Effectively tuning plant growth models with different spatial complexity: A statistical perspective

Yoshiaki Nakagawa, Masayuki Yokosawa, Akihiko Ito, Toshihiko Hara

    Research output: Contribution to journalArticle

    Abstract

    Forest gap models (non-spatial, patch- and individual-based models) and size structure models (non-spatial stand models) rely on two assumptions: the mean field assumption (A-I) and the assumption that plants in one patch do not compete with plants in other patches (A-II). These assumptions lead to differences in plant size dynamics between these models and spatially explicit models (or observations of real forests). Therefore, to more accurately replicate dynamics, these models require model tuning by (1) adjusting model parameter values or (2) introducing a correction term into models. However, these model tuning methods have not been systematically and statistically investigated in models using different patch sizes. We used a simple spatially explicit model that simulated growth and competition processes, and rewrote it as patch models. The patch sizes of the patch models were set between 4 and 1500 m2. First, we estimated the parameter values (the intrinsic growth rate, metabolic loss, competition coefficient, and competitive asymmetry) of these models that best reproduce plant size growth under competition using field data from a Sakhalin fir stand, and compared the parameter values among the models. Second, we introduced correction terms into the patch models and estimated the optimal correction term for reproducing plant size growth under competition using the field data. The estimated parameter values of the patch models for all patch sizes differed greatly from those of the spatially explicit models. Therefore, parameter values should not be shared between spatially explicit models and patch models. In addition, the parameter value sets for the models with small patches differed from those with large patches. This is because parameter values for small patches mainly improve biases of A-II, while those for large patches mainly improve biases of A-I. Therefore, parameter values should not be shared between patch models with small patches and with large patches. The estimated correction term in the patch models with large patches excluded the competitive effects of small and medium-sized plants on their neighbors, even though these effects exist in real stands. This exclusion can be ascribed to the discrepancy between their competition in real plant populations and A-I. Therefore, the competitive effects of small and medium-sized plants should not be included in patch models with large patches. Finally, the reproducibility of the models tuned with correction terms was higher than those with adjusted parameters.

    Original languageEnglish
    Pages (from-to)95-112
    Number of pages18
    JournalEcological Modelling
    Volume361
    DOIs
    Publication statusPublished - 2017 Oct 10

    Fingerprint

    patch size
    parameter
    individual-based model
    size structure
    asymmetry
    effect

    Keywords

    • Competition
    • DGVM
    • Forest dynamics
    • Gap model
    • Individual-based model
    • Size structure model

    ASJC Scopus subject areas

    • Ecological Modelling

    Cite this

    Effectively tuning plant growth models with different spatial complexity : A statistical perspective. / Nakagawa, Yoshiaki; Yokosawa, Masayuki; Ito, Akihiko; Hara, Toshihiko.

    In: Ecological Modelling, Vol. 361, 10.10.2017, p. 95-112.

    Research output: Contribution to journalArticle

    @article{144344c3fad94397aacdbde78671913e,
    title = "Effectively tuning plant growth models with different spatial complexity: A statistical perspective",
    abstract = "Forest gap models (non-spatial, patch- and individual-based models) and size structure models (non-spatial stand models) rely on two assumptions: the mean field assumption (A-I) and the assumption that plants in one patch do not compete with plants in other patches (A-II). These assumptions lead to differences in plant size dynamics between these models and spatially explicit models (or observations of real forests). Therefore, to more accurately replicate dynamics, these models require model tuning by (1) adjusting model parameter values or (2) introducing a correction term into models. However, these model tuning methods have not been systematically and statistically investigated in models using different patch sizes. We used a simple spatially explicit model that simulated growth and competition processes, and rewrote it as patch models. The patch sizes of the patch models were set between 4 and 1500 m2. First, we estimated the parameter values (the intrinsic growth rate, metabolic loss, competition coefficient, and competitive asymmetry) of these models that best reproduce plant size growth under competition using field data from a Sakhalin fir stand, and compared the parameter values among the models. Second, we introduced correction terms into the patch models and estimated the optimal correction term for reproducing plant size growth under competition using the field data. The estimated parameter values of the patch models for all patch sizes differed greatly from those of the spatially explicit models. Therefore, parameter values should not be shared between spatially explicit models and patch models. In addition, the parameter value sets for the models with small patches differed from those with large patches. This is because parameter values for small patches mainly improve biases of A-II, while those for large patches mainly improve biases of A-I. Therefore, parameter values should not be shared between patch models with small patches and with large patches. The estimated correction term in the patch models with large patches excluded the competitive effects of small and medium-sized plants on their neighbors, even though these effects exist in real stands. This exclusion can be ascribed to the discrepancy between their competition in real plant populations and A-I. Therefore, the competitive effects of small and medium-sized plants should not be included in patch models with large patches. Finally, the reproducibility of the models tuned with correction terms was higher than those with adjusted parameters.",
    keywords = "Competition, DGVM, Forest dynamics, Gap model, Individual-based model, Size structure model",
    author = "Yoshiaki Nakagawa and Masayuki Yokosawa and Akihiko Ito and Toshihiko Hara",
    year = "2017",
    month = "10",
    day = "10",
    doi = "10.1016/j.ecolmodel.2017.07.018",
    language = "English",
    volume = "361",
    pages = "95--112",
    journal = "Ecological Modelling",
    issn = "0304-3800",
    publisher = "Elsevier",

