It’s hard to believe that it has been almost 2 years since I was last at Sheskinmore. And I expect that there are many of you who are also facing the same reality – not quite managing to take advantage of the brief episodes of relaxed restrictions to sneak in a socially-distanced visit!
So, what do you do when you are unable to visit your research site, can’t get out to download field instruments that have been logging data since you were last there, and just want to see how the dunes are changing through the seasons? You find a renewed interest in satellite data – i.e. remote sensing! Bare with me … this is going to be a long story!
Satellite imagery has been a core resource that I have used in the past to help outline changes in the shorelines and estuary channels across west Donegal. But as my research interests have evolved, so too has the potential of these datasets. Instruments installed on satellites sense a range of information from the surface of the earth – and they do this on a regular basis, and often over the entire world.
The European Space Agency (ESA) for example operate many missions – satellites that are currently in orbit around the world, upon which there are various instruments sensing information – recording data – about the earth, its land, ice, oceans and atmosphere.
Reflectance
In many cases, the information being ‘sensed’ by the satellites is what is referred to as reflectance – the amount of light (or energy) being reflected by surfaces of the features of the earth. All surfaces reflect light / energy differently.
We see this every day when we look around us – all the colours that our eyes detect are just different ways in which light is being reflected from different materials and surfaces. And these differences are a result of the wavelength of the reflected light – the range of colours that we can see – the rainbow from red through orange, yellow, green, blue to violet cover a range of wavelengths from longer (red) to shorter (violet).
What makes satellite data especially informative is that it also senses beyond the wavelengths of light that we can naturally see with our eyes – for most people, the limits of this visible range are 380 – 780 nanometres (nm) which extends from violet to red. We are affected by shorter wavelengths – for example ultraviolet (UV) light damages our DNA and causes sunburn, and we radiate longer wavelengths – for example infrared (IR) as warm-blooded mammals.
When these wavelengths are sensed from space, they are able to tell us additional information about the earth’s surface beyond that which we can see. Some surfaces reflect a lot of light, others just absorb the light. For example, water surfaces have a relatively low reflectance, but the reflectance is almost zero when you consider light with longer wavelengths – longer wavelength light i.e. infrared – is just absorbed by water and not reflected.
In contrast, the chlorophyll in vegetation (which provides a measure of vegetation health) absorbs visible light with short (blue) and long (red) wavelengths, but is a strong reflector in the green range – and that is why we see healthy vegetation as green. Light in the infrared wavelength range also reflects strongly from chlorophyll-rich vegetation. So this variable reflectance and absorption of different surfaces with different properties across light with different wavelengths gives us a lot of potential information when measuring the surface of the earth from space.
Sentinel-2 data
Sentinel-2 is one of the current ESA missions – twin satellites that orbit the earth re-visiting the same location every 5-10 days. Launched in June 2015, one of the draws of these data is the scale of the measurements on the ground. Much of the satellite imagery available is relatively coarse in its spatial resolution – meaning that each data point can cover 100s m to kms on the ground, capture an average of the reflectance from that large area.
Sentinel-2 collects data in 10m by 10m blocks, meaning that for areas as large as Sheskinmore, there is quite a lot of detail. As shown in Figure 1, at this resolution, we can see the dunes, beaches, lakes, channels, and fields.
The top image – the “true colour” version – is formed from the reflectance data collected across three visible bands:
- Red – central wavelength 664 nm
- Green – central wavelength 559 nm
- Blue – central wavelength 492 nm
It looks just like an aerial photograph – the grass is green, the sand is light beige, the peatland is a brownish colour, and the ponds and lakes are dark. The bottom image – the “false colour” version – is formed from the reflectance data collected across two visible bands, and a longer wavelength band beyond the visible range:
- Near infrared – central wavelength 832 nm
- Red – central wavelength 664 nm
- Green – central wavelength 559 nm
This image is quite unlike anything we see in reality, and that is because it is using reflectance data collected in the non-visible, near infrared range, and representing it as a red colour. So, the ‘green’ areas are now generally red as the reflectance of healthy green vegetation is strong in the infrared range.
The different shades of red represent the differences in degree of chlorophyll in the vegetation – the meadows and pastures bright red (healthy green vegetation) and the peatland are more brown, representing the range of plants and their tendency to not be bright green in colour.
The water bodies – lakes, ponds, and the estuary and coastal water – all show as a dense black, which is caused by the fact that light in the near infrared range is absorbed and not reflected by water. It is this significant contrast that can help us understand, and potentially monitor, environmental change across Sheskinmore.
If you want to access the recent true and false colour imagery for west Donegal, I recently set up a webpage to provide direct access to the most recent cloud-free imagery. But you can also undertake your own explorations of the satellite data through the ESA sentinel hub Playground and EO Browser.
Near infrared reflectance at Sheskinmore
Satellite data can be used for a wide range of applications, but in this post, I wanted to share some simple analyses I have been undertaking to explore the dune ponds at Sheskinmore and consider the possibilities for monitoring.
We have water level loggers installed across the site, measuring the variation in water depth at several ponds and Sheskinmore Lough, along with monitoring weather conditions (measured at our weather station at McGlincheys). There are many ponds in the dunes, and they are quite varied in their hydrological behaviour – some are flooded all year round, whilst others only flood for short periods during the winter or spring. At a time when I can’t travel to Sheskinmore to download these data, satellite imagery can possibly keep me up to date with changes in the ponds, but it might also enable us to monitor more ponds in the future.
Satellite data can indicate whether a pond is wet or dry. For example, in Figure 1 above, it is possible to distinguish the open water of Sheskinmore Lough, as well as open water in some of the dune ponds. These water bodies are particularly prominent in the false colour image as this incorporates the near-infrared reflectance (which is very low as it is absorbed, not reflected, by water).
