Artikel

Beranda Artikel Crowdsourcing Road Defects Monitoring And Road Infrastructure Performance: A Case Stude of Jalan Kita
Beranda Artikel Crowdsourcing Road Defects Monitoring And Road Infrastructure Performance: A Case Stude of Jalan Kita

Crowdsourcing Road Defects Monitoring And Road Infrastructure Performance: A Case Stude of Jalan Kita

  •  25 Sept 2019
  • Article/Article
  • 510 viewed
Crowdsourcing Road Defects Monitoring And Road Infrastructure Performance: A Case Stude of Jalan Kita
Foto: Crowdsourcing Road Defects Monitoring And Road Infrastructure Performance: A Case Stude of Jalan Kita

The idea that a lot of insightful information can be gathered by individual citizens themselves has developed significantly in recent decades, notably with the rise of Wikipedia. In the fields of public policymaking and development, crowdsourcing data sources have been put forward by researchers as an innovative way of getting more granular data or even completely new information about a specific issue (Estelles-Arolas et al., 2012; Noveck, 2009). 
An app called ‘JalanKita’, initially developed by Indonesian Road & Bridge Research Agency in 2014 and officially connected to the database of National Directorate General of Highways in 2017. Similar to other road reporting apps, it aims to encourage citizens to report road defects in order to allow the road authorities for identifying and fixing the defects in timely fashion.

Abstract

This research aims to assess the comparability and the extent of prioritization from the crowdsourced road defects reports in Indonesia with the improvement of the road performance itself. A visual quantification method called Pavement Condition Index will be applied to solve the comparability issue of the crowdsourced reports, followed by a comparison and multiple regression analysis model to analyze its utilization in the road authorities’ maintenance decision making. The research reveals that the majority of reported road defects are requiring heavier maintenance and over-reliance on the existing roughness survey may generate inadequate maintenance options. However, the standardized coefficient produced by the regression model elicits the fact that maintenance decision on the reported sections is still strongly based on the roughness condition (0.847) and the crowdsourced reports (0.020) were not taken as the main consideration.

While officially, many road authorities in the world use the road roughness or International Roughness Index (IRI) to measure the road’s longitudinal unevenness or smoothness (Bergal, 2018; Schmidthuber et al., 2017). Roughness are evaluated and used for managing the road network systems, mostly due to its quantitative nature, which based on profiler measurement; it can be surveyed considerably fast and spatial based, which suitable for national scale survey; it heavily influenced by other parameter, such as crack, rut, and weather events (NCHRP, 2004); lastly it also has high correlation with road user cost, speed, driving comfort, and road safety (Huang, 2004; King, 2014).

However, despite the new step-up in accountability and transparency of public services performance in crowdsourcing the road defects, there are still many cases of road managers, even in Western countries and unfortunately also in Indonesia, that ignore the reports for months and years, regardless of many publications and reports from TV and newspapers (Guy, 2017; Kenney, 2017; Herliansyah, 2018; Shiddiq and Eka, 2018).

Thus, how is the crowdsourced report’s prioritization? Since, the road authorities in Indonesia has been oriented towards roughness, traffic load/volume, and political intervention to formulate road maintenance plans and budget (MPWH, 2018). To answer this question, this research will, first, clarify the comparability between crowdsourced road defects reports and the road roughness data. Then, tries to apply a suitable methodology for assessing the crowdsourced reports, so it can become complementary information for road asset management system. And lastly, evaluates the extent of crowdsourced data utilization in Indonesia for road maintenance decision making.

Literature Review

The improvement of road condition lead to rapid benefits to the road users, namely improvement of social facility access, comfort, speed, safety, and cheaper vehicle operation cost (Burningham & Stankevich, 2005). To preserved these benefits, a well-planned maintenance program must be followed. To this date, the maintenance method is mainly chosen based on pavement age, road condition, and available funds. The most prominent one, road condition, was measured with parameter like surface distress, roughness, and deflection, are used to develop maintenance strategies. Indonesia, as other countries, commonly extrapolate these data into the future to establish the most suitable maintenance plan on annual basis (NCHRP, 1981, MPWH, 2018).

