Introduction

Urbanization, a global socioeconomic phenomenon, has witnessed an unprecedented surge, significantly altering land use and land cover patterns (Li, 2014). With over half of the world's population residing in urban areas and projections indicating a rise to 67.1% by 2050 (United Nations, 2012), the impact of urbanization on local and global environments remains profound, despite urban lands covering only 3% of the Earth's surface (Grimm et al., 2000; Herold et al., 2003; Liu & Lathrop, 2002).

Urban expansion, an essential facet of urban geography and planning, signifies the physical enlargement of urban areas, profoundly shaping landscapes, societies, and policy landscapes. In the broader spectrum of urban development, it intertwines with concepts like urban growth and urban sprawl, constituting a dynamic triad that molds the fabric of modern cities. Scholarly work often differentiates urban expansion from urban growth, where the former characterizes the outward spatial growth of built-up areas, while the latter encompasses both spatial and demographic aspects, reflecting the increase in a city's physical footprint alongside its population (Alberti, 1999; United Nations, 2012). Concurrently, urban sprawl, closely associated with urban expansion, captures the unplanned, disorganized, and often inefficient expansion of cities, leading to fragmented land use and socio-environmental challenges (Ewing et al., 2002; Galster et al., 2001). Studies by Seto et al. (2012) and Angel et al. (2005) underscore the global scale of urban expansion, emphasizing its implications for land use, environmental sustainability, and socio-economic dynamics. Additionally, research by United Nations (2012) and Alberti (1999) has delved into the nuanced interplay between urban growth and its multifaceted dimensions, illustrating the intricate relationship between urban spatial expansion and demographic transitions. In the context of urban sprawl, extensive scholarly discourse, such as the works of Ewing et al. (2002) and Galster et al. (2001), highlights its detrimental effects on land use efficiency, transportation patterns, and community well-being, emphasizing the need for efficient urban planning strategies to mitigate sprawl-related challenges.

Jordan stands amidst this global urbanization trend, grappling with rapid urbanization that has led to escalated population densities within major cities (Jawarneh, 2021; Makhamreha & Almanasyeha, 2011; World Bank Group, 2001). The country has experienced an annual urban population growth rate estimated at 3.8% (Al-Kofahi et al., 2018a; Higgitt, 2009; United Nations, 2015). Additionally, waves of migration from neighboring countries, such as Iraq and Syria, have significantly contributed to this uncontrolled urban growth (Al shawabkeh, 2019; Stave & Hillesund, 2015; World Bank Group 2013; Al Rawashdeh & Saleh, 2006, AL-Bilbisi & Tateishi 2004). Notably, the city of Irbid, positioned in proximity to the Syrian border, has undergone a substantial surge in population density, emerging as the most densely populated city in Jordan (Al-Kofahi et al., 2018b).

Urban expansion, a critical facet of urban development, manifests as the outward growth of built-up areas, transforming landscapes and posing environmental challenges (Seto et al., 2012; Angel et al., 2005). In Jordan, various studies have examined urban expansion dynamics, indicating a significant increase in built-up areas and a decline in agricultural and forested lands (Al shawabkeh et al., 2019; Jawarneh, 2021; Obeidat et al., 2019; Shatnawi et al., 2020). Particularly within Greater Irbid Municipality (GIM), the annual urban expansion rate witnessed a 1% increase from 2003 to 2015, with a concurrent annual agricultural land loss of 2% (Al-Kofahi et al., 2018b). This trend highlights the encroachment of urbanization on agricultural territories, raising concerns about potential land depletion by 2053 (Al-Kofahi et al., 2018b). Several studies have explored urban expansion patterns in Irbid and Amman cities (Al Shogoor et al., 2022; Al-Bilbisi, 2012; Al-Kofahi et al., 2018a; Makhamreha & Almanasyeha, 2011). However, within the specific context of Greater Irbid Municipality (GIM), a noticeable gap exists in the literature regarding comprehensive investigations into the evolving nature of urban expansion, its driving forces, and the intricate interplay between physical, socioeconomic, and neighborhood factors.

This research aims to address this gap by employing rigorous methodologies such as logistic regression and geographically weighted regression (GWR) to unravel the complexities of urban expansion dynamics within GIM focusing on pivotal historical periods in Irbid's chronology and contributing to sustainable urban planning endeavors.

