April 5th, 2013
The article of G. G. Wilkinson is dated, and this is significant in a field that is rapidly evolving. Nonetheless, in my point of view, the author’s argument is still valid today. He talks about uncertainty and data structures in remote sensing and GIS. Sophisticated technologies and remote sensing don’t automatically solve the problem of delimitating boundaries. Even with technology development, classification is still a complex task. It is like trying to create boundaries where the world is actually maybe more like a continuous landscape. We are trying to define distinctive class of land cover or topographic zones for example, but in reality is there a frontier between different types of land? It partially explains why uncertainty is attach to any kind of techniques in remote sensing. Taking the limits of remote sensing techniques into account, the author evaluate different procedure and use of data structure. He thus suggests that part of the further development is to identifying the best techniques and technology development that will allow the best representation of the phenomenon that is intended to be represented by the remote sensing data. Although the problems of errors and uncertainty are unlikely to be solved easily even with technical development in data structures or with visualization techniques such as 3d environment and virtual reality.
April 4th, 2013
Helen Couclelis wrote an article called Certainty of Uncertainty and I think that David J. Unwin is making a similar point. The problem of uncertainty is not merely technical. Uncertainty doesn’t only come from data and information but it is also about geographical knowledge that is sometimes inevitably uncertain. There are things that we simply can’t know. The literature focus on finding technical solutions, but the author explains that “at the heart of all the contributions is a concern for exactly how we can usefully represent our geographic knowledge in the primitive world of the digital computer”.
As mentioned in previous discussion about ontology, we conceptualize the world as field or object based which correspond to raster or vector in GIS. The author shows that both representation comes with specific uncertainties. Furthermore, we discussed how delimitating boundaries is often a difficult task and uncertainty is inevitable. The conclusion is bringing us back to the first discussion in class about GIS as a tool or as science and the determinism of the technology. The author suggest that rethinking the way we use the technology and the way we structure problems and databases is essential to achieve sensitivity in GIS. It is about adapting the technology to represent knowledge in a way that would take into consideration our conceptualization of the world and not merely relying on GIS technology to calculate the world for us.
Couclelis, H. (2003). The Certainty of Uncertainty: GIS and the Limits of Geographic Knowledge. Transactions in GIS, 7(2), 165-175.
April 4th, 2013
Uncertainty lies at the core of GISci where MacEachren et al. acknowledges the GISci community has given more attention to formalizing approaches to uncertainty than in other communities such as information visualization communities (p. 144). The authors go through several examples of how uncertainty can be visualized from changes in hue to symbols with different transparencies to depict where uncertain data may exist. What peaked my interest was the interactive visualization techniques that users can control depictions of uncertainty. Instead of permanently adding a layer of complexity that can obstruct and confuse the readers from what the data is trying to depict, the user is in full control of how much or little information (with regards to uncertainty) is available to them. To me this seems like a better solution than to simply find a single “ideal” ways to represent uncertainty visually in a static manner – especially since every individual will have their own preferences on what they think “best” means (context matters!). What I don’t quite agree with is the authors’ assertion that humans are not adept to using statistical information to make decisions and base on heuristics (based on a study in 1974). Since the quantitative revolution, hasn’t statistics been bought to the forefront of geography such that we may rely on statistics too much at this point? That being said, visualizing uncertainty can take on many forms, from charts, changes in opacity, 3D graphics where the way in which uncertainty should be viewed will ultimately be context specific to meet the goals of the researcher.
