Using large-scale climate drivers to forecast meteorological drought condition in growing season across the Australian wheatbelt
Graphical abstract
Introduction
Australia is a major global wheat (mainly bread wheat and durum wheat) producer and exporter (GLNC, 2020). It produces 3% of the world's wheat production but accounts for nearly 15% of the world's annual global wheat trade (AEGIC, 2019). Thus, Australian wheat production is of central importance to ensure global wheat supply and food security. Wheat in Australia is mainly grown under rain-fed conditions, thus the wheat industry is sensitive to climate disasters, especially drought. Recurring drought events have resulted in large yield losses in the past decades. For example, the 2018 drought resulted in a 53% reduction in winter crop production in eastern Australia compared to the average of past 20 years (ABARES, 2019). Mitigating the impacts of drought on crop production has been a major research focus as early and reliable drought forecasting can assist farmers to undertake management decisions (e.g. sowing time) which will have financial implications. However, forecasting drought events remains a challenge to the scientific community because their triggers are complex and their features are variable in time and space.
Drought is a normal part of the climate in almost all regions of the world, which results from prolonged absence or shortage of rainfall in comparison with normal years (Zarch et al., 2015; Yu et al., 2020). Australia is the second driest continent on earth because it is located under the subtropical high-pressure belt, which prevents the lift of air required for rain (Williams and Stone, 2009). Australia is also a country prone to drought, as rainfall shows great inter-annual variability throughout the continent. It is well established that the variability of Australian rainfall is linked to the climatic anomalies originating from its surrounding oceans, including the Pacific, Indian, and Southern Oceans (Risbey et al., 2009). These climatic anomalies have been described by several dominant large-scale climate drivers. They include the Indian Ocean Dipole (IOD) which is expressed by the sea surface temperature (SST) gradient between the western and eastern tropical Indian Ocean (Saji and Yamagata, 2003), the El Niño Southern Oscillation (ENSO) defined by the SST anomalies in the central-eastern equatorial Pacific Ocean (McBride and Nicholls, 1983), the Southern Annular Mode (SAM) which refers to the atmospheric circulation in the mid- to high-latitudes of the southern hemisphere on interannual timescales (Marshall, 2003).
The teleconnections between large-scale climate drivers and Australia's rainfall conditions are among the strongest in the world (Kirono et al., 2010). Previous studies have suggested that Australia's extreme hydroclimatic events in the past two decades largely resulted from the anomalies of these climate drivers and their interactions (Xie et al., 2019). For example, the 2002–2009 ‘Millennium drought’ was caused by a combination of long-term upward trend of SAM and prolonged lack of negative IOD phase (Ummenhofer et al., 2009). Following the Millennium drought, extreme wet periods (2010−2011) were found to be mostly driven by a sustained strong La Nina event (Luo et al., 2017), together with a concurrent positive SAM event (Gergis et al., 2012). The 2015 drought event originated from a strong El Niño event but was further enhanced by SAM and IOD variability (L'Heureux et al., 2017; Power and Delage, 2018). The drought events of recent decades in Australia occurred either across the entire continent or in specific regions, resulting in severe adverse impacts on crop production (Dijk et al., 2013). Thus, many studies have focused on empirical relationships between rainfall and large-scale climate drivers in order to provide timely and reliable climate outlooks and mitigate drought impacts (Risbey et al., 2009).
It is generally accepted that large-scale climate drivers have varying influences on rainfall in Australia which is dependent on geographic locations and seasons (Risbey et al., 2009). For example, the Nino 3.4 SST variability has been demonstrated to have a predominant impact on austral autumn rainfall in eastern Australia (van Rensch and Cai, 2014). SAM has been shown to mainly influence southern parts of Australia (Meneghini et al., 2007). Positive phases of SAM tend to result in an increase in spring rainfall in southwest Western Australia and New South Wales (King et al., 2014). There is generally, a negative correlation between IOD and rainfall from June to October in Western Australia, Victoria, South Australia, and southern New South Wales (Steptoe et al., 2018). It should also be noted that these relationships do not act independently, and each climate driver usually accounts for <20% of rainfall variability (Gallant et al., 2012; Risbey et al., 2009). Rainfall conditions throughout the Australian continent are generally the result of the synchronization of these climate drivers (Cleverly et al., 2016).
