Modeling the spatial dynamics of land cover transitions and vegetation conditions in Abuja city, Nigeria

Abstract

Purpose

This paper aims to assess the current and future dynamics of land cover transitions and analyze the vegetation conditions in Abuja city since its establishment as the capital of Nigeria in 1991.

Design/methodology/approach

A random forest classifier embedded in the Google Earth Engine platform was used to classify Landsat imagery for the years 1990, 2001, 2014 and 2020. A post-classification comparison was used to detect the dynamics of land cover transitions. A hybrid simulation model that comprised cellular automata and Markovian was used to model the probable scenario of land cover changes for 2050. The trend of Normalized Difference Vegetation Index was examined using Mann–Kendall and Theil Sen’s from 2014 to 2022. Nighttime band data from the National Oceanic and Atmospheric Administration were obtained to analyze the trend of urbanization from 2014 to 2022.

Findings

The findings show that built-up areas increased by 40%, while vegetation, bare land and agricultural land decreased by 27%, 7% and 8%, respectively. Vegetation had the highest declining rate at 3.15% per annum. Built-up areas are expected to increase by 17.1% between 2020 and 2050 in contrast with other land cover. The proportion of areas with moderate vegetation improvement is estimated to be 15.10%, while the proportion of areas with no significant change was 38.10%. The overall proportion of degraded areas stands at 46.8% due to urbanization.

Originality/value

The findings provide a comprehensive insight into the dynamics of land cover transitions and vegetation variability induced by rapid urbanization in Abuja city, Nigeria. In addition, the findings provide valuable insights for policymakers and urban planners to develop a sustainable land use policy that promotes inclusivity, safety and resilience.

Citation

Mshelia, Y.S., Onywere, S.M. and Letema, S. (2024), "Modeling the spatial dynamics of land cover transitions and vegetation conditions in Abuja city, Nigeria", Urbanization, Sustainability and Society, Vol. 1 No. 1, pp. 115-132. https://doi.org/10.1108/USS-12-2023-0026

1. Introduction

Over 50% of the world’s population dwells in urban centers (UN-Habitat, 2020) and is expected to surge up to 70% by 2050 (UN-DESA, 2019). This upsurge of urbanization is majorly credited to the high social and economic mileages of cities (Managi et al., 2019; Latief et al., 2022). Urban expansion often manifest through demographic changes and alteration of underlying natural land surface (Amado De Santis et al., 2023; Pandey et al., 2023). The urban areas cover a relatively small fraction of earth’s terrestrial surface, yet it exerts intense pressure on the natural landscape (UNCCD, 2015). Unrestrained urban expansion is a major factor driving ecological deterioration and reducing the resilience of cities (Sousa et al., 2022). Built intensification leads to loss of underlying natural landscapes, which causes air pollution and urban heat island (UHI) manifestation (Mwangi et al., 2021; Mshelia et al., 2022; Pandey et al., 2022). As a result, concerns have been raised that necessitates urgent need for solutions that will reconcile economic development with ecosystem conservation (Sousa et al., 2022). This call to action resonates with the United Nations Sustainable Development Goals 11 and 15, which cater for sustainable cities and land respectively (Branch, 2019; UN-Habitat, 2020). These can also be attained by guiding the urban spatial trajectory through deliberate and proactive management planning (Nyatuame et al., 2023).

