This study examines how tourism attractions are distributed within Luxor Governorate, located in Upper Egypt along the Nile River. Luxor Governorate is a premium tourism destination due to its famous ancient archaeological and heritage sites that exist in both urban areas (e.g., Luxor, Armant, and Esna) and rural areas (e.g., surrounding villages). The region has been identified as having several world-famous tourist destinations such as the Valley of the Kings, the Karnak. Temple Complex, and Hatshepsut’s Mortuary Temple. Spatially, Luxor Governorate is split between the east and west sides of the Nile with each side possessing a unique combination of tourism capacity, physical planning, and land use characteristics.

For data collection purposes, this study employed two main sources to collect tourism attractions in Luxor Governorate: OpenStreetMap (OSM) accessed on 6 July 2025 and official documentation. The POIs associated with tourism contained in OSM provided a base spatial dataset from which the georeferenced POI data were extracted. Official government sources were also consulted to validate and enhance the data collected through OSM; these included the Centre for Documentation of Cultural and Natural Heritage (CULTNAT) and the General Organisation for Physical Planning (GOPP). Both sources provided authoritative classifications, site descriptions, and spatial verification. The compiled data from these aforementioned sources were then placed into a single geospatial database, which contained detailed information about each attraction (e.g., location, name, and classification). Also, it is worth mentioning that the attractions were categorized into four categories: historical (e.g., Luxor Temple, Hatshepsut Temple, Medinet Habu); cultural (e.g., Luxor Museum, Local festivals, and folklore shows); scenic (e.g., Nile River riverfront, Banana Island, and Theban Mountain backdrop); and leisure (Nile cruises, hot air balloon, and local markets) ¹; ²⁷. Figure 1 below presents the geographic location of the study area, as well as its various tourism attractions.

After conducting the data collection process, there are two main analytical components to the research methodology for this study: (1) an examination of the spatial distribution of tourism attractions, and (2) an assessment and analysis of the factors that affect it. The workflow that integrates these analyses is shown in the following Figure 2.

The second component of the analysis for this study is to find the most important factors that influence the spatial distribution of tourism attractions in Luxor. At first, all factors that could affect the spatial distribution of tourism attractions were collected from previous literature. Next, they were filtered to be relevant to the local context of Luxor and the availability of sufficient data for the analysis. Finally, the importance of these factors was evaluated using both of the tools (Analytic Hierarchy Process, AHP and Geodetector). AHP helped to evaluate the importance of each factor according to the opinion and experience of the experts. Geodetector is a statistical tool that uses the actual data to estimate how much of the pattern of the distribution of the tourism attractions on the map can be explained by each factor. Therefore, the combined application of both tools provides a more complete and balanced evaluation of the factors that are responsible for the spatial distribution of tourism attractions in Luxor.

a- Analytic Hierarchy Process (AHP)

The process of the AHP starts by creating a pairwise comparison matrix = a_ij, where each element a_ij represents the relative importance of factor i versus factor j, as judged by experts (panel list). The matrix is constructed in the following Equation 7:

(7)

This matrix is reciprocal, meaning a_ij = 1/a_ij , and each diagonal entry is 1 because a factor is equally important as itself. To populate the comparison matrix, fifteen experts were selected through purposeful sampling by using specific eligibility criteria to ensure they had appropriate knowledge and expertise. For example, each participant was required to have at least seven years of professional experience in either tourism planning, heritage management or urban development.

The 15 participants included six senior planners from the General Organisation for Physical Planning (GOPP); two tourism development specialists from the Ministry of Tourism and Antiquities; five university professors whose specialisations include regional tourism development and urban planning; and two local government representatives from Luxor Governorate. Each expert independently performed pairwise comparisons of the factors using the standard scale (1-9), where higher values represent a stronger perceived importance of one factor over another. Once the matrix was formed, the geometric mean method was used to compute the relative importance of each factor. For each row, the geometric mean is calculated as Equation 8:

(8)

These geometric means are then normalized to derive the final priority weights W_i , using the equation 9:

(9)

