Elsevier

Science of The Total Environment

Volumes 637–638, 1 October 2018, Pages 1061-1068
Science of The Total Environment

Activity of invasive slug Limax maximus in relation to climate conditions based on citizen's observations and novel regularization based statistical approaches

https://doi.org/10.1016/j.scitotenv.2018.04.403Get rights and content

Highlights

  • An armature naturalist conducted a daily survey to track invasive slug Limax maximus.

  • Bayesian regularization was applied to evaluate the relation to meteorological data.

  • The results showed a superiority of Bayesian regularization to PCA based analysis.

  • The results bring us to quantitative prediction of the number of slug appearances.

Abstract

Citizen science is a powerful tool that can be used to resolve the problems of introduced species. An amateur naturalist and author of this paper, S. Watanabe, recorded the total number of Limax maximus (Limacidae, Pulmonata) individuals along a fixed census route almost every day for two years on Hokkaido Island, Japan. L. maximus is an invasive slug considered a pest species of horticultural and agricultural crops. We investigated how weather conditions were correlated to the intensity of slug activity using for the first time in ecology the recently developed statistical analyses, Bayesian regularization regression with comparisons among Laplace, Horseshoe and Horseshoe+ priors for the first time in ecology. The slug counts were compared with meteorological data from 5:00 in the morning on the day of observation (OT- and OD-models) and the day before observation (DBOD-models). The OT- and OD-models were more supported than the DBOD-models based on the WAIC scores, and the meteorological predictors selected in the OT-, OD- and DBOD-models were different. The probability of slug appearance was increased on mornings with higher than 20-year-average humidity (%) and lower than average wind velocity (m/s) and precipitation (mm) values in the OT-models. OD-models showed a pattern similar to OT-models in the probability of slug appearance, but also suggested other meteorological predictors for slug activities; positive effect of solar radiation (MJ) for example. Five meteorological predictors, mean and highest temperature (°C), wind velocity (m/s), precipitation amount (mm) and atmospheric pressure (hPa), were selected as the effective factors for the counts in the DBOD-models. Therefore, the DBOD-models will be valuable for the prediction of slug activity in the future, much like a weather forecast.

Introduction

The potentials of citizen science have recently attracted notable attention (e.g., Bonney et al., 2009; Suzuki-Ohno et al., 2017). Citizen science can be a powerful approach to monitoring long-term and large-scale observations by using many biology enthusiasts, while valuable detailed information can also be gained from amateur naturalists with higher level of knowledge on the subject. In particular, citizen science is expected to be a powerful tool in resolving the problems of introduced species (e.g., Hester and Cacho, 2017; Pocock et al., 2017).

The increasing numbers of invasive species have resulted in wide-ranging ecological and economic impacts (Simberloff et al., 2005; Lowe et al., 2000; Mack et al., 2000). Many species of terrestrial molluscs have been recognised as some of the most significant and intractable threats to native ecosystems and agriculture (Lowe et al., 2000; Mito and Uesugi, 2004; e.g., Euglandina rosea (Clarke et al., 1984; Kinzie III, 1992), Achatina fulica (Barker, 2002; Thiengo et al., 2007)). The giant garden slug Limax maximus (Limacidae, Pulmonata) is also considered one of the pest species of horticultural and agricultural crops (Barker and McGhie, 1984; Gaitán-Espitia et al., 2012; Kaya and Mitani, 2000; Kozłowski, 2012; McDonnell et al., 2009; Morii and Nakano, 2017; Morii et al., 2016), although rarely recognised as a serious pest (Herbert, 2010). L. maximus is native to Europe and Asia Minor but has now become widespread throughout the world and is found in North America, South America, North Africa, South Africa, Australia, New Zealand and a few other regions (Barker and McGhie, 1984; McDonnell et al., 2009). L. maximus is a generalist and may be cannibalistic; moreover, it can live in modified habitats such as parks, gardens and rubbish dumps (Barker and McGhie, 1984; McDonnell et al., 2009; Kozłowski, 2012).

