Spatial Analysis to Determine Black Spot Area in Kulon Progo Regency, Yogyakarta, Indonesia

Nowadays the number of traffic accidents increase due to the surging volume of road traffic. The efforts to reduce number of accidents is important, especially to identify the location and causes the accidents. Therefore, this study aims to identify black spot areas, which are areas with a high concentration of traffic accidents, in Kulon Progo Regency in Yogyakarta, Indonesia. The method used is spatial analysis by combining analysis of traffic accident data and geographic information system (GIS) data using ArcGIS software. The study focused on provincial road consist of primary collector 2 and 3, as well as national road segments. The analysis of accident data allowed the identification of hotspots where accidents occur most frequently. Hotspot analysis methods such as Local Statistic Analysis and Kernel Density Estimation were used to assess the spatial patterns and clustering of accidents. The study found that the black spot area in Kulon Progo Regency was mainly concentrated on the National Road Segment, with fewer occurrences on the Provincial Road with function of Primary Collector 2 and minimal presence on the Primary Collector 3 Road Segment. This result can be used to develop strategies to mitigate accidents and improve road safety in the area.


Introduction
The escalating frequency of traffic accidents corresponds to the surging volume of human journeys undertaken.Conversely, human travel experiences a marked upswing due to burgeoning economic expansion and the proliferation of vehicle ownership [1].According to data sourced from the World Health Organization (WHO), Indonesia witnessed 17 fatalities per 100,000 inhabitants during the course of 2019 [2].Additionally, the Yogyakarta Police Traffic Unit has reported a surge of 2,480 traffic accident cases between 2021 and 2022 [3].The effort to reduce the number of accidents is necessary due to the increasing number of mobility using road, hence it is important to know where and why the accidents occur [4].Based on the previous research, it has already known that traffic accidents happen randomly and affected by several factors such as traffic volume, weather condition, and road geometric design [5].Black spot area is described as the areas where traffic accidents concentrated [6].Identification of black spot area could be conducted using Geographical Information System (GIS) as already done in Penang, Kuala Lumpur, Pakistan, and Beijing [7][8][9][10][11].Effective spatial analysis methods are employed to identify black spot areas in Kulon Progo Regency.By combining geographic information systems (GIS) and accident data, researchers can pinpoint locations with a high frequency of accidents.This allows authorities to prioritize resources and implement targeted interventions to reduce the occurrence of accidents in these areas.
Apart from length, other parameters in hotspot identification include the crash number threshold and period.The followings are treshold parameters used in different countries such as Belgium has ±3 crashes over 3 years [13]; Denmark has 4 crashes over 5 years [20]; Germany has 5 crashes over 3 years or 3 fatal crashes over 5 years [15]; Hungary has ±3 crashes over 3 years [14]; and Norway has ±4 crashes over 5 years [14].These parameters are important and useful in identifying and analyzing crash hotspots in different countries [12].
This study aims to identify black spot areas on provincial and national road segments in Kulon Progo Regency by combining traffic accidents and the geographic information system (GIS) data in spatial analysis.A comprehensive exploration of spatial analysis methods employed in identifying black spot areas in study area.The process consists of identification, examine the factors that influence the occurrence of these areas, and discuss the significant impact that research findings have on enhancing road safety measures.The results of this study are expected to develop effective strategies and interventions to mitigate accidents and enhance road safety.

Methods
This study was carried out through several stages as shown on Figure 1.First stage is data collection on accident and road network, followed by data categorization, data analysis, and the creation of a hotspot map.It should be noted that this study primarily deals with secondary data.The accident data was acquired from the Police Traffic Unit of Yogyakarta, spanning the period from 2016 to 2018, encompassing attributes such as coordinates, severity classification, and the count of casualties.On the other hand, the road data was sourced from the Indonesia Geospatial Portal in the form of raster data.[21].

Location
The main location of the study was on Kulon Progo Regency, which is located in Yogyakarta, Indonesia.Kulon Progo Regency is situated in the westernmost part of the Special Region of Yogyakarta Province.It shares borders with Central Java Province to the north and west, while the Indonesian Ocean is located to the south.Geographically, Kulon Progo Regency is mostly flat, although it is surrounded by mountains, particularly in the northern region.The coordinates of the regency range between 7°38'42" and 7°59'3" South latitude, and 110°1'37" and 110°16'26" East longitude [22].The map of Kulon Progo Regency could be seen on Figure 2.

