Is it safe to go caving yet? Modelling cave flood events in Yorkshishire

Hall2501

Member
Modelling cave flood events in Yorkshishire

*disclaimer* This idea is just for fun/interest, I don't recommend using any outcome of this project to decide whether it's safe to go caving.

I've recently been wondering whether it would be possible to create some kind of empirical model relating recent rainfall/surface river levels to the water levels in caves and the period of time required for caves to drain to safe water levels after periods of heavy rain. This could allow one work out which caves are safe to go in without relying solely on judgements from people with a good feel for how the cave responds to water. Obviously common sense and judgment from those experienced with the caves should definitely take priority when deciding if a trip should be given the green light as well as forcast weather and other factors but it would at least be interesting to see how well these models match up with how people think the cave stream levels behave during/after heavy rain.

My idea is to make some devices capable of long term data logging of water levels and place them at  points throughout a cave that are prone to flooding/dangerously high water levels after heavy rain (e.g. eureka junction, Easegill or Kingsdale master cave). The device I have in mind would use a type of pressure-based water depth sensor from remote control submersible devices with 0.16mm precision and +/- 4cm accuracy. The device should be able to take a water level reading every 5-10 minutes for several months before the batteries need changing. I would also place a rain gauge above the cave and use data from the UK's Environment Agency flood-monitoring network (local rain gauges/river level monitors) to monitor rainfall/surface water levels. Then use the data I've obtained to create a model of cave flooding events, where you can predict how flooded a cave is or how long it will take to drain based on factors which can be observed without going to the cave to check. It might also be interesting to log surface above cave river water temperatures and cave water temperatures as a proxy for working out whether water can come in direct from surface streams (you would expect this to be colder) or perculation (you would expect this to be warmer, as it's had longer to equilibrate to the constant cave temperature).

This is currently just an idea but lets see how it goes, I'll post any progress updates here. So far I've designed an insitu long term cave water level monitoring device (?120), obtained a surface tipping bucket rain gauge and temperature logging devices.


 

aricooperdavis

Moderator
I tried to convince my university to pay for my PhD doing exactly this, but they said it was too niche, so instead I'm doing the same thing in sewer networks.

The modelling is the interesting bit to me - I'm using neural networks at the moment, and recent advances in machine translation has opened up the field for  sequential/time series modelling considerably. I'd start by having a look at NARX models, and seq-2-seq using something that can capture time series information for the autoencoder, such as LSTMs.

For your data collection, be aware of sensor drift, and of the practicalities of using data from tipping bucket rain gauges, which are notoriously troublesome. How about looking at publically available rainfall radar too?

Would love to chat more about this, and to get involved if you're looking for input!  :beer:
 

andrewmcleod

Well-known member
The depth gauge in OFD has been recording stream depths there for some time now. Combined with the surface rainfall record from the SWCC(?), this should now be an excellent dataset for exactly the sort of analysis you are describing. So to some degree the data already exists for someone to start making a theoretical attack on.

Obviously it would be good to get more data in Yorkshire etc as well!
 

Hall2501

Member
aricooperdavis said:
I tried to convince my university to pay for my PhD doing exactly this, but they said it was too niche, so instead I'm doing the same thing in sewer networks.

The modelling is the interesting bit to me - I'm using neural networks at the moment, and recent advances in machine translation has opened up the field for  sequential/time series modelling considerably. I'd start by having a look at NARX models, and seq-2-seq using something that can capture time series information for the autoencoder, such as LSTMs.

For your data collection, be aware of sensor drift, and of the practicalities of using data from tipping bucket rain gauges, which are notoriously troublesome. How about looking at publically available rainfall radar too?

Would love to chat more about this, and to get involved if you're looking for input!  :beer:

Yeah it would make a great PhD project! Shame you didn't get funding. Although I think it could be done on the side/cheap.

If it gets going I plan on making all the data public so it'd be interesing to see what you could do with the data  :sneaky:

Hadn't considered sensor drift, I guess I could do periodic checks to see if the sensors still calibrated. Rain radar could be good, I wasn't sure how well that translated to rain fall, I though it was just how swole the clouds are?
 