Different stage of maintenance interventions is able to treat single or multiple distress types. Generally, the type of maintenance treatment sorted from the most expensive to the cheapest are Reconstruction, Rehabilitation, Corrective, Preventive, and Routine Maintenance (Qiao et al., 2013). These interventions may be combined and applied multiple times to produce more desirable effects if necessary (Qiao, 2015). However, the budget constraint often compels the road authorities to establish maintenance priorities. The priority for preservation budget shall be offered to the most functionally important road sections and in unacceptable level (Burningham & Stankevich, 2005). This level, usually referred 

as trigger value, are the maximum acceptable distress indices that suggest the uppermost tolerable levels of distress for pavements.

Crowdsourcing may become an appropriate and valuable tool for road preservation plan development. The reasons mainly because the crowd themselves are largely more attentive in the road condition that involve the region where they live, work, and socialize (Erickson, 2010). The quality of these reports is nearly never a problem, since the reporters volunteer out of desire to help the road authorities and their fellow road users (Misra et al., 2014). However, the subjectivity and lack of consistency to the user contribution can be a problem. The reports resulting from crowdsourcing app like JalanKita are mainly showing the visual of the road defects that present.

Thus, a ratings process is needed to visually inspect the distress manifestation on those images of deteriorated pavement surface. Generally, this distress data will be combined with roughness data and/or other variables when used for asset management or reporting purposes (Wu et al., 2010).  They provide the information for pavement performance analysis, and is vital to forecast pavement performance, anticipate maintenance, predict rehabilitation needs, establish maintenance priorities, and allocate funding (Timm and McQueen, 2004).

Table 1. Descriptive scale of Present Serviceability Rating (Nakamura & Michael, 1962)

Surface distress is commonly assessed and summarized using pavement condition indices. The very first indices used by various 

transportation agencies was a simple descriptive combination of ride quality and distress rating, termed as Present Serviceability Rating (PSR) (Nakamura & Michael, 1962). The resulted index was determined based on the experience of observers panels that ride a vehicle in a particular pavement section (Attoh-Okine and Adarkwa, 2013). The mean of each individual rater then established from descriptive scale from 0 to 5, as listed in Table 1.

In 2011, Indonesia has developed its own alternative metrics for road distress, called Road Condition Index (RCI). Similar to PSR, the RCI is a combination of visual description of road surface condition and its ride comforts, as shown in Table 2. The weakness of this two indexes is the same, their liability to bias error. When the severities and extents from distresses are not well defined, it may lead to a confusion on the part of the rater (Attoh-Okine & Adarkwa, 2013). However, later in 2016, the Indonesian Ministry of Public Works had released a more objective visual survey rating manual that adopted from American Society for Testing and Materials (ASTM) with the same name: Pavement Condition Index (PCI) Survey. 

Unlike PSR and RCI, the PCI is rated based on multiple deduct value curves that quantify the type, severity, and extent of road. It provides a measurement of distress observed on the pavement surfaces, which also indicates the structural integrity and its surface functional condition (e.g. localized roughness and safety) (ASTM, 2016). The PCI is not directly measure the structural capacity, skid resistance or roughness, but it becomes an objective tool for road assessment (Hajj et al., 2011).

Research Methodology

The way this research is methodized starts with the notion of how the road network performance was collected. With the rise of crowdsourced road defect reports, the Indonesian road authority was enriched with an innovative method to collect pavement condition data. However, this public driven data has their share of differences with the main road data collection process, which is roughness survey.

 The time and spatial resolution of crowdsourced reports and roughness survey differ greatly due to the core characteristic of each method. The crowdsourced report with its flexibilities can point out the road defect images in every given time frame, but it all depends on the interest of the road users. While the roughness survey can only be done twice a year due to strict government budget, but it can produce a continuous data for the whole road section in the national scale.

The differences between each data collection method emerge the issue of comparability between the two. To clarify this, an initial data processing need to be addressed to the crowdsourced reports, which include the data extraction from the JalanKita website database, spatial association of each reported images with the road section based on their coordinates, and the aggregation process to further analyzed the surface distress data from each reports. Next, a novel core methodology called Pavement Condition Index (PCI) is applied to quantify the visual surface distress shown on the crowdsourced reports.

The time and spatial resolution of crowdsourced reports and roughness survey differ greatly due to the core characteristic of each method. The 

crowdsourced report with its flexibilities can point out the road defect images in every given time frame, but it all depends on the interest of the road users. While the roughness survey can only be done twice a year due to strict government budget, but it can produce a continuous data for the whole road section in the national scale.