The intricate interactions between man and natural subsystems alter land cover (Quintero-Gallego et al., 2018; Assaf et al., 2021). Land cover is a highly dynamic phenomenon that occurs at various spatial and temporal scales (Staab et al., 2015). Land cover change has received increasing global attention as a major topic amongst scientists, policymakers and land managers due to its potential implications on the ecosystems, economies and human well-being (Fan et al., 2016; Spiliotopoulou and Roseland, 2020; Guha and Govil, 2023). Land cover transitions is an important manifestation of anthropogenic land use, which is a critical indicator for understanding the interaction between man and nature at various spatial scale (Abdullahi and Pradhan, 2017; Guha et al., 2022). Understanding the drivers and impacts of land cover change is essential for making informed decisions about land use planning and management (Heymans et al., 2019). Therefore, geospatial technology has the potential to characterize, monitor and understand changes in land cover types overtime (Manakos et al., 2021; Guha et al., 2022). The Earth observation (EO) missions have made available consistent, accurate and cost-effective spatial information that will further contribute toward unraveling the complex man-nature interactions (Jiménez et al., 2022). Google Earth Engine (GEE) is a valuable platform that enables efficient processing of massive geodata (Osman et al., 2023) that is embedded with catalogs of over forty (40) updated satellite imageries, including Landsat, Moderate Resolution Imaging Spectroradiometer and National Oceanic and Atmospheric Administration (NOAA) (Zurqani et al., 2018; Managi et al., 2019; Ermida et al., 2020). In addition, the GEE cloud computing platform includes a variety of geo-processing and advanced analysis tools for public domain. The integration of remote sensing and geographical information system (GIS) can provide insights on land cover trajectories is vital toward making an informed decision (Nyatuame et al., 2023).

An accurate and in-depth analysis of geospatial data using landscape transition matrix provides vital insights on the process of change and potential future trajectory (Pontius, Shusas and McEachern, 2004). Geo-visualization and simulation models of land cover changes such as Conversion of Land Use and its Effects at Small Regional Extent, cellular automaton (CA), Markov chain (MC) and hybrid models have been adopted by scientist and is often used for land cover modeling (Ruben et al., 2020; Xiong et al., 2022). This model is built on the relationships between landscape transitions and its drivers overtime (Zhang et al., 2021). The hybrid CA and MC was adopted in this paper because of its reliability and high accuracy over the other models (Xiong et al., 2022). The model combined the CA’s spatial continuity with the Markovian ability to overcome the constraints of individual models to simulate a long-term future scenario of land cover (Ajeeb et al., 2020; Leta et al., 2021; Nyatuame et al., 2023).

Generally, land alterations and vegetation degradation are among the major issues that concern sustainability of cities (UNDP, 2019). The Normalized Difference Vegetation Index (NDVI) serves as an integral indicator or proxy for vegetation conditions (Conley et al., 2022). For instance, Czekajlo et al. (2020) studied urban vegetation conditions from 1984 to 2016 across Canada and reported an overall decreasing trend. The accurate evaluation of urban greenness can provide insight toward an informed planning and management strategy that will ensure the functionality of essential ecosystem services in cities (Czekajlo et al., 2020). The nighttime lights can be effectively used to map urban areas, population density and economic activity in cities (Liu et al., 2015). Some examples of nighttime light data sources are the Defense Meteorological Satellite Program’s Operational Linescan System nighttime light time series imagery (Liu et al., 2015; Du et al., 2019) and the Visible Infrared Imager Radiometer Suite (VIIRS) day/night band (DNB) for nighttime scenes (Mills et al., 2013; Adelabu et al., 2022). These sources contribute to the comprehensive understanding and mapping of urban landscapes by harnessing the unique information embedded in nighttime light patterns.

Much of the world’s future urbanization is predicted to occur in Africa and Asia (United Nation, 2014). Nigeria, China and India will account for about 35% of the projected growth in population (UN-DESA, 2019). Abuja was inaugurated as the capital of Nigeria in 1991 and has had a remarkable population increase estimated at the rate of 8.32% per annum (Abubakar, 2014; Mshelia et al., 2022). Abuja’s landscape is characterized by built intensification and significant proliferation of informal settlements triggered by economic development (Babandede, 2017). This paper therefore, assesses the dynamics of land cover from 1990 to 2020 and model the change for 2050 scenario as well as vegetation conditions in Abuja city, Nigeria. This paper is motivated by the need to improve the understanding of landscape changes and vegetation conditions toward an informed decision making in cities.

