COMPARISON OF PREDICTOR SELECTION PROCEDURES IN SPECIES DISTRIBUTION MODELING: A CASE STUDY OF Fagus hayatae

Selecting predictors for species distribution models (SDMs) is a major challenge. In this study, we evaluated a comprehensive set of 62 environmental predictors that may be related to the occurrence of Fagus hayatae. We modeled F. hayatae as a case study to compare model performance through different environmental predictor subsets according to three selection procedures, namely correlation coefficients between predictors, contribution level of predictors, and expert choice of biologically relevant predictors. The three selection procedures provided satisfactory results with high performance using about 6-10 valid predictors but had their respective limitations. Consequently, we suggest a synthetic strtegy of predictor selection. Accordingly, the first step was identifying and eliminating ineffective variables with nonidentifiability by using bivariate scatterplots. Next, calculate the correlation coefficients between other candidate predictors. Finally, comprehensively select the applicable environmental predictors with lower correlation coefficient on the basis of highly contribution level and expert knowledge for SDM of target species.

Keywords:
Synthetic strtegy; Effective warmth index; Expert knowledge; Environmental variableTaiwan

Authorship

Cheng-Tao Lin

National Chiayi University, Chiayi, TaiwanNational Chiayi UniversityTaiwanChiayi, Taiwan National Chiayi University, Chiayi, Taiwan

http://orcid.org/0000-0003-2857-1625

Ching-An Chiu ⁺ + cachiu@nchu.edu.tw

National Chung Hsing University, Taichung, TaiwanNational Chung Hsing UniversityTaiwanTaichung, Taiwan National Chung Hsing University, Taichung, Taiwan

http://orcid.org/0000-0002-6134-0530

+ cachiu@nchu.edu.tw

SCIMAGO INSTITUTIONS RANKINGS

National Chiayi University, Chiayi, TaiwanNational Chiayi UniversityTaiwanChiayi, Taiwan National Chiayi University, Chiayi, Taiwan

National Chung Hsing University, Taichung, TaiwanNational Chung Hsing UniversityTaiwanTaichung, Taiwan National Chung Hsing University, Taichung, Taiwan

Figures | Tables

Figures (5)
Tables (2)

Thumbnail

FIGURE 1
(A) Geographic location of Taiwan. (B) Digital elevation model of the study area and spatial distribution (red points) of Fagus hayatae. (C) Partially enlarged occurrence data (red sampling points) of F. hayatae.

Thumbnail

FIGURE 2
Examples of scatterplots of variables: (A) scatterplot matrix between all environmental predictors generated by AcrGIS; (B) TAR vs. DIS plot (for abbreviations, refer to Supplementary Table 1), which reveals a random relationship (|r| < 0.001); (C) ELE vs. Bio1 plot, which presents high correlation (|r| > 0.998); (D) CI vs. TRAI plot, which indicates no discrimination between CI values in sampling points. Blue points represent the values of environmental predictors on all modeling grids; red points represent the values of environmental predictor on all Fagus hayatae occurrence grids.

Thumbnail

FIGURE 3
Comparison of model performance. Model performance was compared using the ROC by AUC and TSS criteria within the (A, B) 10-R subsets, (C, D) 7-C subsets, and (E, F) 7-E subsets. Each box-plot with a mean (dotted line) illustrates results of 20 cross-validation runs based on different predictor subsets by using backward elimination.

Thumbnail

FIGURE 4
Average of and difference in probability maps. The average of (left) and difference in (right) probability maps based on different MaxEnt predictor subsets from three backward elimination procedures.

Thumbnail

FIGURE 5
Comparison of Fagus hayatae predictive maps using R, C, E, and synthetic (abbreviated as S) strategies to select 7 factors. The area enclosed by the black line is the true distribution area of F. hayatae.

Thumbnail

TABLE 1
The total 22 subsets of predictors selected through backward elimination used in different SDM models, labeled with three selection procedures (Pearson correlation coefficient as “R,” training contribution level as “C,” and expert selection as “E”) and appended with the number of predictors.

Thumbnail

TABLE 2
Pearson correlation coefficient (r) between predictors and their contribution percent (CP, %) individually in the R6, C6, and E6 models. Italicization indicates a strong correlation between predictors.).