Pitlamp

Well-known member
I suspect John Gunn may also have a large amount of data already along these lines, for the Peak system. (Some of it was obtained with involvement from the late (and sadly missed) Nigel Ball.

Great initiative by the way Brendan. You quote examples in the Dales, so I'm assuming you're based (or at least do plenty of caving) there.

Could I suggest that you include at least one site which is percolation fed, drains to an immature lower underflow spring - and acts as a flood resurgence when flows are too great for the underflow route to cope? There are several systems of this type to choose from, some of which are extremely dangerous because of these rapid (often unpredictable) flooding characteristics, often because there is hydrological inertia in the catchment. Sleets Gill Cave would be a really valuable site to collect data for.

Incidentally, you don't necessarily need to skimp on equipment; this is exactly the sort of study which the BCRA's CSTRI awards were created to help with (financially). If you're going to do this you might as well do it as well as possible, with financial help from the caving community.

 

alastairgott

Well-known member
(Drafted the below before the last three posts and can't be bothered redrafting  :tease: )


Love it, I was only thinking of something similar for Peak cavern at the weekend, when I couldn't be sure what the conditions were like in the lower bung, If I could have been sure that levels were dropping rather than rising then I might have made a different choice and let a group in on Sunday. but with extra technology you bring yourself closer to the line between safe and unsafe practices.

Hypothetically if I were to study it, I would probably get about 4 float switches and get a diver to place three in the lower bung streamway and place one in Treasury. that way you could monitor the rising of the water in the lower bung streamway, when a float switch becomes engulfed with water it will record on a data logger in treasury chamber.

Much like the use of a pressure(/depth) sensor you will be able to then be able to tell how far the water is backing up in the lower bung, and therefore how close you are to a flooding event.

Though I don't think i would ever get this far, as it does take the adventure and the unknown out of an otherwise thoroughly enjoyable cave.
 

Steve Clark

Well-known member
Have you seen this gadget?

https://reefnet.ca/products/sensus/

It?s about the size of a matchbox. It?s designed to be carried in a diver?s pocket, but you can just leave it in the cave for months. I understand the team surveying sump 3 and the surrounding cave in Cabouy, France have had a couple in the cave on submerged bolts for several years. Catches the flood pulses which are 10m+ high, which is astonishing considering the size of the cave and discharge area.

https://m.facebook.com/thehiddenriverproject/

Look for the post on 18th Oct 2019 for the nice charts with accompanying weather data.

Contact Torsten for more info, he?s been exceptionally helpful with some therion code for me and is really enthusiastic.
 

Hall2501

Member
Pitlamp said:
I suspect John Gunn may also have a large amount of data already along these lines, for the Peak system. (Some of it was obtained with involvement from the late (and sadly missed) Nigel Ball

andrewmc said:
The depth gauge in OFD has been recording stream depths there for some time now.


If data long these lines already exists that's good news for the coding side of things. My modeling skills are pretty basic tbh, my inital plan was to literally overlap rainfall/cave water level data and see how it looked but if Ari can use this to write some clever code that takes in time-series rainfall/cave water level data while I make/test a prototype device the project could be off to a speedy start.

Pitlamp said:
Could I suggest that you include at least one site which is percolation fed, drains to an immature lower underflow spring - and acts as a flood resurgence when flows are too great for the underflow route to cope? There are several systems of this type to choose from, some of which are extremely dangerous because of these rapid (often unpredictable) flooding characteristics, often because there is hydrological inertia in the catchment. Sleets Gill Cave would be a really valuable site to collect data for.

Great idea  :)

Pitlamp said:
Incidentally, you don't necessarily need to skimp on equipment; this is exactly the sort of study which the BCRA's CSTRI awards were created to help with (financially). If you're going to do this you might as well do it as well as possible, with financial help from the caving community.

