Archive for December, 2017

Sakar and al. Animal Movement.

Monday, December 4th, 2017

This paper focuses on movement data from barnacle goose migration patterns. By using migration hotspots with periodicity and directionality from these hotspots they are able to establish the movement patterns of these geese. I’m not sure if geese have a need for privacy but if they do, attaching GPS trackers to them would definitely be an invasion of privacy. This study also highlights the relationship between location information and information about an individual/animal quite well. Location information is a critical piece of information about the identity of what’s moving.

I thought this paper was very well written; I am left with very few questions bout the effectiveness of the methods. Notably, the paper accounts for the effect of weather and ecological stresses from certain hotspots that affect the movement of the geese. Considering that these stresses have already affected movement, it would be interesting to see how the migration patterns will be affected on a longer timeframe. Will urbanization, climate change, atmospheric conditions continue to alter these patterns over time?

Written in 2015, this paper is well up to date in terms of the algorithms, clustering methods and GPS devices used. It would be interesting to compare and contrast the migration of other geese (perhaps the Canada goose) using the same tracking methods. Perhaps there would be fundamental law’s of goose travel that would become more apparent.

Analyzing Animal Movement Characteristics From Location Data (Sarkar et al., 2016)

Sunday, December 3rd, 2017

This paper adopts Periodica algorithm by enhancing the hotspot detection, which accomplished through substituting Kernel Density Estimate (KDE) with Getis-Ord Gi*. The new method is applied in periodical behavior discovery in animal movement. It is a very typical case in spatial data mining for integrating spatial autocorrelation with traditional statistical analysis. As the author mentioned, KDE is criticized because it assume data points are independent and sensitive to the shape of data points. Getis-Ord Gi* perfectly avoids these drawbacks; however, Getis-Ord Gi* is grid-based and still suffering from Modifiable Areal Unit Problem (MAUP). It means the model will have different results when applied in different spatial scales, which represents the granularity here. Considering the scale problems is a significant different between spatial data mining and traditional datamining. Therefore, it is important to explicitly discuss the size of grid in practices, and it is also to necessary to think about time scales. Periodical behaviors may simultaneously happen in many different-length and interlaced periods (e.g., seasonal behaviors and daily behaviors can happen together). Sometimes there is only one kinds of periods we need to consider (e.g., animal immigration), but sometimes we may need to have cross-scale analysis (e.g., human’s periodic behaviors), which will make the situation far more complicated.

Beyond the Periodica algorithm, I believe there could be a better way to discover the periodic behaviors. First, I think no matter what hotspot detection methods used, it never gets rid of arbitrarily determining the hotspots. More instinctively, it means how big a staying region is to represent a periodic behavior is happening. Second, since the points in trajectories are not independent, why we separate them from trajectories to conduct analysis? Can we directly analyze the trajectories even the basis are still points? Do we only care about the periodical behaviors within certain locations and how the directions they traveled? Is there any interesting periodic behaviors during the travelling (e.g., certain routes they always travel)? Simplifying information to binary data also means losing information for further discovering. I’m not arguing we always have to know all the answers of those questions, but when it is necessary, we should have better methods or tools to do the job. Third, in my perspectives, periodical behaviors mean always having some event in a certain time. There is nothing about space. I think most of literature have their clear definitions about periodical behaviors but seem not natural. Mathematically, it is good to make “hard” definitions for analysis, but we still need more discussion about this assumption (e.g., using locations to situate periodical behaviors). Therefore, I argue we should have better solution to substitute the Periodica algorithms if necessary, and I suggest it can start from the concept of clearly separating space from other information in spatial data mining.

