The MWWT – the maximum winter water table – would be the base elevation of any quarry at Straitgate Farm. It is a model, a prediction of what might be happening across some 55 acres, based on water levels recorded in just 6 piezometers. The accuracy of this prediction would matter less if Aggregate Industries were planning to leave a safety margin, an unquarried buffer, above this surface – but it is not.
In July, Professor Brassington warned:
The MWWT for this site has been defined by using a computer model as the number of piezometers (six) are insufficient to cover the quarry area in sufficient detail. Computer models of groundwater systems are good at showing changes in groundwater levels although they are poor at showing the actual amount of such changes. As a result, the computer model derived MWWT surface is unlikely to provide an accurate representation of the real maximum groundwater levels...
I am concerned that there is a very steep hydraulic gradient across the site, from around 152m in the west to less than 135 m in the east, and the limited number of piezometers used to grid the water table surface. Variations in the shape of the water table cannot be contoured based on the number of piezometers used in the application... The steep hydraulic gradient combined with limited monitoring, in my opinion, is likely to result in errors in the actual depth to maximum groundwater across the site.
Even Aggregate Industries’ own consultants – before they whitewashed their reports – warned:
It should be recognised that a variety of possible grid generation algorithms and parameters exist for discrete point data and that each of these can be supplemented to varying degrees by supporting interpretative data and conceptual understanding. The two grids therefore represent just two of many possible interpretations of the data which themselves are based on an incomplete parameterisation of the detailed groundwater dynamics of the Site.
Nevertheless, it has been a struggle to get Aggregate Industries to say anything on the accuracy – or inaccuracy – of its prediction of maximum groundwater levels at Straitgate. The company has continually refused to provide an answer. What could it want to hide? In 2016, we asked Devon County Council:
Since AI now intends to dig right down to the maximum water table, perhaps you could ask Amec to confirm the specific level of accuracy (in +/- m) to which their maximum groundwater contours are mapped?
Aggregate Industries was asked. Silence. In October 2017, the Environment Agency tried again:
Whilst our position remains unchanged from that which we set out in our letter of 01 September 2017 we agreed with the Straitgate Action Group that further information should be requested from the applicant... In particular we agreed that the following information should be requested: ... • A description of the tolerance levels and interpolation method used to produce the ‘Maximum Winter Water Table’ grid;
Again, as we posted at the time, on the matter of tolerances – silence. In November 2017, we wrote to the EA, pointing out that:
Amec have now responded to your letter of 9 October, but unfortunately they still fail to address a number of the fundamental concerns we raised. Your letter specifically asked for "A description of the tolerance levels...". This has not been provided.
Amec say "Four alternative grid interpolation methods were selected... Each of these methods represents an exact interpolator". In this case, all "exact" means is that the derived MWWT surface passes through each of the six maximum recorded points. It does not mean that the interpolation surface is exact anywhere else across the site. In fact, given that two of the interpolation methods were found not to be suitable, and that for the one chosen, Radial Basis Function, "There are five different basis functions... Each basis function has a different shape and results in a different interpolation surface" there is obviously a very high chance that the MWWT surface provided by AI does not reflect reality.
As you are well aware, this would matter less if any quarrying went down no further than 1m above the maximum winter water table. Since this is not what’s proposed, it is crucial to know the tolerance or error levels of the MWWT - as we agreed at the meeting.
For Radial Basis Functions "2.3 ...the error goes to zero when the data samples are getting dense", but in this case the data samples are not "dense"; far from it. Amec has essentially used 6 data points to model almost 60 acres. If the error or level of tolerance is +/- 1m or more, this could have a dramatic bearing on the working method employed.
In response, writing to us in February the following year, the person responsible for the EA’s technical position on Straitgate weighed in to defend the cement conglomerate and its silent consultants:
Your email questions [sic] the term 'exact interpolator'. Exact interpolator here means that the modelled surface will use the values measured at the piezometers. These will not be changed by the interpolation method. The values in between the points for which values are known will of course have to be interpolated, as is the nature of interpolation. It is not possible to drill boreholes at every point across this quarry (or any other). Therefore it is not possible to know, everywhere, what the difference is between the modelled surface and actual groundwater levels. Interpolation and contouring are very commonly used methods in hydrogeology and geology. They are used because it is not possible to see the groundwater surface or underlying geology. Therefore the best that can be done is to obtain information at set points (boreholes) and interpolate between them. We are satisfied that this addresses the issue of 'tolerance levels'.
But clearly that reply didn’t address, or even understand, the issue of 'tolerance levels', and later that month the EA were persuaded as much, and wrote to DCC again:
...the Straitgate Action Group has highlighted to us the incompleteness of the response provided by Aggregate Industries (via AMEC) to our letter of 18 September 2017 with regard to them providing a 'description of tolerance levels'... Section 2.1 of AMEC’s response makes no reference to tolerance levels, the relevance of tolerance levels, or the level of accuracy in the interpolated maximum winter water table (MWWT). We recommend that a description of tolerance levels is therefore requested again from the applicant to support the planning application and to provide clarity in advance of the Planning Committee.
In March 2018, a response was finally provided by Aggregate Industries’ consultants (Amec Foster Wheeler, previously AMEC, now Wood). Now – at the end of 2019, and only through a Freedom of Information request – that response has finally been provided to us. Unfortunately, however, Aggregate Industries and its consultants still haven't come clean on the accuracy of their prediction. Surprised?
