Andries Potgieter

Advances in application of climate prediction in agriculture

Agricultural ecosystems and their associated business and government systems are diverse and varied. They range from farms, to input supply businesses, to marketing and government policy systems, among others. These systems are dynamic and responsive to fluctuations in climate. Skill in climate prediction offers considerable opportunities to managers via its potential to realise system improvements (i.e. increased food production and profit and/or reduced risks). Realising these opportunities, however, is not straightforward as the forecasting skill is imperfect and approaches to applying the existing skill to management issues have not been developed and tested extensively. While there has been much written about impacts of climate variability, there has been relatively little done in relation to applying knowledge of climate predictions to modify actions ahead of likely impacts. However, a considerable body of effort in various parts of the world is now being focused on this issue of applying climate predictions to improve agricultural systems. In this paper, we outline the basis for climate prediction, with emphasis on the El Nino-Southern Oscillation phenomenon, and catalogue experiences at field, national and global scales in applying climate predictions to agriculture. These diverse experiences are synthesised to derive general lessons about approaches to applying climate prediction in agriculture. The case studies have been selected to represent a diversity of agricultural systems and scales of operation. They also represent the on-going activities of some of the key research and development groups in this field around the world. The case studies include applications at field/farm scale to dryland cropping systems in Australia, Zimbabwe, and Argentina. This spectrum covers resource-rich and resource-poor farming with motivations ranging from profit to food security. At national and global scale we consider possible applications of climate prediction in commodity forecasting (wheat in Australia) and examine implications on global wheat trade and price associated with global consequences of climate prediction. In cataloguing these experiences we note some general lessons. Foremost is the value of an interdisciplinary systems approach in connecting disciplinary Knowledge in a manner most suited to decision-makers. This approach often includes scenario analysis based oil simulation with credible models as a key aspect of the learning process. Interaction among researchers, analysts and decision-makers is vital in the development of effective applications all of the players learn. Issues associated with balance between information demand and supply as well as appreciation of awareness limitations of decision-makers, analysts, and scientists are highlighted. It is argued that understanding and communicating decision risks is one of the keys to successful applications of climate prediction. We consider that advances of the future will be made by better connecting agricultural scientists and practitioners with the science of climate prediction. Professions involved in decision making must take a proactive role in the development of climate forecasts if the design and use of climate predictions are to reach their full potential

Spatial and temporal patterns in Australian wheat yield and their relationship with ENSO

Spatial and temporal variability in wheat production in Australia is dominated by rainfall occurrence. The length of historical production records is inadequate, however, to analyse spatial and temporal patterns conclusively. In this study we used modelling and simulation to identify key spatial patterns in Australian wheat yield, identify groups of years in the historical record in which spatial patterns were similar, and examine association of those wheat yield year groups with indicators of the El Nino Southern Oscillation (ENSO). A simple stress index model was trained on 19 years of Australian Bureau of Statistics shire yield data (1975-93). The model was then used to simulate shire yield from 1901 to 1999 for all wheat-producing shires. Principal components analysis was used to determine the dominating spatial relationships in wheat yield among shires. Six major components of spatial variability were found. Five of these represented near spatially independent zones across the Australian wheatbelt that demonstrated coherent temporal (annual) variability in wheat yield. A second orthogonal component was required to explain the temporal variation in New South Wales. The principal component scores were used to identify high- and low-yielding years in each zone. Year type groupings identified in this way were tested for association with indicators of ENSO. Significant associations were found for all zones in the Australian wheatbelt. Associations were as strong or stronger when ENSO indicators preceding the wheat season (April-May phases of the Southern Oscillation Index) were used rather than indicators based on classification during the wheat season. Although this association suggests an obvious role for seasonal climate forecasting in national wheat crop forecasting, the discriminatory power of the ENSO indicators, although significant, was not strong. By examining the historical years forming the wheat yield analog sets within each zone, it may be possible to identify novel climate system or ocean-atmosphere features that may be causal and, hence, most useful in improving seasonal forecasting schemes.

Rainfall Variability at Decadal and Longer Time Scales: Signal or Noise?

Rainfall variability occurs over a wide range of temporal scales. Knowledge and understanding of such variability can lead to improved risk management practices in agricultural and other industries. Analyses of temporal patterns in 100 yr of observed monthly global sea surface temperature and sea level pressure data show that the single most important cause of explainable, terrestrial rainfall variability resides within the El Nino-Southern Oscillation (ENSO) frequency domain (2.5-8.0 yr), followed by a slightly weaker but highly significant decadal signal (9-13 yr), with some evidence of lesser but significant rainfall variability at interclecadal time scales (15-18 yr). Most of the rainfall variability significantly linked to frequencies tower than ENSO occurs in the Australasian region, with smaller effects in North and South America, central and southern Africa, and western Europe. While low-frequency (LF) signals at a decadal frequency are dominant, the variability evident was ENSO-like in all the frequency domains considered. The extent to which such LF variability is (i) predictable and (ii) either part of the overall ENSO variability or caused by independent processes remains an as yet unanswered question. Further progress can only be made through mechanistic studies using a variety of models.