Study area

Irbid Greater Municipality (GIM) positioned in the northwestern part of Jordan, is strategically situated near the Syrian border (Fig. 1). It consists of 23 districts, and an area close to 324 km2. GIM is the second most populous metropolitan municipality in Jordan (907,675 residents), after Amman, and accounts for almost half of the Irbid governorate's population (DOS, 2019). Its geographic location has contributed to its historical importance as a trade route hub connecting various regions. Irbid city serves as a pivotal center within Jordan, renowned for its historical significance and contemporary growth. It embodies a region of dynamic evolution and significant urban development. The socio-economic fabric of Irbid is characterized by a blend of old and new, where traditional practices coexist with modern developments, contributing to the city's unique identity and vibrancy. The topography consists of diverse landscapes, ranging from flatlands to undulating hills, fostering agricultural activity and urban expansion.

Fig. 1
figure 1

Location map of Greater Irbid Municipality

Full size image

Over the years, Irbid has experienced rapid urbanization, evolving from a traditional town into a bustling city. The population growth has been significant, driven by factors such as migration, economic opportunities, and educational institutions, including Yarmouk University and Jordan University of Science and Technology. The presence of academic institutions has played a pivotal role in the city's development. The city's expansion is evident in the development of residential areas, commercial centers, and infrastructure.

Studies highlight the impact of urban sprawl on the region's agricultural lands. The encroachment of urban areas onto fertile agricultural spaces poses challenges for sustaining traditional farming practices and preserving green spaces. Moreover, the influx of Syrian refugees has brought about changes in land use and land cover patterns, necessitating assessments of environmental impact and sustainable resource management. Efforts in spatial planning have been evident through the Irbid master plan. This plan aimed to guide and manage the city's growth, laying out strategies for infrastructure development, zoning regulations, and urban design. However, the evolving dynamics of urbanization and demographic shifts continuously challenge the efficacy of such plans. In essence, Irbid stands as a testament to the ongoing evolution of urban centers, showcasing the complexities of balancing growth with environmental preservation, cultural heritage, and socio-economic progress. The evolution of the city represents a rich blend of influences, challenges, and opportunities that shape its landscape and define its identity within Jordan and the broader region.

Methodology

Data

Various sources of data were used in this study. The data of 1967 are Corona (KH-4B) panchromatic images acquired from the U.S. Geological Survey’s Earth Resources Observation and Science (USGS/EROS). The Corona images have a 1.8 m spatial resolution. Downscaling the Corona images to a 15 m resolution of the Landsat images would make it more difficult to differentiate between urban and non-urban land, as the Corona images have only one band. So urban areas were digitized from Corona images and then rasterized to a resolution of 15 m to address discrepancies in spatial resolution between different images. An aerial photograph of 1978 was obtained from the Royal Jordanian Geographic Center (RJGC) with a scale of 1:20000. Several Landsat images were downloaded from the USGS as shown in Table 1. A digital elevation model (DEM) was obtained from the Alaska Satellite Facility (ASF) website with a spatial resolution of 12.5 m. It was resampled to match the spatial resolution of the satellite images. Highway roads were obtained from GIM for 2019. It was used as a reference for previous years. The socioeconomic centers were digitized from Google Earth, and the city center was obtained from GIM.

Table 1 Description of Landsat images used for mapping the urban expansion
Full size table

Data pre-processing

The Landsat 7 ETM + failed on May 31, 2003, causing all Landsat ETM images to have a wedge-shaped gap on both sides of every scene (Scaramuzza & Barsi, 2004); so, the image for 2011 was fixed by the fix scanline errors tool. Then the data was converted from the Digital Number (DN) to the Top of Atmospheric (TOA) reflectance for images that were downloaded from Landsat Level 1 (1976, 2011) using Eq. (1):

$$TOA=\frac{M*DN+A}{SIN({\varnothing }_{{\text{se}}})}$$
(1)

M is the band-specific multiplicative rescaling factor; A is the band-specific additive rescaling factor; and θse is the local sun elevation angle at the scene center. The values were available in the metadata file of each image. All other images were downloaded from Landsat level 2 as surface reflectance data, radiometrically and atmospherically corrected. After data correction, the individual bands of each satellite image were stacked and clipped according to the boundary of GIM. The panchromatic band with a resolution of 15 m was used whenever possible to improve the images' resolution using the Gram-Schmidt method. When the panchromatic band was unavailable, image output resampling to 15 m was done to standardize the spatial resolution for all data.

