Gains:
- Ability to recognize and manage ethical risks such as farmer data ownership, privacy and algorithmic bias
- Ability to combine all links of the module in a single seasonal end-to-end workflow
- Establishing in a quality system that AI output does not replace competent agronomist approval in safety-critical agricultural decisions
Module Exam
1. What are the anchors of the 'three anchors' validation applied before incorporating AI output into a decision in agricultural engineering?
- A) Size range, agronomic plausibility and field evidence ✔
- B) Popularity of the model, length of answer and language fluency
- C) Price, speed and user interface
- D) Brand credibility, internet connection and file format
Description: Three anchors; order of magnitude, which checks whether the result is within the expected power of 10, agronomic reasonableness, which looks at the consistency of the output with crop and climate, and field evidence, which provides confirmation in the field rather than core/sample. The brand of the source, the popularity of the model, or the fluency of the text are not verification criteria.
2. What does NDVI (Normalized Difference Vegetation Index) basically measure?
- A) Determine the exact moisture percentage of the soil in centimeters
- B) Density and strength of green, vibrant vegetation ✔
- C) Which disease agent is present at the species level?
- D) Final product price after harvest
Description: NDVI measures the density and strength of green, photosynthetic living vegetation by normalizing the difference between near-infrared (NIR) and red band reflectances. Since healthy plants reflect NIR strongly and absorb red intensely, NDVI increases; The value decreases in stressful or sparse cover. It does not directly measure yield, soil moisture or disease type; these require additional data.
3. A satellite image shows an area with a low NDVI value. Which of the following is the best step to take before turning this anomaly into an engineering decision?
- A) Apply fungicide to that area immediately
- B) Doubling the fertilizer dose to the area
- C) Confirm the cause by field control, meteorology and phenology ✔
- D) Ignore the anomaly and apply an average to the entire field
Explanation: Low NDVI may be caused by water stress, disease, nutrient deficiency, infrequent emergence, cloud shade or freshly plowed soil. We can only distinguish this uncertainty through field control, meteorology and phenology; The map alone does not tell why. Applying pesticides or fertilizers directly carries the risk of interfering with the wrong cause.
4. What is the most accurate approach when entering farmer field coordinates, yield and cost data into a publicly available AI tool?
- A) Pasting all data as is with its ID and coordinates
- B) Anonymize identifying information and share only necessary technical data ✔
- C) Loading data into a public table without checking it at all
- D) Donating data to model training without permission from the farmer
Disclosure: Farmer data is commercially and personally sensitive; Coordinate, identity and cost information is a matter of property and confidentiality. The correct approach is to anonymize identifying information with representative labels (Plot-A, X coordinate) and share only the technical data required for analysis. Sharing the data as it is or opening it to education without permission poses ethical and legal risks.
5. The soil analysis report reads pH as 4.8. What does this value mean agronomically?
- A) Neutral and ideal soil for most crops
- B) Strongly acidic; A condition that may restrict food intake and require correction ✔
- C) Strong alkali; a condition requiring liming
- D) It means that the salinity in the soil is extremely high.
Description: pH 4.8 indicates a strongly acidic soil; Under this condition, phosphorus and molybdenum uptake is reduced, the risk of aluminum/manganese toxicity increases, and root development is suppressed in many crops. For most field crops, 6.0-7.0 is the appropriate range. Therefore, a correction such as liming is considered; Acidic soil is not 'ideal' and does not by itself guarantee high yields.
6. AI recommended 60 kg of pure nitrogen per decare for a wheat field; Regional practical and target yield indicates approximately 12-16 kg of pure nitrogen per decare. What is the best behavior?
- A) Implementing the recommendation as is, because the AI calculated
- B) Increasing the dose even more to be safe
- C) See it as an order of magnitude warning and check inputs and local calibration ✔
- D) Dividing the proposal in half without taking into account the soil analysis
Description: The recommendation is approximately four times the expected range; This is an order of magnitude warning. Excessive nitrogen means both economic damage and the risk of depression, disease and nitrate pollution. The correct behavior is to reject the recommendation and check the inputs (unit, yield, target yield, soil N) and local calibration. Blind application causes environmental and agronomic harm.
7. What does 'plant water consumption' (evapotranspiration, ET) used in irrigation decisions mean?
- A) The amount of water entering the soil only through precipitation
- B) Total water loss through evaporation from the soil and transpiration from the plant ✔
- C) Electrical conductivity of the soil
- D) Pressure value in the irrigation pipe
Explanation: Evapotranspiration is the total water loss from the soil to the atmosphere through evaporation and transpiration from the plant. By multiplying the reference ET with the crop coefficient (Kc), the actual water need of the product (ETc) is estimated and irrigation is planned according to this water balance. ET is not rainfall or soil salinity; It represents the outflow side of the water budget.
