural networks operate in a supervised learning framework, leveraging labelled training samples to train the network’s parameters. This characteristic allows for the exploitation of prior knowledge and expert input in the classification process. By utilising a large and diverse training dataset, neural networks can learn to recognise intricate patterns and generalise well to unseen data, enhancing their performance in land cover classification tasks.

The importance of conducting an accuracy assessment in image classification cannot be overstated as it plays a crucial role in establishing the confidence and reliability of the obtained results (Gashaw et al. 2017). A comprehensive approach was employed in this study to ensure a robust evaluation of the land cover output. Specifically, a stratified random sampling technique was implemented, considering the different land cover classes within the study area.

The ground-truth points used for accuracy assessment were thoughtfully distributed across the study area using the stratified random sampling approach. The aim was to ensure that each land cover class was adequately represented, accounting for the inherent heterogeneity of the landscape. High-resolution imagery from Google Earth served as the basis for identifying suitable locations for ground-truth points.

The well-established Cochran’s formula for stratified random sampling (Cochran 1977) was employed to ascertain an appropriate sample size for the study: 1 n=(∑wisi)2[S(O⌢)2]+(1/N)∑wisi2≈(∑wisiS(O⌢))2 $$n = {{{{(\mathop \sum \nolimits^ {w_i}{s_i})}^2}} \over {[S{{(\mathord{\buildrel{\lower3pt\hbox{$\scriptscriptstyle\frown$}}\over O} )}^2}] + (1/N)\mathop \sum \nolimits^ {w_i}{s_i}^2}} \approx {({{\mathop \sum \nolimits^ {w_i}{s_i}} \over {S(\mathord{\buildrel{\lower3pt\hbox{$\scriptscriptstyle\frown$}}\over O} )}})^2}$$ where: –

N represents the number of units in the region of interest,

The formula was pivotal in determining the number of samples required to ensure a representative and statistically significant representation within each stratum. The overall accuracy is given by: 2 Overall accuracy=Total number of correct classified pointsTotal number of points×100 $${\rm{Overall accuracy}} = {{{\rm{Total number of correct classified points}}} \over {{\rm{Total number of points}}}} \times 100$$

The total number of correctly classified points is the number of points with the same class values from the classification output and the ground truth, while the total number of points is the number of random points created. The kappa coefficient was also calculated. Kappa tests are used to measure the accuracy of classification as the test accounts for all elements in the confusion matrix, including diagonal elements (Halmy et al. 2015). The kappa test is a measure between predefined producer rating and user-assigned rating, which can be expressed as follows: 3 K=P(A)−P(E)1−P(E) $$K = {{P(A) - P(E)} \over {1 - P(E)}}$$ where: –

P(A) is the number of times the k-raters agree, and

Meanwhile, user accuracy can be defined as the probability of a pixel on the image representing a class on the ground. The producer’s accuracy indicates the probability of a pixel being correctly classified and is mainly used to determine how well an area can be classified (Pontius, Millones 2011).

The CA-Markov model is well established for use in land cover prediction (e.g., Nath et al. 2020, Beroho et al. 2023, Rani et al. 2023). The process of land cover prediction using the CA-Markov model involved several steps. Firstly, the classified images were converted to an ASCII data format to ensure compatibility with Clark Labs’ IDRISI software. The ASCII file was then imported into the IDRISI environment and converted to an *.rst format. Subsequently, the reclass tool was utilised to reclassify the raster, assigning numerical values to different land cover classes.

A transition probability matrix was generated. The process required selecting the earliest year’s reclassified image as the first image, while the latest year’s reclassified image was selected as the second. The resulting probability matrix was labelled with the prefix ‘output’ and included the conditional probability values. The period forward and the epochal difference between the two images were specified, while the proportional error value was set to 0.00.