The teleconnections between large-scale climate drivers and Australian rainfall provide the scientific basis of data-driven drought forecast models (Abbot and Marohasy, 2014). Lagged values of large-scale climate drivers can be adopted as potential predictors of future drought conditions (Mera et al., 2018; Zhang et al., 2019). In the past few decades, a number of data-driven statistical models were developed within different regions of Australia for the forecasting of rainfall or drought conditions in the next month or season. For example, Mekanik et al. (2016) developed eight adaptive network-based fuzzy inference systems models based on lagged values of single or multiple climate drivers (ENSO, IOD, or Inter-decadal Pacific Oscillation) to forecast spring rainfall in Victoria. Their results suggested that the best performing models were able to forecast spring rainfall in a 10-year test period with acceptable correlation coefficients (r = 0.29–0.66) and low errors (RMSE = 10.9–25.0 mm) in 9 locations of Victoria. Abbot and Marohasy (2014) used artificial neural networks and lagged climate variables to forecast monthly rainfall with 1-month lead time in 3 sites of Queensland and achieved r values of >0.55. The two studies also demonstrated that statistical models developed using climate drivers performed better than the state-of-art dynamical model, the Predictive Ocean Atmosphere Model for Australia (POAMA) developed and run by the Australian Bureau of Meteorology. Physics-based dynamical weather forecasting models are normally considered as the mainstream approach by scientific community. However, dynamical models are usually expensive to operate and implement and rely overly on initial conditions. Despite of substantial technological advances and research efforts, dynamical models still have similar performance on seasonal rainfall forecasts in comparison to simple statistical models (Abbot and Marohasy, 2014). Thus, data-driven statistical models based on large-scale climate drivers can still be used to provide seasonal rainfall forecasts effectively in various regions of Australia.
The Australian wheatbelt produces about 25 million tons wheat per year (ABS, 2019). However, wheat yield varies greatly from year to year and is totally constrained by growing season rainfall. Reliable rainfall forecast across the wheatbelt is a critical first step to help growers reduce yield losses from drought. However, from our knowledge, most of previous studies were conducted in specific locations of a certain region, usually a state, and none has ever taken the Australian cropping areas as the domain. Cropping areas are critical areas that will directly benefit from reliable rainfall or drought forecasts. Furthermore, most studies focused on the prediction of next month or next 3 months and few have ever concentrated on the specific growing season (e.g. Apr.-Nov.) of the Australian cropping areas (Sacks et al., 2010). Growing season drought forecasts will provide more preparation time in comparison with seasonal forecasts for farmers to develop drought mitigation strategies. In addition, most previous studies developed linear models considering up to 9-month lead-time impacts of the climate drivers to find the predictability (Hossain et al., 2018). Given the complex effects of climate drivers on Australian rainfall, those linear models might fail to account for the climate drivers with more lead times. The dominant driver in a certain location and its nonlinear effect on rainfall also remain unknown.
Therefore, from the motivation of better understanding the relationship between growing season rainfall and various lagged large-scale climate drivers, the present study investigated growing season rainfall in the Australian wheatbelt as a case study. Instead of forecasting the absolute rainfall amount, we adopted the Standardized Precipitation Index (SPI) (Deo et al., 2017) as the dependent variable, which made the rainfall conditions comparable across different geographic regions. Moreover, it is also a commonly used meteorological drought index evaluating the degree of aridity for a certain location. We implemented a popular machine learning method, RF, as the regression technique to build forecasting models, instead of traditional linear models. The primary objectives of this study are to 1) develop growing season SPI forecasting models using various lagged large-scale climate drivers for each climate station throughout the Australian wheatbelt, 2) identify the best forecasting model for each location and compare their performance across the wheatbelt, 3) quantify the effects of the dominant climate drivers for each station on determining growing season SPI.