2. Materials and methods

2.1 Study area

The paper focuses on the spatial expanse of Abuja city, Nigeria (Figure 1). The area covers 256 km² that extends between latitude 7°25' and 9°20' north of the equator and longitude 5°45' and 7°39' East of the Greenwich meridians. The city is divided into four (4) zones located within the Federal Capital Territory of Nigeria. Abuja is the fastest growing city in Africa at the rate of 8.1% per annum with over three million inhabitants (Abubakar, 2014). The region lies within the Guinea Savannah region of Nigeria (Gumel et al., 2020). The average air temperature in the region ranges between 21°C and 40°C. The region has tropical wet (rainy) and dry (warm/cold) seasons (Adeyeri et al., 2017). The rainfall season lasts from April to September while the dry season lasts from November to March. Abuja city has experienced a significant up-surge of population and land alteration since it became as the capital city of Nigeria (Abubakar, 2014).

2.2 Data acquisition and preprocessing

This paper used geospatial technology to explore the dynamics of land cover transitions and vegetation conditions in Abuja city. Landsat and NOAA data sets (Table 1) were obtained via GEE platform (https://earthengine.google.org/). The land cover data for 1990, 2001, 2014 and 2020 were obtained from level-1 tier Landsat collection of paths:189/row:54. This collection comprises of the Thematic Mapper ^(TM), Enhanced Thematic Mapper ^(ETM+) and Operational Land Imager ^(OLI). To eliminate phenological variability, the images were chosen during the dry season (Verbesselt, 2010). The VIIRS DNB were obtained from NOAA for the years 2014 to 2022.

The boundary shapefile data were acquired from Abuja Geographical Information System. The reference data for accuracy assessment consists of ground truth data and high-resolution images in Google Earth Pro. The digital elevation model (DEM) was obtained from the United States Geological Survey (https://earthexplorer.usgs.gov/). Other data required for simulation of future land cover dynamics are waterbody, rivers and roads networks were acquired from Open-Street Map. Image preprocessing was performed on the GEE platform (Mshelia et al., 2022). The preprocessing step is important for establishing affiliation between the acquired data and biophysical properties (Wang et al., 2021). The schematic methodology is depicted in Figure 2. An initial correction of atmospheric and radiometric distortion was carried out to enhance spectral characteristics (Padró et al., 2017). The images were georeferenced to the World Geodetic System-1984 and then co-registered to the Universal Transverse-32N. All the images were clipped to the boundary shapefile of Abuja city and stacked for classification.

2.2.1 Image classification and change detection.

Image classification was implemented in the GEE platform based on random forest (RF) algorithm. The pixel-based nonparametric based RF classifier is a powerful and versatile algorithm for image classification, combining accuracy, robustness, feature importance analysis and interpretability (Kwame Tweneboa, 2017). RF classifier has been demonstrated to produce higher accuracies than other pixel-based classifiers such as the maximum likelihood algorithm. A total of 9,205 systematically collected points were sampled to maintain the uniformity of the image classification in ArcMap 10.8.1 (Supplementary I). The paper used 70% of the sample points for training and 30% for classification validation. The spectral characteristics of the Landsat images was classified into five (5) land cover categories, agricultural land, bare land, built-up areas, waterbody and vegetation based on the spectral responses from the Landsat images. The reliability of the classification was performed on each pixel using ground truth and high-resolution imagery on the GEE. An error matrix was generated for each of the reference epochs (1990, 2000, 2014 and 2020). The matrices present the overall accuracy (OA) and the kappa coefficient (K). Kappa Index signifies the agreement between classified outputs and reference data (Koko et al., 2022).

The paper is based on a 4 × 4 majority filter to enhance the cartographic appearance of the classified outputs in ArcMap. The proportion of each land cover category was computed for 1990, 2000, 2014 and 2020. A post-classification comparison (PCC) was performed to detect the extent and rate of change in the land cover categories (Takam Tiamgne et al., 2021). These were achieved by using the intersect spatial analyst tool in ArcMap. The magnitude, percentage and rate change in land cover category were calculated using equations 1–3, respectively.

(1)      M=Y2−Y1

(2)     PC=Y1 − Y2Y×100

(3) RC=A1 − A2n×100

where M denotes the change magnitude of trend; PC denotes the percentage change; RC denotes annual rate of change; Y₁ and Y₂ denote the proportion of land cover in the initial Year and final Year and finally, n is the total number of years between Y₁ and Y₂.