Estimating crop area using seasonal time series of Enhanced Vegetation Index from MODIS satellite imagery

Cereal grain is one of the main export commodities of Australian agriculture. Over the past decade, crop yield forecasts for wheat and sorghum have shown appreciable utility for industry planning at shire, state, and national scales. There is now an increasing drive from industry for more accurate and cost-effective crop production forecasts. In order to generate production estimates, accurate crop area estimates are needed by the end of the cropping season. Multivariate methods for analysing remotely sensed Enhanced Vegetation Index (EVI) from 16-day Moderate Resolution Imaging Spectroradiometer (MODIS) satellite imagery within the cropping period (i.e. April-November) were investigated to estimate crop area for wheat, barley, chickpea, and total winter cropped area for a case study region in NE Australia. Each pixel classification method was trained on ground truth data collected from the study region. Three approaches to pixel classification were examined: (i) cluster analysis of trajectories of EVI values from consecutive multi-date imagery during the crop growth period; (ii) harmonic analysis of the time series (HANTS) of the EVI values; and (iii) principal component analysis (PCA) of the time series of EVI values. Images classified using these three approaches were compared with each other, and with a classification based on the single MODIS image taken at peak EVI. Imagery for the 2003 and 2004 seasons was used to assess the ability of the methods to determine wheat, barley, chickpea, and total cropped area estimates. The accuracy at pixel scale was determined by the percent correct classification metric by contrasting all pixel scale samples with independent pixel observations. At a shire level, aggregated total crop area estimates were compared with surveyed estimates. All multi-temporal methods showed significant overall capability to estimate total winter crop area. There was high accuracy at pixel scale (>98% correct classification) for identifying overall winter cropping. However, discrimination among crops was less accurate. Although the use of single-date EVI data produced high accuracy for estimates of wheat area at shire scale, the result contradicted the poor pixel-scale accuracy associated with this approach, due to fortuitous compensating errors. Further studies are needed to extrapolate the multi-temporal approaches to other geographical areas and to improve the lead time for deriving cropped-area estimates before harvest.

FRAMEWORK FOR FORECASTING THE EXTENT AND SEVERITY OF DROUGHT IN MAIZE IN THE FREE STATE PROVINCE OF SOUTH AFRICA

An effective framework for drought assessment requires a definition of drought severity, a weather-soils database for the relevant region in a geographic information system (GIS), a reliable crop growth model, a method of forecasting daily weather data from the present date till the end of the growing season and a mapping procedure for the graphical representation of a drought situation. The development and main features of such a framework (system) which is already in use in the Free State Province of South Africa, is described. Based upon the phase of the southern oscillation index, it has been applied to quantify and map drought hazard in maize by running maize crop growth models in a GIS. Input and output data for the latter are grouped in 9800 homogeneous natural resource zones. For each, computed maize grain yield forecasts are compared against long-term cumulative probability distribution functions of yield to determine their probabilities of non-exceedence and used to delimit drought severity areas accordingly. The system enjoys wide acceptance and credibility in the province. To date, the results have been well received by a rapidly growing number of users, now totalling 360. Major users are grain merchants, importers and exporters, millers, the provincial government and maize producers. No tests of accuracy of the forecasting system have been possible at this stage because the computation procedures and software have only just been completed. A similar project has, however, yielded promising results.

An integrated, probabilistic model for improved seasonal forecasting of agricultural crop yield under environmental uncertainty

We present a novel forecasting method for generating agricultural crop yield forecasts at the seasonal and regional-scale, integrating agroclimate variables and remotely-sensed indices. The method devises a multivariate statistical model to compute bias and uncertainty in forecasted yield at the Census of Agricultural Region (CAR) scale across the Canadian Prairies. The method uses robust variable-selection to select the best predictors within spatial subregions. Markov-Chain Monte Carlo (MCMC) simulation and random forest-tree machine learning techniques are then integrated to generate sequential forecasts through the growing season. Cross-validation of the model was performed by hindcasting/backcasting it and comparing its forecasts against available historical data (1987-2011) for spring wheat (Triticum aestivum L.). The model was also validated for the 2012 growing season by comparing its forecast skill at the CAR, provincial and Canadian Prairie region scales against available statistical survey data. Mean percent departures between wheat yield forecasted were under-estimated by 1-4 % in mid-season and over-estimated by 1 % at the end of the growing season. This integrated methodology offers a consistent, generalizable approach for sequentially forecasting crop yield at the regional-scale. It provides a statistically robust, yet flexible way to concurrently adjust to data-rich and data-sparse situations, adaptively select different predictors of yield to changing levels of environmental uncertainty, and to update forecasts sequentially so as to incorporate new data as it becomes available. This integrated method also provides additional statistical support for assessing the accuracy and reliability of model-based crop yield forecasts in time and space.