Section snippets
Study area
The study area was the Australian wheatbelt (Fig. 1). It is confined to a relatively narrow band of land to the southwest, southeast, and east of the country with a Mediterranean or temperate climate. Actual crop growing regions across the wheatbelt are about 46 million hectares, accounting for 6% of Australia's total land area (ABS, 2019). Most of Australia's agriculture, in particular its grain production, is conducted in the wheatbelt. The most important crop is winter wheat in most regions
Descriptive statistics of SPI and climate indices
Long term growing season SPI was calculated for all climate stations across the wheatbelt. The distribution of SPI values from all climate stations in each year was illustrated using a box shown in Fig. 2. Red and blue shaded areas in Fig. 2 reflect dry and wet periods in the history. There was great inter-annual variability of growing season SPI in the wheatbelt. Years with serious dry conditions occurred frequently over the past hundred years, such as 1902, 1940, 1982, and 2002. Each period
Discussion
Australia is naturally a drought reoccurring continent. Frequent drought disasters in the past hundred years have caused great losses to agricultural production. In this study, we made use of the prognostic features of large-scale climate indices to forecast drought conditions of wheat growing season in the Australian crop belt. The results of model performance show that the bias-corrected RF model could provide acceptable drought forecasts with r > 0.5 and nRMSE < 23% in most parts of the
Conclusions
The study developed a growing season meteorological drought forecasting model for the Australian wheatbelt using machine learning technique and multiple lagged large-scale climate indices. Observed climate data and regional wheat yield records were adopted as reference data to evaluate the performance of the drought forecasting model. Results indicate that oscillation activities from Australia's surrounding oceans could largely account for growing season drought in eastern wheatbelt. The bias
CRediT authorship contribution statement
Puyu Feng:Conceptualization, Methodology, Software, Validation, Formal analysis, Writing - original draft, Visualization.Bin Wang:Conceptualization, Methodology, Writing - review & editing, Supervision.Jing-Jia Luo:Conceptualization, Writing - review & editing.De Li Liu:Conceptualization, Supervision, Project administration, Writing - review & editing.Cathy Waters:Project administration, Writing - review & editing, Funding acquisition.Fei Ji:Conceptualization, Writing - review & editing.Hongyan
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The first author acknowledges the China Scholarship Council (CSC) for the financial support for his Ph.D. study. Facilities for conducting this study were provided by the New South Wales Department of Primary Industries and University of Technology, Sydney. Thanks to Dr. Jian Liu of Northwest A&F University for helping re-calculating SAM.
References (82)
- et al.
Input selection and optimisation for monthly rainfall forecasting in Queensland, Australia, using artificial neural networks
Atmos. Res.
(2014) - et al.
Managing mixed wheat–sheep farms with a seasonal forecast
Agric. Syst.
(2012) - et al.
Optimal N fertiliser management based on a seasonal forecast
Eur. J. Agron.
(2012) - et al.
Modeling flood susceptibility using data-driven approaches of naïve bayes tree, alternating decision tree, and random forest methods
Sci. Total Environ.
(2020) - et al.
Drought forecasting in eastern Australia using multivariate adaptive regression spline, least square support vector machine and M5Tree model
Atmos. Res.
(2017) - et al.
Using CERES-wheat to simulate durum wheat production and phenology in southern Sardinia, Italy
Field Crop Res.
(2011) - et al.
Machine learning-based integration of remotely-sensed drought factors can improve the estimation of agricultural drought in south-eastern Australia
Agric. Syst.
(2019) - et al.
Predictive soil parent material mapping at a regional-scale: a random forest approach
Geoderma
(2014) - et al.
Seasonal climate forecasts for agriculture: towards better understanding and value
Field Crop Res.
(2007) - et al.
Multiple regression and artificial neural network for long-term rainfall forecasting using large scale climate modes
J. Hydrol.
(2013)
Modeling the joint influence of multiple synoptic-scale, climate mode indices on Australian wheat yield using a vine copula-based approach
Eur. J. Agron.
On modeling reference crop evapotranspiration under lack of reliable data over Iran
J. Hydrol.
Drought assessment and monitoring through blending of multi-sensor indices using machine learning approaches for different climate regions
Agric. For. Meteorol.