To validate the consistency of the expert judgments, the maximum eigenvalue λ_max is calculated using the following Equation 10:

(10)

Also, the Consistency Index (CI) is then computed using Equation 11:

(11)

And the Consistency Ratio (CR) is finally calculated as Equation 12:

(12)

b- Geodetector

Geodetector is a spatial statistical tool used to identify and explain spatial stratified heterogeneity by measuring how strongly an explanatory variable (e.g., influencing factors) accounts for the observed spatial variation in a dependent variable (e.g., tourism attractions). Geodetector was chosen over other methods (OLS regression or Pearson Correlation) since it does not require a linear relationship between variables, nor does it require that all data be either categorical or continuous. Geodetector measures explanatory power by calculating variances within spatially stratified sub-regions (unlike global regression models which calculate associations across entire study areas).

Therefore, Geodetector has the ability to capture local variations and non-stationary patterns that would not be captured by standard models. Additionally, Geodetector’s ability to provide robust results when dealing with multi-collinear data, combined with the complementary nature of Geodetector and the AHP results (which provide data-driven support for expert judgments), makes Geodetector an ideal method to analyse the complex spatial phenomenon of the different kinds of tourism attractions in this study.

As shown in Figure 3, the study area is divided into sub-regions h = 1, 2, …, L. For each, the mean and variance of the dependent variable Y (i.e., tourism attractions) are calculated as and σ²h, respectively. These are then compared to the overall mean and variance σ² for the entire Luxor governorate. The explanatory power of each factor is calculated using the q-statistic, ranges from 0 to 1, the higher q value indicating stronger explanatory power.

c- Multiscale Geographically Weighted Regression (MGWR)

The MGWR model used in this study estimates how the relationship between the dependent variable and a set of explanatory variables changes across space. More specifically, an adaptive bisquare kernel function was utilised, and bandwidth sizes were independently optimised for each covariate based on the Akaike Information Criteria adjusted for small sample size (AICc). The model assumes that the value of the dependent variable (e.g., tourism attractions) at a given location is a function of locally varying coefficients associated with each independent variable (e.g., influencing factors). All MGWR models were run using the “mgwr 2.2.1” Python software package ²³, which utilises the back-fitting algorithm to determine the best fit of local regression coefficients at various spatial scales. The model takes the Equation 13:

Where X_ik is the K explanatory variable at location i, β_k(u_i,v_i) represents the coefficient of that variable at coordinates (u_i,v_i), β_o(u_i,v_i) is the location-specific intercept, and ε_i is the random error term. It is worth mentioning that the adaptive bandwidths were between 30 and 64 nearest neighbours (see Table 4) for the MGWR models. These bandwidths are calculated utilising an iterative golden section search to find the minimum AICc, while also ensuring that each local regression has sufficient spatial resolution to capture meaningful spatial patterns, and does not over-fit the data.

The overall spatial distribution of the four categories of tourism attractions in Luxor shows a statistically significant clustered pattern. The range of values for the Nearest Neighbour Index (NNI) varies from 0 (perfect agglomeration - all the data coincides) to 1.0 (spatial randomness - no data coincidence) and approaches 0 as agglomeration increases and exceeds 1.0 as data disperses. Therefore, it is essential to evaluate both the Z-Score and p-values simultaneously when interpreting NNIs. Although NNIs measure effect size, they do not verify if the observed pattern of agglomeration differs from randomly occurring patterns unless evaluated using Z-Scores and p-values ²².

Following established conventions in spatial statistics, this study interprets NNI values of < .10 as extreme agglomeration, .10 to .25 as strong agglomeration, .25 to .50 as moderate agglomeration, and values > .50 as little agglomeration. All four categories of tourism attractions showed evidence of statistically significant agglomeration (p < .01) with Z-Scores varying from -18.55 to -24.51 (less than 1% probability of occurrence due to random chance).

