Big data insights into pest spread

Pests and diseases reduce agricultural yields and are an important wildcard in the evaluation of future climate impacts. A unique global record of pests and diseases provides evidence for poleward expansions of their distributions.

Food security depends on our ability to effectively manage crop pests (arthropods and pathogens). Because of the important effects of weather variables such as temperature and precipitation on crop pests, scientists have for some time hypothesized that where climate change results in a more (less) favourable environment for pest establishment, losses to unmanaged pests are likely to increase (decrease)¹. But evidence that ranges have shifted under climate change is often anecdotal, and the availability of long-term data sets of pest occurrence is limited^2,3. In this issue of Nature Climate Change, Bebber and colleagues⁴ present an analysis of decades of reported pest distributions, concluding that pests have moved towards the poles over the past fifty years, in line with expectation under climate change.

One of the interesting aspects of this analysis is its reliance on 'big data'. The data set that Bebber and colleagues⁴ analysed, although not challenging in terms of sheer storage and computational requirements, has been assembled over some time as many, many individuals reported where and when they found particular pests. In their popular book, Mayer-Schönberger and Cukier⁵ discuss three aspects of big data that present challenges for scientists. The first is a shift towards using large amounts of data from different sources, often collected for different purposes. The second is an acceptance of 'messiness', where having large amounts of data may make up for introducing increased sources of variability, and potentially even for introducing bias (more on that later). The third is a willingness to accept correlations as the outcome of analyses, rather than necessarily understanding causation.

Big data issues are not new in the context of mapping the geographic distribution of species⁶. Pest risk mapping generally has as a first step either (1) recording where a pest has been observed and the associated environmental conditions, or (2) obtaining estimates of the effects of environmental variables on a pest in controlled experimental conditions such as growth chambers⁷. Both approaches may err by estimating the environment–risk relationship based on specific factors (such as sampling approaches or pest subpopulations) that are not relevant in other areas where risk will be evaluated. A particular problem with the records of presence, as in the data used by Bebber and colleagues, is that there tends to be little information about the uncertainty associated with these records, in contrast to some more standard approaches to meta-analysis.

Bebber and colleagues⁴ take geographic mapping a step further by evaluating how the records of pest distributions have changed over time. An important challenge in interpreting the records in this way is to convincingly demonstrate that sources of bias have been adequately accounted for, or that the analysis is conservative with regard to likely sources of bias. Bebber and colleagues use the assumption that the scientific infrastructure in tropical nations has lagged that in temperate areas to postulate that more limited sampling effort in the tropics may result in a bias towards reporting pests in temperate countries earlier. Consequently, when they find that reporting has increased in temperate areas in recent years, this is conservatively interpreted as evidence that pests are moving further into temperate regions. This is a key argument, but one that is difficult to evaluate because good records of sampling effort corresponding to entries in the CABI database⁸ would be very difficult to assemble. It can be argued, however, that the CABI data set is such a unique entity, as the largest global repository for data related to global pest distributions, that it merits analysis even if we know there is the possibility of little-understood biases.

An understanding of current and future changes in the geographic ranges of agricultural pests is important from several standpoints. Good estimates of current changes allow us to test our understanding of the most important factors influencing pest risk, and improve our ability to provide realistic scenarios for the future. Scenarios of future pest and disease distributions under climate change support prioritization in agricultural research programmes. For example, if it is likely that a particular disease will become more important in a region, crop-breeding programmes can respond by incorporating better resistance to that disease in locally adapted varieties.

The prevalence of crop pests is a function of many factors, so identifying particular drivers of change is challenging⁹ (Fig. 1). If we wish to ask whether crop pests have altered distributions because of climate change, this is complicated by the many other factors that have changed simultaneously, even when long-term data are available¹⁰. It is reasonable to expect that higher temperatures will often reduce limitations on pest overwintering and increase the number of pest generations per year (Fig. 1A). However, other factors simultaneously influence pest risk at any location, including host genotypic and phenotypic resistance, where phenotypic resistance may respond to weather variables such as temperature (Fig. 1B). When multiple observers and multiple levels of sampling effort are involved, this adds another layer of uncertainty when comparing the actual distribution of pests and reported distributions (Fig. 1C).

Figure 1: Bebber and colleagues⁴ address the challenge of evaluating how rapidly crop pests (arthropods and pathogens) have spread to the north or south over recent decades, based on observations accumulated from a wide range of observers.

The maps show an example for a hypothetical pest. A, Weather variables such as temperature, precipitation and relative humidity are important in determining how likely a pest is to become established in a new area, influencing risk factors such as the probability of overwintering (oversummering) and the number of generations per year. B, Other factors that determine how likely a pest is to become established in a new area include: human transportation networks; agricultural management, including planting dates and use of tillage; the distribution of susceptible hosts, where some forms of resistance are temperature-sensitive; interactions with other hosts, arthropods and microbes. C, Other factors that determine how likely an established pest is to be observed include: traits of pest or disease (ease of identification or diagnosis, perceived economic importance, shifts in taxonomic resolution); traits of host (economic importance, hectarage); traits of location (infrastructure, training of personnel, sampling effort). Images from: A,B, Istock/Thinkstock; C, © E. De Wolf.

Bebber and colleagues⁴ provide a stimulating analysis of changes in pest distributions, along with a new set of hypotheses to engage scientists working with pests. Future 'big data' analyses may address the geographic distribution of pest genomes and microbial metagenomes associated with plants and soil, including analysis of the geographic spread of genes important in crop damage. Cell phone availability may facilitate analysis of global digital images of crop damage. Better data archiving systems and more data sharing are needed to support future synthetic analyses. For addressing large-extent questions, we also need advances in methods to evaluate more directly the factors that lead, not only to pest risk, but also to reporting of observations, to support understanding of what variables may be good proxies for sampling effort.