FIGURE 1 (A) Geographic location of Taiwan. (B) Digital elevation model of the study area and spatial distribution (red points) of Fagus hayatae. (C) Partially enlarged occurrence data (red sampling points) of F. hayatae.

FIGURE 2 Examples of scatterplots of variables: (A) scatterplot matrix between all environmental predictors generated by AcrGIS; (B) TAR vs. DIS plot (for abbreviations, refer to Supplementary Table 1), which reveals a random relationship (|r| < 0.001); (C) ELE vs. Bio1 plot, which presents high correlation (|r| > 0.998); (D) CI vs. TRAI plot, which indicates no discrimination between CI values in sampling points. Blue points represent the values of environmental predictors on all modeling grids; red points represent the values of environmental predictor on all Fagus hayatae occurrence grids.

FIGURE 3 Comparison of model performance. Model performance was compared using the ROC by AUC and TSS criteria within the (A, B) 10-R subsets, (C, D) 7-C subsets, and (E, F) 7-E subsets. Each box-plot with a mean (dotted line) illustrates results of 20 cross-validation runs based on different predictor subsets by using backward elimination.

FIGURE 4 Average of and difference in probability maps. The average of (left) and difference in (right) probability maps based on different MaxEnt predictor subsets from three backward elimination procedures.

FIGURE 5 Comparison of Fagus hayatae predictive maps using R, C, E, and synthetic (abbreviated as S) strategies to select 7 factors. The area enclosed by the black line is the true distribution area of F. hayatae.

TABLE 1 The total 22 subsets of predictors selected through backward elimination used in different SDM models, labeled with three selection procedures (Pearson correlation coefficient as “R,” training contribution level as “C,” and expert selection as “E”) and appended with the number of predictors.

Label	Subsets of selected predictors
R61 = C61 = E61	T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, P1, P2, P3, P4, P5, P6, P7, P8, P9, P10, P11, P12, Bio1, Bio4, Bio5, Bio6, Bio8, Bio9, Bio10, Bio12, Bio15, Bio16, Bio17, Bio18, BT, PER, WI, EWI, HI, PS, PW, PSR, PWR, TAR, ELE, SLO, LAT, LON, SR, DIS, ROU, SRR, CUR, SP, SAR, CTI, HLI, TRAI, WLS
R32	P2, P3, P4, P5, P6, P7, P8, Bio4, Bio5, Bio6, Bio12, Bio15, Bio16, Bio17, Bio18, PER, WI, HI, PS, PSR, SLO, LAT, LON, SR, DIS, SRR, CUR, SAR, CTI, HLI, TRAI, WLS
R27	P2, P3, P5, P6, P7, P8, Bio4, Bio5, Bio6, Bio15, Bio16, Bio17, Bio18, PER, WI, HI, PS, LAT, SR, DIS, SRR, CUR, SAR, CTI, HLI, TRAI, WLS
R23	P2, P3, P5, P6, P8, Bio4, Bio6, Bio15, Bio16, Bio17, Bio18, PER, PS, LAT, SR, DIS, SRR, CUR, SAR, CTI, HLI, TRAI, WLS
R21	P3, P5, P6, P8, Bio4, Bio6, Bio15, Bio16, Bio17, Bio18, PER, PS, LAT, SR, SRR, CUR, SAR, CTI, HLI, TRAI, WLS
R18	P5, Bio4, Bio6, Bio15, Bio16, Bio17, Bio18, PER, PS, LAT, SR, SRR, CUR, SAR, CTI, HLI, TRAI, WLS
R12	Bio4, Bio15, Bio16, Bio18, PER, SR, SRR, CUR, SAR, CTI, TRAI, WLS
R9	Bio4, Bio15, Bio18, PER, SR, SRR, SAR, CTI, WLS
R8	Bio4, Bio15, Bio18, PER, SR, SRR, SAR, CTI
R6	Bio15, Bio18, PER, SR, SRR, CTI
C30	T1, T6, T7, T8, P1, P2, P4, P5, P6, P8, P9, P10, P11, Bio4, Bio5, Bio6, Bio9, Bio10, Bio15, Bio16, Bio17, Bio18, HI, PS, PW, TAR, ELE, LAT, LON, TRAI
C28	T6, T7, T8, P1, P2, P4, P5, P6, P8, P9, P10, P11, Bio4, Bio5, Bio6, Bio9, Bio10, Bio15, Bio16, Bio17, Bio18, HI, PS, TAR, ELE, LAT, LON, TRAI
C23	T6, T7, T8, P1, P2, P8, P9, P10, P11, Bio4, Bio5, Bio10, Bio15, Bio16, Bio17, Bio18, HI, PS, TAR, ELE, LAT, LON, TRAI
C15	T6, T7, T8, P1, P8, Bio4, Bio5, Bio15, Bio16, Bio18, HI, PS, LAT, LON, TRAI
C9	T6, T7, P1, Bio5, Bio15, Bio16, HI, PS, LON
C6	T6, P1, Bio5, Bio16, PS, LON
E31	Bio4, Bio5, Bio6, Bio8, Bio9, Bio10, Bio12, Bio15, Bio16, Bio17, Bio18, PER, WI, EWI, HI, PS, PW, PSR, PWR, SLO, SR, DIS, ROU, SRR, CUR, SP, SAR, CTI, HLI, TRAI, WLS
E26	Bio4, Bio5, Bio6, Bio8, Bio9, Bio10, Bio12, Bio15, Bio16, Bio17, PER, WI, EWI, HI, PS, PW, SLO, SR, DIS, ROU, CUR, SP, SAR, CTI, TRAI, WLS
E21	Bio4, Bio5, Bio6, Bio8, Bio9, Bio12, Bio15, Bio17, PER, WI, EWI, PS, PW, SLO, SR, CUR, SP, SAR, CTI, TRAI, WLS
E16	Bio4, Bio5, Bio8, Bio12, Bio15, Bio17, PER, WI, EWI, PS, SLO, SR, SP, CTI, TRAI, WLS
E11	Bio4, Bio8, Bio12, Bio17, PER, WI, EWI, SR, SP, CTI, WLS
E6	Bio4, Bio12, PER, WI, CTI, WLS

