Jennifer Marohasy
The Australian Institute of Marine Science (AIMS) put out a media release earlier today claiming to have scientifically documented changes in coral cover for the northern section of the Great Barrier Reef, particularly in the vicinity of Lizard Island. Claims include that:
The losses of coral we’ve recorded so far are significant. Despite these losses, coral cover on most reefs is at moderate levels of between 10 and 30%.
In fact, the method used to survey the corals makes it impossible to determine whether the loss is statistically significant or not, and the surveys were only of the reef perimeters, specifically avoiding habitats with higher levels of coral cover including the reef crest. So, claims that coral cover is between 10 and 30% really needs to be qualified.
Many textbooks have been written about the type of survey needed to know if there is a statistically significant change in some aspect of the distribution and abundance of an organism. This could be the population of a city, or even the extent of coral at the Great Barrier Reef. Knowing the distribution and abundance of organisms and how this might change over time is fundamental to understanding the world around us. It is fundamental to the biological sciences.
Rarely is it possible to count every person or every coral. Instead, scientists rely on samples that must be representative of the entire population of interest, and samples must be large enough to generate meaningful data that gives unbiased results.
A population will always include some natural variability and knowing how this is distributed in space and time is important before any conclusions are drawn about the significance of a perceived increase or deterioration, for example, in the corals.
Importantly, in order to be able to quantify statistical significance the data needs to be gathered as an actual count. By which I mean the number of corals needs to be counted per unit area, and given the extent to which coral cover changes naturally with habitat it is important that the survey method distinguishes between the different habitats at a reef.
Of concern, the long-term monitoring being undertaken by the Australian Institute of Marine Science (AIMS) of coral cover is by way of perimeter surveys that do not include, for example, the reef crest that is often the habitat with the highest coral cover. This is the equivalent of claiming to know how the population of a city has changed but never actually going into the central business district (CBD) or even the inner-city suburbs or industrial area. The AIMS surveys that have been reported upon in the media today are surveys of only the reef perimeter, which is the equivalent of skirting around the outer suburbs of a city and from this limited information having an opinion about how the population of a city might have changed.
It is also very important to note that the data is not collected as an actual count. Rather AIMS coral cover data is categorical data, with the survey collected by towing a person behind a boat and noting whether the extent of the coral cover falls into one of the following categories: 0-10%, 10-30%, 30-50%, 50-75%, 75-100%.
These categories have a very large range, up to 25%. It doesn’t matter how many surveys are undertaken, given the way the data is collected it is impossible to calculate average coral cover. It follows that it will be impossible to know how average coral cover changes.
Because the AIMS coral cover data is categorical data it is not possible to calculate a standard error or any other such statistic. Calculating an uncertainty value, for example a standard error, depends on knowing the sample mean. The method used by AIMS precludes this, it is therefore not possible to draw any conclusions about the statistical significance of the recent survey results.
First to comment!
Jennifer, with all respect for your amazing work, consider the following. You say:
I fear this is not true. The sample mean can be calculated. Their method increases the uncertainty, but the amount of the increase in uncertainty can also be calculated.
To prove this, I used the following R computer code. Everything after a # is a comment. A line starting with a “>” is an instruction to the computer. A line starting with a “[1]” is the result of the computer calculation. Here’s the result for 100 uniform random values.
Because of the uneven distribution of the bins, there will be a larger error in the data from 50% to 100% than in the values from 0% to 50%. However, even with only 100 samples, the difference is tiny.
The sample uncertainty can be estimated either by the bootstrap method or by monte carlo simulations as I’ve done above, but using the actual data.
My best to you, keep up the excellent work.
w.
Thanks Willis.
I have been reading and rereading the valuable work you did just ten years ago on thunderstorms, published at this excellent blog site, and I have been meaning to email you requesting a copy of the paper mentioned at this link, the technical paper is behind a paywall, https://wattsupwiththat.com/2010/07/24/willis-publishes-his-thermostat-hypothesis-paper/
If you could email it to me jenn at jennifermarohasy.com :-).
Regarding the calculation of uncertainty values from categorical data: I appreciate the code that you have provided here, and I understand the power of statistics even for generating an average world temperature from remodelled values.
But it doesn’t pass what I could consider to be proper ecological methods, it violates too many assumptions.
One of my favourite books is ‘Ecological Methods’ by TRE Southwood. It was published long ago, like so much of the good stuff. My edition was published in 1966. :-).