Water bodies stand out as black (i.e. very low reflectance values) when viewing just the near-infrared reflectance. In the Sentinel 2 data, this data is collected in band 8. In Figure 2, we can see the near-infrared (band 8) working very well across the Tramore dunes in distinguishing between dune ponds that are flooded versus just damp or dry.
- in March 2016 and 2018, two large temporary ponds (marked as P1 and P2) were fully flooded, both showing as a large black patch
- in the summer months, the ponds gradually dry out so that by early autumn (e.g. October 2016 and 2018), there are no large black patches in these pond areas
- interestingly, the example shown here confirms how dry the ponds were in spring 2017, following the dry winter of 2016/2017 that we reported on before – you can see that there are very small black patches in the corner of these two pond areas (not much more than a small puddle), but they are clearly not flooded in the way they were in 2016 and 2018
Monitoring ponds at Sheskinmore
Given that the near-infrared data clearly shows when the ponds are flooded or not, if we extracted the value of reflectance for each pond area, could we generate a dataset that shows extent of flooding over time?
We have some water level data for ponds P1 and P2, shown in Figure 3 downloaded in previous site visits, and we have rainfall data. The water level data for both ponds show wet and dry phases in the spring and autumn respectively, and both show anomalous conditions (low water levels) during the winter of 2016-2017.
Using the rainfall data to predict when the ponds are wet or dry is tricky as it is very dependent on prior conditions – for pond P1 to not flood in the winter of 2016-2017, a persistent period of drier than average conditions was needed – but we only have that one example of a dry winter, so little to go on to use this as an indicator.
Using GoogleEarthEngine coding system, I wrote a script to pull out the near infrared (NIR) reflectance data (representing top of atmosphere (TOA) reflectance scaled by 10,000) for these two ponds, and then I plotted this against the same time frame as the weather/water level data (Figure 4). The reflectance values are relative – high values mean that the surface is a strong reflector in that wavelength; low values indicate that the surface absorbs more light, and isn’t a strong reflector of that wavelength.
In Figure 4, the value of the NIR reflectance fluctuates between low reflectance in the spring (when the ponds are full of water) and high reflectance in the summer/autumn (when the ponds are dry). This is exactly what I had hoped for!
Except … actually, when you look at the detail, the data are quite messy, even when you ignore the records from cloudy days (the grey symbols in the plots). The wet periods seem to have reflectance values of less than 1800 and the dry periods have values above this. But finding a specific threshold value that accurately differentiates between the wet and dry phases a little challenging! There are lots of possible reasons for this, most notably atmospheric interference, the angle of sunlight and the tendency for shadows, and the height of the vegetation, meaning that it sometimes protrudes above an otherwise flooded pond.
Perhaps a relative measurement between an area always vegetated and the ponds would work as a way of reducing these interferences. I extracted the near infrared reflectance both within the pond and from the dune around it and used a ratio (more specifically a normalised difference) to calculate the relative difference between the always dry dunes and sometimes wet ponds.
This relative near-infrared value (rNIR in the last plot in Figure 4) is high when there is a big contrast between the dunes and ponds (i.e. the ponds are full of water), and low where there is minimal contrast (i.e. when both are covered in vegetation and the pond is dry) – and a value of around 0.2 seems to mark the divide in the data (approximately!). This hasn’t solved the problem of the messy data though … so perhaps there is a bit more work to do on this yet!
Using the satellite data to predict pond hydrology at Sheskinmore
Based on the analysis done so far, I thought I would see whether it is possible to use these data to predict when the ponds have been wet or dry over the last 2-3 years (i.e. since the last data download). I’ve already noted that the data are messy – so I appreciate that this approach might not work very well!
Using the approximate cutoff of 1800 based on the data shown in Figure 4, I separated out the dates when the NIR reflectance was less than this (i.e. the ponds would be flooded) from those when NIR was greater than 1800 (i.e. the ponds were vegetated and dry). The result is shown in Figure 5 – the orange stripes show dates when the ponds are likely to be dry, the green stripes are those dates when the ponds are likely to be wet.
Well … as it happens, it seems to have worked quite well! The main periods predicted to be dry are summer to autumn, and those predicted to be wet are winter to spring. But there are also some subtleties in the predictions – the dry period of 2020 for example seems to be shorter than average, and the wet winter of 2019/2020 looks longer than average.
It remains to be seen when I download the loggers again how closely this matches the water level data, but as an exploratory analysis, it hasn’t done a bad job at all! There are a few challenges to iron out, and next it would be good to work in the rain data to see if there is a way to predict the flooded status from that. Lots of questions still, which is never a bad thing!
And for anyone planning a walk out across Sheskinmore, at the current time of writing, these ponds are definitely still flooded!
It has been an interesting side-project – something to keep my mind off pandemics, social distancing and restrictions on movement … plus it keeps me connected to Sheskinmore! Happy to share the code if you are interested in exploring this yourself or taking the analysis further – just get in touch. Or if you are regularly out and about at Sheskinmore, you could start taking a note of when the different ponds are wet or dry.
Let me know if you have any questions!
Friends of Sheskinmore 
How do you access the aerial images?
You can view the Sentinel-2 images on the Sentinel Playground website: https://apps.sentinel-hub.com/sentinel-playground
– you need to zoom on the map to the area you want to look at, and then use the calendar to select a specific date.
Let me know if more guidance on this is needed ok.
Many thanks. I’m in Waterford and would dearly love to visit Sheskinmore, particularly to see the orchids growing there.
Fantastic stuff Helene. Enjoyed the read.