Mapping urban expansion

The Normalized Built-up Area Index (NBAI) proposed by Waqar et al. (2012), was applied to extract the built-up area from a Landsat TM image (Eq. 2) because it demonstrated spectral separability between bare soil and built-up areas (Hazaymeh et al., 2019). The accuracy of the index was verified through calculating the overall accuracy and kappa coefficient for 2003–2020 by random samples to assess the discrepancy between the classification results and the Google Earth images.

$${\text{NBAI}}=\frac{{{\text{B}}}_{{\text{SWIR}}2}-{{\text{B}}}_{{\text{SWIR}}1}/{{\text{B}}}_{{\text{GREEN}}}}{{{\text{B}}}_{{\text{SWIR}}2}+{{\text{B}}}_{{\text{SWIR}}1}/{{\text{B}}}_{{\text{GREEN}}}}$$
(2)

where BGREEN, BSWIR1 and BSWIR2 is the surface reflectance value of the visible green band, first shortwave-infrared and the second shortwave infrared, respectively.

The maximum likelihood classification was applied to the Landsat MSS image because not all bands were available for this index. However, the result of the classification was modified in reference to the 1978 aerial photograph. Each map was classified into two types of land use (1-urban and 0-non-urban). Therefore, six urban maps (1967, 1976, 1987, 2003, 2011, 2020) were created based on multi-temporal remote sensing images and aerial photos. Urban expansion was measured for 5 periods (i.e., 1967–1976, 1976–1987, 1987–2003, 2003–2011, 2011–2020) which were generated by subtracting the end-of-period data from the beginning of the period. The result of the urban expansion maps was data encoded with zero and one, where places urbanized during the period were coded (1), and those not urbanized coded as (0).

The potential driving factors of urban expansion

Previous research has identified several variables and their impacts on urban expansion. The decision of factors selection to include in the analysis depends on a prior understanding of the process behind the urban expansion. According to the literature (Liao & Wei, 2014; Alsharif & Pradhan, 2013; Li et al., 2013), seven variables were investigated that represent physical, socioeconomic, and neighborhood factors (Table 2).

Table 2 List of the selected drivers of urban expansion
Full size table

Physical factors

Elevation and slope were chosen as topographic variables that affect where urban expansion occurs. The slope was extracted from the DEM. Due to the limited size of the city, the climate does not considerably differ throughout the study area.

Socioeconomic factors

To represent socioeconomic factors, we chose two types of proximity variables (distance to socioeconomic centers and distance to roads).

Distance to socioeconomic center

In this study, three socioeconomic centers were defined: the city center, Yarmouk University, and Jordan University for Science and Technology although the latter is outside the study area but very nearby. Proximity maps were calculated as the Euclidean nearest distance using Spatial Analyst in ArcGIS 10.3.

Distance to roads

Several types of roads are typically made for different objectives and can affect urban growth differently. In this study, we considered highways (national-level roads connecting Irbid and other cities) and major roads (city-wide roads separated by middle islands and linking Irbid's social and economic centers). In the beginning, the major roads were digitized for the year 2020. Then, the 2011 road dataset was generated by eliminating the roads not in the images as of 2011. Similar procedures were applied to obtain all road layers using google earth and aerial photos. The Euclidean distances to the two types of roads were calculated using Spatial Analyst in ArcGIS 10.3. Distance to the roads in the final year of the period was used in the regression analysis. For instance, the impact of the proximity to roads on urban expansion from 2003 to 2011 was examined using distance to roads in 2011 because the roads completed in 2011 would impact urban expansion through this period.

Neighborhood factors

Neighboring factors were defined as the proportion of urban land in a surrounding area. According to numerous studies, areas are more likely to be developed if they are surrounded by more urban land (Miller, 2012; Luo & Wei, 2009; Liu & Zhou, 2005). The percentage of urban land within a 3 × 3 pixel window was calculated to examine the influences of local neighborhood factors on urban growth.