The prediction of land cover for 2030 involved a two-step process. Initially, the land cover of 2018, referred to as predicted 2018, was predicted. This prediction served as a basis for comparing the performance of the model with the actual data obtained from Landsat-based classification. The transition probability matrix for 1987–2000, suitability maps and a 5 × 5 contiguity filter were employed, and a comparison was conducted between the classified and predicted maps of 2018.

The classified 2018 map was utilised as a base map for further predictions, after which the transition probability matrix for the period 2000–2018 was employed to simulate the land cover for 2030.

A post-classification detection method was utilised to perform the land cover change detection. The calculate geometry function in ESRI’s ArcMap^TM was employed to calculate the coverage areas of the land cover classes. The resulting data on coverage areas were then transferred to the Microsoft Excel workspace for additional analysis. In parallel, population data from 2006 to 2016 were obtained and used to project the population figures for 2017 and 2018. The projection of population figures followed the formula provided by George et al. (2004): 4 Nt=Pe(r×t) $$Nt = P{e^{(r \times t)}}$$ where: –

Nt represents the future population,

The assessment of land cover change reveals significant transformations in the study area over the years. Figure 3 depicts the land cover maps for 1987, 2000 and 2018, providing a visual representation of the changes that occurred. Furthermore, Table 3 presents the areal distribution of land cover for the three respective years, offering quantitative insights into the changes.

Fig. 3.

Table 4 presents the land cover change at three study intervals: 1987–2000, 2000–2018 and 1987–2018. Figure 4 shows a graphical representation of the percentage changes.

Fig. 4.

Between 1987 and 2000, there was an observable increase in built-up areas, expanding from 168.28 km² (1.51% of the study area) to 201.42 km² (1.80%). This trend continued, with built-up areas further growing to 382.87 km² (3.42%) by 2018. Concurrently, wetlands experienced a notable decline from 2479.71 km² (22.18%) in 1987 to 1102.15 km² (9.86%) in 2000. The decrease continued, reaching 1047.99 km² (9.37%) in 2018. The area of bare land increased from 118.29 km² (1.06%) in 1987 to 185.03 km² (1.66%) in 2000 but subsequently decreased to 153.02 km² (1.37%) in 2018. Vegetation, which exhibited the largest cover throughout the years, expanded from 8373.33 km² (74.89%) in 1987 to 9643.82 km² (86.27%) in 2000. However, vegetation coverage decreased slightly to 9549.34 km² (85.43%) by 2018. Waterbody experienced a marginal increase from 41.18 km² (0.3%) in 1987 to 46.06 km² (0.41%) in 2000 but slightly decreased to 45.38 km² (0.41%) by 2018.

During the period from 1987 to 2000, there was a significant gain in bare land (56.41%), built-up areas (19.69%), waterbody (11.85%) and vegetation (15.17%), whereas wetlands experienced a substantial loss of 55.55%. From 2000 to 2018, there was a decline in bare land (17.3%), wetland (4.91%), vegetation (0.98%) and water-body (1.49%), while built-up areas experienced a substantial increase of 90.09%. From 1987 to 2018, there was an increase in bare land (34.72 km², 29.36%), vegetation (1,176.01 km², 14.04%), built-up areas (214.59 km², 127.52%) and waterbody (4.19 km², 10.20%). However, wetlands experienced a considerable loss of 1431.73 km² (57.74%).

The accuracies for 1987, 2000 and 2018 classifications were 87.85%, 85.33% and 91.27%, with kappa coefficients of 0.84, 0.81 and 0.87, respectively.

The transition probability matrix provides valuable insights into the likelihood of land cover conversions within specific periods in the Anambra River Basin. Figure 5 presents summaries of the probability matrix, focusing on the major land cover change between 1987–2000, 2000–2018 and 1987–2018, respectively.

Fig. 5.

