Modelling the effect of rain on free flow speed and capacity under urban uninterrupted traffic flow conditions

In this study, change on traffic flow conditions under the influence of rain is analyzed at two main arterials in İstanbul, which are O2, O3, O4 Motorways and D100 State Highway. By clarifying all possible effects of rain on traffic flow, it is aimed to improve the success of traffic management implementations (such as variable speed management) in order to reduce delays. Study area is selected in the urban regions, where appropriate traffic data could be obtained, capacity flow is observed and congestion is experimented. The road sections have three or four lanes on the motorways while three or two lanes on D100 State Highway. Generally, the road sections that are located at east and west boundaries of the city, could not be used in the study since they have less demand compared to urban areas. Traffic flow data are obtained via RTMSs (Remote Traffic Microwave Sensors), which are operated by Istanbul Metropolitan Municipality and weather information are gathered via meteorology stations by Disaster Coordination Center of Istanbul. These data are evaluated jointly; meteorology data from meteorology stations that are the closest to each RTMS are utilized. Totally seven different meteorology stations' data are matched with 25 RTMSs. Possible change in traffic flow is analyzed both considering rain precipitation and making comparisons between sunny and rainy days. Change in capacity and free flow speed (FFS) are modelled under the rainfall effect. In this research, 4-month traffic and meteorology data between 1st of November 2015 and 29th of February 2016 are used. First of all, the rainy weekdays that have an influence on traffic flow are determined from this observation period. Furthermore, the sunny days that are as close as possible to the rainy days are selected. By this way, the dataset of rainy and sunny days during 4-month period are finalized. By establishing the dataset with rainy days and sunny days close to rainy days, possible seasonal differences on traffic demand that may cause misevaluation of results are tried to be avoided. Traffic flow and meteorology data are collected at 2-min and 1-min intervals respectively and they are aggregated over 15-min time intervals. Additionally, heavy vehicles are converted to passenger car units (pcu) with 1.5 equivalent as suggested by Highway Capacity Manual (TRB, 2000) and pcu is used for all model estimations. Data with 15-min time intervals are applied similar to previous studies, which allows the change in traffic flow to be seen clearly. Although RTMSs are able to collect volume, speed and occupancy data in 2-min intervals, their accuracies are unknown. In order to test the RTMS data accuracy, cumulative vehicle counts recorded by RTMSs are used. Each RTMS is considered as the border of sections along motorway. Since there is no entry or exit between two RTMSs, total vehicle counts of each RTMS must be almost equal to each other. RTMS records are regarded as correct in the analyses, if and only if the total vehicle counts difference is less than 10%. In contrast to motorway, there is no way to create such sections by using RTMSs along D100 State Highway since several entrance and exit lanes make impossible to apply the same procedure. Alternatively, the compatibility of speed-flow rate diagrams, the existence of missing or incorrect data and the capacity flows are investigated between RTMSs along D100 State Highway. Eventually, it is decided to use 11 RTMSs on the motorways and 14 RTMSs on D100 State Highway to achieve the purpose of the study. According to Highway Capacity Manual (TRB, 2000), FFS reduction is offered as 10 km/h in light rain or snow, 20 km/h in heavy rain and 50 km/h in heavy snow in comparison with clear and dry days for 120 km/h FFS. These proposed reductions are the average of several studies and do not represent exact traffic flow behavior. Consequently, in order to clarify the change in the traffic flow under rain in Istanbul, the most crucial arterials (O2, O3, O3 Motorway and D100 State Highway) are used to estimate the speed-flow rate models. In this study, a novel model is used by considering occupancy, volume and speed information which are collected by RTMSs. Speed-flow rate relationship is analyzed in two different regions which are free flow region and congested region (CR). There are no statistically significant differences between sunny and rainy days' speeds in CR. Therefore, model estimations for CR are done for rainy and sunny days jointly. In order to distinguish free flow conditions and CR, re-scaled cumulative vehicle counts and speed time series are used. The main purpose is to find the point where the slope of the speed-flow curve changed in the cumulative vehicle counts during speed increase or decrease. In the latter case, slope change point is considered