Corresponding author: Regan Early ( r.early@exeter.ac.uk ) Academic editor: John Ross Wilson
© 2018 Regan Early, Pablo González-Moreno, Sean T. Murphy, Roger Day.
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Citation:
Early R, González-Moreno P, Murphy ST, Day R (2018) Forecasting the global extent of invasion of the cereal pest Spodoptera frugiperda, the fall armyworm. NeoBiota 40: 25-50. https://doi.org/10.3897/neobiota.40.28165
|
Fall armyworm, Spodoptera frugiperda, is a crop pest native to the Americas, which has invaded and spread throughout sub-Saharan Africa within two years. Recent estimates of 20–50% maize yield loss in Africa suggest severe impact on livelihoods. Fall armyworm is still infilling its potential range in Africa and could spread to other continents. In order to understand fall armyworm’s year-round, global, potential distribution, we used evidence of the effects of temperature and precipitation on fall armyworm life-history, combined with data on native and African distributions to construct Species Distribution Models (SDMs). We also investigated the strength of trade and transportation pathways that could carry fall armyworm beyond Africa. Up till now, fall armyworm has only invaded areas that have a climate similar to the native distribution, validating the use of climatic SDMs. The strongest climatic limits on fall armyworm’s year-round distribution are the coldest annual temperature and the amount of rain in the wet season. Much of sub-Saharan Africa can host year-round fall armyworm populations, but the likelihoods of colonising North Africa and seasonal migrations into Europe are hard to predict. South and Southeast Asia and Australia have climate conditions that would permit fall armyworm to invade. Current trade and transportation routes reveal Australia, China, India, Indonesia, Malaysia, Philippines and Thailand face high threat of fall armyworm invasions originating from Africa.
Agriculture, biological invasion, climate envelope, crop pest, ecological niche model
Fall armyworm, Spodoptera frugiperda (J.E. Smith) (Lepidoptera: Noctuidae) is native to the Americas. The moth lives year-round from as far south as La Pampa, Argentina, to as far north as southern Florida and Texas, USA and undergoes seasonal migrations as far north as Québec and Ontario (Figure
Fall armyworm distribution data. Red points are the complete set of presence locations used to make models. Black points in the USA are not part of the year-round native distribution and were not included in models. Grey areas are the geographically unrestricted background from which pseudo-absences could be drawn. Red lines indicate the geographically restricted background (500 km radius around presence locations) within which pseudo-absences could be drawn.
In January 2016, major outbreaks of armyworms were reported in South West Nigeria and Ghana and shortly after in Benin, Sao Tomé and Togo (
Fall armyworm’s year-round distribution is expected to be restricted to relatively warm and moist areas, as it cannot survive cold temperatures by entering diapause (
As fall armyworm has huge potential to affect staple and economic crops globally, we urgently need information on the pest’s potential distribution and environmental limitations. Such information would assist national and regional pest risk assessments and appropriate management strategies in several ways: by quantifying the agricultural areas within Africa that are at risk from year-round populations or seasonal migrations, by informing the likelihood of seasonal migrations outside Africa, and by classifying the likelihood of establishment if fall armyworm is transported into other parts of the world. This information would also help target awareness raising and monitoring for early detection. Early detection of infestations is extremely beneficial as chemical insecticides are only effective while the larvae are small (
Here, we first reviewed what is known of the environmental controls on the fall armyworm’s life-cycle and herbivory, particularly on maize, the crop most economically important and threatened by the moth in Africa (
To forecast a species’ potential range, it is necessary to consider the environmental factors that are needed for the species to complete its life cycle. These factors could directly limit the target species’ distribution and are often termed ‘proximal’ variables. Using these variables increase accuracy and biological realism of projections of species distributions following invasion (also termed ‘transferability’(
Empirically measured environmental effects on fall armyworm life cycle. Summary of data from literature of temperature, moisture, soil and host plant effects on fall armyworm survival and developmental time and observations of abundances in the field under different conditions and seasons. All studies were conducted with populations from the Americas. Rectangles are life stages, ovals are processes. Orange arrows represent effects of fall armyworm on maize or vice versa. The occurrence of a 7th instar is not universally reported. Only direct effects (measured or imputed) of the environment and host on fall armyworm were included. Unless otherwise noted, moths for experiments were reared on fresh maize or an artificial diet. RH is relative humidity. More information is given in Suppl. material
Based on the life-history and environmental requirements of fall armyworm, and in light of climatic conditions in the Americas and Africa, we selected the following climatic variables:
– SumWet, total amount of precipitation in wettest three months of year (intensity of rainy season, when most food available and population growth is fastest)
– LenWet, number of months when rain is greater than average (length of rainy season, when most food is available and population growth is fastest)
– SeasPpn, seasonality of precipitation (difference in rainfall between rainy and dry season)
– MinTemp, mean temperature of the coldest month of the year (the lowest limit for growth)
We also initially used GDD13.8, annual growing degree days above a lower development threshold of 13.8 °C (minimum temperature for survival). We selected 13.8 °C as the lower development threshold (
In addition to climatic variables, we also used ‘Forest’, the proportion of each 10 arc-minute grid-cell that is covered by trees. This is because fall armyworm is only reported from agricultural areas, though there may be many areas covered by forest that are climatically suitable for the species, but from which it is not reported or is absent due to a lack of host plant. We therefore expected a high forest cover to indicate environmental unsuitability. Without the forest variable, climate conditions alone would have been less able to discriminate between suitable and unsuitable locations. We used forest rather than crop or pasture land as forest is relatively easier to delineate than grassland using satellite data. Forest cover was drawn from the European Space Agency’s Global Land Cover 2000 project at 1 km (https://www.esa-landcover-cci.org/).
Presence records for the Americas were obtained from three sources. 1) Global Biodiversity Information Facility (ww.gbif.org) in November 2016. Records that did not have coordinates but did have location descriptions were georeferenced with accuracy equal to the climatic grid data. 2) Review of literature on Fall Armyworm in the region. 3) Consultation between CABI and local experts in several countries. In the southern USA, occurrences south of 27 degrees in Florida and south of 31 degrees in Texas were considered to be year-round populations and were included as presence data points (
Distribution data were filtered so that only one presence was recorded in each climatic grid-cell, resulting in 240 presences in Africa and 167 presences in the Americas. Due to the recent dedicated searches, the African distribution was better sampled than the American distribution. The difference in sampling intensity between the two continents led to concerns that SDMs might be over-fitted to the well-studied locations in Ghana and Zambia. This would underestimate the suitability of areas outside Ghana and Zambia, particularly areas that are environmentally similar to the native range, but different to the Ghanaian and Zambian range. We therefore sub-sampled several proportions of the African distribution (5, 10, 20, 30, 50 and 70%) and used four sub-samples at each proportion to construct alternative SDMs.
In order to select pseudo-absences, we used two approaches to delimit the geographic background in the Americas and Africa to which fall armyworm could reasonably be expected to disperse without human assistance (Figure
In total, we ran models with 488 datasets: six different levels of sub-sampling of African presences from the entire background (5, 10, 20, 30, 50 and 70%) each with 20 repetitions of sub-sampling, one dataset of all presences using the entire background, and one dataset with the geographically restricted background and for each of which we randomly sampled pseudo-absences four times.
In order to estimate parameter values for the environmental variables, and to predict the potential year-round distribution of fall armyworm, we created an ensemble SDM (
We used internal validation to evaluate SDM accuracy, splitting each of the 488 distribution datasets randomly so that 70% of the presence and pseudo-absence points were used to calibrate the models. These models were used to predict suitability at the 30% remaining validation distribution data points. The Area under the Receiver Operating Curve (AUC) and True Skill Statistic (TSS) were used to judge how accurately the models predicted the validation data (
For each of the distribution datasets, we constructed an ensemble forecast for the global terrestrial surface. Ensembles were made using models (from the 488 distribution datasets) for which validation TSS ≥ 0.4. Models with TSS ≥ 0.4 are considered to have ‘moderate’ performance (greater than ‘fair’, but less than ‘substantial’ (
In order to investigate the effect of biased recorder effort and geographic background on environmental suitability, we compared the agreement of the global ensemble projections made using each distribution dataset by using Cohen’s Kappa and balanced accuracy (
In order to determine whether the environment in the geographic region from which distribution data were drawn is representative of the entire global terrestrial surface, we calculated the Multivariate Environmental Similarity Surface (
The importance of environmental variables for fall armyworm’s range was calculated using all of the distribution data in a given dataset and using all models, regardless of TSS score. For any given environmental variable, that variable was randomised, an SDM was made with the shuffled dataset and the Pearson’s correlation (r) calculated between the SDMs with original and shuffled data. Importance is calculated as 1-r, so a value 0 indicates the variable has no influence on the SDM.