In the analysis of the transition probabilities between 1987 and 2000 (Fig. 5A), it is observed that the probability of a waterbody remaining as a waterbody was 46%. Meanwhile, the probability of bare land converting to vegetation was relatively high at 81%. Additionally, the probability of vegetation remaining unchanged was 90%, indicating considerable stability in vegetation cover during this period. There was a relatively low probability of vegetation transitioning to built-up areas (2%). In terms of wetlands, there was a 16% probability of them remaining unchanged, while the probability of wetlands converting to vegetation was notably high at 82%. Figure 5B shows that between 2000 and 2018, the probability of wetlands remaining unchanged was 64%, while the probability of wetlands changing to vegetation was 32.10%. The probability of built-up areas remaining unchanged was 53%, while the probability of built-up areas changing to vegetation was 45%. Figure 5C shows that between 1987 and 2018, vegetation had the highest probability of 92% remaining as vegetation. In contrast, waterbody, built-up area, wetland and bare land had probabilities of 73%, 51%, 43% and 17%, respectively, to remain unchanged.

The land cover transition matrices presented in Tables 5–7 provide detailed information on the transformations between different land cover classes during specific time intervals: 1987–2000, 2000–2018 and 1987–2018.

Analysing the land cover change between 1987 and 2018 (Table 7), it is evident that the wetland area experienced the most significant transformation, with a total of 1819.46 km² being converted to various land cover types, including waterbodies, built-up areas, vegetation and bare land. During the same period, 103.098 km² of bare land transitioned to built-up areas and vegetation. Moreover, approximately 277.59 km² and 122.78 km² of vegetation changed into built-up areas and bare land, respectively. These transformations highlight the dynamic nature of land cover within the Anambra River Basin, illustrating the substantial shifts between different land cover categories over the examined period.

Specifically examining the land cover transitions between 1987 and 2000 (Table 5), it can be observed that a significant portion of wetlands, measuring 1761.91 km², transformed into vegetation. Similarly, a total of 427.42 km² of vegetation was converted into wetlands between 2000 and 2018 (Table 6).

The validation of land cover prediction is a crucial step in ensuring the credibility and applicability of the CA-Markov model’s outputs. By comparing the predicted results with classified land cover information, potential limitations and areas of improvement for the model can be identified. This information can be valuable for refining the model and enhancing its accuracy in future applications.

A comparison was made between the predicted land cover output for 2018 and the classified land cover derived from imagery to assess the accuracy of the land cover prediction generated by the CA-Markov model. The prediction for 2018 was obtained using the transition probability matrix calculated for the period from 1987 to 2000. The results of this validation process are presented in Figure 6.

Fig. 6.

Figure 6 provides a clear overview of the comparison between the classified and predicted values for different land cover categories, namely waterbodies, wetlands, vegetation, bare land and built-up areas. The analysis reveals a close alignment between the classified and predicted values for waterbodies, wetlands, vegetation and bare land. These categories demonstrate a consistent correspondence between the predicted land cover and the observed land cover derived from the imagery data.

However, there is a notable difference in the predicted area for built-up areas compared to the classified values, with a discrepancy of 182.89 km². This disparity can be attributed to the model’s limited sensitivity in accurately capturing small and scattered patches of built-up areas. The model’s performance in predicting the extent and distribution of built-up areas appears less accurate than other land cover categories.

In terms of overall performance, the accuracy assessment indicates a satisfactory outcome for the CA-Markov model in simulating future land cover. The model achieved an overall accuracy of 91.60%, reflecting the percentage of correctly predicted land cover categories and a kappa coefficient of 0.87, which measures the agreement between predicted and observed land cover beyond what could be expected by chance alone.

The future land cover prediction for the study area was conducted using the CA-Markov chain model. The resulting distribution of land cover classes for the year 2030 is visualised in Figure 7, providing a comprehensive overview of the projected changes in land cover across the study area.

Fig. 7.

Figure 7 illustrates the spatial distribution of different land cover categories, including water-bodies, wetlands, vegetation, bare land and built-up areas, as projected by the CA-Markov chain model for 2030. This visualisation offers valuable insights into the expected changes in the land cover composition and patterns over the specified time frame, aiding in understanding the future landscape dynamics and potential environmental impacts.