2.2.2 Simulation of future land cover dynamics.

Hybrid simulation model comprising of CA and MC was used to simulate the 2050 scenario of land cover change. The CA-MC model is embedded in the TerrSet, which enables the prediction of future scenarios of land cover changes by computation of land transitions between an initial Time-1 and later Time-2 and then extrapolation of the observed changes into the future (Ruben et al., 2020; Xiong et al., 2022). The model’s inputs are from broad category of location and topographic factors (Figure 3). The locational factors are derived using the Euclidean function in ArcMap to obtain layers for distance to existing built-up surfaces, distance to major roads, distance to railway and distance to urban center. The topographic factor is calculated from DEM to produce slope. Model calibration and validation were performed to simulate land cover for the year 2050. The Markovian model was used to generate the transition probabilities and transition area using Markov model (Ruben et al., 2020). The stochastic logic of Markov is based on the principle of progression of land cover transition (Rahman et al., 2017). CA models are mathematical models that simulate the potential transition maps considering the spatial structure and state of neighborhood (Xiong et al., 2022). Suitability maps were acquired using multi-criteria evaluation and Fuzzy membership function (Table 2). Constraints are the Boolean criteria that hinder urban growth (Supplementary II). A standard (5 × 5) contiguity filter was applied to retrieve the transition probability and areas matrix (Supplementary III).

The simulated output was validated using the actual and simulated land cover for 2020 (Supplementary IV). Validation of the simulation model is significant for land cover prediction (Khan et al., 2020). The validate module in the TerrSet software was used to calculate diverse components of the kappa Index of Agreement (KIA) including the kappa standard (K_standard), kappa for grid-cell level location (K_location) and kappa for no information (K_no), which allow the determination of the overall weaknesses or strengths of the results based on both quantity and location. An accuracy higher than 80% infers confidence in the simulation model (Rimal, 2020).

2.2.3 Analysis of Normalized Difference Vegetation Index and nightlight band.

Vegetation and nighttime light were computed from Landsat 8 and VIIRS-DNB, respectively, using JavaScript library in the GEE platform (https://code.earthengine.google.com/). NDVI is considered suitable for vegetation monitoring due to its sensitivity to high biomass regions. The trend of NDVI was computed for the period between 2014 and 2022. The magnitude of persistent increase and decrease of vegetation status was derived using the Theil-Sen (TS) slope estimator (Liu et al., 2015). The direction of trend was performed using the nonparametric Mann–Kendal test (Adenle et al., 2020). The thresholds of vegetation conditions were categorized into three (3) namely; degradation (significant decrease, p < 0.05), stable (no significant change) and improvement (significant increase, p < 0.05). The DNB is an effective data source for urban mapping, which widely reflects energy use, population density and economic activity (Mills et al., 2013; Liu et al., 2015).

3. Results and discussions

3.1 Land cover change dynamics

The RF classification produced an OA for the years 1990 (93.76%), 2001 (90.40%), 2014 (86.70%) and 2020 (88.64%). The kappa coefficient (K) ranged between 0.85 and 0.9 for all the reference epochs (Table 3). According to Leta et al. (2021), kappa values above 0.8 denote strong agreement in land cover classification (Kwame Tweneboa, 2017). Therefore, accuracy of the classified land cover output is deemed suitable for analysis and simulation. There are five major land cover types in Abuja city namely bare land, agricultural land, built-up, water body and vegetation.

The evolution of classified land cover types in Abuja city is illustrated in (Figure 4), which shows a progressive change dynamic in land cover spatial distribution over the span of 30 years. The city mainly expanded outward toward the northwest and southwest regions.