Accidents data
According to the data obtained from the Police Traffic Unit of Yogyakarta, between 2016 and 2018, there were a total of 1,478 recorded accident cases on the road segments in the Kulon Progo Regency as shown on Figure 3. Out of these cases, 112 resulted in fatalities [3].The accident data includes various attributes such as the accident coordinates, severity level of the accidents, and the number of victims.In this study, the severity index of the victims is categorized into three levels using a risk-weighted score approach: death, heavy injury, and light injury.[24].
Table 1.exhibits the attribute table that was employed for projecting the accident points onto the map, in conjunction with the road data.Conversely, Figure 4. presents a comprehensive summary of the total number of avictims, categorized according to the severity level.

Road data
This study has specific focus on segments of provincial road consist of primary collector 2 and 3, as well as national road segments.The classification of road types was established according to the Governor decree of the Special Region of Yogyakarta No. 117/KEP/2016, which outlines the designation of collector roads 2 and 3 within the primary road network.[25].Based on Law No. 2/2022 on 2 nd revision of Law no.38/2022, provincial road is devided into three functions, those are primary collector 2 and 3, and also strategical road [26].Figure 5. illustrates the intricate road network of Kulon Progo Regency, wherein the segments highlighted in a light blue hue represent the collector 2 roads, the segments in a dark blue shade denote the collector 3 roads, and the segments in an orange hue signify the national roads.

Accidents poin map
The data pertaining to accidents and the road network were merged to create a comprehensive point map of accidents.This merging process was carried out using ArcGIS Pro 3.1.2,utilizing the WGS 1984 UTM Zone 49S coordinate system for projection.Figure 6a. to 6c. presents the accidents data spanning from 2016 to 2018, projected onto the primary focal road segment examined in this study which is in segments of provincial road consist of primary collector 2 and 3, as well as national road segments, Kulonprogo Regency, Yogyakarta.Subsequently, all the accidents points from the three-year dataset were consolidated into a single layer as displayed in Figure 6d.Based on the projection of accidents data in Figure 6d., it is evident that a significant number of accidents have taken place in the focal road segment.This projection provides valuable insights into the frequency of accidents in this particular area, enabling us to identify hotspots where accidents occur most frequently.This information is crucial in formulating effective strategies to reduce the number of accidents in the identified areas.

Accidents cluster analysis
Clustering is the methodology employed to categorize vast quantities of data into distinct subsets, thereby rendering the data more comprehensible, utilizing a distance function.[27,28].Within ArcGIS Pro 3.1.2,there exist two geoprocessing tools that can be utilized to conduct cluster analysis.The manifestation of local spatial patterns may or may not coincide with the global manifestation.While it is conceivable for the local pattern to be congruent with the global indicator, there exist situations where the local pattern may be perceived as an anomaly that the global indicator would fail to detect.Furthermore, there may be instances where a handful of local patterns diverge from the overarching spatial trend [29].In the domain of spatial analysis, it becomes crucial to juxtapose the results of localized spatial analysis with those of global spatial analysis.

Cluster and outlier analysis: Anselin Local Moran's I
Cluster and outlier analysis is geoprocessing tools that could be used to identify statistically hot spots, cold spot, and spatial outlier from the given set of weighted features [30].Hot spot aids in locating areas with statistically significant high values relative to the predicted average by examining the spatial distribution of data points.In contrast, cold spot focuses on finding regions where a phenomenon does not concentrate or cluster.The Getis-Ord Gi* statistic, which determines z-scores for each site depending on its proximity to adjacent locations, is one often used measurement.Hot spots are those with positive Z-scores, and cold spots are those with negative Z-scores.The technique of Local Moran I proves invaluable in examining developmental assessments, particularly in investigating the dynamics of core-periphery relationships and identifying the interconnectivity among distinct clusters of settlements encompassing individuals from neighboring micro-regions or counties.This analytical approach illuminates the intricate network of provider-receiver connections within these cohesive groups [31].
As evident from Figure 7a. to 7d., the findings of the localized spatial analysis reveal a clustering pattern among the accident points.The cluster analysis outcomes are represented by a spectrum of colors ranging from blue to red, indicating the level of clustering for each accident point.Figure 7a. to 7d. comprises four images that depict variations in severity-weighted analysis for conducting the localized spatial analysis.Directing our attention to the red circle, which signifies areas with a high frequency of accidents, we can preliminarily conclude that the red points may represent black spots in terms of severity weighting.Notably, the results of the localized spatial analysis indicate a higher concentration of clustering accident points in the death-weighted and slight-weighted analyses compared to the heavyweighted analysis.
The findings of this comparative analysis endeavors to ascertain whether the localized hotspots align with the broader spatial trends observed in the global analysis.Through this comparison, we can evaluate whether the identified hotspots are integral components of larger clusters or if they represent local anomalies.Consistency in clustering patterns across different spatial scales furnishes compelling evidence of the presence and importance of these hotspots.It aids in substantiating the reliability and robustness of the localized spatial analysis results [29,32].Based on the localized spatial analysis conducted using severity-weighted analysis, the highest z-score for clustering accident points was obtained.Table 2. provides information that the most clustered data is derived from the death-weighted local spatial analysis, while the most randomized data is associated with the heavy-weighted analysis.The local analysis yielded a bandwidth (radius) value of approximately 1.675, which can be rounded to 1.7.For a more comprehensive depiction of the results, please refer to Figure 8.