April 4th, 2013
Brivio et al. provides a case study where the integration of GIS and RS is able to compensate for limitations that may exist in each technology. The study provides a good example of how these two closely related fields can combine together to produce a more realistic representation of various phenomenon. While this case study specifically used additional GIS data as a supplementary component to improve on the RS classification of flooded areas, RS data can similarly be used to as a tool to produce GIS data (ex. land cover classification dataset derived from remote sensing data). However while there are many advantages in integrating the two, several issues come to mind. RS data is pixel based, while spatial data can be vector or raster based. To have to convert one to the other in order to do analysis would compound issues of accuracy and uncertainty. We know RS is already well acquainted with their own issues related to scale, noise and technological limitations, but these issues can quickly get amplified, and I can imagine that recognizing these sources of uncertainty will be difficult once the data thoroughly entangled in one another. Also, what kind of data models is required for this integration? Spatial data is generally represented in 2D, while RS hyperspectral cubes are in several dimensions. For the researcher whose interested in integrated such technologies, they have to be well versed in the inherent issues that each type of data presents to provide a comprehensive analysis – definitely no small feat.
April 4th, 2013
MacEachren et als article provides a thorough overview of the current status of uncertainty visualization along with its future and its challenges. It seems to be established that uncertainty visualization is more useful at the planning stage than at the user stage of an application. This makes me think back to an earlier discussion on temporal GIS. We talked about how the important aspect of temporal GIS was in its analytical capabilities, rather than in its representational capabilities. While I do not deny the positive effect on analysis that visualization might have, I question if it should be the aspect of uncertainty that is given the most attention.
Two of the challenges proposed by the article are developing tools to interact with depictions of uncertainty and handling multiple kinds of coexisting uncertainty. Might representation in some instances prove more troublesome than its worth? Might representational practices at times be obfuscative of data that might be understood as just data? I want to note that I am asking these questions in earnest, not rhetorically. Which I guess boils down to a question I have probably asked all semsester: how do we evaluate what is important enough or useful enough to invest time in?
April 4th, 2013
Unwin‘s 1995 paper on uncertainty in GIS was a solid overview of some of the issues with data representation that might fly under the radar or be assumed without further comment in day-to-day analysis. He discussed vector (or object) and raster (or field) data representations, and the underlying error inherent in the formats themselves, rather than the data, per se.
While the paper itself is clear and fairly thorough, I can’t help but question whether error and uncertainty are worth fretting over. Of course there is error, and there will always be error in a digital representation of a real-world phenomenon. Those people, such as scientists and policy makers, who rely on GIS outputs, are not oblivious to these representation flaws. For instance, raster data is constrained by resolution. It is foolhardy to assume that the land cover in every inch of a 30-meter grid cell is exactly uniform. It is also wrong to suggest that some highly mobile data (like a flu outbreak) would remain stationary over the course of the interval between sensing/mapping. There are ways around this, such as spatial and temporal interpolation algorithms and other spatial statistics, and I feel like estimates are often sufficient. If they aren’t, then perhaps the problem isn’t with the GIS, but rather in the data collection. Better data collection techniques, perhaps involving more remote sensing (physical geography) or closer fieldwork (social geography) would go far in lessening error and uncertainty.
With all of that said, I am not about to suggest that GIS is perfect. There is always room for growth and improvement. But, after all, the ultimate purpose of visualizing data is for understanding and gaining a mental picture of what is happening in the real world. An error-free or completely “certain” data representation is not only impossible within human limitations, but it is not particular necessary.
April 4th, 2013
No matter how good technology becomes, we will always face challenges in data uncertainty and error; the question is, can we develop appropriate techniques to mitigate the effects of these noises, and come away with the correct signal. As MacEachren et al. (2005) point out in their article titled “Visualizing Geospatial Information Uncertainty”, we use this information to base decisions off of, and the uncertainty is inherent in the data and must be taken into account.
There are multiple dimensions of uncertainty, as the authors point out, ranging from credibility of a source to precision of a physical variable, and these will compound, effecting the amount of correctness the end result will have. They function across many scales, including the direct attribute of the information, the specific context or location of the information (which may not be what you want to apply the information to), as well as temporally. It all seems very complicated when examined through this framework… but it is important to take these into account in order to have confidence in your product.