Thoughts on Toch et al

Sunday, December 3rd, 2017

This article gave me a much-needed injection of nuance into my views on privacy data. I found the concept of high and low entropy locations quite interesting, and how levels of comfort were correlated to these notions. By allowing the participants how to control their privacy settings based on time and location was quite very exciting. There is a narrative of location-tracking devices and apps imposing an authoritarian top-down imposition of privacy guidelines which is quite worrisome, so it comes at a pleasant surprise to see privacy guidelines being manipulated by the user rather than the creator. This potentially has a ‘win-win’ effect of reaping the benefits of location-tracking while phasing out moments in which one feels uncomfortable.
In a vacuum I like this quite a bit, but it sounds that this kind of narrative could potentially normalize the idea of location-based sharing. That is the direction in which we’re heading as each generation becomes more comfortable with this potential lack of anonymity. Additionally, by reaping in statistical data for different areas, we would be able to seemingly create models that determine which areas have high and low entropy. Wouldn’t we risk falling to certain biases based on the participants themselves? The positionality of those collecting the data must be considered, and even then, it would be impossible to account for everyone’s preference on a statistical basis. Statistical tools are generalizing by nature, so is it safe or ethical to infer too much data on what the data that it produces?
On a more positive note, I find the concept of low and high entropy in the context of location-sharing a potentially interesting tool of analyzing space. While in some space people would want to share their location publicly versus private spaces. What urban implications may this have? Could make for an interesting study.

Empirical Models of Privacy in Location Sharing (Toch et al, 2010)

Sunday, December 3rd, 2017

In this paper, the authors propose a model for privacy location sharing, and investigate the relationships between the sharing behaviors and location characteristics and tracking methods. During the modeling process, it is meaningful to introduce entropy and apply it in later statistical analysis. However, I think that the other settings may lead to bias when conclude the results. The most influential factors may be the investigating system. How to communicate with participants and what they know about this project can lead to different location sharing behaviors. In my perspectives, the best datasets for analyzing privacy in location sharing are produced by users in daily life. The empirical environment is not natural for users and likely to change their behaviors. For example, we may doubt whether participants are more willing to share locations because they get pay from this project. We also don’t know whether they share locations on purpose in this project, which is not the natural states. Therefore, limitations can include that data come from particular applications and devices provided by researchers.

 

It is tricky to measure how comfortable people are willing to share their locations. Hence, it is necessary to ensure the natural inputs from participants (i.e., lessen the systematic errors). There are two possible ways we can improve privacy analysis in location sharing. In one hand, we should collect datasets from people’s daily life, which ensure the randomness of data. In the other hand, we can have more comprehensive data, participants, and context to simulate the natural environment rather than rely on a simple model.

On Sarkar et al. (2014) and movement data

Sunday, December 3rd, 2017

I thought Sarkar et al.’s “Analyzing Animal Movement Characteristics From Location Data” (2014) was super interesting, as I don’t have a very strong background in environment and I didn’t know about all of the statistical methods involved in understanding migratory patterns via GPS tracking. The visualizations were super interesting, like the Rose diagram to show directionality and the Periodica method to then determine hotspots. I also appreciated the macroscopic viewpoint of this article; though the inclusion of equations is important for replication and critical understanding, it is also important to discuss the outputs and limitations of the equations at a larger level, in order to better understand results. It is especially useful for those without deep math backgrounds, like myself, to understand the intentions of using these certain equations without having the math background of being able to visualize output.

As interesting as it was to learn about the incredible utility of understanding migratory patterns of animals, I couldn’t help but think about applications to human geography. I wonder if these patterns are already being used to extrapolate information on someone based on their location, as the Toch et al. (2010) paper on privacy for this week hinted at. If they are different, it would be interesting to evaluate the utility of the methods to study human spatial data patterns as applied to animal migratory patterns. Since Sarkar et al. used unsupervised classification to learn new patterns, it seems like the combination of methods used could apply (and probably do apply) to human spatial data mining. As terrifying as that is.