Statistical analysis

Data sampling

For each period, samples were created with a distance between each point of 200 m through the fishnet tool in ArcGIS 10.7, to minimize spatial autocorrelation. We then coded the points that were urbanized during the period as (1) and those that were not urbanized as (0). The number of points with no change (i.e., coded as 0) was much larger than the number of points with change (i.e., coded as 1). To obtain an equal number of samples of locations with change and no change, another random sampling procedure was run on points coded as 0 (Cheng & Masser, 2003; Monteiro et al, 2011). Consequently, the sample points used in the later logistical regression were 500, 938, 1704, 1174, and 1232 for the periods of 1967–1976, 1976–1987, 1987–2003, 2003–2011, and 2011–2020, respectively.

Modeling of spatial relationships

A logistic regression (LR) model was built for each time period to test the effects of the selected variables on urban expansion. The logistical regression model (Eq. 3) was formed as:

$${\text{logit}}\left({\text{y}}\right)-{\text{log}}\left(\frac{{\text{y}}}{1-{\text{y}}}\right)={\upbeta }_{0}+\sum\limits_{{\text{i}}=1}^{{\text{n}}}{\upbeta }_{{\text{i}}}{{\text{x}}}_{{\text{i}}+\upvarepsilon }$$
(3)

where y is the probability of urbanization, β0 is the intercept, β is a vector of estimated parameters, x is a vector of driving factors, and ε is a randomly distributed residual error.

The logistic regression model is a global model that produces a set of constant regression coefficients for the entire region, showing that the driver factor has the same impact on urban expansion everywhere. However, it may be more complex, which is different from the actual situation. Additionally, it assumes that model residuals are independent of each other. Given these two considerations, the parameters predicted in these models may need to be more accurate and complete (Shafizadeh-Moghadam & Helbich, 2015). Contrarily, the geographically weighted regression (GWR) has demonstrated success in handling the issue of spatial variation and spatial autocorrelation (Brunsdon et al., 1996; Fotheringham et al., 2001). The geographically weighted logistic regression (GWLR), a local form of logistic regression, is a better fit than the logistic regression and addresses the issue by considering the spatial properties of exploratory variables (Luo & Wei, 2009; Mirbagheri & Alimohammadi, 2017). Rather than producing a single equation for the whole study area, the geographically weighted logistic regression creates a different equation for each sample. According to how each independent variable affects each sample point, the parameter value of the independent variables likewise varies. Equation 4 is the GWLR formula:

$${\text{ChangeProbi}}=\frac{e\left({\beta }_{{\text{o}}}\left({u}_{{\text{i}}}{v}_{{\text{i}}}\right)+\beta 1\left({u}_{{\text{i}}}{v}_{{\text{i}}}\right)x{1}_{{\text{i}}}+\cdots ..{\beta }_{{\text{k}}}\left({u}_{{\text{ki}}},{v}_{{\text{ki}}}\right){{\text{X}}}_{{\text{ki}}}\right)}{1+e\left({\beta }_{{\text{o}}}\left({u}_{{\text{i}}}{v}_{{\text{i}}}\right)+\beta 1\left({u}_{{\text{i}}}{v}_{{\text{i}}}\right)x{1}_{{\text{i}}}+\cdots ..{\beta }_{{\text{k}}}\left({u}_{{\text{ki}}},{v}_{{\text{ki}}}\right){{\text{X}}}_{{\text{ki}}}\right)}$$
(4)

where (ui,vi) are the location coordinates in space of point i.