Personally, i have experienced a lot of uncertainty while trying to create a global map of administrative subdivisions. Every County collects data at different resolutions and time, however these countries are supposed to be contiguous as we well know. The borders do not always align, but who is right? Furthermore, this issue is compounded when you consider the global land mass as a whole. We want to have an accurate total area of land surface, however if you trust each country to represent their land correctly and then end up with an incorrect total, who is wrong? Where do you remove land? Where do you add it? These are some of the challenges I have faced with uncertainty, and I was not qualified to make the adequate decision.
What I didn’t do at the time was try to quantify and visualize the uncertainty, which as the authors say, is crucial to making sure the data is useable, and that you are confident it is correct for answering the questions you are trying to answer.
April 4th, 2013
Remote Sensing and GIS technology has changed significantly since Wilkinson (2007) wrote his review on how the two fields overlap. Hyperspectral imagery is now commonplace, and the software is well equipped to deal with it. Currently, we still struggle with handling error and uncertainty, but there are prescribed ways for dealing with each issue. Atmospheric conditions, topography, angle, sensor, and georeferencing are now done to eliminate some of the error caused through data collections. Things like fuzzy logic help to deal with uncertainty, although it remains an issue. As data collection techniques further improve, our ability to deal with this uncertainty will become less and less important.
Most of the current issues still lie in data models. The complementary nature of GIS and Remote Sensing is evident, however these two technologies speak different languages in situations where we expect them to communicate and enforce their complimentary relationship. This becomes even more difficult when we try to represent more complex relationships that are no longer 2-dimensional with hierarchical classifications. Personally, I find that the 2 commercial softwares for each technology interact quite well when performing simple tasks, like making a supervised classification and turning that into a GIS layer. However, when the data becomes more complex, and the classifications with them, the ability of the softwares to communicate with each other becomes increasingly bad.
April 4th, 2013
Since the paper by Wilkinson in 1996 many satellites have been put into orbits and several million GBs of satellite image have been collected. But more importantly, with the coming of the digital camera there has been an explosion in the amount of digital images that have been captured. Consequently, people were quick to spot the opportunity in leveraging the data from the images; hence a lot of research has been conducted in the image processing domain (mainly in biometrics and security). This being said, some of the most successful approaches in other domains have not been as well, when applied to satellite images. And the challenges outlined in the paper still hold true today.
According to my understanding this is mainly because of the great diversity in satellite images. The resolution is only one part of the equation. The main problem lies in the diversity of the things being imaged. This makes it very difficult to come up with training samples that are a good fit. Thus, traditional Machine Learning techniques based on supervised learning have a hard time. Moreover, the problem is compounded by the fact that when we are classifying satellite images, we are generally interested in extracting not one, but several classes simultaneously with great accuracy. However, the algorithms do perform well when classification is performed one image at a time but significant human involvement is needed to select good training samples for each image. But to the best of my knowledge no technique exists which can completely automatically classify satellite images.
April 4th, 2013
Brivio et als paper presents a case study integrating Remote Sensing and GIS to produce a flood map. After explaining methodology and results of other methods, the paper finds the integrative method to be 96% accurate.
This speaks to the value of interdisciplinary work. While RS applications on their own proved inadequate, a mixing of disciplines gave a fairly trustworthy result. While I understand the value of highly specialized knowledge, having a baseline of capability outside of one’s specific field is useful. I remember in 407 Korbin explaining that knowing even a bit of programming can help you in working with programmers, as understanding the way that one builds statements as well as the general limits of a given programming language will give you an idea of what you are can ask for. The same is true for GIS/RS. Knowing how GIS works and what it might be able to do is useful for RS scholars in seeking help and collaboration and vice versa. I think McGill’s GIS program is good in this respect. I got to dip my toes into a lot of different aspects of GIS (including COMP) and figure out what I like about it. If I end up working with GIS after I graduate, I know that the interdisciplinary nature of the program will prove useful.