The differences between each data collection method emerge the issue of comparability between the two. To clarify this, an initial data processing need to be addressed to the crowdsourced reports, which include the data extraction from the JalanKita website database, spatial association of each reported images with the road section based on their coordinates, and the aggregation process to further analyzed the surface distress data from each reports. Next, a novel core methodology called Pavement Condition Index (PCI) is applied to quantify the visual surface distress shown on the crowdsourced reports.

The now quantified and comparable reports data then assessed together with other variables (e.g. roughness, budget, political intervention) to understand the prioritization of each method in the Indonesian Road Authority’s maintenance decision making. The assessment mainly target the recurrent road defects reports that happened in multiple road sections. Specifically, it consists of qualitative approach through semi-structured interviews and statistical method of Multiple Correlation analysis. Through those stages of prioritization assessment, the overall analysis is concluded to propose suitable recommendations to better guide the road defects report crowdsourcing positioning in the road preservation planning cycle.

The majority of this research is based on quantitative methods, while the qualitative one is utilized to clarify and to strengthen the reasoning of the results. The quantitative portion involves the quantitative descriptive analysis to depict the general characteristic of the crowdsourced reports, the Slovin’s formula to define the required sample for comparison and its following statistical analysis, the cross tabulation analysis to directly compare the Crowdsourced Reports (PCI) and Roughness (IRI), and lastly the multiple correlation analysis to understand the crowdsource report’s prioritization.

While the qualitative methodology comprises of content analysis to break down the comparability issue of crowdsource report and roughness data, and also semi-structured interviews conducted in the fieldwork to investigate both causal-relationship of recurrent road defects phenomena and the extent of JalanKita’s data utilization in the current maintenance decision making.

Crowdsourced Data Extraction, Association, and Aggregation

As this research is revolved around the crowdsourced reports and roughness performance, the secondary datasets become the core sources of the unravelment. The crowdsourced report database was obtained through the JalanKita’s authorized administrator, which led by the Indonesian Road and Bridge Research Development Center. While the roughness and the budget allocation details were acquired from the Indonesian Directorate General of Highways. The reliability and the validity of each database were assured by its respective publisher, and their detailed datasets are listed in Table 3.The original JalanKita datasets is in the form of spreadsheet. There are a total of 6404 reports that spanned from April 2017 until November 2018. Afterwards, the spreadsheet is validated to remove the duplication of the data, whether it is a report images duplication or simple data duplication errors. After the initial validation, the number of the reports is significantly decreased 

to 5547 reports. However, the validated datasets are still containing the reports of every road jurisdiction in Indonesia, while the scope of this research is to assess the utilization of road condition data in the National road networks. Thus, a further data processing is needed to set aside the extraneous reports from the report associated with the National road sections.

To extract the crowdsourced reports that associated with the National roads only, a GIS-Spatial-Join processing are conducted. The crowdsource spreadsheet datasets are firstly transformed into a shapefile database by simply correlating each reports with its coordinate. Then, the association process is handled by overlaying the crowdsourced report nodes with other secondary datasets, the National Roughness Condition shapefile, as shown in Figure 3. 

Figure 3. Spatial join process in GIS to associate crowdsourced reports with national road sections

Rating Surface Distress from Visual Defect Reports

The condition index measured by PCI is calculated based on 3 main parameters, which are Distress Type, Distress Density, and Distress Severity. Firstly, the Distress Type are manually determined by visually examine the pavement surface, or in the case of crowdsourced reports, the pavement images sent by the public. To ensure the uniformity, the PCI guidebook describe an extensive explanation and visual examples of each distress types. For both rigid and flexible pavement, there are a total of 38 distress type that regulated in the PCI manual published by ASTM and Indonesian Ministry of Public Works.

Not only description of each distress variation, the manual also provides comprehensive information to measure the second parameter, the level of Distress Severity. This information is presented in the form of image examples and supplementary detailed dimension tables (as shown in Figure 4). Generally, there are 3 stages of distress severity for each distress type, namely Low Severity, Medium Severity, and High Severity. 