8. An engineer who saw symptoms on the leaf gave the photo description to AI and received a preliminary diagnosis. Which is mandatory before translating this diagnosis into a pest control decision?
- A) Confirming the agent in the field/laboratory and selecting a licensed and labeled drug ✔
- B) Taking several different medications together, just to be safe
- C) Apply the active ingredient immediately whatever AI says.
- D) Choose the cheapest pesticide and apply it to the entire field
Explanation: Preliminary diagnosis from the image alone is not sufficient due to different factors (fungal, bacterial, viral, nutrient deficiency, abiotic stress) that cause similar symptoms. The agent should be confirmed with a field sample and, if necessary, laboratory diagnosis, and then only a drug licensed for that agent and that product and suitable for its label should be selected. Broad-spectrum pesticide application without verification creates the risk of resistance and residue.
9. What is the purpose of the GDD (Growth Degree Day) concept, which is frequently used in agricultural decision support?
- A) Following the pH value of the soil
- B) Predicting plant development stages and pest emergence by heat accumulation ✔
- C) Measuring irrigation pipe flow rate
- D) Calculating the market price of the product
Description: GDD is the total heat accumulated by subtracting a base temperature from the daily average temperature and is used to estimate the development (phenology) stages of the plant and the emergence time of some pests. Thus, spraying, fertilization and harvest windows can be planned according to heat accumulation rather than calendar. GDD is not soil pH or direct rainfall amount.
10. Which of the following is the most accurate statement for a yield prediction model established with machine learning?
- A) Predicts the future precisely and without error
- B) It is based on past patterns; Must be given with uncertainty range and verified ✔
- C) It always works with the same accuracy in conditions other than training data.
- D) He is the sole final decision maker without agronomic information
Description: A machine learning model is based on patterns it sees in historical data; Confidence decreases when input conditions fall outside the training data (new variety, unusual climate). Therefore, the output should be given with an uncertainty range, not a point, and should be verified with historical yield and benchmarks. The model does not know the future with certainty and does not work with the same accuracy under all conditions.
11. What is the best practice before running an AI-generated Python/pandas analysis code?
- A) Applying the code directly to the entire data set without reading it
- B) Testing with a small sample data with known results and checking formulas and units ✔
- C) Considering that just working without errors is sufficient.
- D) Putting the output into the report without checking it at all
Description: While the AI-generated code may appear correct, it may contain unit, column name, index formula, or filter logic errors. Good practice is to manually run the result against a small sample of data you know, compare it to the expected output, and check formulas like NDVI against the source. Applying code to big data without seeing it moves silent bugs to production.
12. Why is the drug administration notebook critical in traceability records?
- A) Just to get a tax deduction
- B) Provides verifiable record for MRL compliance, certification and batch traceback ✔
- C) Only for comparison with neighboring fields
- D) To prepare advertising brochures
Description: Medication administration book; It documents which product, which active ingredient, in which dose and date it was used, and the waiting period between the last application and harvest. This recording is mandatory for maximum residue limit (MRL) compliance, good agronomy/organic certification, and batch trace back if a problem occurs. It is not kept solely for cost tracking or advertising.
13. An AI tool confidently reported a pest not found in the area as a 'common threat'. Which concept can best explain this situation?
- A) Sensor calibration error
- B) Hallucination: production of information that does not exist in a safe language ✔
- C) Network connection interruption
- D) Lack of rainfall data
Explanation: It is called hallucination when the model produces information that does not actually exist or does not match the context, as if it were real, fluent and confident. The filter of agronomic plausibility (is the region's climate and crop pattern suitable for that pest?) and field evidence catch such errors before signature. This is not a sensor malfunction or unit error.
14. How should the role of artificial intelligence be defined in agricultural decisions critical to safety and quality (e.g. pesticide dosage, fertilizer prescription, irrigation plan)?
- A) The competent authority that alone makes the final decision
- B) Automated system that replaces competent agronomist approval
- C) An assistant who produces drafts and suggestions and leaves the decision to the expert ✔
- D) Accurate source of information that does not require verification
Description: Artificial intelligence; It is an assistant that produces outlines, summarizes data, and proposes hypotheses. The final say in critical decisions belongs to the competent agricultural engineer who is familiar with field observation, local calibration and legislation; AI output does not replace this approval. When it comes to product, environmental and food safety, the responsibility lies with the signing expert.