Gains:
- Ability to automate environmental data cleaning and analysis tasks with AI-powered Python code
- Ability to verify AI-generated code with unit checking, testing and edge cases
- Ability to apply principles of protecting confidential/personal environmental data and responsible AI use
Module Exam
1. You asked the AI about the discharge limit of a wastewater treatment plant and the model answered '120 mg/L for COD, according to Table 19 of the Water Pollution Control Regulation'. Which is the most correct behavior?
- A) Verifying the value and the referenced table from the official text of the current regulation ✔
- B) Since AI is a current model, writing the value directly to the discharge permit
- C) Just convert the unit from mg/L to g/L and use
- D) Ask the AI 'are you sure?' and be satisfied with the answer 'yes'
Explanation: LLM can produce breakpoints and table/item numbers that are statistically plausible but not real (hallucination). The discharge limit varies depending on the sector, receiving environment and current regulation; Using the value and the referenced table without verifying it from the official regulation text (Official Gazette) will lead to non-compliance and penalty.
2. What would typically be expected if the F/M (nutrient/microorganism) ratio is too low in an activated sludge system?
- A) A situation where the organic load is very high and the sludge remains young
- B) Extremely aged sludge and an operating condition dominated by internal respiration ✔
- C) A situation where ventilation becomes completely unnecessary
- D) Proof that discharge limits are automatically maintained
Explanation: F/M ratio expresses the organic load per mass of microorganisms. Very low F/M is an operating condition where microorganisms starve and switch to internal respiration, the sludge becomes excessively old, and needle formation/dispersed sludge and sedimentation problems can generally be observed. If AI interprets a F/M value, it is necessary to verify it with sludge age, SVI and laboratory data.
3. You used AI to estimate the annual NOx load using the emission factor in a flue gas emissions calculation. What should you do before you can safely report the result?
- A) Round the result up and enter it directly into the emissions inventory
- B) Just correct the colors of the chart and publish the report
- C) Using the result even if CO2 was calculated by mistake instead of NOx
- D) Print the used emission factor, source, unit and activity data and verify the calculation ✔
Explanation: It works with the logic of emission = factor × activity size. AI may use a wrong emission factor (wrong fuel, wrong unit, outdated source). It is necessary to reproduce the calculation by clearly printing the factor used, its source, unit and activity data and comparing it with an official emission factor source (e.g. EMEP/EEA or national guidance).
4. Which of the following is the prioritization order according to the 'waste hierarchy' principle in waste management?
- A) Disposal > recovery > recycling > reuse > prevention
- B) Recycling > disposal > prevention > reuse > recovery
- C) Prevention > reuse > recycling > recovery > disposal ✔
- D) Reuse > disposal > prevention > recycling > recovery
Explanation: The waste hierarchy, from most to least, is: prevention (reduction) > reuse > recycling > (energy) recovery > disposal (landfill). Even if AI produces a draft waste management plan, it must be verified that this sequence and proposed disposal method complies with legislation and waste characterization.
5. AI wrote in a draft EIA report that the 'environmental impact of a project is negligible'. What is the best engineering attitude for this output?
- A) Not writing the impact judgment in the report without verifying it with field data, modeling and expert evaluation ✔
- B) Since AI is neutral, put the statement as is in the conclusion
- C) Just rewrite the sentence in more formal language
- D) Skipping the impact assessment completely and applying directly for permission
Description: Impact assessment in EIA; based on field data, receiving environment sensitivity, cumulative effects and regulatory thresholds. AI can produce a judgment such as 'negligible' without grounding it in data. This should be taken as a hypothesis/sketch; The effect size should not be written as a final judgment in the report without being verified by measurement, modeling and expert evaluation.
6. In a PM10 time series from the continuous monitoring station, the AI flagged a sudden 'overshoot' (above threshold value). Which verification step is the most critical?
- A) Sending the sign directly to the official as an official exceedance notification
- B) Checking raw data, sensor QA/QC flags, calibration and neighbor station data ✔
- C) Smoothing the graph and hiding the overshoot
- D) Raising the threshold and ensuring that the exceedance is not visible
Explanation: Sudden jumps in sensor data may be a real contamination event, but may also be a sensor malfunction, calibration drift, maintenance window or data transmission error. Before accepting the overshoot as real, it is necessary to verify it by looking at the raw data, sensor QA/QC flags, calibration log and neighboring station/meteorology data; otherwise a false alarm or missed event will occur.
7. What do 'Scope 2' emissions mean in corporate carbon footprint calculations?
- A) All indirect emissions resulting from supply chain and product use
- B) Direct emissions from fuel combustion in the facility
- C) Indirect emissions of purchased electricity, steam and heating/cooling ✔
- D) Only emissions from waste disposal
Description: Scope 1 covers direct emissions (fuel combustion in your own facility, vehicle fleet), Scope 2 covers indirect emissions of purchased energy (electricity, steam, heating/cooling), Scope 3 covers other indirect emissions in the value chain (supply, transportation, product use). Although AI produces a draft inventory, the emission factors and scope classification it uses must be validated.
8. When remote sensing gives a high value of NDVI (Normalized Difference Vegetation Index), what does this typically indicate?
- A) Deep and clean water body
- B) Bare rock and soil
- C) High air pollution concentration
- D) Dense and healthy green vegetation ✔
Description: NDVI is calculated as the ratio of the difference of the near infrared and red bands to the sum of them and is usually between -1 and +1. High positive values indicate dense, healthy green vegetation; water and bare soil give low or negative values. Even if the AI makes an NDVI interpretation, cloud, shadow, season, and sensor calibration effects must be verified with ground truth.
9. You asked the AI a question about environmental legislation and the model gave a clear limit value with reference to a specific regulation article. Which is mandatory before using this information in an official application?
- A) Confirming the substance number, limit value and validity status from the current official source ✔
- B) Write the article number given by AI on the application petition as it is.
- C) Just rewrite the sentence in a formal style
- D) Taking the average because AI gives more than one answer
Explanation: LLMs may produce non-existent article numbers, repealed regulations, or incorrect breakpoints and submit them without current date information. Legislative reference and limit value cannot be used in official transactions without being confirmed with the effective date from the current official source (Official Gazette, legislation portal).
10. Which is true before using an AI-generated Python script to clean and trend monitoring station data?
- A) Putting the trend result directly into the report as the code works without errors
- B) Shorten variable names and delete comment lines
- C) Uploading the script with all raw tracking data to a public service
- D) Verify with tests with known results, unit checking and edge cases, and protect data confidentiality ✔
Description: AI code may contain incorrect unit conversion, incorrect missing-data filling, incorrect date parsing, or in edge cases silent bugs. Code; validated with small test data with known results, volume/size checking, and edge cases (missing value, negative read, double record); Additionally, personal/confidential data should not be sent to external services without permission. The result is confirmed by engineering judgment.