I'm keen enough to make/test a prototype device myself anyway but if it works and the BCRA are happy to pay for multiple devices and better sensors, then better richer data can't be a bad thing.

alastairgott said:
(Drafted the below before the last three posts and can't be bothered redrafting  :tease: )


Love it, I was only thinking of something similar for Peak cavern at the weekend, when I couldn't be sure what the conditions were like in the lower bung, If I could have been sure that levels were dropping rather than rising then I might have made a different choice and let a group in on Sunday. but with extra technology you bring yourself closer to the line between safe and unsafe practices.

Hypothetically if I were to study it, I would probably get about 4 float switches and get a diver to place three in the lower bung streamway and place one in Treasury. that way you could monitor the rising of the water in the lower bung streamway, when a float switch becomes engulfed with water it will record on a data logger in treasury chamber.

Much like the use of a pressure(/depth) sensor you will be able to then be able to tell how far the water is backing up in the lower bung, and therefore how close you are to a flooding event.

Though I don't think i would ever get this far, as it does take the adventure and the unknown out of an otherwise thoroughly enjoyable cave.

Yeah I looked at a few different methods, I figured a pressure sensor would be small, energy efficent and with the lack of moving parts fairly roubust if the conditions get rough but didn't consider signal drift so will see how much of an issue that is. 

Yeah, certainly would take the exitiment away a bit depending on how much you trusted the data/model  ;)
 

aricooperdavis

Moderator
blhall195 said:
andrewmc said:
The depth gauge in OFD has been recording stream depths there for some time now.

If data long these lines already exists that's good news for the coding side of things. My modeling skills are pretty basic tbh, my inital plan was to literally overlap rainfall/cave water level data and see how it looked but if Ari can use this to write some clever code that takes in time-series rainfall/cave water level data while I make/test a prototype device the project could be off to a speedy start.

I have seen a data file linked from one of the C&KSs but haven't had a play with it - I'll stick it on the to-do list though. :)
 
Pitlamp said:
I suspect John Gunn may also have a large amount of data already along these lines, for the Peak system. (Some of it was obtained with involvement from the late (and sadly missed) Nigel Ball.

Great initiative by the way Brendan. You quote examples in the Dales, so I'm assuming you're based (or at least do plenty of caving) there.

Could I suggest that you include at least one site which is percolation fed, drains to an immature lower underflow spring - and acts as a flood resurgence when flows are too great for the underflow route to cope? There are several systems of this type to choose from, some of which are extremely dangerous because of these rapid (often unpredictable) flooding characteristics, often because there is hydrological inertia in the catchment. Sleets Gill Cave would be a really valuable site to collect data for.

Incidentally, you don't necessarily need to skimp on equipment; this is exactly the sort of study which the BCRA's CSTRI awards were created to help with (financially). If you're going to do this you might as well do it as well as possible, with financial help from the caving community.
You beat me to suggesting Sleets Gill John. Dave Hodgson prompted me a number of times to find some cash to fund a project monitoring rainfall on Parsons Pulpit and loggers monitoring water levels in the cave to establish the flooding pattern to rainfall. It is a sobering sight watching water flow out of the cave entrance knowing that it represents about 100 feet head of water over the cave system below. It would definitely get my support considering what has happened in the past.
 

mikem

Well-known member
Similar idea for kayakers (black vertical bars are from rain gauge, grey bars is met office forecast, blue line is river level from EA Gauge & dashed is guesswork formula of what it's going to do - it's not terribly reliable, but can be helpful & needs to be above 0.7 to say YES, but river can be done lower):
http://isthedartrunning.co.uk/
 

aricooperdavis

Moderator
I've just had a read of that article I referenced above and it's a superb piece of work.

Downloading and visualising the data does already illustrate the issues I was mentioning about tipping bucket raingauges though - the duration highlighted on the right doesn't seem to be recording any rainfall, and the duration highlighted in the centre seems to have rainfall depths far lower than one might expect given the response of the cave. This could be because that precipitation was falling frozen and not correctly operating the TBR, because high winds were causing shielding of the collection vessel etc.