as the beginning of congestion, while in the former case slope change point is considered as the end of congestion. In this way, by considering each day's traffic flow data free flow region and CR are distinguished precisely. FFR model estimations are conducted by dividing it into two sub-regions. Under free flow conditions, it is known that, until some flow rate, speeds remain constant and then they start to decrease with increasing flow rates. This boundary, which is dividing free flow region into two regions, is named as breaking flow rate. Before this boundary, the speed-flow rate measurements are named as free flow region 1 (FFR1) and after this boundary they are named as free flow region 2 (FFR2). Model estimations are conducted separately for these two regions. In order to determine the breaking flow rate, a method is used in which the flow rates are sorted from smallest to largest and then correlation coefficients are calculated by adding a continuous pair of data starting from the first pair. In this way, breaking flow rate is determined, and hence FFR1 and FFR2 regions are distinguished. Model estimation in FFR1 is used to find FFS of a road, while FFR2 model is used to find capacity. This modelling approach, which is applied for both sunny and rainy days, allows the traffic flow modeling according to weather conditions. Significant variables reflect traffic flow like FFS on the basis of the breaking flow rate and capacity is calculated with the help of models. In order to determine whether to use data from multiple traffic sensors in the same model or not, ANOVA and Kruskal-Wallis tests according to the occupancy measured in free flow conditions at both road sections are performed. Traffic measurements from multiple sensors are considered to represent the same observation space. However, the groups could not be created using all of the 11 traffic sensors used on the motorways and 14 traffic sensors on D100 State Highway. For this reason, the traffic flow change under the influence of rain for each traffic sensor is investigated separately. Different models are estimated for 25 traffic sensors and evaluations are conducted according to road characteristics. In the analyses, the rainfall effect is investigated in two different classes. Light rainfall is considered in which rain is measured up to 0.6 mm/m2 for 15 minutes. If rainfall is more than this value, it is included in the models as medium-heavy rainfall. Road surface conditions are considered as wet for 1-time interval (15-min) after the rain has stopped. As a result, the effect of rain on traffic flow is investigated in three groups as wet road surface, light, and medium-heavy rainfall. The measurements of these three groups are evaluated jointly as rainy days and compared with sunny day model. Model results are compared primarily to understand the difference between sunny and rainy days. The effects of all precipitation (wet road surface, light and medium-heavy rainfall) are revealed and compared on the basis of sunny days. It is found from the model results that the motorway capacity decreased by 7.88% and FFS by 8.61% on rainy days as on average. At D100 State Highway, capacity decreased by 9.07% and the FFS by 8.16% during rainfall. It is understood from the results that, more decrease in capacity appears at D100 Highway and FFS at motorways. The rainfall quantity and wet road surface model results are compared with the sunny days and the change in the traffic flow is examined in details. On motorways, the capacity decreased by 5.03% for wet road surface, 8.44% for light rainfall and 13.19% for medium-heavy rainfall. These reduction rates are found as 6.45%, 9.62% and 13.40% for D100 State Highway, respectively. These results show more capacity reduction on D100 State Highway. At D100 State Highway, FFS on the wet road surface decreased by 6.00%, 8.23% for light rainfall and 11.07% for moderate-heavy rainfall compared to sunny days. These decreases are 6.37%, 8.32% and 13.01% respectively for motorways, and more decreases are calculated at motorways among all groups. Breaking flow rate reduction is calculated as 20.0% for D100 State Highway while 13.2% for motorways in comparison to sunny days. It is also found that, as the road gradient increase, for both road types, FFS is decreased. No significant relationship is found between percent reduction in FFS and gradients for both road types. For the motorways, with the increasing gradient, both capacity and capacity reduction is increased while gradient has no effect at D100 State Highway. In this study, speed models, which are the function of flow rate, are developed for different road sections. Under the adverse environmental conditions caused by rain, it is seen that the change in traffic flow is revealed, the decreases according to the amount of rainfall are modeled and an easily applicable model is proposed for different road types.