Climate change and Australian wheat yield

Abstract

Nicholls replies — Godden et al. state that I proxied non-climate factors “by a linear trend in yields”. I did not and my conclusions were not dependent on the functional form of any relationship between non-climate factors and yield. They suggest that forcing the intercept to zero and fitting with ordinary least-squares biases my estimates. I repeated the analysis without forcing a zero intercept, with no significant effect on the results. The climate-yield change relationship is so strong (see Fig. 2 of Ref. 1) that other forms of fitting produce a result very similar to ordinary least-squares. This relationship dominates the effects of other variables and provides a way to estimate the effect of climate on yield change. Godden et al. believe that detrending with first differences is inappropriate. As I noted, repeating the analysis on residuals from fitted non-linear trends produces very similar results.

Main

Gifford et al. suggest that the temperature-yield relationship simply reflects the effect of rainfall on yield. Although rainfall was related to yield and temperature, the temperature-yield relationship was stronger. Even if the temperature-yield relationship partly reflects the influence of rainfall, there is a direct temperature effect as well. They believe the correlation between minimum temperatures and yield is unlikely to reflect the effect of frost damage. Yet there is a substantial literature showing concern about the effects of frost on Australian wheat yield.

Gifford et al. propose the use of multiple regression of absolute yield on climate variables. Many other factors affect yields. These would confound the use of such a regression of absolute yield on climate variables. De-trending, either through first-differencing or calculating residuals from fitted non-linear trends (I did both), is necessary to avoid this confounding. They suggest that regression on absolute yields would avoid underestimating the impact of the atmospheric CO₂. I did not need to, and did not, estimate the CO₂ impact on yield, but did check that the increasing CO₂ had not confounded my analysis. It had not.

Godden et al. restate my comment that farmers varying their inputs in response to climate might lead to an overestimate of the effect of climate. For instance, in a year with poor yield the farmer might decide not to harvest. Next year, with good climate conditions and good yield, the farmer harvests. The observed increase in yield, from one year to the next, could be attributed to increased harvest effort. However, the ultimate cause of the increased yield is the better climate, even if this is mediated partly by farmers' responses.

The comments of Godden et al. and Gifford et al. do not negate the conclusion that climate trends seem to have led to increased Australian wheat yield. Their comments indicate the difficulties in calculating the precision of the estimated yield increase owing to climate trends. Because of these difficulties, I did not attempt to calculate the precision of the estimate.