Looking for connections, correlations and relationships in the big picture with the aid of advanced statistical methods and data mining can help agriculture to prepare for climate change and other big issues.
Predicting the weather can be difficult enough but how about predicting crop yield depending on weather conditions and many other factors in a changing climate? That sounds nearly impossible – but it is not. Scientific studies from Aarhus University, in agreement with several global studies, show that predictions can become much more accurate and robust by using an ensemble of models compared to using single models.
- Uncertainty in climate impact studies is a major concern for policymakers for making accurate predictions of food security. Part of this uncertainty lies in the crop models that have been widely used for climate assessment studies in agriculture. Our study shows that uncertainty can be reduced by comparing several process-based models (ensembles), says postdoc Behzad Sharif from the Department of Agroecology at Aarhus University.
Global warming is the big game changer and has led to a situation where normal weather conditions are not so normal anymore. Although many efforts are being made to mitigate global warming and its effects, it is crucial that farmers adapt their agricultural practices to the new climate conditions.
It can be a challenge to sift through the myriad of factors that may or may not have an effect on crop yield, some of which are, in turn, affected by climate change.
Scientists can already put numbers on the effects of changes in temperature and rainfall on crop yields. The question for crop growers, advisers and researchers is where else they should set their sights. A better understanding of which climate factors affect crops and how they do so can help the agricultural sector make the right decisions with regard to options such as tilling system, crop variety and pest management.
Data mining of field data
It seems that answers can be found by mining agricultural data using field observations to project crop yield responses. Data mining means sorting through data to identify patterns and establish relationships by analysing data from different perspectives and summarizing it into useful information.
Behzad Sharif has used advanced statistical methods to crunch the numbers and figure out which elements are important to factor into the equations used in the prediction models. To test this, he and his colleagues investigated the sensitivity of yield change predictions of climate input variables, levels of model complexity, and regression techniques using a large dataset of winter oilseed rape.
They used 42 different regression models to mine the dataset that included 689 observations of winter oilseed rape yield from replicated field experiments conducted in 239 sites in Denmark, covering nearly all regions of the country from 1992 to 2013.
Teasing out the relevant factors
Behzad Sharif found that for every 1°C increase in temperature during flowering, there was an increase in yield ranging from 0 to 6 percent, depending on the choice of regression method. He also found that an increase in precipitation during the autumn and winter had an adverse effect on yield. For every 1 mm/day increase in precipitation, the yield decreased by 0 to 4 percent. Soil type was also important for crop yields with lower yields on sandy soils compared to loamy soils. Later sowing was found to result in increased crop yield.
In this way, Behzad Sharif and his colleagues could home in on which climate factors have an impact on the yield of winter oilseed rape. Researchers can then focus their attention on analysing why and how these particular factors act on yield. Perhaps the temperature in May affects pollinators, and perhaps the amount of rainfall during the winter affects the risk of disease in the crop.
- Many factors might be taken into account, but work is needed to improve the models for exploring genotype x management x environmental interactions in climate change studies. Some biotic interactions, in particular with pests and diseases, are notoriously difficult to handle in process-based simulation crop models. Statistical models developed from large datasets could therefore offer useful alternatives to process-based models for predicting crop yield, says Behzad Sharif.
The statistical method that Behzad Sharif has developed can also be applied to data mining of other agricultural production systems. In fact, with effect from February 1, 2017 Behzad Sharif will be working for a commercial company doing data mining for analysis of the entire food value chain.
Read the scientific article Comparison of regression techniques to predict response of oilseed rape yield to variation in climatic conditions in Denmark published in the European Journal of Agronomy.
For more information please contact:
Postdoc Behzad Sharif, Department of Agroecology, email: email@example.com
Professor Jørgen E. Olesen, Department of Agroecology, email: firstname.lastname@example.org, telephone: +45 8715 7778, mobile: +45 4082 1659
Climate-Smart Agri-Food Systems is one of the research areas in which the Department of Agroecology is particularly strong and from which results are delivered in line with national and global societal challenges and goals.