April 4th, 2013
Geospatial analysis can be no better than the original inputs, much like a computer is only as smart as its user. In the field of remote sensing, this ideology may be on its way to becoming obsolete. Brivio et. al show from a case study of catastrophic inundation in Italy that they can compensate for the temporal disparity in the capturing of remotely sensed data and the peak point of the flood, a few days before.
The analysis, however, was not completed with the sole use of synthetic aperture radar images. Had it not been for the integration of topological data, it is unlikely that one would be able to obtain similarly successful results.
With any data input, temporal or spatial resolution are limiting factors. Brivio highlights this by acknowledging the use of NOAA thermal infrared sensors, which have a finer temporal resolution, while lacking in spatial resolution. Conversely, the SAR images used in the case study analysis have a relatively higher spatial resolution, but produces longer temporal intervals.
Given Brivio et. al’s successful prediction of flooding extent, it may mean that, if need be, it is advantageous to choose an input with a finer spatial resolution in exchange for a coarser temporal resolution, complementing the temporal delay with additional inputs to compensate.
Break remote sensing down into it’s two main functions: collection and output. One will inevitably lag behind the other, but eventually the leader will be surpassed by the follower. Only for it to happen again some time down the road. Much like two racers attached by a rubber band.
What all of this means for GIS; eventually the output from remote sensing application will surpass the computing power of geographic information systems. At which point, the third racer, processing, will become relevant, if he isn’t already.
April 3rd, 2013
Brivio et al.’s article “Integration of remote sensing data and GIS… for mapping of flooded areas” presents the very common process of using RS data and GIS to map flooding and flood plains. Although the article presents how the integration of RS and GIS can accurately map a flood with a concluded method accuracy of 96%, it only looks at a single event and study site. From my experience, this is not always the case, as integration methods, even if they are the same, often vary in accuracy from one location to another. Furthermore, event duration, intensity and geologic substates often interfere with flood area prediction from RS data and GIS, as variations can modify water location within minutes to hours. To clarify, one area may be flooded at certain points during the flood period while during other periods dry (i.e. it may transition from wet to dry to wet), which interferes with accuracy of the RS data and GIS prediction. Fundamentally, water changes how the surrounding environment reacts, modifying where floods are. As floods react to the environment, often areas become flooded for only minutes and as such, are never recognized as a flooded area, in both GIS predictions and RS data, as well as human reports (although they were flooded; but only for minutes).
To better predict flood area, TWIs (topographical wetness index) and DEMs (digital elevation models) when compared to flow paths (cost-distance matrix), may in fact, better predict flooded areas when used in conjunction with RS data then just the integration of RS data to cost-distance matrixes. In addition, more data sets and studies would further help to create a more general integration protocol and predictive area estimates for floods. To elaborate, the techniques in the article work well on the study area by may not work on other floods, therefore by adding more data from more types of floods, the technique could be adapted to other situations. The result of multiple integrations with multiple data sets would also reduce error and produce greater accuracy. The “Big” question, however that will still remain unanswered from this article is: how can we account for ecosystem and flood variability within GIS and RS data sets?
April 1st, 2013
In any data related field great efforts are put into ensuring the quality and integrity of the data being used. It has long been recognized that the quality of results can only be as good as the data itself, moreover, the quality of data is no better than the worst apple in the lot. Hence, for any data intensive field great efforts are put into data pre-processing to understand and improve the quality of the data. GIS is no exception when it comes to being cautious about the data.
The various kinds of data being handled in GIS makes the problem of errors more profound. Not only does GIS work with vector and raster data, it also needs to handle data in forms of tables. Moreover, the way the data is procured and converted is also a concern. Many a times data is obtained from external sources in the form of tables of incidences that have some filed(s) containing the location of the event. Usually this data was not collected with the specific purpose of being analysed for spatial patterns, hence, the location accuracy of the events are greatly varied. Thus, when these files are converted into shapefiles, it inherits the inaccuracy inbuilt in the data-set.