Figure 4. Visual examples that provided by PCI manual to determine Distress Severity level (KPUPR, 2016; ASTM, 2016)

The third parameter, Distress Density, which defined as a percentage of the Surface Distress Area divided by the Total Area of the Sample Unit. The calculation of Surface Distress Area in the crowdsourced report’s images can be quite intricate. In this research, it is measured by estimating the road defects dimension which relative to the dimensions of the nearest standardized object that present, e.g. cars, kerbs, or road markings. While, the Total Sample Area is determined based on the number of sample unit and the minimum sample unit to be surveyed. The standard sample unit is equal to 3.6 m x 50 m, as visualized on Figure 5. 

Figure 5. Example on dividing a road link into sample units (KPUPR, 2016; ASTM, 2016)

In this research, the main analysis is tried to compare between Roughness Survey datasets with the Crowdsourced Images themselves. The smallest roughness datasets have an average dimension of 100 m x 3.6 m or equals to 2 sample units. Thus, based on the PCI manual, the minimum sample unit to be surveyed is 1. In other words, 1 crowdsourced report image is enough to represent a sample unit’s PCI to be compared with 1 roughness dataset (per 100 m).

Lastly, to determine the distress deduction value, which describe as the extent of deterioration caused by the surface distress towards the pavement service life. The deduction value is derived from the plotted combination of the 3 parameters mentioned-above into the Deduct Value Curves. This deduction value then used to subtract the final PCI number resulting to the visual based pavement condition, as demonstrated in Figure 6. 

Figure 6. PCI calculation example

Results and Discussion

In this research, an attempt has been made to compare the crowdsourced road defect reports based on its roughness condition and its PCI value. Based on their value, then each reports are identified of their respective maintenance and urgency stage. As summarized on Figure 9, the now quantified crowdsourced samples are grouped based on their PCI category and cross-compared with their surveyed roughness (IRI) categories. The most alarming situation that can be observed in the chart is that the majority of the crowdsourced reports’ PCI are categorized as requiring structural and reconstruction/recycle maintenance. Despite surveyed as Good or Fair roughness, more than 50% of the road defects that reported by the public are identified as needing heavier maintenance than their roughness condition indicates. These findings surely can ignite road managers’ perspective on how they should view and decided the maintenance options for the reported roads.

The comparison shown in Figure 9. demonstrates that by associating the crowdsourced reports with an index (PCI), the defect photos will be comparable and can complement the roughness data. The road authority will be able to analyze crowdsourced reports to decide the most suitable maintenance plan and the budgets required. The PCI method provides the objective basis for assessing those reports, so that the maintenance decision does not need to wait for the roughness survey that conducted every six months. In doing so, the road managers may execute more optimum and timely reactive maintenance to satisfy the general public. 

Figure 9. Comparison summary to assess the urgency of the crowdsourced reports for maintenance

Reactive Maintenance Quality

To clarify the reactive maintenance issue, this research also analyzes the recurrent defects that reported in JalanKita. There are three samples of many similar cases that had been evaluated. First is the case in Durenan-Pligi. There was a road defect report in November 2017 at sta-26.8 until sta-26.9 (100m section), as shown in Figure 10. In 2017, the annual budget for the whole section (30.40km long) was Rp 1.17 billion or 53.72% of East Java average budget. This budget was increased significantly in 2018 with Rp 6.46 billion or 174.99% of East Java average budget. Based on the surveyed roughness, the road condition was known as slightly improved from 3.53 in July 2017 to 3.26 in January 2018, which categorized as Good and only need Routine Maintenance.

However, there was a second defect report in February 2018 in the same 100m section. When visually observed and quantified with PCI, it was shown that near the patching area there was an alligator crack. As compared from the 1st and the 2nd defect photos, the crack had widely spread, worsened, and even some potholes were formed. Based on PCI value, this road section condition has deteriorated to Fair condition and required Structural Maintenance. Later, this situation was reflected on the newest Roughness survey on January 2019 (IRI 2018-2). The 100m section of Durenan – Pligi had declined to Fair roughness condition, with an IRI of 5.59. 

Figure 10. Location, Roughness, and PCI value of recurring road defects sample #1

The second sample case is located in Kediri City Border – Tulungagung Municipality Border road section, specifically from sta-2.6 until sta-2.7 (100m section). As shown in Figure 11, there were crowdsourced road defects reports in the same spot in September 2017 and February 2018. As a side note, the annual budget for the whole section (14.69 km) in 2017 was Rp 1.1 billion or 50.49% of East Java average budget. In 2018, the budget was multiplied to Rp 4.2 billion or 115.82% of East Java average budget.