The GWLR estimates the observation parameter by giving a weight to all observations based on Euclidian distance. Naturally, closer distances have higher weights. The performance of the logistic regression and GWRL model was assessed using the Percent Correct Predictions (PCP) calculation. The samples were classified into predicted 0 and 1 based on the underlying model, and then the predicted samples were compared with actual 0 and 1. PCP was calculated as the number of correctly predicted cases divided by the total number of cases. The relative operating characteristic (ROC) technique was also used to analyze the prediction accuracy of the regression model and the probability that results from it. The ROC technique has been used in land use/cover change modeling studies and is regarded as a reliable way to validate models (Arsanjani et al., 2013; Hu & Lo, 2007; Pontius & Schneider, 2001; Wang & Mountrakis, 2011). The ROC measures the relationship between anticipated and actual changes. This technique determines the percentage of true and false positives for a range of thresholds and relates them to each other in a chart (Pontius & Schneider, 2001). The ROC calculates the area under the curve (AUC), which ranges from 0.5 to 1. A value of 0.5 indicates the random assignment of the probabilities, i.e., the expected agreement is due to chance. A value of 1 indicates perfect probability assignment (Pontius & Schneider, 2001).

Results and discussion

Mapping urban expansion

Over a period of 53 years, the GIM has witnessed significant urban expansion (Fig. 2). The area of urban land in 1967 accounted for 2.45% of the total land area (Table 3), while in 2020, the urban land comprised 22.58% of the entire study area, meaning that it has increased about nine times over this period. The highest urban expansion was in period 3 (1987–2003) with 99% and in period 1 (1967–1976) with 98%, when compared with the beginning year of each period (1987 and 1967). This is followed by period 2 (976–1987) 65%, period 4 (2003–2011) 18%, and lastly period 5 (2011–2020) 19% (Table 3). The overall accuracy ranged from 90–91% and coefficient kappa ranged from 80 to 82% from 2003–2020 (Table 4).

Fig. 2
figure 2

Spatial distribution of urban land in GIM over the period of study

Full size image
Table 3 Urban land and urban expansion GIM from 1967 to 2020
Full size table
Table 4 Confusion matrix for urban expansion GIM from 2003–2020
Full size table

As previously stated, the years of study were chosen based on the establishment of important urban centers in the GIM or in reference to political crises in the region that imposed immigration to Jordan. The largest urban expansion was in period 3 (1987–2003), which means that the establishment of JUST (1987) and the Iraq war (1991) had the most significant impact. Another reason could be that this period was the longest period among all periods of study, as it spanned 16 years compared to the other periods which represent durations in the range of 8–11 years.

Spatiotemporal relationships between urban expansion and driving factors

A multicollinearity was found between Yarmouk University and the city center, as well as between slope and elevation (the variance inflation factor (VIF) was greater than 10), therefore the city center and elevation were excluded. Additionally, the slope was excluded as it had no effect on urbanization throughout the study periods. As for the distance to the socioeconomic factors, several distances (1, 2, 5, and 10 km) were thoroughly investigated. The distance of 2 km was adopted as the best distance that optimized the interpretability of both the logistic regression (LR) and geographically weighted logistic regression (GWLR) models. Furthermore, both models were decided to take into account the same deriving factors.

Logistic regression results

The variance inflation factor (VIF) for all factors was less than five, so multicollinearity is not a concern. The best fitting model was chosen according to the percentage of correct predicted (PCP), Akaike information criterion (AIC), and the relative operating characteristic (ROC), where it turns out that the model performs well, and selected factors clearly explain the variation of urban expansion from a global perspective. A spatial autocorrelation which violates independence was found in the LR models for the periods 1, 2 and 3 using Moran's I (Table 5).

Table 5 Comparison between LR (logistic regression) and GWR (geographically weighted regression) based on ROC (relative operating characteristic), AIC (Akaike information criterion), and PCP (percentage of correct predicted)
Full size table

Results in Table 6 show the odds ratio values where the odds is \(\left(\frac{{\text{y}}}{1-{\text{y}}}\right)\) as defined in Eq. 3 above. The odds value compares the proportion of the event's probability to its probability of not happening (Bland & Altman, 2000). The percentage of urban areas (urban proportion) within a 45 window only has a negative effect on urban expansion in period 5 and has an odds ratio of 0.04. This result indicates that the prospect of urban expansion in a region will be 25 (1/0.04) times larger compared with a region that has a 1% higher urban proportion. The socioeconomic factors also showed significance in urban expansion, but their effects varied over time. The model demonstrated that major roads have a significant effect on urban development, which showed consistently negative relationships with urban expansion and had the most significant effect in the second period, where the odds ratio was 0.158. This result indicates that the odds of the urbanization expansion in an area nearer to major roads is 6.33 (1/0.158) times as great as that of a region 2 km further away from it. This is followed by the periods 5,1, 3, and 4, with odds ratios of 5.59, 4.98, 2.72, and 2.04, respectively.

