Uncertainty in your GIS work

Think about your own field data, or spatial data from other sources such as satellite imagery.

  • What types of uncertainty exist in them?
  • Are these inevitable, i.e. caused by the inability to measure precisely?
  • Are they deliberate generalisations or simplifications?
  • How might they affect what you can do with these data, for example in your analysis, map-making, or making management recommendations?
  • Is there anything you can do to reduce the sources of uncertainty?

Part of the uncertainty in satellites can come from lack of precision, misalignments in their trajectory and clouds that make measurements difficult. I think we can try to avoid them by choosing images of the precision we need or from different moments when the weather does not affect the measurement.


In collecting water samples, we always go back to the same locations and roughly at the same time in the month. This could create both spatial and temporal uncertainties as our data may not necessarily give a proper overview of the entire sea area we wish to sample. Perhaps we could diversify the time of the month and locations we collect our water samples.

I am still very new to this so unsure, but ill have an attempt. Using satellite as mention gives a snapshot of what the world looks like at a given time. Often the imagery may have been taken at one point and uploaded at another. This could skew data if one is trying to show for example real time forest degradation.

The biggest source of uncertainty in my data is temporal uncertainty. This is because survey flights are not always conducted at the exact same time of day (although we try to) and the weather is inevitably not exactly the same for every flight (we count more elephants when it is cool and cloudy than when it is hot and sunny). Therefore, this uncertainty is inevitable even though we try to limit the scope of it. When writing flight reports, we always take into account this uncertainty by adding notes to indicate if it was hot or cold, sunny or cloudy, and morning or evening, etc. We reduce this source of uncertainty by trying to fly the surveys at the same time of day, at the same time of the year, and with the same exact transect lines, but of course the weather is a somewhat unpredictable variable so this is not always possible.

1 Like

IN my camera trap data that is used alongside other data such as spoor transects and direct observations, in order to determine home ranges and territories the uncertainty exists in saying ’ how many times do I need to see this cheetah at this ocation to decide this is in ots home range versus territory?
The uncertianties are inevitable - there is no way we can ever know the boundaries for sure. We need to simplfy for conservation management purpouses and needs.
Maps need to show the general areas that are considered to be within a territory or home range so that when we introduce new species to teh area so as to build gene pools and augment mating bonds (African wild dog) we know where is the best place to introduce the new individuals without disrupting what already exists.
The more data we collect the less uncertianty but the ore hours in the bush and the more expenditure - so we have to weight up what we want from the data - collect as much as we need to answer our question and balance the budget.

1 Like

as a newbie, i assume that the challenge will be the updating of the satellite and the objects at a given time

We do collect some sound data from Swift Acoustic Recorder we install in the field. When we run the gunshot detector on the sound, we do get some possible detections which we then listen to pick out the real possible gunshots. After listening to all, we score(rate) the sounds to determine which is more likely a gunshot and which is not.

The more people that listen to a particular sounds will give different ratings at to how close the sound is to a gunshot.
Increase in the number of gunshots in the park means increase hunting pressure. The increase in hunting pressure might not necessarily mean decrease in wildlife population.

Uncertainty: When I work different cities, the boundary between them and the agglomeration is often fuzzy and not easy to tell which house is still part of the city and which one is part of a village.
If there is a digitalized map about the boundaries like in case of European Urban Atlas then it is evitable, however, there are regions what do not have detailed digitalized maps and therefore these mistakes can happen.
Generalization: in case of urban green space we generalize. We do not make difference between forests, meadows and lawns.
While I do analyse different vegetation that makes my analysis more simple. However, if I would like to make difference between vegetation I would have to refresh my polygons from airial photos.

A common discussion I have is comparing the difference between “precision” and “accuracy” (the two are different). Is it better to be precisly wrong or vaguley accurate ?.Often, geospatial data is given in too much precisioni (ie. a co-ordinate location to many decimal places giving a precision of at millimeters or smaller) while the accuracy of the measurement if often to within metres (eg. accuracy of mobile phone GPS location is normally +/- 5m, but to give this to 6 decimal places is over-egging the pudding somewhat!.
There is another interesting discussion to be had over maps and the cartographic decisions made at different scales of maps which frequently result in features not really being where they are shown on the map (issues of aggregation, simplification and displacement). Happy to discuss this further :slight_smile:

Uncertainty in satellite images is mixed pixels. sometimes it is difficult to get pure pixels (spectral information) of targets like plant species and water. one way of getting around this problem is to use statistical models to correct for uncertainity. Generalization is done using indices like NDVI. Generalization and simplification are important if one uses large amount of data.