The triggers of intense slug activity during shorter time scales have been thoroughly studied and used to control slugs representing horticultural and agricultural pests (e.g., Barnes and Weil, 1944, Barnes and Weil, 1945; Crawford-Sidebotham, 1972; Dainton, 1954a, Dainton, 1954b; Grimm et al., 2000; Kozłowski et al., 2011; Pearson et al., 2006; Young et al., 1991). In particular, the impacts of temperature and moisture on slug activity have been investigated in several slug species, where high temperature and moisture were found to be correlated with a high slug activity (e.g., Crawford-Sidebotham, 1972; Kozłowski et al., 2011). The negative effects of wind speed were also suggested in several studies (e.g., Dainton, 1954b; Young et al., 1991). The influences of other meteorological factors, such as precipitation amount (Barnes and Weil, 1945; Choi et al., 2004; Grimm et al., 2000; Willis et al., 2006) and light intensity (e.g., Hommay et al., 1998b; Pearson et al., 2006), were also surveyed, but the surveys were not comprehensive enough to determine a consistent rule for slug activity. It has been reported that the activities (intensity of moving and feeding behaviours) of some slug species respond to different optimum weather conditions (Hommay et al., 2003); therefore, it is important to investigate the relationships between weather conditions and slug activity for each slug species.

The weather conditions correlated with L. maximus activity during shorter time scales are relatively unknown especially in the field, although it has been suggested that higher temperature, higher humidity, and lower light intensity can increase the activity levels of this slug species (Hess and Prior, 1985; Prior and Grega, 1982; Rollo, 1982). The strong correlation of environmental factors with each other can lead to failure when estimating the regression parameters of each explanatory variable against a response variable. This problem, called multicollinearity, has been one of the major concerns in statistics, especially in ecological studies. The most prevalent method used to eliminate multicollinearity is principal component analysis (PCA), which is a procedure that converts a set of variables into another set of uncorrelated variables (Pearson, 1901; Hotelling, 1933), and conducts maximum likelihood estimation using principal components (PCs) as the predictor of the model. However, this method has two potential disadvantages. First, it is theoretically impossible to avoid information loss, because a smaller number of components is used rather than the original variables. Second, and more importantly, this transformation renders the results unreadable (Abdi and Williams, 2010). These issues can be crucial for the interpretation of results for both academic and pragmatic points.

On the other hand, the rapid developments of recent statistics and computing methods have brought another approach to resolve the problem of multicollinearity. Meteorological factors are a typical example of variables that exhibit strong correlations; thus, we adopted Bayesian regularization regression using three prior distributions, Laplace, Horseshoe and Horseshoe+, which enable us to simultaneously estimate the coefficients and select important variables. This method not only is robust against multicollinearity but also can incorporate many explanatory variables relative to sample size with only a slight increase in computation cost.

Recently, L. maximus was introduced to Japan, and it is now rapidly spreading (Hasegawa et al., 2009; Iijima et al., 2013; Morii and Nakano, 2017; Morii et al., 2016). L. maximus was first discovered in 2006 in Ibaraki Prefecture and since then it has been discovered in four Japanese prefectures. Naturalised populations of L. maximus on Hokkaido Island, Japan were first discovered in 2012 in the isolated natural forests of Mt. Maruyama in Sapporo City, and they have now become common in this area (Fig. 1; Morii and Nakano, 2017; Morii et al., 2016). Several observations that L. maximus fed on agricultural products were also reported in Japan (Iijima et al., 2013; Morii and Nakano, 2017). In the present study, we investigate the environmental factors that are correlated with the intensity of L. maximus activity and create a model to predict slug activity using continuous observations at Mt. Maruyama by a single amateur naturalist and recent regularization based statistical analyses.

Section snippets

Study site and observations

An author of this paper, S. Watanabe, observed L. maximus in the Maruyama Forest around Mt. Maruyama, which is in the southwest part of Sapporo City, Hokkaido Island, Japan (Fig. 1). Watanabe recorded the number of L. maximus individuals along a census route (approximately 1 km length) from the entrance of the Maruyama Forest (43.05316°N, 141.31115°E, alt. 30 m; Site A in Fig. 1) to the top of Mt. Maruyama (43.04728°N, 141.31632°E, alt. 225 m; Site B in Fig. 1) every morning (5:00–6:00) for two

Bayesian regularization based statistical analyses

All of four HMC chains were well converged (R^<1.03). The selected explanatory variables did not differ between models based on two-years integrated and 2016 only versions (Table 2 vs Table C.1; Table 3 vs Table C.2), and WAIC scores were much lower in models of two-years integrated versions (Table 1). Therefore, we only mention about the models based on two-years integrated data below.

WAIC scores and tendencies of selected variables were similar within OT-, OD- or DBOD-models with Laplace,

Discussion

WAIC scores improved in the novel regularization based models with Laplace, Horseshoe and Horseshoe+ distributions compared to traditional PCA based approaches. Furthermore, it was much easier to interpret results using our new methods than with PCA based models. Few suggestions about predictors against the slug activities were indicated in PCA-OT-model and PCA-OD-model, because all factors (year, month, month^2 and other meteorological indicators) were selected as more or less reliable

Authors' contributions

Y.M. developed the experimental design; S.W. collected the data; Y.O. analysed the data; Y.M. and Y.O. wrote the manuscript.