Has the fall armyworm invaded where we expect it to, based on the native distribution in the Americas?
To answer this, we calculated the niche expansion between the Americas and Africa, using the methodology developed by
In order to illustrate the potential for fall armyworm to spread from Africa to other parts of the world, we first identified the countries most likely to act as sources for fall armyworm and the countries most vulnerable to fall armyworm establishment, as those with > 33660 km2 of suitable climate (i.e. 100 × 10 arc-minute grid-cells with climate suitability > 0.5). This resulted in 64 countries being identified as sources or vulnerable. We then examined two major pathways for invertebrate introduction: trade and passenger air travel. Trade is one of the main drivers of plant pest introduction globally (
Passenger air travel is suggested to be the route by which fall armyworm was first introduced to Africa and is thought to be important in insect introductions (
The most commonly studied relationship between life-history stage and environment is the effect of air temperature on larval and pupal survival and development rates. The minimum temperature for development was reported between 8.7 °C and 13.8 °C (
The importance of moisture and precipitation is complex. Precipitation and irrigation have a direct negative effect on larval and pupal survival. Heavy rainfall fills the maize whorl with water, in which larvae float, until it overflows and the larvae are spilled out or drown (a process which is helped by wind gusts, A. van Huis 1981). Rainfall and irrigation are thought to trap moths and drown them in their pupation tunnels, with the effects being stronger in more friable soils, when rainfall can also cause the tunnels to collapse (
Research on the effect of host plant on fall armyworm populations is limited to maize. There is evidence that the maize growth stage positively impacts development speeds (i.e. development is fastest for larvae eating mature leaves) (
There are two genetically distinct fall armyworm ‘strains’, which specialise on maize and rice (
Has the fall armyworm invaded where we expect it to, based on the native distribution in the Americas?
All but 3% of the recorded current distribution in Ghana and Zambia is found in climate and forest conditions that match the native range, i.e. there is virtually no orange area in Figure
Similarity of environmental niches in the Americas and Africa. The outlines represent the environmental conditions available in the Americas (purple, solid line) and Ghana and Zambia (green, dashed). The larger purple shaded area represents the conditions the species occupies only in the Americas. The yellow shaded areas represent the conditions the species occupies in the studied African range that lie within the American environmental niche (niche stability). The green area with dots shows the part of the African range that is found in different environmental conditions to the native range (niche expansion). The shading under the yellow/green area is the density of the species’ occurrence in the African range.
Internal cross-validation indicated that TSS scores were ‘moderate’ and AUC scores were ‘fair’ (Table
Agreement between the ensemble projections resulting from sub-sampled and complete datasets was ‘moderate’ to ‘substantial’ (Cohen’s kappa values Table
The dataset using 100% of the data and the entire geographic background gave SDMs that had the highest AUC and TSS scores (Table
Global suitability for fall armyworm and likely invasion routes. a Potential global distribution of fall armyworm, as predicted by an ensemble of SDMs constructed using all distribution data and with four pseudo-absence datasets. SDMs were permitted into the ensemble if the TSS from internal cross-validation was ≥0.4. The ensemble was calculated as the mean of projections from all permitted SDMs, each model weighted by the cross-validated TSS b uncertainty in projections, as calculated by the variation between all projections included in the ensemble c value of all exports from 2012–2016 from source sub-Saharan African countries climate to vulnerable countries outside sub-Saharan Africa. The top 5% of trading relationships between these countries are shown and the five colour categories represent 20% quantiles of export values d number of passengers in 2013 travelling from source sub-Saharan African countries with their final destination in vulnerable countries outside sub-Saharan Africa. The top 13% of travel routes between these countries are shown and the five colour categories represent 20% quantiles of passenger numbers.