Spatial autocorrelation: Global Moran's I
The Spatial Autocorrelation (Global Moran's I) tool is designed to evaluate the existence of spatial autocorrelation by considering both the spatial distribution of features and their associated attribute values.This tool proves particularly beneficial in ascertaining whether the observed pattern of features exhibits clustering, dispersion, or randomness present in the data [29].Moran's I serves as a metric that spans from -1 to 1, wherein positive values indicate spatial clustering (the similarity of feature values among neighbouring features), negative values indicate spatial dispersion (the dissimilarity of feature values among neighbouring features), and values approximating zero imply a random pattern.By considering both the spatial locations and attribute values of features, it bestows valuable insights into the nature of spatial autocorrelation [33].
The outcomes of Global Moran's I analysis conducted on the accident data are presented in Figure 9.
The results are shown separately for different severity levels of accidents.Figure 9a.represents the analysis results for slight injury accidents, Figure 9b.for heavy injury accidents, and the final figure depicts the analysis for fatal accidents.From the analysis, it can be concluded that fatal injury accidents exhibit a clustering pattern.This is indicated by the z-score value of 1.856747, which is the highest among all severity levels.The presence of a high z-score suggests that the occurrence of fatal accidents is not a random event, but rather forms a distinct spatial cluster.
In contrast, the analysis of accidents with severity levels other than fatal reveals that their occurrences are random.This conclusion is supported by the sequence of z-score values presented in Table 3., where the z-score for fatal accidents stands out as the only significant clustering result.These findings highlight the importance of distinguishing between different severity levels of accidents when analyzing their spatial patterns.The clustering of fatal accidents suggests the presence of specific factors or conditions that contribute to their higher occurrence in certain areas.Understanding these patterns can aid in the development of targeted interventions and measures to reduce the number of fatal accidents and improve overall road safety.
When performing a global Moran's analysis on a dataset, one statistical metric used to assess spatial autocorrelation is called the Z score.It determines whether similar values are significantly distributed or clustered in the geographic area under investigation.The Z score is determined by comparing the observed and expected values of the Moran's I statistic under the assumption of a random geographical distribution.Positive Z scores show positive spatial autocorrelation, which means that similar values are more likely to be close together.A negative Z score, on the other hand, denotes a dispersed pattern and shows negative spatial autocorrelation.The presence of spatial autocorrelation in the data is often indicated by a Z score larger than 1.96 (at a significance level of 0.05) [29,32,33].

Distance band (radius)
Both cluster analyses yield a z-score value that can be utilized as an indicator for determining the severity level of accidents, which can then be weighted for blackspot analysis in the subsequent stages of analysis.Figure 10.provides a comparison of the z-score values, highlighting the consistency of death injuries forming a cluster.Consequently, we can determine that the severity level of fatal accidents should be employed as the weighted-severity level in the blackspot analysis.
Figure 10.Z-score value of cluster analysis.
In hotspot analysis, the determination of the distance band is based on the average value of the highest z-score obtained from the analysis [34].In this specific analysis, the average z-score value is calculated to be 1,766.To determine the distance band, this average z-score value is rounded down to the nearest whole number.In this case, the rounded down value of 1.766 is approximately equal to 1.5.This value represents the distance band in which the clustering of accidents is considered significant.These distance units are typically measured in kilometres.The selection of the appropriate distance band is crucial in hotspot analysis as it helps define the spatial scale at which clustering patterns are examined.By considering the average z-score value, the distance band ensures that the analysis captures meaningful spatial relationships and identifies significant clusters of accidents [29].