1 Like

The biggest source of uncertainty in my theoretical rare plant species data would be temporal, I think. It is inevitable that you would be unable to accurately estimate vegetation cover, given that with rare plant species especially, presence/life stage may vary significantly from week to week. This may be especially true if we don’t completely understand the life cycle of that species. If targeted species are missed when surveying then they could be found outside the projected boundary. Thus, in terms of management recommendations, we would want to create, for example, a 30m buffer outside the projected area. To reduce uncertainty in the future it would be best to go out and survey several times at different time points.

Some of the bat locations that I recorded for my project were roost sites- which should have low uncertainty other than the level of precision of the GPS unit used to record the sites. However, other locations were recorded when bat calls were seen on an automatic acoustic monitor. These monitors could be picking up bats from up to 30m away in the right conditions, and so these locations are not as accurate as a roost site location. I think this kind of uncertainty is inevitable when trying to record bat locations.
There also seemed to be a decent amount of uncertainty when I was using remote-sensing environmental data to make species distribution models. Although I did not create the mean annual temperature, elevation, slope, river, etc. rasters that I used in my model, it was clear that they were not completely accurate. For example, some of the roost sites we recorded were quite close to a small stream, but this small stream was not recorded on the river raster that I was able to download. I think some uncertainty in environmental rasters is also inevitable, but the degree of uncertainty of each layer should be considered and reported.

As we track Painted dogs/African wilddogs using VHF Telemetry we may assume Painted dogs (Lycaon Pictus) utilize certain vegetation type which may not be the case, their spatial distribution may be larger than what we collect as we don’t follow them 24/7. To reduce the sources of spatial uncertainty GPS satellite collars may help but so far we don’t have a working GPS.
Population size of painted dogs may be changing rapidly but we do not track all packs everyday so we wouldn’t know how quick the population size is changing.
If there’s someone who has an idea please leave a comment. and if anyone has used GPS collars which have lasted over a year on African wild dogs please get InTouch.

Hi Prichard, Yes, there is certainly issues around the way satellite data is collected, processed and made available to the users and it depends on what source of data is being used. Have a broader think though about other data you may use within the GIS such as your own survey data - how precises is this in terms of geographic location? if ‘classifying’ data , how certain are you of the classes of data, is the data continously variable (eg. height) and if so, how is this collected (eg. are the data collection samples appropriate to the variability in the data?

@M_gog may be able to help on this one. Plenty of experience of GPS collars for tracking animals in Africa. One for you Hollie :slight_smile:

We have a fair amount of uncertainty particularly as we are doing invasive species management and there is a temporal distortion. A lot of this revolves around our main source of data being citizen science fuelled with member of the public reporting squirrel sightings which can be unpredictable in how long it takes to be reported and not entirely accurate. However, it is certainly better than the alternative (not having any information) to make sure we focus are efforts at the key hotspots - where grey squirrels are dispersing or still present in high numbers. I think it is just something we often have little control over but need to ensure we take note of it - so we know the restrictions our data set has and that it will likely never be a perfect representation

Related to determining forest loss: looking at sattelite imagery and categorising data found there- sometimes land cover might be original forest - or it might be some kind of agricultural crop/agroforest that just as green as primary forest… sometimes hard to tell the difference… so if you are using sattelite data to determine forest loss - uncertainty would exist if you cannot verify via a different method. Need to be familiar with the data and familiar with what is happening on the ground as much as possible.

thank you, i get it now

When collection data on habitat type extent and degradation, my data is dependent on 1) landowner perception of the extent of their habitat types and areas of degradation which is anecdotal, potentially biased or inaccurate this creates uncertainty in the extent of habitat types and the severity of the degradation. 2) I also go into the field to undertake a rapid infield verification ecological assessment however this leaves uncertainty as this is a low spatial and temporal extent as l I only see a snapshot of the extent of the degradation for example. 3) satellite imagery could have a low spatial resolution if not updated seasonally or after rainfall events for instance. Vegetation classes could also be spatially uncertain at boundary level as well as temporally uncertain due to seasonality and vegetation cover or absence. These uncertainties don’t have to be inevitable however would take a lot a corroboration of various sources to increase the detail whereby the accuracy of the sources will still remain out of our control or knowledge of the integrity of the data. I believe uncertainty is inherent and we need to be accepting of this- taking it with a ‘pinch of salt’ until you can corroborate on the ground?