The trajectory datasets record a series of position information at different times, so they become the new data sources to study the laws of human mobility. As a main form of social remote sensing data, trajectory datasets also bring a new individual viewpoint to study geographical phenomena. With the emergence of big data, trajectory data mining becomes a hot topic in geographical information science, urban computing and other correlative disciplines. In this paper, we gave a brief review on trajectory data mining and its applications in cities. First, we listed the data sets frequently adopted by human mobility research, gave the classification and their typical applications using FCD data, mobile phone data, smart cards data, check-in data, etc. Then, we summarized its application in solving cities’ problems from four aspects: (1) the identification of urban spatial structure and function unit; (2) the patterns recognition of human activity and the behavior prediction of human movement; (3) the traffic time estimation and the anomaly detection of intelligent transportation; (4) other applications in urban computing such as in urban air and noise pollution, disaster prevention and rescue, even in intelligent tourism and information recommendation. At the end, we pointed out the challenges and further research directions of trajectory data mining.
Floating Car Data (FCD) has been widely applied into traffic supervision, smart travelling, urban planning and so forth. Map-matching is one of the key technologies of FCD, for current map-matching algorithms, it is difficult to improve their map-matching efficiency considerably with a guaranteed accuracy. To solve this problem, our research proposes a map-matching model based on Hidden Markov Model (HMM), and makes a variety of improvements compared with the traditional model: (1) in addition to the position information, it introduces the heading angle variable to emission probability calculation, and discusses its influences on model accuracy and how to set a reasonable weight; (2) it divides road network according to a square grid, constructs candidate road segments searching algorithm based on hash index, and then discusses the optimization approach of the candidate road segment collection; (3) the numbers of segments in the path is used as the measurement for transition probability computation instead of the practical length, which simplifies the calculation procedure; (4) by preprocessing the road net, it constructs a road segment transition matrix according to the characteristic that floating cars have a limited scope of space activities in a given time, which realizes the fast calculation of road segment transition probability and reduces the time complexity of road matching calculation to a significant extent. We have applied this map-matching model in analyzing Beijing taxis’ trajectory data, in which the sampling time interval varies from 1 s to 120 s. The result demonstrates that this model is practicable, the required road segment transition matrix can be constructed in affordable space cost, and its efficiency is improved significantly with the condition that the accuracy meets the application requirements, which makes the model more applicable for massive FCD map-matching. As a conclusion, the proposed model has a high application value for multiple cases.
Floating car data (FCD), which is the trajectories of vehicles, are automatically collected by huge quantities of commercial vehicles which are equipped with GPS devices. Exploring and exploiting such data is essential to understand the dynamic aggregation patterns of trajectory data. However, the existing methods of spatial density clustering mainly focus on undirected data, and it is difficult to effectively find the characteristics of trajectory data. We contribute to the literatures on FCD trajectory data mining by presenting a novel method called directed density clustering method (D-OPTICS), which is formulated based on the spatial density clustering algorithm (OPTICS). In our method, the directed density is computed by a fan-shaped neighborhood region, and the density connectivity is restrained by its direction information. Then, the base clusters are generated using the curve analysis of reachable distance. Finally, the D-OPTICS cluster method is formed by the optimization method of spatial grid and cluster polymerization. This method can be naturally applied to FCD trajectory data mining, and it is also appropriate for handling other directed spatial data. It can be employed to discover the spatio-temporal distribution characteristic of traffic trajectory, and then be adopted to extract the structure information of complex road network. The experiments, with massive floating car data of Fuzhou city, show that the D-OPTICS can cluster directed spatial data effectively, and is useful to uncover the inherent distribution characteristic of the massive trajectory data. Based on its clustering result, the topology information of road network can be extracted. In this work, we extracted the topology graph for the complex road network of Fuzhou city. The experiment results also show that the algorithm can automatically determine the number of clusters, and it is found that the algorithm is not limited to globular cluster data and is capable to deal with clusters of arbitrary shapes. The key contribution of this method is that it takes the direction information into account and it can also be effective in reducing the problems caused by traditional clustering algorithms which may incorrectly merge or decompose thus naturally produce large clusters and noise data. Meanwhile, the result of performance experiments shows that, compared with DBSCAN and OPTICS, the proposed method is more suitable for large-scale data processing.