Machine learning approaches for spatial modeling of agricultural droughts in the south-east region of Queensland Australia
Sci. Total Environ.
Spatiotemporal patterns of maize and winter wheat yields in the United States: predictability and impact from climate oscillations
Agric. For. Meteorol.
A quantile–quantile plot based pattern matching for defect detection
Pattern Recogn. Lett.
Multi-climate mode interactions drive hydrological and vegetation responses to hydroclimatic extremes in Australia
Remote Sens. Environ.
Global synthesis of the impact of droughts on crops’ water-use efficiency (WUE): towards both high WUE and productivity
Agric. Syst.
Meteorological drought forecasting based on a statistical model with machine learning techniques in Shaanxi province, China
Sci. Total Environ.
Australian Bureau of Agricultural and Resource Economics and Sciences, Australian Government
Australian Bureau of Statistics, Australian Government
Australian Export Grains Innovation Centre
Quality and potential utility of ENSO-based forecasts of spring rainfall and wheat yield in south-eastern Australia
Aust. J. Agric. Res.
Random Forest
Mach. Learn.
An asymmetry in the IOD and ENSO teleconnection pathway and its impact on Australian climate
J. Clim.
The importance of interacting climate modes on Australia’s contribution to global carbon cycle extremes
Sci. Rep.
Random forests for classification in ecology
Ecology
The millennium drought in Southeast Australia (2001–2009): natural and human causes and implications for water resources, ecosystems, economy, and society
Water Resour. Res.
Earth System Research Laboratory, National Oceanic and Atmospheric Administration
Impacts of rainfall extremes on wheat yield in semi-arid cropping systems in eastern Australia
Clim. Chang.
Greedy function approximation: a gradient boosting machine
Ann. Stat.
Understanding hydroclimate processes in the Murray-Darling Basin for natural resources management
Hydrol Earth Syst Sc
On the long-term context of the 1997–2009 ‘Big Dry’ in South-Eastern Australia: insights from a 206-year multi-proxy rainfall reconstruction
Clim. Chang.
Grains & Legumes Nutrition Council
The Elements of Statistical Learning
A tripole index for the interdecadal pacific oscillation
Clim Dynam
Planetary-scale atmospheric phenomena associated with the southern oscillation
Mon. Weather Rev.
Long-term seasonal rainfall forecasting: efficiency of linear modelling technique
Environ. Earth Sci.
Long-term seasonal rainfall forecasting using linear and non-linear modelling approaches: a case study for Western Australia
Meteorog. Atmos. Phys.
Analyses of global sea surface temperature 1856–1991
Journal of Geophysical Research: Oceans
Extreme rainfall variability in Australia: patterns, drivers, and predictability
J. Clim.
Cited by (37)
Evaluating the future total water storage change and hydrological drought under climate change over lake basins, East Africa
2024, Journal of Cleaner ProductionShort-term drought Index forecasting for hot and semi-humid climate Regions: A novel empirical Fourier decomposition-based ensemble Deep-Random vector functional link strategy
2024, Computers and Electronics in AgricultureLong-term spatiotemporal characteristics of meteorological drought in China from a three-dimensional (longitude, latitude, time) perspective
2024, International Journal of Applied Earth Observation and GeoinformationEvaluating the predictability of eight Atmospheric-Oceanic signals affecting Iran's Droughts, employing intelligence based and stochastic methods
2023, Advances in Space ResearchCitation Excerpt :The results indicated that the negative phases of NAO and AO simultaneously lead to wetness in both countries, with the difference that this wetness occurs in Turkey during the winter and spring, but in Iran, it is limited to winter. The use of climatic signals and time lags as the input leads to an increase in precipitation and drought prediction accuracy worldwide (Cheng et al., 2020; Feng et al., 2020; Mera et al., 2018; Zhang et al., 2019). Tosunoglu et al. (2018) investigated the spatiotemporal correlation between the atmospheric-oceanic oscillations and meteorological droughts in Turkey.