As presented in Table 2, historical tourism attractions exhibit the strongest clustering, with a nearest neighbour index of only 0.004 and a remarkably high z-score of -22.86. This explains that Luxor’s ancient monuments (e.g., temples, tombs, etc.) are densely concentrated, especially in the East and West Bank archaeological zones. Cultural tourism attractions were also found to be strongly clustered (NNI = 0.122, z-score = -18.55), which suggests many cultural attractions (e.g., museums, cultural centres, and folklore venues) are co-located near the primary historical landmarks or are located in urban centres close to those attractions. The leisure tourism attractions demonstrated a similar level of clustering (NNI = 0.202), which indicates that many of the leisure-related tourism attractions (e.g., shopping, dining, recreational activities, etc.) may be located in more commercially viable or accessible areas, but the majority are still within walking distance to the most popular tourist locations. Lastly, while scenic tourism attractions are still clustered (NNI = 0.301), they demonstrate the least value of density of all four categories and therefore have the widest geographic dispersion along the Nile River and riverbanks. Generally, all four categories of tourism attractions have shown consistently negative Z-Scores and all p-values are less than .01, thus validating that the spatial distributions of the four categories of tourism attractions are not the result of random variation, and that the NNIs demonstrate increasing degrees of agglomeration The detailed results of the nearest neighbour analysis are presented in Appendix (A) and the summary results of the nearest neighbour index (NNI) are presented in the following Table 1.

To further analyse the spatial logic and directional patterns of tourism attraction clusters in Luxor, Figure 4 shows the results of the Standard Deviational Ellipse (SDE) and Centre of Gravity. Both analyses resulted in the identification of a clearly defined North-South axis for all four categories of tourist attractions, which run almost directly along the Nile River corridor. The two categories of historical and cultural attractions (red and blue), are both centred on Luxor City and the Qurna area and demonstrate compact ellipses which indicate strong centralisation of these attractions around the major historic zones of Luxor (e.g., Karnak, Luxor Temple, and the Valley of the Kings). The relatively narrow and concentrated nature of the orientations of these two types of attractions also indicates the high density and spatial compactness of the historically layered archaeological sites. In contrast, leisure and scenic tourism attractions (pink and black) have slightly larger ellipses which extend further to the South than those other two categories, indicating a greater spatial distribution and lower clustering intensity. These two categories tend to follow areas which are more accessible or natural scenery, especially along the riverbanks, viewpoints, and in rural leisure spots.

Building upon the established directional and cluster trends from the NNI and SDE analyses, the KDE was used to explore how concentrations vary by type of tourism attractions. As shown in the density maps (Figure 5), there are varying degrees of concentration for each category of attractions. For example, historical attractions (Figure 5b) have a dense, highly concentrated hotspot at Qurna and Luxor city, which corresponds with the well-preserved archaeological zones. This confirms that historical tourism is driving much of the development and movement of visitors within Luxor and is responsible for an extreme amount of attractions with a maximum intensity value of 144. Cultural attractions (Figure 5c) have a similar but slightly less concentrated pattern of density centres at both Luxor and Esna. This distribution reflects the close relationship between cultural destinations and major heritage landmarks and the many traditional markets, cultural folklore, and local art galleries.

Unlike the other two tourism attraction categories (historical attractions and cultural attractions), the leisure attractions category (Figure 5d) has both a high level of concentration as well as a wider spatial distribution. The core area for leisure attractions is located in Luxor and Qurna, but there are many secondary clusters that are located along the southern part of the Nile River, particularly near Armant and El-Tod. Additionally, the leisure attractions recorded the highest maximum density of all four categories at 258 attractions due to several factors, such as the popularity of Nile cruises, which is considered the main transportation link between Luxor in the north and Aswan in the south. Finally, the scenic spots category (Figure 5a) has the widest geographic distribution with many moderate-density zones throughout the entire governorate (from Esna in the south to Luxor in the north). This pattern demonstrates how visually appealing and environmentally attractive the Nile River corridor is. Scenic spots had a lower peak density of 240, however due to their linear and dispersed patterns of clustering reflects tourists’ interest in natural landscapes, riverbank views, and panoramic vistas rather than specific or fixed monuments.