Thanks for getting the conversation going, and for reliably being a contrarian.
The statistical technique is called ordinal regression using “Anderson’s [1] grouped continuous response variable”. See my article with Myron Zalucki in the Encyclopedia of Environmetrics on defoliation where you can substitute percentage coral cover for percentage defoliation classes without loss of generality. Using the technique one can estimate a regression that relates mean percent cover to predictor variable(s) or simply estimate a mean and standard deviation of the underlying continuous percent cover distribution. Although it is a rapid assessment method there are important caveats such as allocation of a subset of the sample views to the wrong (usually adjacent) ordinal class, potential observer bias in allocations over the sample giving bias in parameter estimates, and managing between-observer variation that inflates uncertainty of estimates (see below).
From; Defoliation, Steven G. Candy & Myron P. Zalucki,Volume 1, pp 479–484,in
Encyclopedia of Environmetrics,(ISBN 0471 899976),Edited by Abdel H. El-Shaarawi and Walter W. Piegorsch, John Wiley & Sons, Ltd, Chichester, 2002.
“Per cent defoliation classes correspond to Anderson’s [1] grouped continuous response variable, while severity scores are an assessed ordered response. The midpoints of the per cent defoliation classes or the ordinal severity scores can be used as the response variable in standard analysis of variance (ANOVA)/regression analyses. However, this
approach may result in a seriously underestimated error variance since it ignores the implicit variation in actual defoliation levels. A more statistically rigorous approach is to consider as the response variable the counts of sample units (e.g. leaves, whole plants) in
each ordinal class for each given value of the predictor variables. Statistical models are available that treat the class counts as arising from a multinomial sample with ‘cut-points’ which can be estimated along with other model parameters. Such models are referred
to as ordinal regression (see Ordinal response) or threshold models [38, 45].”
Important caveat relevant to percentage coral cover ordered categorical data:
“For visually assigned per cent defoliation classes the nominal values for the
cut-points are the upper limits of the classes, and these can be used as initial values in the estimation of the ‘true’ values. Retaining the nominal values of the cut-points may be adequate if these values have been validated by direct measurement using leaf area meters (e.g. [43]). For operational disease loss assessment using subjective severity scores, training of assessors is essential to achieve accurate and consistent
estimates [40].”
I would be concerned with systematic error in deciding which bin was assigned to a given observation.
There is the issue that the abundance of the corals varies consistently between different habitat types from consistently about 15% cover in say the back lagoon to say 100% cover at the reef crest and sometimes down a reef front. With a perimeter survey they completely cut across the patterns so obvious if you think first in terms of within and between habitat variability. :-).
Willis,
Here is my interpretation of uncertainty in the mean of the categories. Jennifer has said the ranges of the different categories.
0-10%, 10-30%, 30-50%, 50-75%, 75-100%
The range in each category is:
±5, ±10, ±10, ±12.5, ±12.5
For grins, let’s assume the 1σ values are 34% (68% / 2) of the range:
±2, ±3, ±3, ±4, ±4
The combined uncertainty would be:
√(2² + 3² + 3² + 4² + 4² ) = ±7
This would indicate that any mean calculated from the range values would have an uncertainty of ±7. That is, “mean value” ±7.
I think your calculation only gives the SEM, which indicates how accurate the mean value is. It is not the dispersion of the measurement values that can be attributed to the mean which is the measurement uncertainty. I would also say that what I have calculated is purposely small.
Without a calibration study to determine how accurate each bin estimate is, there is no good way to truly estimate the actual Type B uncertainty value that should be used.
Thank you for this thoughtful reply.
Calculate the mean of the sample set, ok. Calculating the mean of the reef given the nature of the sample set, I do not see that happening.
The value of the mean of the sample set is irrelevant since it is not a full coverage of the reef and any conclusions drawn will likewise be inconclusive.
Yes. The first thing that I was taught in zoology at uni in the 1980s was to get some idea of the distribution of the organisms of interest within and between different habitat types. Then I spent 7 years working as a field biologist in different parts of Africa and I honed that skill, as my job was often to collect a maximum number of specimens of a particular insect species always within a limited amount of time. 🙂
We’re all aware by now that these AIMS studies of GBR coral status are written for unvetted publication in The Guardian.
(Clarification – I apologize unreservedly for saying that the AIMS reports are unvetted by The Guardian.
Upon reflection, I’m absolutely positive that The Guardian reporters actually dictate the report to the AIMS scribes.)