Three Putative Types of El Niño Revealed by Spatial Variability in Impact on Australian Wheat Yield

The El Nino-Southern Oscillation (ENSO) phenomenon significantly impacts rainfall and ensuing crop yields in many parts of the world. In Australia, El Nino events are often associated with severe drought conditions. However, El Nino events differ spatially and temporally in their manifestations and impacts, reducing the relevance of ENSO-based seasonal forecasts. In this analysis, three putative types of El Nino are identified among the 24 occurrences since the beginning of the twentieth century. The three types are based on coherent spatial patterns (footprints) found in the El Nino impact on Australian wheat yield. This bioindicator reveals aligned spatial patterns in rainfall anomalies, indicating linkage to atmospheric drivers. Analysis of the associated ocean-atmosphere dynamics identifies three types of El Nino differing in the timing of onset and location of major ocean temperature and atmospheric pressure anomalies. Potential causal mechanisms associated with these differences in anomaly patterns need to be investigated further using the increasing capabilities of general circulation models. Any improved predictability would be extremely valuable in forecasting effects of individual El Nino events on agricultural systems.

From rainfall to farm incomes—transforming advice for Australian drought policy. I. Development and testing of a bioeconomic modelling system

In this paper we report the development of a bioeconomic modelling system, AgFIRM, designed to help close a relevance gap between climate science and policy in Australia. We do this by making a simple econometric farm income model responsive to seasonal forecasts of crop and pasture growth for the coming season. The key quantitative innovation was the use of multiple and M-quantile regression to calibrate the farm income model, using simulated crop and pasture growth from 2 agroecological models. The results of model testing demonstrated a capability to reliably forecast the direction of movement in Australian farm incomes in July at the beginning of the financial year (July-June). The structure of the model, and the seasonal climate forecasting system used, meant that its predictive accuracy was greatest across Australias cropping regions. In a second paper, Nelson et al. (2007, this issue), we have demonstrated how the bioeconomic modelling system developed here could be used to enhance the value of climate science to Australian drought policy.

Quantification of wheat water-use efficiency at the shire-level in Australia

In eastern Australia, latitudinal gradients in vapour pressure deficit (VPD), mean temperature (T), photosynthetically active radiation (PAR), and fraction of diffuse radiation (FDR) around the critical stage for yield formation affect wheat yield and crop water-use efficiency (WUE = yield per unit evapotranspiration). In this paper we combine our current understanding of these climate factors aggregated in a normalised phototermal coefficient, NPq = (PAR· FDR)/(T · VPD), with a shire-level dynamic model of crop yield and water use to quantify WUE of wheat in 245 shires across Australia. Three measures of WUE were compared: WUE, the ratio of measured yield and modelled evapotranspiration; WUEVPD, i.e. WUE corrected by VPD; and WUENPq, i.e. WUE corrected by NPq. Our aim is to test the hypothesis that WUENPq suits regional comparisons better than WUE or WUEVPD. Actual median yield at the shire level (1975–2000) varied from 0.5 to 2.8 t/ha and the coefficient of variation ranged from 18 to 92%. Modelled median evapotranspiration varied from 106 to 620 mm and it accounted for 42% of the variation in yield among regions. The relationship was non-linear, and yield stabilised at ~2 t/ha for evapotranspiration above 343 mm. There were no associations between WUE and rainfall. The associations were weak (R2 = 0.09) but in the expected direction for WUEVPD, i.e. inverse with seasonal rainfall and direct with off-season rainfall, and strongest for WUENPq (R2 = 0.40).We suggest that the effects of VPD, PAR, FDR, and T, can be integrated to improve the regional quantification of WUE defined in terms of grain yield and seasonal water use.

Determining crop acreage estimates for specific winter crops using shape attributes from sequential MODIS imagery

There are increasing societal and plant industry demands for more accurate, objective and near realtime crop production information to meet both economic and food security concerns. The advent of the Moderate Resolution Imaging Spectroradiometer (MODIS) satellite platform has augmented the capability of satellite-based applications to monitor large agricultural areas at acceptable pixel scale, cost and accuracy. Fitting parametric profiles to growing season vegetation index time series reduces the volume of data and provides simple quantitative parameters that relates to crop phenology (sowing date, flowering). In this study, we modelled various Gaussian profiles to time sequential MODIS enhanced vegetation index (EVI) images over winter crops in Queensland, Australia. Three simple Gaussian models were evaluated in their effectiveness to identify and classify various winter crop types and coverage at both pixel and regional scales across Queenslands main agricultural areas. Equal to or greater than 93% classification accuracies were obtained in determining crop acreage estimates at pixel scale for each of the Gaussian modelled approaches. Significant high to moderate correlations (log-linear transformation) were also obtained for determining total winter crop (R2 = 0.93) areas as well as specific crop acreage for wheat (R2 = 0.86) and barley (R2 = 0.83). Conversely, it was much more difficult to predict chickpea acreage (R2 ≤ 0.26), mainly due to very large uncertainties in survey data. The quantitative approach utilised here further had additional benefits of characterising crop phenology in terms of length of growing season and providing regression diagnostics of how well the fitted profiles matched the EVI time series. The Gaussian curve models utilised here are novel in application and therefore will enhance the use and adoption of remote sensing technologies in targeted agricultural application. With innate simplicity and accuracies comparable to other more convoluted multi-temporal approaches it is a good candidate in determining total and specific crop acreage estimates in future national and global food security frameworks.