Perhaps they were unsure what 'tolerance' meant. AFW said:
We assume that in this context tolerance is used to mean an allowable amount of variation of a specified quantity (for example in the dimensions of a machine or part) or the level of "uncertainly" around the MWWT.
for which they obviously needed help from a dictionary:
But here’s the thing: Aggregate Industries’ consultants do accept that the MWWT contours are not accurate. We’ll say that again for anyone drifting off at this point: Aggregate Industries’ consultants accept that the MWWT contours are not accurate:
...the MWWT grid is faithful to the actual water levels measured at the piezometers which existed on the site when the historically high levels of winter 2013/14 were experienced. Therefore, at these locations the uncertainty is zero and the grid is accurate to the measured data.
At all other locations – i.e. virtually the whole 55-acre site where there is not one of the 6 piezometers – we can therefore infer that the uncertainty is not zero. Hallelujah. Are we getting somewhere now? The question is, if the uncertainty is not zero, what is the scale of the uncertainty, and how does it vary across the site? Uncertainties will be less in locations close to piezometers, and greater further away. Here are the locations of the 6 piezometers used in 2015, and the MWWT contours interpolated from the maximum groundwater levels recorded by them:
But if the uncertainty is not zero, why have Aggregate Industries’ consultants not built a margin of error, a margin of safety into the company's prediction of the MWWT contours, the base of any quarry? Oh, but we have, AFW argued:
As the MWWT was subsequently based on the higher of the two interpolated surfaces (i.e. Radial Basis interpolation) this could be considered to represent the upper limit of interpretation, with the Kriging interpolation representing the lower limit of interpretation and therefore the difference between these two realistic grids could be considered as reflection of the ‘tolerance’.
However, even AFW doesn't sound convinced: it could be considered a reflection. Really? Is that the best they can come up with? Aggregate Industries’ consultants are no doubt clever people. They will know the reasoning behind that statement is flawed. They will know that just because radial basis interpolation has resulted in a slightly higher surface than kriging for these 6 points, does not mean it is accurate. Likewise, they will know that there is no mathematical textbook that says the lower limit of interpretation of radial basis can be determined by kriging. The argument is hokum.
All AFW was attempting to do was to obfuscate, to confuse, to muddle the EA. All AFW was doing was showing the difference between how two techniques (kriging and radial basis) have interpolated 6 measurements to predict groundwater levels across 55 sloping acres. All AFW was showing was the difference between one inaccurate model and another inaccurate model. This is not a tolerance, it is a difference.
What AFW has NOT done, for very obvious reasons, is to consider – at least in public, for the benefit of locals, the Council and the EA alike – the inherent inaccuracies of each of those techniques.
And there will be interpolation errors in both radial basis and kriging. Our email to the EA above looked at radial basis, and the fact that the error only goes to zero when the data samples are getting dense. But Aggregate Industries’ consultants also claim to have employed kriging. They will therefore know that kriging comes with an advantage, an advantage widely known about. Here are four sources – all saying much the same thing:
For example:
The figure below, created in 3D analyst, illustrates variation in kriging predictions, showing the range of prediction error (sticks) over kriging predictions (surface). A stick’s color changes according to the prediction error value. As a rule, prediction errors are larger in areas with a small number of samples.
So why, given that Aggregate Industries’ consultants would have known the scale of the potential errors across the site, did they chose not to provide this information to the EA when asked? Why instead, did they chose to mislead, by just comparing the results of two different modelling methods? What are we to infer? That the kriging errors are of such a magnitude to make the MWWT contours look absurd?
But there’s more. Modelling – despite the impression given by Aggregate Industries’ consultants – is not an exact science. For kriging:
Because the output is a geostatistical layer, it’s dynamic, meaning you can change its output type as prediction, errors of prediction, probability or quantile. Or you can even go back into the geostatistical layer and change the parameters if you don’t like the optimized output.
Is "change the parameters if you don’t like the optimized output" ringing any warning bells?
In other words, not only would Aggregate Industries’ consultants have seen where their predictions were poor, they would have also produced a variety of interpolated surfaces with a range of different predictions. And yet what do we have? We have supposedly bright people quoting differences between one guesstimated surface and another guesstimated surface to cm precision, e.g.
the largest positive value (at the northeast boundary of the extraction area) is 0.08m
This is clearly nonsense. Beyond the locations of the piezometers, these models can hardly predict reality to the nearest metre, let alone cm. And of course, to prove this, in 2018 the MWWT failed in four locations, by almost 1.6 m in one location and almost 1.3 m in another.
Has the EA been persuaded by the arguments put forward by Aggregate Industries’ consultants, following their earlier (mis)understanding of tolerances? It would appear so. In an email from the EA in June 2018:
We have reviewed the document and we are satisfied that it answers our questions about the derivation of the Maximum Winter Water Table grid.However, the EA has been spurred into calling for daily groundwater monitoring – not weekly as previously suggested by Aggregate Industries. The EA has also now requested:
that groundwater levels in the piezometers constructed in 2016 and 2017 are estimated for the 'high water table' period in 2013/14. This 'hindcasting' will provide an indication of groundwater levels in these piezometers prior to their construction and improve confidence in the MWWT grid in this area of the proposed quarry. Correlation of the contemporaneous datasets from 2016 to the present should assist in this process.
Can anybody trust Aggregate Industries’ consultants to do that properly – based on their reluctance to come clean on tolerance levels? Any upwards movement of the MWWT loses sand and gravel resource – aka profit – for its client. As far as Aggregate Industries is concerned:
AIUK would be willing to do this if permission is granted.
In other words, after determination, away from prying eyes. We shouldn't be surprised by Aggregate Industries' attitude anymore, not after AI put a stop to public scrutiny of groundwater data for Straitgate application.
In summary, it's simple: To allow for the uncertainty in Aggregate Industries' predictions, an unquarried buffer should remain above the guesstimated MWWT. Prof Brassington argued "an unquarried buffer of at least 3 m is left above the maximum water table to minimise the negative impacts."