Table 6 The odds ratio of the global regression models
Full size table

The distance to the highways negatively affected urban expansion after 1987 and had the most significant effect during period 4, where the odds ratio is 1.68, which means there is a 68% increase in the odds of urban expansion for every 2(km)-distance closer to highway roads. The odds ratio for the city center was 1.24 during period 1, which means the there is a 24% increase in the odds of urban expansion for every 2(km)-distance closer to the city center. The effect of the city center was not studied during the other periods because it showed a high correlation with Yarmouk University (YU), where R2 was greater than 0.80 and VIF larger than 10. The impact of YU has been studied since its foundation in 1976. The distance to YU showed negative effects on urban expansion during period 2, where the odds ratio was 1.27, which means there is a 27% increase in the odds of urban expansion for every 2(km)-distance closer to YU. In period 4, YU showed a positive effect with an odds ratio of 1.078, meaning there is a 7.8% increase in the odds of urban expansion for every 2(km)-distance closer to YU. The influence of JUST has been evident since its inception (1987) and continues to influence urbanization positively. It had the most significant effect during period 4, where the odds ratio is 1.37, reflecting that the 37% increase in the odds of urban expansion for every 2(km)-distance closer to JUST. Initially, the elevation factor was excluded because it formed a strong relationship with the slope. Still, after the experiment, the slope did not show an effect during all periods, which means that this factor does not affect the urban expansion pattern, so it was left out also. However, there is a difference in the elevations and slopes in the study area, where the elevation ranges between 149 – 1094 m, and the slope between 0—225.59%. No effect of slope appeared on the possibility of urban growth. This could be due to the low value (cost) of lands with rough terrain.

Geographically weighted logistic regression results

We used the geographically weighted logistic regression to the same sample points for each period. The logistic GWR model outperformed the global logistic model. The GWR model has significantly better goodness of fit than the global model, as shown by the decline in the values of AICc and the rise in R2. The logistic GWR model has improved prediction accuracy, according to increases in ROC, and spatial dependency of residuals has been decreased, as seen by the drop in the residuals' Moran's I. The relationships between urban expansion and driving factors show spatiotemporal variation, which is proved by varying the coefficient (β) across the study area. A t-statistic were created for each sample point by dividing the coefficient (β) by the standard error then using these values, a continuous surface was interpolated using the inverse distance weighted (IDW). It was found that there is a negative relationship between the distance to the major roads and the urban expansion in the GWLR model during all periods. This is similar to the results of the global logistic regression. In period 1, the negative effect for major roads decreased toward the north, and the negative impact on the northeast was higher than in the northwest, while from period 2 to period 3, the negative effect was higher in most northern regions and decreased toward the south. From period 4 to period 5, the negative impact was higher in the middle of the area and decreased toward the south and north. The t-statistical surface showed that the effect of the distance to the major roads in period 1 was significant in the entire study area, while in period 2 was a minimal area in the northeast that was not significant. In period 3, the factor was significant in the vicinity of Irbid and the south of the study area. In periods 4 and 5, small regions in the northwest and southeast appeared insignificant (Fig. 3). The effect of the highway roads on urbanization appeared in 1987 in the global logistic regression, while in the GWLR model, its effect appeared since period 1, as it was significant in the region's south and small parts in the north of the GIM. In period 2, it was significant in the north and southeast, and in period 3, it was significant from the GIM’s center and towards the south. Periods 4 and 5 were significant in the GIM's center and extended to the north. These places represented strong negative relationships with urban expansion, except for the second period, a positive effect appeared, but it was mostly negative in all periods (Fig. 4).

Fig. 3
figure 3

GWLR results of the estimated coefficient (β), and t-statistic surfaces for distance to major roads

Full size image
Fig. 4
figure 4

GWLR results of the estimated coefficient (β), and t-statistic surfaces for distance to highways

Full size image

The neighborhood factor showed a negative relationship with urban expansion since 1987, as it was significant in Irbid city and its environs during periods 3 and 4, while in period 5 it was statistically significant in most of the area and the strongest negative relationship was in the Irbid city and its environs (Fig. 5). This factor was significant during period 5 in global logistic regression.