Lastly, Local Moran’s I spatial autocorrelation analysis results indicate various levels of clustering and spatial outliers distribution of tourism attraction densities throughout the Luxor Governorate. As shown in the following Figure 6, the majority of the governorate falls into the “Low-Low” category (dark blue), which indicates low tourism values in surrounding areas. These areas are primarily found in Luxor governorate’s south and central parts (e.g., Al-Adaimah, Al-Deir, and Asfun). The fact that these areas are generally underdeveloped in terms of tourism may create opportunities for investment or planning focus. In contrast, few areas show the exact opposite pattern. For example, Esna has been highlighted in red as a “High-High” cluster; an indication that it is a local tourism hotspot with high attraction density surrounded by similarly high-value areas, thereby making it a key tourism hub. Additionally, areas identified in light red (e.g., Al-Toud and Al-Dabaiya) are classified as “High-Low”; an indication that while they have high tourism values, they are surrounded by areas with significantly lower tourism values. Due to this, they can be viewed as being situated as potential strategic nodes linking tourism-based urban centres with the surrounding rural tourism areas. Finally, the grey areas are considered “Not Significant”, meaning there was no discernible tourism-related spatial pattern evident. The grey areas can therefore be seen as transitional zones between tourism hub areas.

The next step after assessing the spatial distribution of tourism attractions in Luxor, is to assess the factors that affect and support the development in tourism attractions. Table 2 shows an overview of the 16 selected factors in this study. These selected factors have been categorised into five main categories: urban (four factors), social (three factors), infrastructural (three factors), environmental (one factor), and economic (five factors).

The different categories represent the dimensions of how tourism developments occur geographically within the governorate. For a detailed view of the spatial distribution of these factors, see the mapping of the 16 selected factors in Appendix (B).

After identifying the 16 selected influencing factors, the study applied AHP and Geodetector analyses to determine the relative importance of the factors in shaping the spatial distribution of tourism attractions in Luxor. The AHP analysis found that the 10 tourism experts ranked the following as the most important factors: proximity to the nearest urban centre (PUC), with a weight of 0.10. This was followed by the GDP index (0.08), regional services centrality (RSC, 0.08), road buffer (RBUF, 0.07), tourism workforce localisation (TWL, 0.07), and connectivity index (CONN, 0.07).

The Geodetector analysis also identified the following six factors as the top influences in terms of creating the patterns of tourism attractions in Luxor: regional services centrality (RSC), GDP index (GDP), proximity to the nearest urban centre (PUC), population (POP), urbanisation level (URB), and tourism workforce localisation (TWL). These six factors are associated with q-values of 0.20, 0.17, 0.12, 0.11, 0.10, and 0.10, respectively. To ensure comparability, the AHP weights and q-values for each factor were normalised to a common 0-1 scale. The normalised scores were then averaged to generate a balanced ranking that reflects both expert-based priorities (from AHP) and empirical spatial influence (from Geodetector). The final results revealed that the most influential factors affecting the distribution of tourism attractions in Luxor are: regional services centrality (RSC), GDP index (GDP), proximity to the nearest urban centre (PUC), tourism workforce localisation (TWL), urbanisation level (URB), and the clean and healthy environment index (CHEI). The following Table 3 presents the final rankings of the most influential factors based on the combination of both results.

To gain an improved understanding of these most influential factors, MGWR was used. Table 4 presents the outputs from the local regression for each factor (e.g., mean, minimum, maximum, and SD of coefficients), as well as the statistical significance and bandwidths associated with each factor. This table also provides visual evidence for how the effect of each factor is different across different areas of Luxor governorate. Spatial representations of the variation in the coefficient values of each factor can be seen in Figure 7. Areas with stronger positive influence (green) or negative influence (red) are clearly observed, showing the unevenness of the spatial impacts of each factor across Luxor governorate. It is also important to note that only statistically significant effects at p < 0.05 have been emphasised by using a hatched diagonal area representation, while non-significant areas have been represented as completely transparent (and therefore, indicate no significant effect).