Figure 11. Location, Roughness, and PCI value of recurring road defects sample #2

Regardless, after being maintained with patching and overlay, the surface smoothness can only last for a short period. As recorded in the July 2017 roughness survey, it has 6.71 IRI and deteriorated to 9.26 IRI in January 2018 survey. When analyzed with PCI method, it was observed that the overlaid layer had corrugated and become rippled/wavy. Some broken patching also exposed the alligator cracks underneath the upper pavement layer and several cracks had deteriorated into small potholes. The PCI value has worsened from 66 (Fair condition) to 52 (Poor condition), which required Reconstruction/Recycling maintenance. This condition was mirrored again in the newest roughness survey on January 2019 (IRI2018-2), as the 100m section’s roughness keep deteriorating and recorded as 9.83 IRI (Poor roughness). 

 

Figure 12. Location, Roughness, and PCI value of recurring road defects sample #3

Lastly, the case of Tanjung Bumi Seaport – Bangkalan/Sampang Municipality Border road section, specifically at sta-14.7 until sta-14.8 (100m section). As seen in Figure 12, in 2017 the upper layer was slided, while in 2018 it was shown that a heavy alligator cracking was present and causing a shallow but rather wide pothole on the same spot. Important to be noted, the annual budget for the whole section (37.00 km) in 2017 was Rp 432 million or 19.81% of East Java average budget. While in 2018, the budget is slightly decreased to Rp 695 million or 18.84% of East Java average budget.

Crowdsourced Report’s Prioritization

This research has conducted a multiple regression analysis to assess the JalanKita’s reports prioritization in the Indonesian road authority’s maintenance decision making. Specifically, it attempts to understand the relationship between variables that strongly affecting maintenance to improve road condition, which administratively represented by IRI changes value. Those affecting variables are: initial roughness value of the reported road sections (Initial IRI), the crowdsourced report PCI condition (PCI), and the annual budget on the reported sections (Rupiah). The statistical results for the analysis are shown on Table 4. The table shows the multiple linear regression model summary and overall fit statistics. The adjusted R² of the model is .712 with the R² = .715. This means that the linear regression explains 71.5% of the variance in the data. Also, the Durbin-Watson “d” is equal to 1.648, which between the two critical values of 1.5 < d < 2.5. Therefore, it is assumed that there is no first order linear auto-correlation in the multiple linear regression data.

Table 4. Multiple regression model summary

The next output is the F-test of ANOVA. The linear regression’s F-test has the null hypothesis of that the model explains zero variance in the dependent variable or R² equal to 0. Also, the F-test is proven to be highly significant, with F = 260.381 > Ftable [3,312,0.05] = 2.62. Thus, it can be assumed that the model explains a significant amount of the variance in IRI changes.

The third table shows the multiple linear regression estimates, including the intercept and the significance levels. In the Enter-Method of multiple linear regression analysis, it is found that the Initial IRI has the highest impact when compared with PCI and Annual Budget. It comparison is based on the standardized coefficients of beta=-.847 versus beta = -.020 and -.010 respectively. This can be interpreted as: for every 1-unit increase in Initial IRI value, there will be -.847 decrease/improvement in IRI Changes. The coefficient for Initial IRI (-.847) is significantly different from 0, because its p-value (Sig.) is 0.000, which is smaller than 0.05. While, the coefficient for PCI (-.020) and Annual Budget (-.010) is not significantly different from 0, because their p-value are .527 and .740 respectively, which is larger than 0.05.  The information also checks the multi-collinearity in the multiple linear regression model. The tolerance value is bigger than 0.1 (or VIF < 10) for all variables, which means that no multi-collinearity was shown.