Furthermore, Table 8 presents a detailed summary of the projected land cover change between 2018 and 2030. It provides quantitative information regarding the transitions and transformations anticipated to occur within this period. The table shows the magnitude of changes for each land cover category, shedding light on the expected gains, losses and shifts in land cover across the study area.

The relationship between land cover types and population growth in the study area was examined using Pearson’s correlation coefficient, and the results are summarised in Table 9. This analysis aimed to understand the association between population dynamics and different land cover categories within the basin.

The findings revealed a strong positive correlation (r = 1.0) between population and built-up areas, indicating that as the population increases, there is a corresponding increase in built-up areas, suggesting urbanisation and the expansion of human settlements. The strong positive association implies that population growth is a driving force behind the expansion of built-up areas, possibly due to the need for housing, infrastructure development and other urban amenities to accommodate the growing population.

On the other hand, a strong negative correlation (r = −0.99) was observed between population and bare land, waterbody, wetland and vegetation, implying that these land cover types tend to decrease as the population increases. The negative association suggests that population growth has a detrimental impact on these land cover categories.

The Anambra State Government should develop a comprehensive land use plan for the river basin in collaboration with local communities. This plan should designate protected zones around key waterbodies and wetlands, such as the Otuocha wetland, to restrict development in these ecologically sensitive areas. Strict enforcement of zoning regulations is necessary, for example establishing at least 500 m buffer zones along riverbanks where agriculture and settlements are prohibited (Ziegler 1993, Chung 1994, Huang, Daberkow 1996, Nel 2016).

Local non-profits should launch community outreach campaigns to educate farmers on sustainable techniques and discourage activities like deforestation close to riverbanks. Awareness programmes in schools can foster conservation values in younger generations. Broadcasting public service announcements on local radio can also reach a broad audience with messages on sustainable land use practices (Sola 2014).

The Anambra State agricultural agencies should actively promote climate-smart techniques by providing training, materials and financial incentives to farmers. These could include agro-forestry, no-till farming and using cover crops. Pilot projects demonstrating sustainable practices can be implemented in key farming areas like Ayamelum (National Research Council 2010).

A multi-stakeholder forum involving government, communities, Non-Governmental Organisations and local researchers should be established to oversee land use planning and programmes in the basin. This can enhance coordination, knowledge sharing, monitoring and transparent decision-making. The forum should meet regularly to review land use regulations, conservation initiatives and emerging best practices (Hossain 2015).

While this study provides valuable insights into land cover change and their relationship with population growth in the Anambra River Basin, it is important to acknowledge several limitations. These limitations present opportunities for future research and improvement in land cover analysis and prediction.

Firstly, the accuracy and reliability of the land cover analysis are influenced by certain data limitations, including classification errors and temporal gaps in imagery data. To enhance the accuracy of the analysis, future studies could incorporate multi-temporal datasets and advanced remote sensing techniques. Additionally, the inclusion of ground-truth data and field surveys for validation and calibration would be beneficial.

Secondly, the CA-Markov model used in this study assumes that past land cover conditions and transition probabilities are the sole factors influencing land cover change. This approach neglects the potential influences of socioeconomic factors, policy interventions and climate change. To improve the realism and accuracy of the model, it would be valuable to integrate additional variables and drivers into the analysis.

Furthermore, it is worth noting that this study focused solely on the Anambra River Basin between 1987 and 2018. Expanding the spatial scope and extending the temporal scope of future research would capture larger-scale processes and interactions, providing a more comprehensive understanding of land cover change.

Although the study addresses environmental implications, it is important for future research to assess the socioeconomic and ecological impacts of land cover change using impact assessment methodologies. This broader perspective would provide a more holistic understanding of the consequences of land cover change.

Finally, stakeholder engagement and participatory approaches should be emphasised to align land cover analysis and management strategies with the needs of local communities. Incorporating the knowledge and perspectives of local communities would enhance the effectiveness of land cover analysis and contribute to more sustainable land management strategies.