Summary statistics for Species Distribution Models (SDMs). ‘Dataset’ indicates the percentage of the African distribution data that were sub-sampled. AUC and TSS indicate predictive accuracy. Mean (±standard deviation) TSS and AUC values are averages calculated by internal cross-validation for all SDMs constructed with each dataset, excluding SDMs that were discarded in making the ensemble due to low predictive accuracy (TSS <0.4). TSS values between 0.4 and 0.5 are often considered ‘moderate’(
Pseudo-absence restriction radius | Dataset | TSS | AUC | Spearman’s ρ | Cohen’s Kappa |
---|---|---|---|---|---|
None | 5% | 0.52±0.06 | 0.79±0.05 | 0.88±0.12 | 0.49 |
None | 10% | 0.52±0.07 | 0.80±0.05 | 0.87±0.15 | 0.50 |
None | 20% | 0.54±0.09 | 0.80±0.06 | 0.89±0.09 | 0.57 |
None | 30% | 0.55±0.06 | 0.80±0.04 | 0.87±0.12 | 0.59 |
None | 50% | 0.51±0.07 | 0.79±0.05 | 0.89±0.11 | 0.75 |
None | 70% | 0.55±0.07 | 0.91±0.05 | 0.91±0.09 | 0.76 |
None | 100% | 0.55±0.07 | 0.81±0.06 | 0.89±0.08 | NA |
500 km | 100% | 0.48±0.07 | 0.78±0.04 | 0.79±0.23 | 0.41 |
MinTemp was the most important environmental variable, followed by forest and SumWet (Figure
MESS indicated very few areas in which environmental conditions had no analogue in the training region (Suppl. material
Importance of variables for Species Distribution Models (SDMs) of fall armyworm in the Americas and Africa. Colour codes indicate the percentage of the African distribution that was sub-sampled or the pseudo-absence selection background. Error bars are standard deviations of the results across all SDM techniques and distribution datasets.
Countries vulnerable to fall armyworm (outside South America) that receive the greatest value of commodities exported from African fall armyworm source countries are China, India Indonesia and, to a lesser extent, Australia and Thailand. Countries vulnerable to fall armyworm (outside South America) that receive the greatest number of passengers embarking from African source countries are Australia, China, India, Indonesia, Malaysia and the Philippines. These countries are likely to be the most imminently threatened by fall armyworm invasion.
SDM results were encouragingly accurate and indicated that much of sub-Saharan Africa is highly suitable year-round for fall armyworm, from the Saharan belt to South Africa. Within this region, much of Congo, DRC, Gabon and Cameroon have low suitability (though uncertainty is high in some of these areas). Low suitability in these countries is likely because of extensive forest cover. However, this does not mean that pockets of suitable habitat in those countries will not be severely affected, given the ability of fall armyworm to travel long distances (see below for further discussion of forested areas).
Much of Northwest and Northeast Africa has low suitability (<40% probability of occurrence, Figure
In currently un-invaded portions of Africa, there are pockets of high suitability in Morocco’s productive agricultural regions, as well as the Libyan coast. Transportation to North Africa (countries with part of their land mass north of the Sahara) via trade or air transportation routes from sub-Saharan Africa is less likely than transportation outside Africa (Suppl. material
Low-suitability areas in sub-Saharan Africa may still experience infestation from migrating fall armyworm during some seasons.
Research into seasonal migration and population dynamics within sub-Saharan Africa is clearly needed. Understanding the potential for annual migrations both within and beyond the year-round range requires forecasts of the speed, direction and heights of prevailing wind during periods when fall armyworm populations are large. We also need to know the migration capacity of African fall armyworm populations. The propensity of individuals to migrate and the length of time for which adults can fly varies within populations, often genetically (
Much of South and Southeast Asia and areas of Australia are highly suitable for fall armyworm year-round. Natural dispersal towards this region is considered unlikely, as the distance is over 2000 km, further than fall armyworm is recorded to have travelled in the Americas. Nonetheless, the southwest monsoon blows from Africa to India beginning in June and is a possible route by which fall armyworm adults could arrive in India by their own dispersal. There are important invasion routes from Africa into South and Southeast Asia and Australia (Figure
As conditions outside the predicted year-round range (for example, Europe) might be suitable in certain seasons, improved predictions of seasonal suitability could be achieved with demographic modelling of data from lab or field trials or from statistical modelling (e.g.