One of the things to remember however is, that the aim of GIS is to abstract reality to a form which can be understood and analysed efficiently. Thus it is important not to lay too much emphasis on how accurately the data fits the real world. The emphasis on the other hand should be to find out the level of abstraction that is ideal for the application scenario and then understand the errors that can be accepted at that level of abstraction.
March 22nd, 2013
Goldberg, Wilson, and Knoblock (2007) note how geocoding match rates are much higher in urban areas than rural ones. The authors describe two routes for alleviating this problem: geocoding to a less precise level or including additional detail from other sources. However, both these routes result in a “cartographic confounded” dataset where accuracy degrees are a function of location. Matching this idea — where urban areas and areas that have been previously geocoded with additional information are more accurate than previously un-geocoded rural areas — with the idea that geocoding advances to the extent of technological advances and their use, we could state that eventually we’ll be able to geocode everything on Earth with good accuracy. I think of it like digital exploration — there will come a time when everything has been geocoded! Nothing left to geocode! (“Oh, you’re in geography? But the world’s been mapped already”).
More interesting to think about, and what AMac has already touched on, is the cultural differences in wayfinding and address structures. How can we geocode the yellow building past the big tree? How can we geocode description-laden indigenous landscapes with layers of history? Geocoding historical landscapes: how do we quantify the different levels of error involved when we can’t even quantify positional accuracy? These nuanced definitions of the very entities that are being geocoded pose a whole different array of problems to be addressed in the future.
March 21st, 2013
No process in GIS perfect. There are always limitations, many of which can be ignored, while others must at least be acknowledged. The application of the results of an analysis can have a drastic impact on whether errors are ignored, acknowledged, or painstakingly resolved. Consider the difference between geocoding addresses at the national level to analyze socioeconomic trends. The power of numbers will outweigh the error generated during the processing, but it is enough to acknowledge the limitation. On the other hand, the example of the Enhanced 911 system requires that all addresses be geocoded as precisely as possible for times of emergency.
As a way of increasing the accuracy of geocoding processes, would it be sufficient to input a number of intermediary points as a way to accommodate for the uneven distribution of addresses within a given address segment. It would essentially act as the middle ground between leaving the results entirely up to geocoding and digitizing all addresses manually. After all, why do the corners of blocks have to act as the only reference point? It’s possible that there is an inherent topology that would be lost if this was to be implemented, but I cannot speak to that.
While reading Goldberg et al., one geocoding nightmare kept running through my head. It surprised me that it was not touched on directly. How has Japan addressed the situation? As an OECD country, it likely possesses sufficient GIS infrastructure. If I’m not mistaken, though, house addresses are not based on location so much as time. Within prefectures, addresses are assigned temporally, whereby the oldest structure has a lower value than a newer structure, even if they are immediately adjacent, two structures can have significantly different addresses. Just a thought.
March 21st, 2013
The last few years have seen tremendous growth in the usage of Spatial Data. Innumerable applications have contributed to the gathering of spatial information from the public. Application’s people use every day like Facebook and Flickr have also introduced features with which one can report their location. However, people are not generally interested in geographic lat-long. Names of places make more sense in a day to day life. Hence, all the applications report not the spatial co-ordinates but the named location (at different scale) where the person is. The tremendous amounts of location information generated have not gone unnoticed and several researches have been conducted to leverage this information. But, one issue that is frequently overlooked in researches that use these locations is the accuracy of the geocoding service that was used to get the named locations. Not only is displacement a problem but scale at which the location was geocoded will also have an effect on the study. The comparison of the various accuracy of the available geocoding services done by Roongpiboonsopit et. al. serves as a warning to anyone using the geocoded results.