Trajectory clustering, which aims to uncover the meaningful spatial distributions and temporal variations of moving objects, is of much importance in understanding potential dynamic mechanisms and predicting future development. However, placing many focuses on locational changes, many studies have made limited use of the time dimension in trajectories. This paper presents a density-based clustering method, which integrates time and space information in identifying significant migrating paths from trajectory datasets. Definition of temporal distances between any line segments decomposed from trajectories as well as the criterion of distance threshold selection is provided in detail. The experiments conducted on ocean eddies in the South China Sea demonstrate the effectiveness of this method in obtaining spatiotemporal migrating patterns. The migrating paths in the results are shortened, or separated into parts, or they turn insignificant as the effect of including time component in density clustering, which reveal more specific movement characteristics in the temporal domain covered by spatial clustering. This advantage facilitates the analysis of objects moving along the same path while displaying distinct time patterns.
When the GPS or a network signal is disturbed or obscured, the precision and autonomy of navigation will be greatly reduced. The integrated navigation technology can compensate for this defect. In view of the application requirements of real-time, quickness and accuracy of trail data and road data during the integrated navigation process, this paper puts forward a real-time, high accuracy algorithm which is entitled as an extraction algorithm of track features based on trend set of heading angle variable. The algorithm picks up the calculation and analysis of heading angle in trail data points for navigation, assigns values to the confirmed trend set and the pending trend set innovatively, which effectively improve the real-time aspect and the accuracy of this algorithm in the navigation trajectory data extraction, ensuring the efficiency and precision of the integrated navigation. This algorithm also provides analysis and judgment on trend set status for the invalid features which is caused by external factors (e.g. the overtaking, the emergency avoidance and the drivers’ driving habits). The invalid features will be filtered and eliminated using the buffer of the pending trend set, ensuring the accuracy of the feature extraction. The vehicle trajectory data in Beijing is used for conducting experiments. The results show that the extraction algorithm of track features based on trend set of heading angle variable has obvious advantages in real-time application and extracting effect, and the buffer judging with the pending trend set is able to rule out some invalid features and interferences caused by external factors. The proposed algorithm is not only simple and feasible, but also has high efficiency and strong maneuverability.
As a complement for urban public transportation, taxi plays an increasingly important role in people’s life, which is also taken as one of the most important windows opening to the public. With the rapid development of economy, traffic condition is becoming more and more terrible, which causes the heavy traffic situation in many cities in China. Taxi is a type of transportation resourece that is dynamic and unbalanced in different road networks and at different time, which faces a lot of problems, such as the difficulty in finding a taxi for a passenger or finding a passenger for a taxi driver. This makes taxi transportation to be poorly efficient, and negatively affects the performance of government. It is helpful to learn about the dynamic space in the city, and the patterns of citizens’ working, living and travelling after studying the features and rules of taxis’ pick-up and drop-off distribution. Moreover, it is helpful to learn about the “hotspots” in the city, which represents the areas with huge volumes of taxis’ pick-up and drop-off activities. Based on taxis’s GPS trajecotries big data, this paper puts forward a new model called Line Density Model (LDM) to detect the space time distribution pattern of taxis’pick-up and drop-off activities, in which there are linear trends existing within the taxis’ pick-up and drop-off, and the “hotspots” exhibiting linear trends near the road network in the city. Finally, Wuhan city is taken as the testing area, and the experimental result shows that taxis’ pick-up and drop-off distribution is unbalanced in different areas and at different time, which helps to understand the dynamic and the pattern of the public’s working, living and travelling, and gives a reference to find the “hotspots” at different time in Wuhan city.
It is very difficult to analyze every vehicle trajectory carefully due to the huge number of data. Therefore, it is necessary to divide the space of one city into a collection of smaller areas, among which we can analyze and exploit the vehicle trajectories. Unfortunately, the existing partitioning methods have many disadvantages which may hinder the progress of our study. For example, some traditional partitioning methods based on the Euclidean distance don’t take into account the spatial characteristics of the roads in the city, which may cause a variety of man-made rigid division. Meanwhile, some other partitioning methods ignore the density distribution of taxi’s track points. With further research on trajectory data, the use of traditional space partitioning methods has difficulty meeting the demands of spatio-temporal trajectory data analysis. As a result, we propose a new Voronoi subdivision algorithm on road network which considers the density of taxi’s OD points and the behavior characteristics of taxis. The main body of the algorithm consists following steps. First, the road network should be divided into a series of edges by their intersections. After that, the edges of the road network are subdivided into small linear units. Next, we produce n×n sized regular grids as the space constraints and choose the generating elements in every grid to make them distributing uniformly in space. Then, we can set different speed values for different generating elements and let them spread to the surrounding roads at different speed. Finally, we can get the road network partitioning results consistent with the density distribution of OD points. A series of city sub-regions can be obtained based on the result of network partitioning. Then, we can analyze the track data in these sub-regions with the help of spatio-temporal data visualization methods, such as color sorting, flow graphs, constructing graph structure, etc. At last, we developed an experimental system to generate the network Voronoi diagram, on which we verify the algorithm and analysis methods presented in this paper by testing with the real Beijing taxi trajectory data of one day. Results of these experiments showed that the information about the trajectory data can be obtained intuitively with the help of network Voronoi diagram and the use of various visualization methods for spatio-temporal data.
