Data Reliability

Verification is important.

By Marvin F. Glotfelty, RG

Groundwater professionals use a variety of tools that enable us to do both simple things like turn a wrench to tighten a bolt and complex things such as calibrating a numeric computer model to simulate a regional aquifer system.

These various tools allow us to perform physical tasks, complete analytical processes, and collect required data. Collection of data is almost invariably needed, whether we’re designing a relatively simple residential water well or conducting a complex regional hydrogeologic analysis.

Proper data collection and analysis are vital for us to make good decisions and draw meaningful conclusions that will support the success of our projects. While there are numerous pitfalls with the analysis portion of any project, we should not overlook the importance of collecting reliable and relevant data upon which our analyses will rely. Therefore, we are duty-bound to authenticate and verify our data prior to its evaluation.

Much of the data upon which we rely is routinely double-checked in accordance with established protocol. For example, water quality reports from analytical laboratories typically include a quality assurance/quality control (QA/QC) section that provides verification of the laboratory’s procedures and documents the accuracy and reliability of the results.

In addition, there is often sampling protocol that are implemented to confirm and verify the integrity of the water samples that have been collected, such as trip blanks, equipment blanks, chain-of-custody documentation, etc.

However, such data verification may not be available in many situations. When necessary, uncertain information can be appropriately used as an assumption to evaluate the non-unique outcome of potential conditions or even probable conditions that may occur but remain unverified. Uncertain information, though, does not rise to the standard of reliable and verified data.

Dealing with Uncertainties

In the groundwater profession, we often need to account for uncertainties and incomplete information by applying assumptions. However, we should maintain a clear division between those assumptions (uncertain information) and actual data (verified information) in our minds, as well as in our communications with clients, regulatory agencies, and others.

Some common sources of unverified data that are encountered in the groundwater industry are inaccurate reports from various sources, specific capacity values, and in some cases, information generated by artificial intelligence.

Proper data collection and analysis are vital for us to make good decisions and draw meaningful conclusions that will support the success of our projects.

Inaccurate Reports

We occasionally encounter erroneous records from a variety of sources, sometimes including public information clearinghouses upon which we rely for groundwater data.

I experienced this years ago while I was undertaking a groundwater supply study of an area about 100 square miles. I was compiling and reviewing driller’s logs for the study area from the files of the Arizona Department of Water Resources (ADWR) where such information is cataloged in my state.

I noticed that driller’s log forms for multiple wells that had been submitted by one driller were identical—with the exact information except for differing dates, even including the same smudges along the margins on each submittal form.

The well locations spanned dozens of miles across which the geology and subsurface hydrology were quite variable, yet each report indicated the exact same geology, water table depth, etc. Thus, it appeared that this driller had “dog housed” the reports and did not take the time to provide actual site-specific information.

This is both unfortunate and rare. During my career of more than 40 years, I’ve found that water well drillers—as a group—are some of the most conscientious and ethical professionals I’ve ever encountered. However, there are always exceptions and outliers, and it only takes one or two bad players out of the multitude of well drillers (or hydrogeologists, or any other category of groundwater professional) to blemish the quality of an important public database.

As with essentially all states, ADWR does not have the financial resources to double-check or field-verify all the submittals they receive for every water well, so such inaccurate records do exist within public files, albeit infrequently.

Figure 1. Example of step-rate pumping test data, showing decreased specific capacity values at higher pumping rates.

Specific Capacity Values

A common source of potentially unreliable data is due to the nature of specific capacity values, which makes them vulnerable to misrepresentation of a well’s characteristics. Specific capacity is the ratio of a well’s flow rate (gallons per minute or gpm) to the water-level drawdown (feet) that results from that pumping rate. The specific capacity value is usually reported as gpm/ft, and it provides a good and commonly used indicator of the well’s efficiency.

Trends of specific capacity change that occur over time are good indicators of operational conditions or emerging problems, such as how a new well’s development is progressing, effectiveness of well rehabilitation in an older well, or decline of a well’s efficiency due to scale accumulation or clogging.

The Achilles’ heel of relying on specific capacity trends is that since each specific capacity value is a ratio, it will be sensitive to the well’s flow rate (gpm) and the duration of pumping. At higher flow rates—even in the same well with otherwise unchanged conditions—the specific capacity will generally be less.

This is exemplified by the step-rate pumping test data shown in Figure 1. The well was pumped at five pumping rates (steps) for a four-hour period during each step. Although nothing in the well’s performance was otherwise altered, the well’s specific capacity progressively declined as the pumping rate was increased, so changes in flow rate from 1000 gpm to 2500 gpm resulted in a drop in specific capacity by more than 13 gpm/ft (dropping from 62.2 to 49.1 gpm/ft).

Similarly, the specific capacity is influenced by the duration of well pumping. If a well is pumped for a few hours, the specific capacity will generally be representative of the long-term drawdown that would occur after 24 or more hours of pumping, provided that a hydrologic boundary such as shallow bedrock or a clay formation hasn’t been encountered by the expanding drawdown cone.

However, if a specific capacity value is based on a short pumping period (say 10 to 20 minutes), then the specific capacity may not be representative of the well’s efficiency. This means that if a comparison is made between specific capacity values that were measured after different pumping durations, then the trend of specific capacity values may not accurately reflect any fluctuations (or consistencies) of the well.

Thus, it is critical to verify that reports of specific capacity trends (upward or downward) are based on consistent field measurements that were obtained at similar pumping rates and after similar pumping durations.