Acknowledgements

We would like to thank M. Kyono for serving as an intermediary for the researchers and citizens. We also thank A. Terui for technical support, E. Hasegawa, M. J. J. E. Loonen and U. E. Schneppat for scientific discussions and helpful comments, in addition to T. E. Squires and T. Léger for English revisions and insights on the manuscript.

Research data

All the data used in http://dx.doi.org/10.17632/5sy773j6xs.1.

References (61)

  • G.M. Barker et al.

    The biology of introduced slugs (Pulmonata) in New Zealand. 1. Introduction and notes on Limax maximus

    N. Z. Entomol.

    (1984)
  • H.F. Barnes et al.

    Slugs in gardens: their numbers, activities and distribution. Part 1

    J. Anim. Ecol.

    (1944)
  • H.F. Barnes et al.

    Slugs in gardens: their numbers, activities and distribution. Part 2

    J. Anim. Ecol.

    (1945)
  • A. Bhadra et al.

    The horseshoe+ estimator of ultra-sparse signals

    Bayesian Anal.

    (2017)
  • R. Bonney et al.

    Citizen science: a developing tool for expanding science knowledge and scientific literacy

    Bioscience

    (2009)
  • C.M. Carvalho et al.

    The horseshoe estimator for sparse signals

    Biometrika

    (2010)
  • S. Chatterjee et al.

    Regression Analysis by Example

    (2000)
  • B. Clarke et al.

    The extinction of endemic species by a program of biological control

    Pac. Sci.

    (1984)
  • T.J. Crawford-Sidebotham

    The influence of weather upon the activity of slugs

    Oecologia

    (1972)
  • B.H. Dainton

    The activity of slugs I. The induction of activity by changing temperatures

    J. Exp. Biol.

    (1954)
  • B.H. Dainton

    The activity of slugs. II. The effect of light and air currents

    J. Exp. Biol.

    (1954)
  • J.D. Gaitán-Espitia et al.

    Repeatability of energy metabolism and resistance to dehydration in the invasive slug Limax maximus

    Invertebr. Biol.

    (2012)
  • W. Greene

    Accounting for Excess Zeros and Sample Selection in Poisson and Negative Binomial Regression Models, Working Paper 94–10

    (1994)
  • B. Grimm et al.

    Daily activity of the pest slug Arion lusitanicus under laboratory conditions

    Ann. Appl. Biol.

    (2002)
  • B. Grimm et al.

    Daily activity of the pest slug Arion lusitanicus Mabille

    J. Molluscan Stud.

    (2000)
  • K. Hasegawa et al.

    Discovery of naturalized populations of the introduced slug, Limax maximus in Tsuchiura City, Ibaraki Prefecture, Japan

    Chiribotan

    (2009)
  • D.G. Herbert

    The Introduced Terrestrial Mollusca of South Africa

    (2010)
  • S.D. Hess et al.

    Locomotor activity of terrestrial slug, Limax maximus: response to progressive dehydration

    J. Exp. Biol.

    (1985)
  • S.M. Hester et al.

    The contribution of passive surveillance to invasive species management

    Biol. Invasions

    (2017)
  • M.D. Hoffman et al.

    The No-U-turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo

    J. Mach. Learn. Res.

    (2014)
  • Cited by (11)

    • A review on machine learning forecasting growth trends and their real-time applications in different energy systems

      2020, Sustainable Cities and Society
      Citation Excerpt :

      The various training models (Khwaja, Zhang, Anpalagan, & Venkatesh, 2017): like as LMBNNs, BRBNNs, Gradient descent with adaptive learning rate and momentum backpropagation, and resilient backpropagation used to reduce the network error and improve the forecasting accuracy. The BRBNNs was employed to assess the invasive slug activity with environmental conditions using a novel regularisation statistical technique (Morii, Ohkubo, & Watanabe, 2018). This research (Pereira, Antoni, & Leclère, 2015), gives an overview of the issues regarding how to automate the choice of a suitable regularization variable when determining the inverse problem in acoustics.

    • An ecological study of eosinophilic meningitis caused by the nematode, Angiostrongylus cantonensis (Chen, 1935) (Nematoda: Metastrongylidae)

      2019, Parasitology International
      Citation Excerpt :

      These data may indicate that wind has the greatest effect on slug populations toward the beginning lag. Later on, the number of slugs may decrease with time [18]. Note that the number of EOM patients varied from month to month.

    View all citing articles on Scopus
    1

    These authors contributed equally to this work.

    View full text