Strong opinion, what is the factual basis for this assertion.
As an outsider, I am looking to see the point of all this measurement and striving for calculated results.
If climate boiling fanatics claim reef damage directly connected to global temperature rise, is there not some simpler sampling method available.
Coral polyps are unlikely to be all living in little houses. Do they not swim about at some stage of the life cycle.
What is wrong with taking water samples and counting the live polyps – proof of life can then be published and the certainty gained by the observers allows them to be adamant and persistent in all contradiction of false claims.
If the coverage measurements are determined by “towing a person behind a boat”, I wonder what verification methods are employed to ensure that the person being towed behind the boat is accurately reporting the corral coverage. Perhaps they are just not very adept at estimating coverage percentage? Or, perhaps they have a systemic bias in believing that the coverage of the reef is reducing, which may lead them to (consciously or unconsciously) lowball the coverage measurements. A visual observation requires an entirely disinterested, though attentive, observer who has no interest in one outcome over another. The people who are employed to perform these surveys are, I would strongly suspect, very much on board with the idea that the Great Barrier Reef is under threat. There is also the conflict of interest in that a finding that the reef is not in danger could put their ongoing funding in danger.
Is there not some technological method that could empirically determine corral coverage? I know aerial surveys are problematic due to limited depth visibility, but perhaps some sort of hyperspectral photography, towed under water at a known depth and speed could be used to capture coral coverage and determine it in an empirical and auditable way.
I don’t think anyone is worried about corrals.
O.K. ?
At least no one has to worry about the Clantons anymore 🙂
The Golden ones have a really good buffet.
Yes, but they will get you a Tombstone much more quickly.
Yep.
If you’re engaged & paid to find white or algae covered coral patches, working where & when there’s no one else anywhere around within cooee, you’d have to be a piss-poor investigator if you couldn’t come up with some.
I have wondered about possible technological operations partly because there are an increasing number of diver type surveys, experts or not. Community health has also changed from pollution to various parameters which seem to require some subjective decisions and descriptive field studies have given way to at least some degree of statistical analyses. There are interesting fish comparisons on both natural and artificial reefs where some have used poisons which greatly changes analysis since it produces many more numbers of cryptic species. There is also the devil of lack of work at night, sometimes acoustics has been used there.
Can’t comment specifically on coral reefs and not familiar with Southwood’s Ecological Methods, but based on working with a few good statisticians who dealt with different assumptions have found these useful.
Hairston, N . G . 1989. Ecological Experiments, Purpose, Design, Execution. Cambridge University Press, Oxford.
Peters, R. H. 1991. A Critique for Ecology. Cambridge Univ. Press. 366pp.
Hurlbert, S . H . 1984. Pseudoreplication and the design of ecological field experiments. Ecological Monographs 54:187-211.
As well as differentiating between beige (most coral colour) and white (lacking zooxanthellae).
‘Scientifically documented’? The surveyors must have been joking. There was nothing scientific about this survey the way it was described. How much of the reef was actually covered, relative to the entire reef (?)… about 1E-10%? The Great Barrier Reef is bloody big. And towing some individual behind a boat? You must be joking.
A thousand small drones with cameras could have done the job so much better.
Elon’s Space-X would be a good choice to do GBR surveys.
Put up a low-orbit SE>NW travel satellite with top-shelf glare-filtering powerful zoom cameras and just keep recording.
CoP29 is raising $trillions for saving the planet I hear.
(well, vague promises anyway 🙁 )
Ahhh . . . who or what distills the video data and GPS location data from those “thousand small drones” down to usable numbers of % coverage by location”?
Who fuels/recharges and flies those “thousand” small drones off of, and back on to, mobile marine vessels at sea, since the distance from Australian coast to nearest GBR border varies from 16 km to 160 km?
As always, the devil (and costs) are in the details.
So is the dude running the boat wearing a mask for some legit reason? Or, considering it was 2021, is it just Covid virtue signalling?
ha ha. and he probably is trying to not get sunburnt. :-).
Not funny, Jenn. 🤠 I wore a wide brimmed hat (either the Gulf of Mexico or the Pacific) and my neck and lower part of my face got burnt, that I do easily. I wasn’t warned.
It can be brutal! When I’m going to be over water for the day, I will put on a layer of Butyl Methoxydibenzoylmethane etc. sunscreen (Banana Boat is the brand), then a thick layer of a zinc-based screen (Surf Mud is the brand) over the top, and on top of those two layers I will wrap a very large scarf around my face leaving just a slit for my eyes before tucking it under and/or over my hat. :-). As a kid I used to get burnt all the time. Chemical peels will remove damaged skin from years ago. :-).