The Taiwan Strait is a shortcut linking the Far-east with Southeast Asia. After the implementation of direct cross-strait transportation between mainland China and Taiwan on December 12th, 2008, more and more commercial vessels sail across the strait, causing frequent conflicts between involved ships who undertake continuous and expeditious transit passages through the strait. To minimize the possibility of collisions between ships passing the strait, mainland China and Taiwan should work cooperatively to create a ship routing system to separate vessels and navigate the crossing and encountering situations occurred in the strait. Recent advances in ship tracking and telemetry technology help to collect the movement data more efficiently and accurately. The shore-based Automatic Identification System (AIS) network maintained by China Maritime Safety Administration has observed tens of thousands of seagoing commercial ships travelling annually through the Taiwan Strait. It is obvious that the ship tracks have increased tremendously in the strait. These advances would be useful for delineating Principal Fairways (PFs) in the crowded strait-corridor. In this paper, a space-use method found in the habitat evaluation of animal is applied to extract PFs in the crowded strait. Based on ship trajectory observations of the main traffic flows (transit-passage) and direct cross-strait transits, maritime traffic features of the strait are characterized in the form of probability density with the application of statistical inference methods. Bringing the layer of popular direct cross-strait lanes to the iso-surface of main traffic flow's skeleton, all conflict areas are extracted as the Precautionary Areas (PAs) of China's Coastal Ships Routing System Plan in Western Taiwan Strait. For direct cross-strait transportations, by linking the centers of PAs in the Western Taiwan Strait with the official pass points outside the western Taiwan harbors, this paper recommends the applicable direct cross-strait routes to reduce the risk of conflicts in the Taiwan Strait.
When a drifter is influenced by an eddy, its movement will be changed and the drifter may come back to the location where it stayed before. Then there will be loops appeared in the trajectories of the drifter. So we can extract eddies from the trajectories of the drifter by recognizing the loops in the trajectories. Based on this principle, this paper improved the ALIS method proposed by Dong[1] which neglects the complex structure in the results from the extraction of loops, although the complex structure may contain the movement of an eddy. We named this new algorithm AILIS (an improved automated loop identifying scheme). AILIS can further improve the extraction results by judging whether the trajectory segments in loops have self-intersection, and then it can track some parts of the movement of an eddy by judging the similarity between two eddy transient states. This paper made a comparison experiment between ALIS and the algorithm from Li[2] and also made an verification using the results from HD (Hybrid Detection) and HT (Hybrid Tracking) algorithms. The experiment result show that the algorithm proposed by this paper can obtain more transient states and movements of eddies, providing an important approach to obtain the physical parameters of the eddy.
Some high-resolution land cover maps are not free or available for direct use due to its economic value, the impact of weather or its confidentiality. As a downscaling classification method, sub-pixel mapping (SPM) can produce classification data with spatial resolutions finer than the original input data. We aim to explore the consistency between SPM results and classification data extracted from high-resolution remote sensing images on their accuracy and spatial characteristics. Two experiments were performed: one is in Jinnan District, Tianjin City with Landsat-5 TM image, and the other is in Haidian District, Beijing City with HJ image. Results show that the overall absolute accuracies of SPM results produced by TM and HJ images are 84% and 82% respectively. The overall relative accuracies of Landsat-5 and HJ SPM results were 82% and 77% by taking high-resolution classifications as reference. Furthermore, the overall structures and proportions based on the results using the proposed method are similar with high-resolution classifications. Therefore, with the absence of high-resolution land cover map, results generated by SPM could provide an alternative for land cover data source.