The final results table exhibits that the linear relationship between the PCI and the Annual Budget is proven to significant, because the Sig. value (-3.584E-5) is less than .05. The linear relationship is happened to be pretty weak, since the Pearson Correlation 'r' value (-.037) is close to 0. Also, their linear relationship is negative, because the Pearson Correlation 'r' value is negative. Similarly, the association between the PCI and the Initial IRI is significant because the Sig. value (.000) is less than .05, but the linear relationship is not as weak because the Pearson Correlation 'r' value (.247) is almost quarter to 1, and the linear relationship is positive because the Pearson Correlation 'r' value is positive. Lastly, the association between the Initial IRI and Annual Budget is also significant with its Sig. value (.000) that less than .05, furthermore their linear relationship is also weaker than the aforementioned coefficients because the Pearson Correlation ‘r’ value is only -.243, and the linear relationship is negative because the Pearson Correlation ‘r’ value is negative.  On the side notes, the analysis also checks for normality of residuals with a normal P-P plot. The Figure V.14 shows that the points generally follow the normal (diagonal) line with no strong deviations. This indicates that the residuals are normally distributed.

Conclusion

In this research, it is revealed that there are some complemental characteristics between the crowdsourced road defect reports and the administrative roughness in depicting the road surface condition. On the one hand, roughness survey provides an objective, standardize and a historical measurement of the road network condition. Therefore, it is highly suitable for the annual road maintenance planning and budgeting that conducted by the road authority. While on the other hand, the crowdsourced road defect reports provide the most precise location and the most up-to-date conditions of the pavement surfaces. Thus, it is able to fill in the informational gaps that present in the biannual roughness survey, which is crucial for timely proactive maintenance in preventing greater pavement deterioration.

This research also demonstrates the utilization of Pavement Condition Index method to quantify the road condition through visual data provided by the crowdsourced reporting. PCI is able to measure the observed distresses on the reported images to generate objective and rational basis for maintenance needs and priorities. In addition, this research had assessed the urgency of crowdsourced reports for maintenance by comparing reports sample with their correlated roughness data. The correlated roughness of the reported location was chosen based on the report’s submission time and its distance with the nearest road section. However, not all of the crowdsourced report category is comparable with the roughness characteristics. The most likely reports to be relevant are the ones with Damaged Road category, which also forms the majority of the incoming reports.

Based on the comparison, the majority of the crowdsourced reports appear to be requiring heavy maintenance, such as structural maintenance and pavement reconstruction. Unfortunately, this circumstances also present in the road sections that surveyed as Good or Fair roughness. To further clarify the situation, this research also had assessed several recurring road defects spots to show the significance of the crowdsourced reports. It seeks to display not only the numerical correlation between PCI and IRI value, but also the physical interaction between the crowdsourced reports with the road authority’s observation that reflected in roughness survey months later. The analysis had emerged the notion that over-reliance on roughness survey may generate inadequate maintenance option, which reflected on recurrent road defects and increasing deterioration rate of pavement surface in later years.

Therefore, by associating each reports with PCI value and section-location (nearest distance to roughness survey section), the crowdsource reports can mitigate unnoticed deterioration from the roughness survey. The PCI method can provide more objective basis for crowdsourced reports’ prioritization, so that the maintenance decision does not need to wait for the biannual roughness survey (every 6 month). The crowdsourced reports also become comparable to be processed in the road network analysis to complement the roughness data for maintenance planning and budgeting that conducted by the road authority.

However, this research unveiled that the crowdsourced reports were still not taken as the main consideration for maintenance decision making. A multiple correlation analysis was conducted to understand the relationship between variables that strongly affecting maintenance to improve road condition. Based on the standardized coefficient of the regression model, every 1-unit increase in initial roughness value will result in -0.847 of decrease or improvement in road condition. While, the crowdsourced report and annual budget only affecting in -0.020 and -0.010 of road improvement respectively. In other words, the maintenance decision that occurred on the reported sections still strongly based on the roughness condition. The crowdsourced report was positioned as the second consideration, whilst the annual budget happened to be the least concern of the road authority.

Recommendation

From all of the points described in the results and discussions, it is evident that the most fundamental step to going forward is to develop a legal framework for the crowdsourced road defects reporting. It is expected, since in JalanKita case, the notion of “innovation before regulation” indeed thrived and gained enough momentum. However, the nature of Indonesian public services is still oriented towards governance by rules. Every tier of related stakeholders is longed for proper legal backing to support their action. This is an important step to ensure that everyone is on the same page and willing to engage in collective action to make the policy happen. The road authority can translate this vision into a comprehensive set of regulations, ranging from enforcement, financing, digital infrastructure, digital knowledge, and incentives for related actors.