On Toch et al. (2010) and location-sharing

Sunday, December 3rd, 2017

This article was super interesting, as I didn’t know too much about the actual mechanics behind location sharing (ie. “Creating systems that enable users to control their privacy
in location sharing is challenging” (p. 129)). Their ideas of identifying privacy preferences based on the locations that people go to was confusing (and was not really ameliorated by the end). Perhaps it’s because I don’t understand Loccacino, particularly because of technology constraints from 7 years ago (did they or could they collect data then? All the time, or just when you wanted to share your location like “At the mall”?), underlined by the wonderful image of the “smartphone” (p. 131). Like some of my classmates noted, this seemed very similar to Find My Friends, and perhaps that’s why I didn’t understand how this worked, what the line was between actively volunteering and passively volunteering location.
Further, I had some issues with the participant pool that they used. The researchers relied on a set that was 22/28 male and 25/28 student and then were surprised that “the study revealed distinct differences between the participants, even though the population was homogenous”. As evident from spatiotemporal GIS & feminist GIS, women interact with spaces differently than men. Further, age of participants, as another classmate noted, is crucial: a 50-year-old staff member or student will go different places than a 22-year-old student. Not to mention, analysis of age could determine why there was a big difference in sharing (or if there was not a difference). Also, I was interested in seeing the differences between people between mediums, as some people used phones and some used laptops, and phones are way easier to pull out and share info on than laptops, especially in social gatherings or public spaces. They acknowledged this difference as being 9 mobile & 5 laptop users being “highly visible” (p. 135), but I would be more interested in seeing the differences between the two mediums first and seeing activity levels as a whole for the two mediums, rather than continuing to equate the two, especially since laptops and phones were not distributed equally among participants. I think this study would be interesting to redo today, but with more information about participants and more controls throughout the study (or at least, fixing for differences among participants and modes of participation).

Duckham and Kulik (2006) – Location privacy

Saturday, December 2nd, 2017

In this chapter, Duckham and Kulik outline and compare four approaches to location privacy protection: regulation, privacy policies, anonymity, and obfuscation. The growing presence of locationally-aware devices (and applications) have increased both the richness of personal location data being gathered, but also the range of actors with access to it.

While measures can be taken to limit the ease of subsequent use, an individual’s location data gains a significant amount of meaning when it is considered in a wider context. The extent to which anonymised data can be used to infer identity by relating trajectory information with other locations and events raises a pressing concern in location privacy.

I would guess that a significant proportion my own sets of preferences and personalities could be guessed from my life’s trajectory data alone, given sufficient context (which might include the trajectory data of other people, or easily obtained information about places and events). For instance, my own musical taste could probably quite easily be deduced directly from the concerts and festivals I’ve attended over my life, from the concerts and festivals attended by others whose trajectory intersects my own, and indirectly from other inferred personality traits.

As outlined in the reading, the meaning that can be derived from aggregated data may be greatly understated existing privacy policy and regulatory standards. Addressing this issue and limiting growing opportunities for privacy breaches will require case studies that further illustrate the predictive power of location-based data.
-slumley

Laube and Purves (2011) – Temporal granularity and cow trajectory data

Saturday, December 2nd, 2017

In this article, Laube and Purves explore the influence of temporal scale on the analysis of ecological movement data. In particular, they vary temporal granularity and record the effects on various movement parameters (speed, direction, periodicity) for a high temporal resolution cow trajectory dataset.

Their results show that assessment of these parameters was inconsistent between scales, meaning that the determined speed of the cows depended largely on how finely resolved the timesteps taken were. Thus in general, researchers should be explicit about the temporal scale and range of their analysis; failing to do so could lead to unclear or irreproducible results.

I think in some sense Laube and Purves boil the problem down into two. First, a ‘coastline paradox’ style problem where varying the temporal unit of measurement (e.g. of speed) leads to different speed estimates, though presumably at some point the trajectory lengths tend towards a finite limit in a more meaningful sense than they do for lengths of coastline. Secondly, at smaller temporal scales, uncertainty in GPS measurements become significant. Navigating and accounting for these issues present a challenge and opportunity for GIScientists.
-slumley

Location Privacy and Location-Aware Computing, Duckham and Kulik (2006)

Saturday, December 2nd, 2017

Duckham and Kulik (2006) introduce the importance of privacy in location-aware computing, and present emergent themes in the proposed solutions to related concerns. In their section contextualizing privacy research, the authors present privacy and transparency as opposing virtues (p. 3). I’m curious about the distinction that would motivate the valuation of one over the other. For instance, many would feel uncomfortable with the details of their personal finances being public (myself included), but would advocate for the openness of business or government finances, or even those of the super-rich. Is power the distinguishing characteristic? Perhaps concerns for person wellbeing or intrusive inferences are less applicable to large organizations, but how do we explain the public response to the Panama or Paradise Papers?

