JEI期刊论文：通过机器学习和扩散核方法揭示欧亚水獭（Lutra Lutra）与环境因素相关的空间分布模式研究

Posted 2021-04-03 JEI期刊

原文下载链接:

http://jeionline.org/share/202000443.pdf

原文信息

Title: Spatial Distribution Patterns of Eurasian Otter (Lutra Lutra) in Association with Environmental Factors Unravelled by Machine Learning and  Diffusion Kernel Method

Authors: S. Hong^{1, 2}, T.-S. Chon^{2, 3*}, and G. J. Joo^1*

1. Department of Biological Sciences, Pusan National University, Busan 46241, Republic of Korea

2. Department of Forest Environment Protection, College of Forest and Environmental Sciences, Kangwon National University, Chuncheon 25931, Republic of Korea

3. Ecology and Future Research Association, Busan 46228, Republic of Korea

Abstract

In South Korea, the endangered Eurasian otter (Lutra lutra) populations have been recovered throughout the country. To examine the status of otter populations, we monitored spraint densities at 250 ~ 355 sites annually from 2014 to 2017 in the Nakdong River basin. The diffusion kernel method was applied to both binary and continuous spraint data. Two geographical populations were identified: northern and southern populations. The northern population continuously increased over a broad area from north to south during the study period. In contrast, the southern population narrowly dispersed, limited by its location in an industrial area. The spatial self-organizing map (Geo-SOM) revealed associations between spraint densities and environmental factors by correlating the geographic locations of the sampling sites. Both populations were negatively affected by anthropogenic factors, including proximity to factories and human population density. However, cumulative association of all environmental factors, including landscape, food sources, and anthropogenic factors, were noted in 2016 after which otter populations fully recovered. Population development stabilized while exhibiting an overall high association with environmental factors. The results of the diffusion kernel method and data variation according to the Geo-SOM consistently presented substantial change in population dispersal (i.e. the merge of two subpopulations, and complete associations between spraint and environmental factors). The combination of the diffusion kernel method and Geo-SOM was effective in portraying temporal changes in population states in association with environmental factors based on spraint data in the last phase of full recovery.

Keywords: dispersal; recovery; spraints; habitat preference; conservation; Republic of Korea

 

文中重要图表信息

Figure 1. Study area. 250 sites between 2014 and 2016, and 355 sites during 2017 in the Nakdong River basin. Red circles (47 sample sites) represent the consistent sites throughout survey periods (2014 ~ 2017). Light green (203 sample sites) and light red circles (308 sample sites) represent the survey sites from 2014 to 2016 and 2017, respectively. Dark and light blue lines show the main river and tributaries of local rivers near the east coast, respectively. The upper left panel shows South Korea, and the light green area in the rectangle represent the Nakdong River basin.

Figure 2. (a) Four maps of binary data using diffusion kernel estimations. Arrows indicate origin of S population. (b) Four maps using diffusion kernel estimations with additive barriers (urban area densities). (c) Urban area densities. Red and green colours represent high and low levels of urbanisation, respectively. White dotted circles in the fourth map of panel (a), first, third, and fourth maps of panel (b), and map (c) indicate the industrial belts (Gumi, Daegu, and Changwon).

Figure 3. U matrices for N population in (a) 2014, (b) 2015, (c) 2016, and (d) 2017; and S population in (e) 2014, (f) 2015, (g) 2016, and (h) 2017. Sites clustered with high abundance of spraint densities in (i) N population and (j) S population. Hexagons ranging from light to deep purple and outlined in black indicate the strongest spatial clusters from earlier (2014) to later (2017) survey years, corresponding to study sites with the same-coloured circles in legend labels (i and j). Open white circles outlined in corresponding colours for each year indicate highest regional spraint density clusters (c, d) and sites (i). The arrow indicates the origin of the S population.

Figure 4. Component maps for N population in (a) 2014, (b) 2015, (c) 2016, and (d) 2017; and S population in (e) 2014, (f) 2015, (g) 2016, and (h) 2017. Red and blue rectangles outside of the abbreviated environmental factors indicate positive and negative relationships with spraint densities, respectively. All arrows correspond to the coordinate information, latitude (LAT) and longitude (LON), with higher spraint density areas. The values of all variables on the component planes were normalized between 0 (blue) and 1 (red).

Figure 5. Spraint densities comparing the average (black line), S population (dotted red line), and N population (dotted blue line). Lightly coloured areas represent the standard error of each line. 

Figure 6. (a) Four maps using spraint densities data applying diffusion kernel estimations. (b) Four maps using diffusion kernel estimations with additive barriers (urban area densities). (c) Urban area densities. Red and green colours represent high and low levels of urbanisation, respectively. Arrows in the first map of upper panel (a) indicate the origin of each population, and third map of panel (a, b) indicate the combined area between the two populations. Dotted arrows in the fourth map of (a, b) suggest small regional populations. White circles in land cover map (c) represent the areas corresponding to all arrows of the third and fourth maps (a, b).

     ▼

更多精彩推荐，请关注我们

▼