The proportion of each land cover shows an increasing and decreasing trend from 1990 to 2020 (Figure 5). The dominance of agricultural land and vegetation was replaced by built-up areas. Built-up areas experienced a sustained increase over the span of 30 years, while vegetation, bare land and agricultural land have decreased during the period. Similar studies in the region suggest that the rapid built expansion is due to population growth and economic development (Koko et al., 2022). In 1990, agricultural land covered 44.50%, followed by vegetation (27.45%), built-up (15.90%) and bare land (11.85%). By 2020, built-up increased to 55.45% while agricultural land, bare land and vegetation declined to 19.83%, 5.21% and 19.21%, respectively, with over 70% of the loss observed in natural landscapes. The proportion of waterbody is the lowest as compared to the other land cover types, but did not show any significantly change over the same period.

The PCC results illustrate the extent and annual rate of change in land cover categories from 1990 to 2020 (Table 4). Built-up increased from 15.9% to 55.5% at the rate of expansion of 4.12% per annum. However, bare land, agricultural land and vegetation declined by 24.72%, 8.16% and 6.64%, respectively. The rapid decline in the other land cover could mainly be attributed to built-up expansion resulting from massive migration due to the economic prospects and better infrastructure in Abuja city. The results indicate that waterbody remained relatively unchanged across over the period. The highest declining annual rate of change was observed in vegetation at 3.15% per annum. Adenle et al. (2020), which is attributed to human-induced activities from 2003 to 2018 that led to the negative decline of vegetation.

Further analysis of the transition matrix reveals that 70.93% of land cover has transitioned from one category to another between 1990 and 2020 (Mshelia et al., 2022). Transition matrix helps to expand the understanding of evolving dynamics of land cover change (Pontius et al., 2004; Osman et al., 2023). The proportion of agricultural land experienced highest conversion (44.18%) to built-up (23.07%) and vegetation (8.10%). Most vegetation (27.51%) was mainly converted to built-up (12.76%) and agricultural land (7.25%) while bare-land (12.16%) was largely converted to built-up (7.90%) and vegetation (2.32%). Waterbody (0.23%) was largely converted to vegetation (0.04%) and bare land (0.02%). Generally, the result reveals that the high rate of built-up expansion (4.12%) has consequently intensified net transfer amount from agricultural land, vegetation and bare-land. Asabere et al. (2020) opined that built intensification contributes to the decline of natural landscapes like vegetation and agricultural land. The proportion of land cover that remained unchanged over the study period is presented diagonally in bold font (Table 5). The results show that about 29.07% of the land cover remained unchanged between 1990 and 2020. Built-up experienced the highest persistence (11.97%), followed by agricultural land (10.10%) and vegetation (6.19%). The transition of built-up into other land cover types is significantly low. The persistence of vegetation mainly occurred in the Northern and Eastern regions of the city, which can be traced mainly to the reserved areas (Permana et al., 2017). The continued loss of vegetation degrades the ability of the urban ecosystem to provide essential services that can ameliorate UHI effects (Mwangi et al., 2021; Pandey et al., 2022). Mshelia et al. (2022) highlighted the significance of land cover in the patterns of surface temperature of cities.

3.2 Future land cover changes

The CA-MC model was used to simulate the land cover for 2020 to ensure the accuracy of prediction output. The percentage of probable future change is illustrated in the transition probability matrices. The results for simulation error show that 83% of the total land cover area is consistent. KIA shows a good agreement, with the validation results being K_no (0.8633), K_location (0.8616) and K_standard (0.8225). Therefore, there is a strong agreement between the simulated and actual 2020 land cover. Figure 6 shows the predicted result of land cover in Abuja city for 2050. The KIA results proved that the model is consistent and reliable in forecasting the 2050 land cover. The weight criterion showed that the main drivers of urban expansion are primarily shaped by access to major road, proximity to urban center and the existing built-up surfaces (Table 2). The simulation result shows that there will be a significant built-up expansion at the expense of other land cover types by 2050. By 2050 (Figure 7), the distribution of agricultural land will be 58.63 km² (10.43%), bare land 31.79 km² (5.65%), built-up 407.42 km² (72.46%), waterbody 63.19 km² (11.24%) and vegetation will be 1.23 km² (0.22%). The trend of built-up expansion will mainly be at the expense of vegetation and agricultural land. This finding resonates with results from previous studies that confirmed built-up as the major driver of land cover change (Ibrahim et al., 2018; Acheampong and Asabere, 2021; Wang et al., 2021). Additional findings indicate that agricultural land will experience decrease of about 52.65 km² (9.4%), vegetation will decrease by 45.09 km² (8.05%) and bare land will slightly increase by about 2.54 km² (0.44%).