Duckham and Kulik (2006) also posit that greater familiarity and ubiquity of cheap, reliable location-aware technologies will increase public concern for privacy (p. 4). I’m not so convinced–in fact, is it not the opposite? It would seem that during their inception concern for privacy was much higher than it is now. I would argue the pervasiveness of location-aware technologies has generated a reasonable level of comfort with the idea that personal information is always being collected. I would imagine this is evident in the differential use of location-aware technologies in people that have grown up with them.

I appreciated the authors’ discussion of location privacy protection strategies. They provided interesting critique of regulatory, privacy, anonymity, and obfuscation approaches. I would add to the critique of regulatory or policy frameworks based on “consent” that participation in such technologies is becoming less and less optional. Even when participation is completely optional, consent is often ill-informed. It’s clear that the question of privacy in location-aware computing is one with no clear answer.

Movement GIS and Cows (Laube and Purves, 2011)

Saturday, December 2nd, 2017

What a cross-sub-disciplinary article! In their efforts to address temporal scale issues with regard to movement data, Laube and Purves (2011) nicely tie together many of the GIScience topics that we have covered in this course.

This article reaffirms my view of movement GIS as a subdomain of temporal GIS. Broadly, temporal GIS looks to integrate time into spatial analyses, and movement is an excellent example of where this needs to be done. Moreover, movement is directly connected to many of the theories and implementation strategies behind time geography. Laube and Purves’ research on cows’ movement patterns nicely reflects the interconnectedness between space and time, which is a foundational idea behind time geography. While I understand that the scope of this paper might not have allowed for it, I would have liked to see more references to temporal GIS and time geography literature.

It is interesting to think about the tension between the conceptual simplicity of movement data and the practical complexity of extracting meaningful knowledge from such data. As in this case, movement can often be represented by a series of points, each with an ID, a timestamp, and a set of coordinates. Each point represents a static state of an entity and movement can be inferred by combining points to form a trajectory. This reflects a significant source of uncertainty, as the Euclidean distance between two points may not be an accurate depiction of the movement that took place (which was nicely pointed out in this article). Furthermore, analyzing movement patterns to learn about meaningful behaviours is challenging as many behaviours (like cows grazing) take place at one space over time. This paper’s focus on “granularity grief” also nicely reminds us that behaviours are dependent on temporal scale.

With the popularity of data sources from things like location-based services and social media, we are flooded with movement data at incredibly high spatial and temporal granularities. I think that the field of movement analysis will be an incredibly important direction for future research in GISciences.

Animal Movement Characteristics from Location Data, Sarkar et al. (2014)

Saturday, December 2nd, 2017

Sarkar et al. (2014) present an analytical framework for making inferences about animal movement patterns from locational information. The article was an insightful show-don’t-tell introduction to how movement research could be applied beyond the domain of GIScience. Also, I think this may be one of the first articles we’ve looked at with an explicitly ecological application of GIScience research… A welcome addition!

It’s becoming increasingly evident how these GIScience topics we’ve discussed in class interact to provide a better understanding of how geospatial information is analyzed and represented. I appreciated the authors’ discussion of uncertainty in the Li et al. (2010) algorithm for detecting periodicity. I found myself tempted again to assume that increasing temporal resolution is the best way to minimize this sort of uncertainty. Even withholding concerns for feasibility, ultimately I’m not convinced that this really does more than mask the problem. The detection of periodicity through cluster analysis resembles aggregation techniques for reducing the influence of outliers on uncertainty in the resulting periods, but I am still a little unclear on how the temporality of the location data was incorporated into the clusters. Does the Fourier analysis account for points near in space but distance in time? Perhaps the assumption of linearity is enough in the assessment of migration patterns.

