Discussion

The results show that increasing irrigation quantity has a significant effect positive effect on yield. They also demonstrate that changing irrigation dates in a deficit irrigation framework can have a strong effect on yield as well.

First, yields from 6 out of the 8 10mm irrigation regimes were deemed significantly higher by the Welch’s t-test higher than the baseline, and all 8 25mm regimes were deemed significantly higher than the baseline. The increase is not uniform across growing seasons, with clear nonlinear behavior related to the strong external influence of the climate, manifesting itself as all intervals converging during the maximum and the minimum yields of the modeled growing seasons. This in turn suggests that while deficit irrigation can increase the minimum yield during drought seasons, it cannot offset drought effectively. That being said, increased irrigation has a clear visual effect on reducing the variance of the mean yield, suggesting that there is a significant mediation effect on the variability of the climate.

While the coding structure used in this report allows one to test a large number of intervals there are several drawbacks to the randomization method used. First, the sample space does not include all possible 2 date combinations: the random baseline used in this report only draws from the set of dates with the first date within the interval (Jan 15, Mar 15) and the second date within the interval (Mar 16, May 15). This may exclude key possible optimum dates such as concentrating watering during the grain filling stages. Further, dates are drawn with replacement, which means that it is possible to obtain repeat dates and thus constrains the number of runs that can be tested at any one time, as the number of repeated values will quickly increase as more dates are picked. This is a rather harmless effect for the visual analysis primarily under consideration here using about 8% of possible 2-date combinations, but it remains a weakness that would need to be addressed for any follow-up.

The Welch’s t-test is a powerful test robust to Type-1 errors, especially among samples with variable variances and sample sizes [Derrick and White, 2016]. However, in this case its application is not entirely appropriate, due to the fact that the intervals contain dates which are also contained in the randomization baseline. Nonetheless, as a first-pass analysis of the significance of the results, it is a useful contribution to the visual analysis. This means that

Yield was the primary output discussed in this report, but the changing irrigation dates and depths had significant effects on the biomass, the canopy cover and the water flux variables. Early irrigation dates disproportionately increased the biomass and canopy cover, and even allowed water to percolate into the ground. At no point was this water recovered through capillary action later in the season, suggesting that it is essentially lost. Further, the increased biomass did not efficiently translate into higher yields, leading to a relative decline in harvest index and poor yield performance compared to later irrigation dates. The top irrigation dates in the intervals and the random baseline all included irrigation within the 110 day interval, and the top results occurred with around the irrigation pair (32,102) suggesting that early irrigation does have a significant positive impact on yield.

This analysis resulted from an attempt to implement the Aquacrop-OSPY package for the Chapter 7.7 Food and Agricultural Organization exercise.