Aerosol optical thickness (AOD) can represent the attenuation of solar radiation caused by aerosol, and reflect the atmospheric turbidity or air quality conditions as one of the most important parameters. It is an excellent data source for atmosphere and air quality studies as the atomosphere environment is deteriorating in recent years. Nowadays, there are many kinds of AOD satellite products with good spatial coverage which can be employed to atmospheric environmental researches and the related studies. However, the accuracy and certainty of these satellite products can not meet the increasing demands. Fortunately, the AOD retrieved from AErosol RObotic NETwork (AERONET) exhibits better quality, although their spatial representativeness are localized. With the help of quadratic polynomial interpolation and regression analysis methods, we employed the universal kriging method to integrate both types of AOD products (satellite AOD and AERONET AOD) for analyzing Eastern China in November 2008. Results showed that the quadratic polynomial interpolation method is much better than Angstrom on band interpolation in this study. The comparison between AERONET AOD products and satellite AOD products showed that the fused results (integrated with AOD products) retrieved from universal kriging data fusion method presented a better spatial resolution, wider coverage and higher accuracy. For East China, AOD values in the northern part are higher than the southern part in November 2008; for Yangtze River Delta region, the northeastern of Anhui and the eastern part of Shanxi hold higher AOD values, while the AOD values in southern part of Jiangsu are lower. In addition, this paper has confirmed that the universal kriging data fusion method can provide good fused AOD products in the cases that AERONET AOD data is inadequate and the transit time of different sensors is asynchronous. The AOD data fusion system proposed in this paper can provide better AOD products for relevant scientific researches.
As a research hot-spot of modern geomorphology, landform recognition and classification are important in various study areas such as ecological environment, hydrology and geological structure analysis. Traditional recognition methods, which are inadequate to solve the linear inseparable problem of pattern recognition, exhibit a low accuracy in landform recognition. As a dynamic information processing system, neural network is capable to deal with linear inseparable in landform recognition. Slope spectrum is an effective method to reflect the macro terrain features with quantitative micro-terrain-indicators. It has been receiving widespread attentions in geomorphology. This paper introduces an automatic recognition method based on slope spectrum and neural network. Using DEM data of eight sample areas with different loess landform types in Shaanxi Province, ten small watersheds and their slope spectrums are extracted for each of the eight sample areas. Then, we calculate the slope spectrum indices of these eighty small watersheds and use the indices to construct BP neural network for loess landform automatic recognition. Among the eighty small watersheds, 60% of them are randomly selected as training samples and 40% of them are selected as verification samples. Recognition results show that the accuracy rate is 70% on average for the eight sample areas, and it would be raised to 80% or 85% when the landform types of Loess Hilly-gully or Loess Hill-ridge are eliminated from the eight sample areas respectively. This study indicates that slope spectrum is capable of handling the linear inseparable problem in landform recognition.
Leaf area index is an important parameter for evaluating vegetation ecological conditions and estimating crop yields. Thus, the estimation of LAI has always been a hotspot of quantitative remote sensing research. A growing number of studies have focused on estimating the leaf area index (LAI) of vegetation using several traditional vegetation indices(the Normalized Difference Vegetation index (NDVI), the Ratio Vegetation Index (RVI), and the Enhanced Vegetation Index (EVI)). These vegetation indices were all based on the data of single view zenith angle, which limited the accuracy of LAI estimation. In this article, we compared the sensitivity of the three vegetation indices for crop canopies, and then put forward a new vegetation index named Multi-angle Normalized Difference Vegetation Index (MNDVI) based on CHRIS/PROBA data which includes information with respect to five different view zenith angles. Using the ground crop LAI data obtained in Zhangye city from Gansu Province in June 2008, this paper compared the estimation models of LAI based on the four vegetation indices including the three traditional indices (NDVI, RVI and EVI) and MNDVI. The result shows that: compared with the traditional ones, MNDVI has a much better correlation with LAI, and the correlation coefficient R2 of the LAI calculation model reaches up to 0.716. Besides, in order to verify the accuracy of LAI retrieval model based onMNDVI, this paper calculated the RMSE between the estimated LAI using MNDVI model and the ground-measured LAI, finding that the RMSE was 0.127, which was averagely 33.3% lower comparing with methods using traditional vegetation indices.