March 21st, 2013
Roongpiboonsopit and Karimi provide a very interesting study on the quality of five relatively known geocoding services. Google Maps is something I use very often, however I never really critically thought about the possible errors that may exist and their consequences. A study such as this allows us to understand the underlying parameters that go into these geocoding services and how they may differ from provider to provider. One aspect that was really interesting to me was the difference in positional accuracy of different land uses. Obviously, there tends to be an “urban bias,” of sorts, when geocoding addresses. As a result, one is more likely to get an incorrect result when searching for address in rural/suburban areas. While this makes sense due to spatial issues, I thought that this could theoretically be extending to other criteria. While LBS becomes more popular and geocoding increases in importance, will certain companies offer “better” geocoding services to businesses that are willing to pay for it? For example, Starbucks could make a deal with Google to ensure that all of their locations are correctly and precisely geocoded. Taking it to the extreme, Google could even make a deal to deliberately sabotage the addresses of other coffee shops. While I think this specific case may be unlikely, it does raise issues about having completely authoritative geocoding services. As we increasingly rely on these geocoding services, the companies offering them have a large influence on the people who use them.
This leads into the idea of possibly relying on participatory services, such as Open Street Map. OSM has made leaps and bounds in terms of quantity and quality of spatial data over the past few years. I am curious to see how it would match up with the five services in this paper. OSM relies on the ability of users to edit data if they feel it is incorrect. Therefore, the service is theoretically consistently being updated depending on the number of users editing a certain area. As a result, errors may be less likely to be consistently missed, as with the case of a more authoritative geocoding service. It would also be interesting to see the type of buildings that may be geocoded more or less accurately. As we continue to enter this age of open and crowd sourced spatial data, I believe it has the potential to provide us with even better services.
March 21st, 2013
Geocoding, like many of the concepts that we study in GIScience, is very dependant on the purpose of the process. The act of geocoding is often confused with address matching, which is sometimes correct, however it can also be georeferencing any geographic object and not just postal codes. This implies that the perception of geocoding will affect the ways that we go about doing it.
There are many ways to geocode, as described by Goldberg, Wilson, and Knoblock, and no single one of them is universally correct. Each method uses different algorithms to try to match some identifier to a geographic reference. For example, it might find the length and endpoints of a street and then use a linear interpolation to find the location of a given postal code. The geographical context also bears a great importance in determining which algorithm to use. For example, the method described above may work better in a city with short, rectangular blocks, however it may be less applicable in rural China. These are some of the things that one has to consider when choosing a method of geocoding.
The future of geocoding is perhaps less certain than many of the other GISciences, because as technology and georeferencing becomes more ingrained in our society, the algorithms used to match these objects with a geographic location will become less important. Things like GPS are becoming more and more commonplace in many appliances, however this brings up questions of privacy. Ultimately, the future of geocoding will be a balancing acts of tradeoffs between public acceptance of technology and the development of more powerful and purpose-driven algorithms.
March 21st, 2013
Geocoding is the fascinating process of associating an address or place name with geographic coordinates. Traditionally, Geocoding was solely the realm of specialists, requiring a specific set of skills and equipment. However, with the advent of modern technology, including Web 2.0 applications, Geocoding is now easier than ever for the everyday user. However, despite the multitude of Geocoding services, such as Google and MapQuest, each service uses different algorithms, databases, etc. to code their locations. Therefore, users might not be aware of which services offer the best quality results, or on the contrary, may offer innacurate results. The quality of Geocoding results may in turn affect subsequent decisions, modeling, analysis, etc.
Overall, one of the biggest problems facing Geocoding is the accuracy of the results. In particular, one problem mentioned by the authors was the poor accuracy of addresses located in rural, agricultural, and urban areas. On the other hand, most urban locations tended to be geocoded similarly across platforms. In addition, it was also interesting to note that several platforms consistently offered more accurate results: Yahoo!, MapPoint, and Google. It would be fascinating to investigate what type of geocoding algorythms, databases, etc. these services use, and if they are similar or relatively different.
Another fascinating trend to consider is the future of direction of Geocoding. One possibility could be the standardization of geocoding databases, algorithms etc. On the other hand, this in turn may lead to redundancies in geocoding services, which might not be a realistic outcome. Overall, the future of Geocoding as a useful tool is heavily dependent on how useful and accurate the results can be.