Figure 2. Generalized geologic cross-section (mapped by geologists based on verified field data) of the Grand Canyon. (B.F. Gootee, AZ Geological Survey OFR 19-02)

Artificial Intelligence

AI has emerged as one of the most important and commonly used sources of information and data analysis. In our everyday life, we regularly encounter sources of information or analyses that have been impacted by AI. For example, both Siri (Apple’s intelligent assistant) and Alexa (Amazon’s intelligent assistant) utilize AI to comprehend and process voice commands, and self-driving cars such as Waymo use AI to process the data from cameras and sensors on the vehicle to provide for safety and accident prevention.

Figure 3. Inaccurate AI-generated images that were produced in response to the query: “Make a geologic cross-section of the Grand Canyon.”

Within our groundwater industry, I’ve been told by groundwater modelers with whom I work that AI is a useful tool for such things as developing algorithms to manipulate or filter information within large database files. I’ve also encountered some groundwater professionals who periodically rely on AI to provide narrative or content ideas when they are writing technical memos or reports. And please know that as an old-school guy, I can attest this column was written entirely by me with no AI contribution!

Interaction with AI has become somewhat effortless due to automated software applications (chatbots) that have been developed to mimic human conversation. These chatbot programs include such versions as Chat GPT, Microsoft Copilot, Jasper AI, Zoho Analytics, Google Cloud AI, Google Gemini, IBM Watson, and others.

Chatbots provide human-like conversational responses that facilitate back-and-forth interaction, which allows the user to refine his and her line of questioning so that the best possible response will be provided from AI.

While AI is a useful and powerful tool, it is just that: a tool that enables us to enhance and manipulate data sources. AI is not an actual source of measured data or professional experience.

Conventional computer software may sometimes generate inaccurate results if there’s a flaw (bug) in its code, but AI works differently than conventional software. AI uses powerful search engines to access a multitude of data and reports that are available on the internet and social media, and then based on those data, it generates a prediction as to what the response to the query should be.

Unfortunately, AI predictions are sometimes wrong. Inaccuracies with AI-generated results can occur for several reasons, but the most striking source of inaccurate AI responses is probably hallucinations. An AI hallucination is a seemingly plausible response to the query, which is actually false. These hallucinations can involve faulty responses, citations of academic or legal reports that never existed, or even quotations from nonexistent people.

Just this year there have been AI hallucinations rise during legal proceedings in Utah (www.sltrib.com) and New York (www.reuters.com) when attorneys failed to verify the accuracy of legal briefs that had been prepared by their law clerks using AI-generated case law citations. Those lawyers inadvertently submitted legal briefs to the court that referenced nonexistent or misapplied legal case citations. When these fallacies were discovered, those attorneys were sanctioned by the courts.

Such unintended errors can also happen to those of us in the groundwater industry. To avoid such errors, the onus is on us to diligently confirm and verify all the information upon which we rely.

Putting AI to the Test

Figure 4. Schematic diagram of a subduction zone. Source: www.usgs.gov.

A colleague of mine recently put the reliability of AI’s geologic answers to the test. To assess the accuracy of AI responses in geology, he sent queries to a few different chatbots with geologic queries.

The first query to AI was: “Make a geologic cross-section of the Grand Canyon.” The Grand Canyon in Arizona is one of the most thoroughly mapped geologic features on earth, so its cross-sectional profile and stratigraphy are well known, as illustrated in Figure 2. The AI responses from various chatbots (Figure 3) are artistically attractive, but technically completely wrong.

The second query from my colleague to AI was: “Make a geologic cross-section of a tectonic subduction zone.” A tectonic subduction zone (Figure 4) is a geologic feature that results from the convergence of two tectonic plates where the oceanic plate dives under the continental plate and is subducted. Melting of the subducting plate causes bodies of magma (molten rock) in the subsurface to float upward and create volcanic mountains like the Cascades in the Pacific Northwest, the Aleutians in Alaska, or the Andes in South America.

The AI-generated cross-sections of a subduction zone generated via various chatbots (Figure 5) are unquestionably false. The left-hand image in Figure 5 shows the continental plate being subducted beneath the oceanic plate, whereas the opposite occurs as the oceanic plate is actually subducted beneath the continental plate. The right-hand image in Figure 5 is entirely fabricated and bears absolutely no resemblance to tectonic subduction zones as they are currently understood by geologists.

Figure 5. Inaccurate AI-generated images that were produced in response to the query: “Make a geologic cross-section of a tectonic subduction zone.” The image on the left is close, except that it is the oceanic plate that is subducted under the continental plate. The image on the right bears no resemblance to geologic reality.

As we conduct hydrogeologic studies, design water wells, or manage groundwater facilities, we routinely rely on data obtained from other sources rather than from our own direct measurements or experience. This third-party data is often completely valid and reliable, but since we cannot verify its accuracy, we
have the responsibility to confirm and authenticate the data—whether it was obtained via public records, reported specific capacity values, AI-generated information, or other sources.

If we maintain an unwavering determination to confirm and verify all the information we receive from other sources, we can expect our input data to be reasonable and valid. Therefore, we can also expect our analytical results, infrastructure designs, and professional opinions to be accurate and reliable.

Learn The Art of Water Wells

Get The Art of Water Wells by Marvin F. Glotfelty, RG, a 2019 book from NGWA Press that is a comprehensive overview of well systems and delivers practical information applicable to real-world situations. The book is ideal for everyone working in the groundwater field. It provides practical information of water wells—covering everything from site selection to design, drilling methods, economics, and more. Click here to order it, call (800) 551-7379, fax (614) 898-7786, or email customerservice@ngwa.org.

Marvin F. Glotfelty, RG, is the principal hydrogeologist for Clear Creek Associates, a Geo-Logic Associates Co. He is a licensed well driller and registered professional geologist in Arizona, where he has practiced water resources consulting for more than 35 years. He is author of The Art of Water Wells (NGWA Press, 2019) and was the Groundwater Foundation’s 2012 McEllhiney Lecturer. Glotfelty can be reached at mglotfelty@geo-logic.com.