The distinction between directionality and periodicity as components of movement was insightful. Typically I would think about the significance of movement as it relates to the physical space, but Sarkar et al. demonstrate how inferences from orientation and temporality of movement can be insightful on their own.

Thoughts on Laube et al (2011)

Friday, December 1st, 2017

I have never before considered the complexity of measuring movement. A significant portion of the authors’ work seemed to be dealing with the issue of positional accuracy of a GPS transmitter. Since the data being collected is so simple in form (tuples containing id, x, y, timestamp), all subsequent calculations are determined by the changes in the x and y-coordinates, assuming data is collected at precise intervals. The techniques used in this study, in particular the explicit definition of segmentation values and their influence on trajectories, are only relevant when the data consists of these three dimensions (euclidean position and time).

Technology has advanced to the point where temporal GPS data can be supplemented with local measurements. Accelerometers record local movement, and could be used to simplify data manipulation in the lab. The “rules” defining movement trajectories and stationary segments can be incorporated into a GPS receiver/accelerometer device. Rather than “pruning” subtrajectories in short time intervals or small distances, location data might only be recorded when local movement has been measured to be over a threshold distance from the last recorded location. Accelerometers might also be able to help “smooth” trajectories by vectorizing movement at one recorded location to lessen uncertainty of the next recorded location.

Using a dynamic temporal scale, where location data is supplemented by acceleration vectors, might make most of this research irrelevant.

Privacy in Location Sharing (Toch et al. 2010)

Friday, December 1st, 2017

Prior to reading this article I felt quite comfortable in the privacy literature in preparation of my talk on VGI, however Toch et al.’s (2010) gave many new insights in the field of locational privacy.

The first of these (and the most marking to me) was the concept of ‘locational entropy’, being the measure of diversity of visitors in a given location. I found this metric quite scary as it becomes quite clear to infer attributes of a location (i.e. private residence, business building, university campus) simply based on the location and time stamp of citizens visiting them. I found this parallel’s the ‘How Fast is a Cow’ reading prescribed for Rudy’s movement GIS talk, as it infers behaviour/attributes from limited data though in the traditional GIS method of layering location, time, and geographic context.

Furthermore, locational entropy is then used to infer social networks, by linking social media (Facebook) to the accounts, and seeing areas in common users may visit. I have heard that apps like Tinder already use algorithms like this to pair users who ‘go out’ more with like people, as well as users who may visit certain similar areas frequently and meet a similar social profile (inferred by the software of course).

In this paper I would have liked to see the possibility of differences in volunteered location between different age groups as well as cultures, as I can imagine that although Toch et al.’s findings on high entropy areas being considered less private may work in the chosen study area, it could be further complicated and user dependent in different circumstances. Furthermore, I feel a newer paper could be made, as although this paper only came out 7 years ago, so much has changed in subtleties such as the option to disclose locational information during certain hours of the day largely disappearing from most apps, and privacy sections moving from overt web page options to the subtleties of user terms of service and cell phone plans in more and more coercive forms of VGI.

-MercatorGator

Thoughts on “How fast is a cow? Cross-Scale Analysis of Movement Data”

Friday, December 1st, 2017

It was interesting to see a direct link to Scale (the presentation I gave this week) in the very first paragraph of this paper. It just reaffirmed how scale is a central tenet of many different sub-fields of GIS, from uncertainty to VGI to movement data in this particular case.

The authors enumerate the many different factors which influence the collection of movement data, from sampling method to measurement of distance (euclidian vs. network) and the nature of the space being traversed. One of the concerns they highlight is “sinful simulation” and this reminds me of our discussions of abstraction pertaining to algorithms, agent based modelling, and spatial data mining. For all these methods, the information lost in order to model behaviour or trends is always a concern and I wonder what steps are taken to address the loss of spatial or other crucial dimensions for movement data.

