Modelling the climate, water and socio-economic drivers of farmer exit in the Murray-Darling Basin

Abstract

Absolute farm numbers all over the world have been decreasing over time, and in many countries, this is a source of concern for rural communities. In particular, the Murray-Darling Basin (MDB) in Australia has faced considerable change in the form of increased temperatures and drought severity, reduced irrigation water diversions, declining real agricultural commodity prices, and declining rural community services. This study applies spatial regression modelling at the regional level to assess the impact of weather, economic, and water factors on net farmer number changes over a 20-year period from 1991 to 2011, with climate risk measured using data from 1961 onwards. Our analysis suggests that the direct drivers of farmer exit in local areas were climatic (e.g. increases in maximum temperature and increased drought risk (through decreased long-term precipitation skewness and increased long-term precipitation kurtosis)) and socio-economic (e.g. decreases in commodity output prices, increased urbanisation and higher unemployment). Contrary to the current narrative, changes in irrigation water diversions and water trade movements had no significant impact on MDB farmer exit.

Appendices

Appendix 1

Fig. 3

Annual maximum temperature in the MDB (1910–2018) Source: BOM. Available at http://www.bom.gov.au/climate/change/#tabs=Tracker&tracker=timeseries&tQ =graph%3Dtmax%26area%3Dmdb%26season%3D0112%26ave_yr%3DT

Fig. 4

Annual rainfall in the MDB (1900–2018). Source: BOM. Available at http://www.bom.gov.au/climate/change/#tabs=Tracker&tracker=timeseries&tQ= graph%3Drain%26area%3Dmdb%26season%3D0112%26ave_yr%3DT

Appendix 2

Table 3 Factors affecting farm exit: a synthesis of the literature

Appendix 3. Farmer exit over 5-yearly periods in the MDB

Fig. 5

Net farmer number change in the MDB over 5-year periods, 1991–2011. Source: Specialised data requests from ABS Agricultural Census, 1991, 1996, 2001, 2006 and 2011. Authors’ mapping

Appendix 4. Additional testing on climate risk variables

To further check the robustness of climate conditions’ effects on farm exit, we employed climate risk variables, along with climate condition variables as a sensitivity check. The moment-based approach was used to construct climate risk variables, namely, coefficient of variation (CV), skewness and kurtosis of temperature and rainfall, using average daily maximum temperature and annual total rainfall data over the previous 30 years, following Antle (1983, 1987) and Koundouri et al. (2006). Inclusion of the higher order moments is important as farmers generally exhibit downside risk aversion (Garrido and Zilberman 2008), and increasing occurrence of climate extremes is generally observed (Easterling et al. 2000).

To assess the impacts of the stochastic structure of climate on farm exit decisions, Table 4 presents seven models using various combination of moments of maximum temperature and annual total rainfall while keeping other covariates the same. Model 1 is the main model from Table 2, while models 2–6 use various combinations of the climate risk variables in the SDM regression. Model 7 is the water variables only (namely water diversions and water trade variables) fixed-effects model. The significance and magnitude of our key covariates such as average daily maximum temperature, commodity price index and other covariates are stable, indicating the robustness of our main results. In addition, the skewness models show that an increase of skewness of total rainfall significantly increases farm numbers. Skewness, which measures the degree of asymmetry of a distribution around its mean, is often used to capture the exposure to downside risk in the literature, the increase of which means reduction of downside risk (Kim and Chavas 2003). In our study, the downside risk is the possibility of water deficiency, or drought risk. When the skewness is negative, the long tail towards the left indicates there is certain probability that drought would happen. As the skewness increases, the probability of drought occurrence becomes smaller. Thus, lower rainfall skewness (increased drought risk) reduces farmer numbers. The skewness of rainfall of nearby SLAs also encourages continuance of farming in own SLA. Moreover, rainfall kurtosis was negatively associated with net farmer numbers. The kurtosis refers to the size of the tails on a distribution that measure the number of events outside of the normal range. Therefore, the increase of rainfall kurtosis means the increase of probability of extremely large or small rainfall (namely flood or drought). As such, the increase of rainfall kurtosis would decrease farmer numbers (and hence increase farmer exit). Therefore, the extreme rainfall such as flood or drought plays a more important role on farmers’ decisions of entry or exit than total rainfall in our models. Finally, model 7 provides the results of a water-only regression using fixed-effects modelling.

Table 4 Robustness check of main results using moment-based climate risk measures (n = 996) on MDB net farmer numbers

Common approaches to evaluate the marginal effects associated with independent variables include specifying one unit change or one standard deviation change in the independent variables. Since different units measure the independent variables, we chose to predict farmer number changes as a result of a change in one standard deviation of our independent variable of average daily maximum temperature. Figure 6 presents the observed (1991–2011) and predicted⁹ (2016–2041) farmer number with temperature over time, assuming one standard deviation increase of temperature between 2011 and 2041.¹⁰ It is seen that total MDB farmer numbers would drop from nearly 70,000 in 2011 to just over 31,000 in 2041 due to the assumed change in temperature alone, everything else held constant. While the change is dramatic, it should not be taken at face value as all other factors held constant here will actually vary in complicated ways to jointly affect farm number dynamics.

Of course, these numbers do not account for farmer adaptation, nor does it mean that there will be a consequent decrease in agricultural production and value, as it is known that over time farmers have and will adapt to changing circumstances, but nevertheless they do indicate that economic and temperature factors will continue to decrease absolute farmer numbers to some extent.

Fig. 6

Observed (1991–2011) and predicted (2016–2041) MDB farmer numbers with one standard deviation change of maximum mean temperature in key long-term trends. Note: 95% confidence intervals in shaded area