ICAR-IASRI MSc and PhD Crop Forecasting, Weather-Based Models & Non-linear Growth Models: practice questions

1 · Farmers' eye estimates and appraisal-based forecasts · hard

In double sampling, farmers' appraisals average 24.0 q/ha in a large first-phase sample. In a sub-sample, appraisals average 22.0 q/ha and crop-cutting yields average 25.0 q/ha, with a regression slope of yield on appraisal of 0.9. What is the regression estimate of mean yield?

1. 27.0 q/ha
2. 27.3 q/ha
3. 26.8 q/ha
4. 25.0 q/ha

2 · Agro-meteorological and composite models · hard

When does combining forecasts from weather, remote-sensing and biometric models reduce error most?

1. When all three models have exactly the same errors
2. When the sources carry different information and their errors are not highly correlated
3. When the combination weights are chosen to fit the test years
4. When one source is far worse than the other two

3 · Remote-sensing-based yield forecasting with vegetation indices · hard

Which index was designed to reduce the influence of soil background on vegetation indices in sparse canopies?

1. Simple Ratio of red to blue reflectance
2. Land Surface Temperature index
3. Normalised Difference Water Index (NDWI)
4. Soil-Adjusted Vegetation Index (SAVI)

4 · Non-linear growth models: logistic, Gompertz, Monomolecular and Richards · medium

In the Richards model Y = K(1 + b exp(-rt))^(-1/m), which special case is obtained with m = 1?

1. The logistic model
2. The Gompertz model
3. The monomolecular model
4. The exponential model

5 · Non-linear growth models: logistic, Gompertz, Monomolecular and Richards · hard

Why is a fitted asymptote K unreliable when growth data cover only the early, nearly exponential phase?

1. Few data lie near the plateau
2. Early growth data always contain negative values
3. The rate parameter becomes exactly zero in early data
4. The asymptote can only be estimated by linear regression

6 · Fitting non-linear models: iterative estimation and goodness of fit · medium

Why is R-squared a weak criterion for comparing non-linear growth models?

1. It decreases whenever a parameter is added
2. It is always exactly 1 for non-linear models
3. It lacks the usual sum-of-squares decomposition
4. It can only be computed for linear models with two parameters

7 · Crop simulation models and their statistical calibration · easy

What do crop simulation models such as the CERES models in DSSAT compute?

1. A single regression of yield on total rainfall
2. Daily crop growth from weather, soil, management and cultivar data
3. Only the market price of the crop at harvest
4. Area under the crop from satellite images alone

8 · Crop simulation models and their statistical calibration · easy

Which crop simulation model was developed by FAO with a focus on yield response to water?

1. AquaCrop
2. WOFOST
3. InfoCrop
4. CROPGRO

9 · Machine-learning and hybrid forecasting models · medium

A random forest forecasts district yields with very high accuracy under random 10-fold cross-validation, but performs poorly in the next season. What is the most likely reason?

1. The next season had no weather data at all
2. Random forests cannot use weather variables
3. Cross-validation always underestimates accuracy
4. Folds leaked correlated years and districts

10 · Evaluation of forecasts and their use in policy · hard

A government is deciding whether to permit wheat exports using a forecast. Why should the forecast be issued with an uncertainty range?

1. A range is required only when the point forecast is negative
2. Uncertainty is irrelevant once a model is validated
3. Ranges always make forecasts more accurate
4. The costs of over- and under-estimating differ, and decision-makers must weigh the risk of each