Generally, this finding has significant implications on the future distribution of vegetation in Abuja city. This indicate significant pressure on the natural landscape, which could lead to further ecological deterioration and reduction in thermal comfort of the city (Mshelia et al., 2022). Therefore, proper planning and management of the urban landscape are inevitable in order achieve sustainable development.

3.3 Vegetation trend and conditions

The prevailing vegetation conditions in Abuja city is shown in (Figure 8). The NDVI trend analysis has shown a significant decline in the distribution of vegetation in Abuja city from 1990 to 2020. A significant vegetation degradation can be observed along the central region of the city, with stable areas sparsely scattered across the exterior regions and vegetation improvement observed in the North-Eastern region.

There is persistent increase and decline in the annual cumulative NDVI at pixel level from the TS slope estimator, which highlights the vegetation status for 1990–2020 in Abuja city (Table 6). About 259.44 km² (46.76%) of the city show significant degradation (p < 0.05) of varying degree, about 83.81 km² (15.11%) have experienced significant improvement (p < 0.01) and about 211.63 km² (38.14%) show no significant change over the period. Generally, the result indicates that a higher expanse of vegetation in Abuja city is degraded. This finding align with previous study that reported a similar scenario of vegetation degradation resulting from rapid urbanization (Liu et al., 2015; Guha and Govil, 2023).

Figure 9 shows the linear correlation between average NDVI and average DNB in Abuja city from 2014 to 2022. The results show that earlier epochs have a significant positive correlation but gradually weakens in later years in Abuja city. The result reveals that the higher the DNB value, the more urban vegetation greenness. However, the gradual decline in the correlation in later years indicates an increase in urbanization. This result aligns with previous study that reported on a similar scenario of vegetation degradation resulting from rapid urbanization (Liu et al., 2015; Du et al., 2019). They suggest that such patterns can be found in warm regions similar to Abuja city. This finding suggests that vegetation degradation can be linked to urbanization in Abuja city.

4. Conclusion

The paper used geospatial techniques to gain insights on the dynamics of land cover changes and vegetation conditions in Abuja city, Nigeria. GEE cloud computing platform was used to process the Landsat imageries (1990–2020), NDVI (2014–2022) and the VIIRS nighttime DNB data set (2014–2022). However, the current study’s limitation is the use of images from different time intervals due to the difficulty in obtaining high-quality Landsat images from the study area during the study periods. The RF classification output was satisfactory and deemed suitable for the analysis. Similarly, the hybrid CA-Markov model obtained great performance in producing the probabilistic future scenario of land cover changes at an acceptable accuracy. However, the paper was limited by the coarseness of spatial resolution and availability of long-time nighttime lights data set. Therefore, future studies should explore the integration of VIIRS DNB composites with other data sets like high-resolution satellite imagery.

The paper has demonstrated that built-up areas have experienced a sustained increase that is parallel to other landscapes over the span of 30 years from 1990 to 2020 in Abuja city. Built-up expanded outwards toward the northwestern and southwestern regions. The expansion of built-up has consequently intensified the net-transfer amount from other natural landscapes like agricultural land and vegetation. The built-up expansion primarily depends on accessibility of road networks, proximity to urban center and other built surfaces. The predicted land cover scenario suggests that built-up expansion will further continue at the expense of vegetation, which will worsen the prevailing vegetation conditions that is currently about 40% degraded. The findings suggest that rapid built-up areas and vegetation degradation is linked to urbanization in Abuja city. Rapid urbanization leading to vegetation loss will have a negative effect on urban thermal comfort and livelihoods. Therefore, an integrative land management strategy that combines both spatial planning and environmental governance is crucial to averting these detrimental effects. Future research will need to monitor the effects of urban expansion on the microclimate of cities to implement effective mitigation measures.