Fig. 5
figure 5

GWLR results of the estimated coefficient (β), and t-statistic surfaces for urban proportion

Full size image

As for the effect of the city center in period 1, it showed significant a negative effect in the entire study area, and the lowest effect negative was in the center of the area and increased towards the south and north of the area (Fig. 6). YU showed a significant negative effect in period 2 in most of the region while, the most significant negative impact was around YU, while in period 3, it only showed statistical significance around YU and in the south of the region but in the south the effect was positive. In the fourth and fifth period, it did not represent in most of the region a statistically significant effect (Fig. 7). JUST showed a significant negative effect in the northwest, while in most of the GIM, it showed a statistically significant positive impact in period 3. However, in period 4 and period 5, there was no statistically significant negative effect, while it was only positive (Fig. 8).

Fig. 6
figure 6

GWLR results of the estimated coefficient (β), and t-statistic surfaces for distance to the city center

Full size image
Fig. 7
figure 7

GWLR results of the estimated coefficient (β), and t-statistic surfaces for distance to Yarmouk University

Full size image
Fig. 8
figure 8

GWLR results of the estimated coefficient (β), and t-statistic surfaces for distance to Jordan University of Science and Technology

Full size image

There was a negative relationship between urbanization probability and distance to the city center, distance to major roads, and distance to highways. The distance to YU also showed a negative relationship in period 2. This may be because locations close to socioeconomic centers provide more opportunities for access to social and economic resources (e.g., employment and urban infrastructure). However, a positive effect appeared for YU in period 4. The reason may be that the area has become full of buildings. On the other hand, the neighborhood factor from period 3 to period 5 showed that people tend to live farther away from urban locations may be due to the residents’ desire for more private space and avoiding social and visual interaction (Reilly et al., 2009). JUST showed a positive effect since 1987 except for a small area in the northwest. The area of the University of Science and Technology itself was not an attraction, but because of it, there was an increase in urban expansion in the city of Irbid.

Conclusions and policy implications

Based on multitemporal remote sensing images and aerial photos, we have observed a significant urban expansion in the city of Greater Irbid Municipality (GIM) between 1967 and 2020. During this period, the urban land area of GIM expanded approximately ninefold. Notably, recent trends in urban growth have been directed towards the southern regions of the city, encompassing areas such as Al-Sarih, Al-Husun, and Naima. This observation holds significant implications for urban planners and environmental managers who must gain a comprehensive understanding of the factors influencing this urban expansion. Failing to comprehend the historical and present dynamics of urbanization may lead to unforeseen challenges in the future. To investigate the driving factors behind urban expansion, we employed logistic regression (LR) and geographically weighted logistic regression (GWLR) analyses. Notably, the coefficients for these driving factors exhibited spatial variation across the region, with GWLR allowing us to account for this variability more effectively. Furthermore, the factors contributing to urban attraction and repulsion demonstrated variations over different time periods.

In earlier periods, Yarmouk University played a pivotal role as an influential factor in urban expansion, while in recent times, proximity to Yarmouk University and JUST University, as well as a higher percentage of existing buildings in an area, tended to deter urbanization. These findings highlight the evolving nature of urban development drivers within the study region. From a local perspective, coefficients displayed notable spatial disparities in both magnitude and sign. Utilizing GWLR over LR proved to be a more robust approach, as evidenced by its smaller AIC (Akaike Information Criterion) and lower Moran's I of residuals. Consequently, GWLR not only better identifies regional variations but also enhances the overall reliability of the model.

This study holds significant value for guiding current and future urban planning practices in Greater Irbid Municipality. The insights derived from this research offer an opportunity for the city to proactively address identified challenges and avoid potential pitfalls in its urban development endeavors. By comprehending the intricate dynamics of urban expansion, Irbid can better shape its sustainable and resilient future. The specific contribution of this study to the broader scientific discipline lies in its careful examination of evolving urban development drivers over time, employing GWLR to capture spatial variations effectively. This not only advances our understanding of local urbanization but also provides a methodological framework for comprehensively studying urban expansion in similar contexts worldwide.