Irrigation Effects on Hydro-Climatic Change: Basin-Wise Water Balance-Constrained Quantification and Cross-Regional Comparison

Abstract

Hydro-climatic changes driven by human land and water use, including water use for irrigation, may be difficult to distinguish from the effects of global, natural and anthropogenic climate change. This paper quantifies and compares the hydro-climatic change effects of irrigation using a data-driven, basin-wise quantification approach in two different irrigated world regions: the Aral Sea drainage basin in Central Asia and the Indian Mahanadi River Basin draining into the Bay of Bengal. Results show that irrigation-driven changes in evapotranspiration and latent heat fluxes and associated temperature changes at the land surface may be greater in regions with small relative irrigation impacts on water availability in the landscape (here represented by the Mahanadi River Basin) than in regions with severe such impacts (here represented by the Aral region). Different perspectives on the continental part of Earth’s hydrological cycle may thus imply different importance assessments of various drivers and impacts of hydro-climatic change. Regardless of perspective, however, actual basin-wise water balance constraints should be accounted to realistically understand and accurately quantify continental water change.

Appendix: Significance Testing

The null hypothesis tested is that there has been no change in the long-term average values of investigated variables from the time period in the beginning to that in the end of the twentieth century, with the periods being those listed in Table 1 for each region. The null hypothesis is expressed as $\mu_{\text{b}} = \mu_{\text{e}} = \mu$, where $\mu_{\text{b}}$ and $\mu_{\text{e}}$ are the average values of each investigated variable in the beginning and the end of the twentieth century, respectively. The hypothesized same average value μ for the two periods is estimated from available data for the end-of-century period, that is, as $\mu = \mu_{\text{e}}$. The alternative hypothesis that there is significant change in the long-term average values between the two averaging time periods is expressed as $\mu_{\text{b}} \ne \mu_{\text{e}}$.

The standard normal test variable is as follows:

$$ z = \frac{{\left| {\overline{x} - \mu } \right|}}{\sigma /\sqrt n } $$

which is normally distributed with mean 0 (when the null hypothesis is true) and variance σ ² /n, where σ is the standard deviation of each investigated variable, estimated consistently with μ from available data for the end-of-century period with n being the sample size (number of years with data in that period), $\overline{x}$ is the sample mean value in the beginning-of-century time period and $\left| {\overline{x} - \mu } \right|$ is the absolute value of the difference between $\overline{x}$ and μ. The value of the standard normal test variable z is computed and listed for different investigated variables in Table 3, along with the confidence level (p) at which the null hypothesis of no change is rejected, and hence, the change in variable average value indicated by $\left| {\overline{x} - \mu } \right|$ is significant.

Table 3 Parameters for testing the null hypothesis of no change, with notation as explained in the Appendix text, and with $\left| {x^{*} - \mu } \right|$ being the absolute value of the (positive and negative) limits of the 80 % confidence interval ($\frac{{\left| {x^{*} - \mu } \right|}}{\sigma /\sqrt n } = 1.282$) around μ