Figures

Figure 1.

Location of Abuja city, Nigeria

Figure 2.

Methodology for assessing land cover dynamics and vegetation conditions

Figure 3.

Locational and topographic factors

Figure 4.

Land cover classification for Abuja city, Nigeria (1990–2020)

Figure 5.

Proportion of land cover in Abuja city, Nigeria (1990–2020)

Figure 6.

Predicted land cover scenario for 2050

Figure 7.

Change dynamics of predicted land cover (2020–2050)

Figure 8.

Vegetation conditions (2001–2020)

Figure 9.

The correlation between mean NDVI and DNB in Abuja city, Nigeria

Table 1.

Satellite characteristics and sample points for land cover classification

Data set	Sensor	Resolution	Acquisition date	Sample points	Accuracy (%)
Landsat 5	TM	30 m	February12, 1990	2,313	93.7
Landsat 7	ETM+	30 m	January, 9 2001	2,275	90.4
Landsat 8	OLI/TIRS	30 m	January, 5 2014	2,256	86.8
Landsat 8	OLI/TIRS	30 m	February, 7 2020	2,361	88.6
NOAA	VIIRS-DNB	–	2014–2022	–	–

Source: Created by authors

Table 2.

Factors standardization

Factors	Membership function	Control point	Weight
Distance to urban centers	Linear decreasing	0–3,000 m	0.3452
Distance to roads	J shaped decreasing	0–3,000 m	0.226
Distance to built-up areas	Linear increasing	0–3,000 m	0.2146
Slope	Sigmoid decreasing	0°–20°	0.0608

Source: Created by authors

Table 3.

Accuracy assessment of land cover classification

Year	User accuracy (%)					Producer accuracy (%)					Overall accuracy (%)	Overall kappa
	AG	BL	BU	VG	WB	AG	BL	BU	VG	WB
1990	0.83	0.82	0.89	0.97	1	0.94	0.82	0.86	0.98	1	93.76	0.96
2001	0.82	0.87	0.95	0.94	1	0.82	0.86	0.95	0.98	1	90.4	0.91
2014	0.82	0.84	0.89	0.82	1	0.87	0.88	0.84	0.90	1	86.70	0.85
2020	0.88	0.87	0.91	0.89	1	0.83	0.82	0.92	0.83	1	88.64	0.87

Notes:

AG = agricultural land; BL = bare land; BU = built-up; VG = vegetation; WB = waterbody

Source: Created by authors

Table 4.

Rate of land cover change from 1990 to 2020 (%)

Land cover	1990–2001	2001–2014	2014–2020	1990–2020	Annual rate
Agricultural land	−25.33	1.77	−1.16	−24.72	−2.67
Bare-land	6.85	−6.06	−7.43	−6.64	−2.71
Build-up	19.03	9.62	10.9	39.55	4.12
Vegetation	−0.68	−5.18	−2.3	−8.16	−3.15
Waterbody	0.13	−0.15	−0.01	−0.03	−0.42

Source: Created by authors

Table 5.

Land cover transition matrix between 1990 and 2020

Land cover		2020					Total
		Agric. land	Bare land	Built-up	Vegetation	Waterbody
1990	Agric. land	10.10	2.90	23.07	8.10	0.01	44.18
	Bare land	1.32	0.62	7.90	2.32	0.01	12.14
	Built-up	0.78	0.40	11.97	2.79	0.00	15.92
	Vegetation	7.25	1.27	12.76	6.19	0.04	27.51
	Waterbody	0.00	0.02	0.00	0.04	0.19	0.23
Total		19.45	5.21	55.7	19.41	0.25	100

Source: Created by authors

Table 6.

Trend of Enhanced Vegetation Index (2001–2020)

Mann–Kendal significance	Proportion
	Area (km2)	Area (%)
Degradation (a significant decrease, p < 0.05)	259.44	46.76
Stable (no significant change)	211.63	38.14
Improvement (a significant increase, p < 0.05)	83.81	15.11
Total	554.88	100.00

Source: Created by authors

Supplementary material

The supplementary material for this article can be found online.