    }

    TY - JOUR

    T1 - Effectively tuning plant growth models with different spatial complexity

    T2 - A statistical perspective

    AU - Nakagawa, Yoshiaki

    AU - Yokosawa, Masayuki

    AU - Ito, Akihiko

    AU - Hara, Toshihiko

    PY - 2017/10/10

    Y1 - 2017/10/10

    N2 - Forest gap models (non-spatial, patch- and individual-based models) and size structure models (non-spatial stand models) rely on two assumptions: the mean field assumption (A-I) and the assumption that plants in one patch do not compete with plants in other patches (A-II). These assumptions lead to differences in plant size dynamics between these models and spatially explicit models (or observations of real forests). Therefore, to more accurately replicate dynamics, these models require model tuning by (1) adjusting model parameter values or (2) introducing a correction term into models. However, these model tuning methods have not been systematically and statistically investigated in models using different patch sizes. We used a simple spatially explicit model that simulated growth and competition processes, and rewrote it as patch models. The patch sizes of the patch models were set between 4 and 1500 m2. First, we estimated the parameter values (the intrinsic growth rate, metabolic loss, competition coefficient, and competitive asymmetry) of these models that best reproduce plant size growth under competition using field data from a Sakhalin fir stand, and compared the parameter values among the models. Second, we introduced correction terms into the patch models and estimated the optimal correction term for reproducing plant size growth under competition using the field data. The estimated parameter values of the patch models for all patch sizes differed greatly from those of the spatially explicit models. Therefore, parameter values should not be shared between spatially explicit models and patch models. In addition, the parameter value sets for the models with small patches differed from those with large patches. This is because parameter values for small patches mainly improve biases of A-II, while those for large patches mainly improve biases of A-I. Therefore, parameter values should not be shared between patch models with small patches and with large patches. The estimated correction term in the patch models with large patches excluded the competitive effects of small and medium-sized plants on their neighbors, even though these effects exist in real stands. This exclusion can be ascribed to the discrepancy between their competition in real plant populations and A-I. Therefore, the competitive effects of small and medium-sized plants should not be included in patch models with large patches. Finally, the reproducibility of the models tuned with correction terms was higher than those with adjusted parameters.

    AB - Forest gap models (non-spatial, patch- and individual-based models) and size structure models (non-spatial stand models) rely on two assumptions: the mean field assumption (A-I) and the assumption that plants in one patch do not compete with plants in other patches (A-II). These assumptions lead to differences in plant size dynamics between these models and spatially explicit models (or observations of real forests). Therefore, to more accurately replicate dynamics, these models require model tuning by (1) adjusting model parameter values or (2) introducing a correction term into models. However, these model tuning methods have not been systematically and statistically investigated in models using different patch sizes. We used a simple spatially explicit model that simulated growth and competition processes, and rewrote it as patch models. The patch sizes of the patch models were set between 4 and 1500 m2. First, we estimated the parameter values (the intrinsic growth rate, metabolic loss, competition coefficient, and competitive asymmetry) of these models that best reproduce plant size growth under competition using field data from a Sakhalin fir stand, and compared the parameter values among the models. Second, we introduced correction terms into the patch models and estimated the optimal correction term for reproducing plant size growth under competition using the field data. The estimated parameter values of the patch models for all patch sizes differed greatly from those of the spatially explicit models. Therefore, parameter values should not be shared between spatially explicit models and patch models. In addition, the parameter value sets for the models with small patches differed from those with large patches. This is because parameter values for small patches mainly improve biases of A-II, while those for large patches mainly improve biases of A-I. Therefore, parameter values should not be shared between patch models with small patches and with large patches. The estimated correction term in the patch models with large patches excluded the competitive effects of small and medium-sized plants on their neighbors, even though these effects exist in real stands. This exclusion can be ascribed to the discrepancy between their competition in real plant populations and A-I. Therefore, the competitive effects of small and medium-sized plants should not be included in patch models with large patches. Finally, the reproducibility of the models tuned with correction terms was higher than those with adjusted parameters.

    KW - Competition

    KW - DGVM

    KW - Forest dynamics

    KW - Gap model

    KW - Individual-based model

    KW - Size structure model

    UR - http://www.scopus.com/inward/record.url?scp=85028056403&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=85028056403&partnerID=8YFLogxK

    U2 - 10.1016/j.ecolmodel.2017.07.018

    DO - 10.1016/j.ecolmodel.2017.07.018

    M3 - Article

    AN - SCOPUS:85028056403

    VL - 361

    SP - 95

    EP - 112

    JO - Ecological Modelling

    JF - Ecological Modelling

    SN - 0304-3800

    ER -