Accidents hotspot analysis
Hotspot analysis is a widely used spatial analysis tool available in the ArcGIS Pro application.It offers two primary hot spot analysis tools known for their accuracy and effectiveness.The Local Statistic Analysis: Getis-Ord Gi* tool is used to measure the degree of clustering or dispersion of features within a dataset.This tool provides insights into the presence and significance of spatial clustering patterns, allowing for a deeper understanding of the data.The Kernel Density Estimation tool, on the other hand, is focused on estimating the intensity or concentration of a phenomenon within a dataset.This tool helps identify hotspots or cold spots by visualizing the intensity of the phenomenon across the study area.It provides valuable information about the spatial distribution and density of the features, aiding in the identification and analysis of hotspots within the dataset [29,32,35].Both the Local Statistic Analysis: Gedis-Ord Gi* and Kernel Density Estimation tools are valuable for hotspot analysis as they offer different perspectives on the spatial patterns and clustering of features.

Local statistical analysis: Getis-Ord Gi*
The results of the Local Statistic Analysis: Getis-Ord Gi* using accident data weighted by death severity are depicted in Figure 11.The visualization in the Figure 11 showcases a gradient of colours ranging from blue to red, which serves to represent the spectrum of hot spots to cold spots.This colour scheme effectively illustrates the intensity and distribution of the analysed data, enabling the identification of areas characterized by high or low levels of clustering or dispersion.Based on the analysis conducted using the Local Statistic Analysis: Getis-Ord Gi* (Figure 11.), it is evident that the occurrence of black spot areas is predominantly concentrated in the national road segment.On the other hand, there is a limited number of black spot areas observed in the provincial collector 3 road segment.When comparing the results of the Local Statistic Analysis: Getis-Ord Gi* and Kernel Density Estimation analysis, as depicted in Figure 13., it becomes apparent that the black spot areas identified by both methods exhibit consistency when overlaid.This consistency implies that both analysis methods effectively identify and highlight the same areas as black spots.When multiple analysis techniques produce consistent results, it enhances confidence in the identified areas as reliable black spots.
After analysing the data, it can be deduced that the primary area of focus in Kulon Progo Regency is the National Road Segment.This segment shows a higher number of occurrences compared to other road segments.On the other hand, the provincial Collector 2 Road Segment has relatively fewer instances of black spot areas, while the provincial Collector 3 Road Segment exhibits the smallest extent of such areas.The consistent identification and highlighting of black spot areas by both the Local Statistic Analysis: Getis-Ord Gi* and Kernel Density Estimation analysis methods provide a robust foundation for subsequent actions.These areas can be prioritized for targeted interventions, such as implementing safety measures, improving infrastructure, or increasing enforcement, to address the underlying issues contributing to accidents or incidents.This convergence in findings between different analysis methods demonstrates the reliability and validity of the identified black spot areas, enabling stakeholders to make informed decisions and allocate resources effectively to improve safety and mitigate risks in these specific locations.

Conclusions
The study's analysis has yielded noteworthy findings that significantly contribute to our understanding of accident severity and distribution in Kulon Progo Regency.The application of the Anselin Local Moran's I and Spatial Autocorrelation: Global Moran's I in the Cluster and Outlier Analysis has been instrumental in identifying the severity level of accidents.The z-score values obtained from these methods serve as a reliable indicator for blackspot analysis, providing a quantitative measure of the gravity of accidents in specific areas.
Moreover, the study has determined the distance band value for hot spot analysis by averaging the zscore values obtained from both cluster analyses.The value, rounded down to approximately 1.5 from 1,766, represents a significant distance band within which accident clustering is considered significant.This finding is crucial in identifying and predicting accident-prone areas based on spatial proximity.Our analysis has further revealed that accident concentration in Kulon Progo Regency is primarily found on the National Road Segment.This road segment has been identified as a significant black spot area, indicating a high severity of accidents.In contrast, the Primary Collector 2 Road Segment and the Primary Collector 3 Road Segment have fewer and minimal occurrences, respectively.
In conclusion, these findings provide valuable insights into the spatial distribution and severity of accidents in Kulon Progo Regency.The data underscores the significance of the National Road Segment as a major black spot area, shedding light on the areas that require immediate and targeted intervention to enhance road safety.

Figure 4 .
Figure 4. Number of victims based on severity level.

Figure 5 .
Figure 5. Provincial and national road segments of Kulon Progo Regency.

Figure 13 .
Figure 13.Overlay the result of Local Statistic Analysis: Getis-Ord Gi* and Kernel Density Estimation.

Table 2 .
Z-score value of clustered accidents data from local spatial analysis using severityweighted.