TABLE 2 Pearson correlation coefficient (r) between predictors and their contribution percent (CP, %) individually in the R6, C6, and E6 models. Italicization indicates a strong correlation between predictors.).

R6 model	Bio15	Bio18	PER	SR	SRR	CTI
Bio15		-0.34	-0.09	-0.02	0.00	-0.01
Bio18			0.23	0.00	0.00	0.02
PER				-0.22	-0.08	0.07
SR					0.08	0.08
SRR						-0.05
CTI
CP (%)	33.8	18.2	46.7	1.1	0.1	0.0
C6 model	T6	P1	Bio5	Bio16	PS	LON
T6		0.40	0.90	0.32	0.36	0.37
P1			0.34	0.59	0.53	0.56
Bio5				0.30	0.34	0.36
Bio16					0.68	0.93
PS						0.59
LON
CP (%)	22.7	12.7	10.6	28.4	11.6	14.0
E6 model	Bio4	Bio12	PER	WI	CTI	WLS
Bio4		0.20	0.54	0.88	0.08	0.01
Bio12			-0.62	0.42	0.00	0.17
PER				0.44	0.07	-0.20
WI					0.09	-0.03
CTI						-0.16
WLS
CP (%)	14.9	38.2	20.2	26.3	0.1	0.4

How to cite

copy

UFLA - Universidade Federal de Lavras Universidade Federal de Lavras - Departamento de Ciências Florestais - Cx. P. 3037, 37200-000 Lavras - MG Brasil, Tel.: (55 35) 3829-1706, Fax: (55 35) 3829-1411 - Lavras - MG - Brazil
E-mail: cerne@dcf.ufla.br

Acompanhe os números deste periódico no seu leitor de RSS

Versão para download de PDF

PDF

Inglês

Versões e tradução automática

Versão original do texto

English

Tradução automática

Google Translator
Microsoft Translator

Como citar

RIS

BIBTEX

Outros formatos de citação e exportação:

SciELO - Scientific Electronic Library Online
Rua Dr. Diogo de Faria, 1087 – 9º andar – Vila Clementino 04037-003 São Paulo/SP - Brasil
E-mail: scielo@scielo.org

Leia a Declaração de Acesso Aberto

Métricas

COMPARISON OF PREDICTOR SELECTION PROCEDURES IN SPECIES DISTRIBUTION MODELING: A CASE STUDY OF Fagus hayatae

PlumX

Mensagem

[1] + cachiu@nchu.edu.tw