Another common theme discussed by the authors is the issue of relativity and absoluteness. In their decision to focus on temporal scale, they reiterate that as with slope, “there is no true speed at a given timestamp” (403) because this is dependent on the speed at adjacent points, and is relative. But they say that the speed is dependent on the scale at which it it measured and this confused me because whether they measured it in cm/s or inch/minute, is it the unit which they are using to speak about granularity? Because if so, then regardless of the unit of measurement the speed should be the same. I wonder what they mean by there is no absolute speed at a given timestamp if they are referring to it in terms of a scale issue and not a relative measurement/sampling issue.

The authors contend that the nascency of the field of movement data analysis means that researchers rarely question the choice of a particular temporal scale or parameter definition, and this is definitely an important issue as we have seen with the illustration of the MAUP and gerrymandering. The fact that all these subfields are subsumed within the umbrella of GIS, and that researchers tend to have some “horizontal” knowledge about how methods have been developed and critiqued in other fields, hopefully means that they can adopt the same critical attitude and lessons learned from the past towards this new domain of research.

-futureSpock

How fast is a cow? (Laube 2011): Moo-vement GIS

Friday, December 1st, 2017

Although I’ve never contemplated the speed at which a cow moves at, in this in depth paper I realize (as with the many subsets of GIS we’ve investigated in this course), that movement too has a very nuanced methodology to be done correctly. I found it refreshing to finally see genuine critique of GPS accuracy in a GIS paper, as we often find these instruments to be ‘accurate’. Attempting to counter this uncertainty with increased temporal scale brings up the day old GIS problem of accuracy vs. precision, where in collecting lots of data is somehow seen to offset the accuracy issues, and precisely return incorrect results. I found the measures of sinuosity and turning angle, interesting proxies to determine not only the speed cows actually move at, though also inferring behavior (i.e. grazing). This begs the question of when movement information is collected on individuals, and whether whoever’s collecting the data infers someones behaviour solely on movement (i.e. RFID trackers in passports to detect ‘loitering’). This would be a complete breach of privacy in my opinion, and an example of coerced VGI, which movement GIS could easily be used in (which becomes apparent when the author brings up the innocent commercial shopping cart example besides the tracking of individuals).

Lastly, I find movement (in very much the same vein as temporal GIS) a key study in GIScience, that comes with a humble name though contains lots of variables to consider to avoid the inherent uncertainty that comes with both temporal and geographic scale, as well as instrument error. Lots of uncertainty still remains in the authors work (such as what thresholds to pick to omit information?) which only gets more complicated when you consider that the information gathered (although not very rich in it’s attribute data (x/y coordinates and a time stamp), can quickly become big data in the millions for just three cows, and be used for so much more than tracking movement, though also inferring behaviour, and possibly predicting movement and behaviour if gathered on a regular enough basis.

-MercatorGator

“Empirical Models of Privacy in Location Sharing” (2010)

Friday, December 1st, 2017

This was an interesting article dating from the early days of location-sharing mobile technology, before the widespread use of smartphones. Location sharing in Locaccino, the application used by participants, was based on time, location, and network rules. The authors concluded that these specific privacy settings allowed users to feel more comfortable with the locations they shared. Location sharing applications today feature very few of these customizations. Usually the only restriction that can be set in an app such as Apple’s Find My Friends is a network one, i.e. one can choose the friends with which to share. Locaccino is also different in its reliance on “requests” rather than a stream of location pings, with the most recent location always displayed to friends. The nature of location requests is analyzed considerably in this study, and user’s social groups are easily identified in this analysis.

Today location-sharing tends to be embedded in other social platforms, where network settings are predetermined by a user’s list of friends, and everyone in the list is allowed to access the location (in the cases of Snapchat and Facebook). Facebook allows for customization of this list; users can create specific lists of facebook friends with which to share location, or among academic or work networks, which can include non-friends. The only time-based customization allowed is a restriction to location-sharing “while using the app” versus “always”.

Most of my friends perceive location-sharing on social platforms as invasive. My friends who use Apple’s Find My Friends, which allows sharing only between specifically invited iPhone users, typically share locations with family and spouses only. I think that the results of this research are relevant today, and location-sharing platforms might be better received if they allowed users to tailor their location-sharing settings.