The results of Species Distribution Modelling were encouragingly accurate. AUC values from cross-validation were well within the range usually considered acceptable for SDM studies of invertebrates (
Forest, MinTemp (coldest annual temperature) and SumWet (rainfall during the wettest three months) were consistently identified as the environmental variables that most affected fall armyworm’s distribution. The importance of MinTemp supports the existence of a hard polewards geographical boundary, caused by one or more months where temperature drops below a threshold. This suggests that climate warming could expand the potential range of fall armyworm. SumWet was consistently more important than LenWet (rainy season length) or SeasPpn (the contrast between the rainy and dry seasons). In order to understand if this is due to indirect (i.e. through host plant growth) or direct effects, one could use structural equation modelling incorporating the yield of key host plants, or incorporate life-history parameters from Suppl. material
Rapid evolution of climate tolerances can occur in pest insects following invasion, i.e. a ‘niche shift’ (
It is interesting to note that fall armyworm was not high on a recent list of pest species likely to invade West Africa (in the lower 50th percentile for Ghana, Nigeria and Togo (
Very little research has been done into differing climate tolerances between maize and rice strains and there is insufficient information on their respective distributions to apply SDMs to each strain (see Methods). Slight differences in basal temperatures of the two strains result in approximately one more generation of the rice strain per year at the optimum temperature of 25 °C, i.e. 12 generations (
Diet can affect temperature tolerances, and indeed the temperature threshold for development was several degrees lower when fall armyworms were fed leaves from early vegetative maize plants than when fed leaves from late vegetative or reproductive plants (
Given the likely onward spread of fall armyworm, a united international response is clearly needed and is indeed emerging. In Africa, the Food and Agriculture Organization of the United Nations is coordinating responses to fall armyworm, providing support for early warning tools, farmer field schools on integrated pest management, and a food security risk assessment model. A research consortium uniting Africa and Asia has recently been launched. Led by CGIAR, the Fall Armyworm R4D consortium aims to develop integrated pest management solutions including host plant resistance, environmentally safer chemical pesticides, biological and cultural control methods and agronomic management.
The accuracy of SDM results and the similarity of the environments occupied in the native and invaded range support the robustness of the SDM approach. Temperature of the coldest month and the amount of rain during the rainy season are the most important climatic limits of fall armyworm’s year-round distribution. Much of sub-Saharan Africa can host year-round fall armyworm populations and seasonal migrations are likely to take place along the Nile into Northeast Africa. The likelihood of seasonal migrations beyond this range seems to be low. South and Southeast Asia and Australia, are highly suitable for fall armyworm. Trade and passenger air travel routes indicate parts of this region into which African populations are particularly likely to be transported. There is therefore considerable potential for near global invasion and seasonal migration of fall armyworm. Vigilance is needed to monitor for the onward invasion of fall armyworm via potential migration routes into North Africa and South Asia and on some high-risk trade and air travel routes. Management decisions would be improved by further research on fall armyworm’s seasonal migration and population dynamics and the environmental dependency of interactions with other species.
We are very grateful to Stephanie Wheeler for georeferencing, Dean Paini for information on fall armyworm introduction likelihood and Jason Chapman for discussion on migration. The authors gratefully acknowledge the financial support of BBSRC GCRF IAA sub-award SW-07640, the UK Department for International Development (DfID) and PRISE project (UKSA IPP Call 1). We wish to acknowledge the support of our Plantwise donors: DfID (UK), SDC (Switzerland), DEVCO (European Commission), DGIS (Netherlands), IFAD, Irish Aid and ACIAR (Australia). We would also the like to thank the Ministry of Food & Agriculture, Ghana and Ministry of Agriculture and Livestock and Ministry of Local Government and Rural Development, Zambia, for the extension service providers in their function of plant doctors for gathering plant clinic data and the in-country plant clinic data managers for managing and uploading the data. CABI gratefully acknowledge the core financial support from our member countries (and lead agencies) including the United Kingdom (Department for International Development), China (Chinese Ministry of Agriculture), Australia (Australian Centre for International Agricultural Research), Canada (Agriculture and Agri-Food Canada), Netherlands (Directorate-General for International Cooperation) and Switzerland (Swiss Agency for Development and Cooperation).
Author contributions
RE conceived the study, collected data, performed analysis and drafted the paper. PG and SM contributed to the acquisition and interpretation of data for the project and revised the manuscript.
Data availability statement
Some distribution data from South America analysed during this study are included in the Supplementary Information files. This does not include data from Plantwise clinics in Bolivia, Honduras, Nicaragua and Peru, due to data sharing restrictions. Some other distribution data are available from CABI’s Plantwise programme but restrictions apply to the availability of these data, which were used under licence for the current study and so are not publicly available. Data may be available from the authors upon reasonable request and with permission of Plantwise. All other data used are publicly available from the referenced data sources.