Rather than using your own TBR you could exploit some of the openly available Met Office Data available on DataPoint, and past data is available on MIDAS. These are likely to be more reliably installed, maintained, and monitored than something you can install yourself.




 

Attachments

  • TBR_issues.png
    TBR_issues.png
    53.1 KB · Views: 241

mikem

Well-known member
As Ari points out there are some major issues with using rainfall data, tipping collectors being notoriously unreliable (all gauges under-record with snow, but tippers may over-record in high winds, as don't need to fill before tipping, or under if rain blown almost horizontally in different direction - dartcom, which is used for link I gave for kayaking used to do this, but has now been replaced with a more reliable type, but I can't find the post they made at time to say what). Also very few rain gauges are placed in upland areas, which generally get more rain than towns where the data is collected - kayaking is lucky that dartcom is in the top of catchment for river; but will even self placed gauges be in the right location for where the rain that affects the cave will actually fall? (Dartmoor being granite, the water stays on the surface, rather than disappearing underground, & potentially taking a different route to what we might expect).

Rainfall predictions are also almost impossible to accurately quantify in millimetres (last 4 days on Dartmoor all had much higher predictions than actually occurred & you could watch the peaks disappear on chart as time approached, but there is no record of this in the final public model). Snow can either melt immediately, or some time later, depending on temperature, or even quicker if rain falls on it.

Then the previous weather massively affects what actually happens on / under the ground - previously dry weather & heavy rain results in almost instantaneous runoff, as does already saturated ground (plus the affects of snow mentioned above). But steady rain, of the same amount, may mostly soak into the ground (especially where there are peat bogs above the cave) & slowly release into the system.

I think basing it on river flows alone would make a more accurate model, but very few of the currently available river gauges are placed in suitable locations (most being well downstream of the places we are interested in, so only tell you what has happened, rather than what is coming). But if there is a suitable delay in cave flooding, compared to the nearest river, then they are useful. Some EA sites will only update at widely spaced intervals during day, unless flooding is imminent & then they become more regular, so data may not be available when we make our decision (this can be seen by comparing river Dart charts on rainchasers website to the EA flooding site, as rainchasers only takes the point info, when it updates, whilst EA backfills all the previous data).

EA gauges are also often downstream of confluences where rainfall in widely dispersed areas may affect the level, so you don't know if rain has actually fallen in part of catchment you are interested in, or on one of other tributaries.

All in all, a list of potentially dangerous sites & looking at a detailed weather forecast / synoptic chart are better options for making your decision (although I agree that it is an interesting project).
 

mikem

Well-known member
EA did make some predicted river levels publicly available, but they seem to have discontinued that, presumably because it isn't yet possible to model accurately (I'm sure they are still trying to model it, just not making public), however, the main map does show what areas are currently at risk, by using orange, rather than the standard blue, dots:
https://flood-warning-information.service.gov.uk/river-and-sea-levels
 

Hall2501

Member
aricooperdavis said:
I've just had a read of that article I referenced above and it's a superb piece of work.

Downloading and visualising the data does already illustrate the issues I was mentioning about tipping bucket raingauges though - the duration highlighted on the right doesn't seem to be recording any rainfall, and the duration highlighted in the centre seems to have rainfall depths far lower than one might expect given the response of the cave. This could be because that precipitation was falling frozen and not correctly operating the TBR, because high winds were causing shielding of the collection vessel etc.

Rather than using your own TBR you could exploit some of the openly available Met Office Data available on DataPoint, and past data is available on MIDAS. These are likely to be more reliably installed, maintained, and monitored than something you can install yourself.

I think you're right, it'd probably be best to combine rainfall data from a veriaty of sources to get a more accurate idea of how much rain had actually fallen. I can get data from the environmental agencies rain monitoring stations which are near but not on top of the caves.

There don't seem to be many in wales unfortunatly but plenty in the dales: https://environment.data.gov.uk/flood-monitoring/assets/demo/index.html

Here's data I've gotten data out from the past month for a station near Inglelton.
rainfall-data-1.png
 
Top