Vegetation phenology reflects the response of a terrestrial ecosystem to climate change. It is critical to quantitatively explore the relationships between vegetation dynamics and temperature as well as precipitation. We examined the vegetation-climate relationship using the monthly maximum values of normalized difference vegetation index (NDVI) derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) between 2000 and 2006 in the subtropical region of Fujian. The dataset was also used to calculate phenoloical metrics including the start of season (SOS), the end of season (EOS) and the length of season (LOS) for each year using TIMESAT. The results indicates that the NDVI of each forest types were significantly positive correlated to the monthly mean temperature under no time-lag condition (R2=0.72-0.79, p<0.01). This suggests that the current temperature condition dominated over rainfall amount in affecting vegetation growth. There were significant log-linear relationships between NDVI and rainfall with a 2-month time-lag for each forest types (R2=0.54-0.75, p<0.01), implying that the vegetation growth does not respond immediately to rainfall but to the precedent cumulative rainfall. The phenoloical analysis showed that the SOS began from early to mid April (calendar day 98~103), the EOS appeared in mid-November (calendar day 316~321) and the LOS lasted 213~223 days. The LOS of southern subtropical forests was longer compared to the central subtropical forests; and the LOS of hardwood forests was longer than the conifer forests under similar climate regime. It is possible that the stability of hardwood forests was generally higher and the fluctuations of environmental factors did not limit the growth of hardwood forests compared to the conifer forests. The inter-annual varieties of SOS and LOS were significantly related to spring temperature (February-April), in which the higher spring temperature was related to the earlier SOS (R2=0.83, p<0.01) and consequently the longer LOS (R2=0.80, p<0.01). The trend analysis of LOS for four forest types revealed a significant weak increase between 2.4 days to 3.1 days during the study period.
As an important land cover type of Earth's surface, the surface water has a huge effect on the water cycle, climate change, environment and human activities. Thus, it is significant to analyze the distribution of the surface water and its change trend in time series. This study focuses on the use of the Automated Water Extraction Index (AWEI) to quantitatively evaluate the total surface water area in South America based on Landsat TM imagery and the MODIS digital product acquired in 2011. Meanwhile, the seasonal area fluctuation of the surface water in each typical climatic zone is monitored and the resultant data is used to analyze why the fluctuation occurred. The known data show that the total area of surface water in South America, mainly concentrated in the Amazon River basin, the Parana River basin and the Patagonia Plateau, is 305,000 km2 and the water rate is 1.69%. Based on our study, the number of lakes in South America is 9,579 while the total area is 142,000 km2, which takes 46.42% of the total area of surface water; the area of waterways is 157,000 km2, which takes 51.56% of the total area of surface water; and the reservoirs and ponds, which takes 2.01% of the total area surface water, is 6144.8 km2. From the perspective of climate zones, the area of surface water in the tropical zone of South America was influenced obviously by the change of rainy seasons, and within which the tropical desert climatic zone was the most affected one; in the temperate zone, the area of surface water didn’t change a lot because it has distinct seasons; in the plateau mountain climate zone, the area didn’t change a lot due to its unique climate feature; in both subtropical monsoon wet climate zone and Mediterranean climate zone, the surface water hardly changed with seasons but it exhibited obvious variations in winter and summer seasons. From the perspective of countries in South America, Brazil has the largest area of surface water (147,000 km2) which takes 48.17% of the total area of surface water in South America and its water rate was 1.72%; Argentina has the second largest area (34,000 km2) which takes 11.24% of the total surface water area and its water rate was 1.23%; while, Venezuela has the largest water rate (3.08%). The surface water in each typical climatic zone of South America has seasonal area fluctuation on several levels, which may relate to climate change and human activities. Further studies are needed to explain this issue in future.
Although atmospheric environment monitoring based UAV is a cutting-edge technology, it has become an important method for environmental protection departments. It is significant for protecting and emergency monitoring of sudden atmospheric pollution incidents. Qilu chemical industrial zone with high-risk region of contaminative chemical gas emission is took as the study area. In this paper, we combined the rotor UAV system (HY-1) that equipped with high resolution CCD and the pollution gas monitor with in-situ measurements for three pollution gases of NO, CO and SO2. The results of the three pollution gas concentration distributions at the heights of 150 m/200 m, 250 m, 350 m and the ground level showed that the pollution gases in the chemical industrial zone present a horizontal difference and vertical diffusion characteristics. Analyzing the horizontally and vertically abnormal values of pollution gas concentrations from UAV can be efficient in supporting pollution pre-investigation. This research is a practical application of UAV in atmospheric environment monitoring by environmental protection departments, which can provide a technical support for regular supervision of pollution gas emission in chemical industrial zones and related enterprises.