I thought sunburn was a requirement down under.
I remember being so burnt as a kid one day that I nearly passed out … after sailing all day without any protection. It can be incredibly painful and debilitating, bad sunburn.
None can escape the dooming no matter how small the polyps-
Worrying sign for Great Barrier Reef
Although results from the full assessment of the Marine Park won’t be available until mid-2025, the ongoing impact of bleaching on the Reef is a sign of what may come in the next few months.
“Its resilience is being severely tested,” said Dr Manuel Gonzalez Rivero.
It’s not just Australia feeling the brunt; the ongoing impact of climate change threatens “reefs around the world”.
For the love of GOD! Counting corals has not, is not and will not
ever make the slightest difference to the numbers, health or distribution of corals anywhere in the world ever.
The only thing it DOES do is employ people who like blathering about something they can never have control over.
Just leave the damn things alone. They will be fine. The fact that I am helping pay these peoples wages is beyond irritating!
According to Google’s AI bot:
“Corals have existed on Earth for over 500 million years, with the oldest known corals dating back to the Cambrian period. However, the modern reef-building corals that we see today have only been around for about 240 million years.”
To put those 240 million years of existence into perspective, the last time Earth experienced a “hothouse” condition (no ice at either pole) was during the Paleocene-Eocene Thermal Maximum (PETM), which occurred approximately 56 million years ago. And Earth entered the current Quaternary ice age some 2.6 million years ago, with tens of glacial-interglacial cycles occurring since then. And modern reef-building corals survived all of that just fine.
As you imply, and to paraphrase a famous line from George Carlin, the comedian, “Besides, there is nothing wrong with the corals, nothing wrong with the corals. The corals are fine; the people are f***** !”
(ref: https://genius.com/George-carlin-the-planet-is-fine-annotated )
The GBR did not exist 25000 years ago and only came into existence after the Younger Dryas when the global melt occurred and sea level off the coast of QLD became high enough to allow coral growth.
The greatest threat to the GBR is decreasing sea levels not increasing OHC, which humans have nothing to do with anyway.
The sad thing is Jennifer’s work will be ignored by the MSM. The only thing which will curtail the global boiling madness and grift is if the MSM turns or is bought out by a sceptical billionaire. Elon? Gina?
I didn’t know the discussion about corals was restricted to be only about Australia’s Great Barrier Reef.
There are many coral reefs in Australia other than the GBR, including:
— Ningaloo Reef, a 260 km long World Heritage-listed reef on the west coast of Australia
— Shark Bay, a World Heritage-listed reef in Western Australia with ancient stromatolites, seagrass beds, and dugongs,
— Lord Howe Island, a World Heritage-listed reef 600 km off the coast of New South Wales having the southernmost-located coral in the world.
— Coral reefs off Cocos-Keeling, Christmas Island, and Norfolk Islands.
Hmmm . . . who knew . . . I thought that climate alarmist today were generally concerned about global seal level rise (SLR), not decrease.
Let’s user their coral counting method and claim Detroit is an abandoned city.
Biggest coral ‘eva’ found recently in South West Pacific. Super.
Let’s see . . . the description underneath the Australian Institute of Marine Science “tweet”-with-attached-photo near the top of the above article states:
“. . . the adjacent shallower reef crest that at John Brewer Reef had more than 100% coral cover at the time of this tweet.”
More than 100% coral cover? . . . now that’s record setting!
Yep. On the reef crest you will often find the corals packed in, and one on top of each other; like a bustling metropolis. if you have corals growing over each other and coral cover is normally calculated on a % area basis, then we have more than 100%. :-).
Jennifer, c’mon . . . while I do appreciate your article and its main message that I agree with, this is not scientifically accurate: one cannot have more than 100% of areal coverage.
In fact, you state in your third paragraph above the last photo in your above article:
“Rather AIMS coral cover data is categorical data, with the survey collected by towing a person behind a boat and noting whether the extent of the coral cover falls into one of the following categories: 0-10%, 10-30%, 30-50%, 50-75%, 75-100%.” It seems that data collection methodology does not allow for coral coverage greater than 100%.
The same reasoning applies to cloud coverage, where we don’t exceed “completely overcast” skies, never getting reports of, say, 135% overcast conditions!
Fair enough. Thanks for caring.