Protected-area boundaries as filters of plant invasions

Abstract

Human land uses surrounding protected areas provide propagules for colonization of these areas by non-native species, and corridors between protected-area networks and drainage systems of rivers provide pathways for long-distance dispersal of non-native species. Nevertheless, the influence of protected-area boundaries on colonization of protected areas by invasive non-native species is unknown. We drew on a spatially explicit data set of more than 27,000 non-native plant presence records for South Africa's Kruger National Park to examine the role of boundaries in preventing colonization of protected areas by non-native species. The number of records of non-native invasive plants declined rapidly beyond 1500 m inside the park; thus, we believe that the park boundary limited the spread of non-native plants. The number of non-native invasive plants inside the park was a function of the amount of water runoff, density of major roads, and the presence of natural vegetation outside the park. Of the types of human-induced disturbance, only the density of major roads outside the protected area significantly increased the number of non-native plant records. Our findings suggest that the probability of incursion of invasive plants into protected areas can be quantified reliably.

Figure 1

Mean annual river run off in Kruger National Park (KNP), pattern of non-native plant records relative to KNP boundary and segments (lower left inset), and location of KNP within South Africa (upper left inset).

Figure 2

Number of presence records of non-native plants relative to the distance from the boundary of Kruger National Park toward the interior of the park: (a) 52,000 m and (b) 3000 m. Fitted curves are LOWESS regression models on square-root number of records (backtransformed for visualization) with no patterns of residuals. In (a) span of smoothing = 0.1; equivalent number of parameters (ENP) in curvilinear regression = 29.8; explained variance r²= 0.82; model chosen by deletion test (F =1.16; ENP = 313.2, 29.8; p =0.13) against starting model with span = 0.01 and r²= 0.94. In (b) span of smoothing = 0.5; ENP = 6.4; r²= 0.96; model chosen by deletion test (F =2.37; ENP = 15.1, 6.4; p =0.07) against starting model with span = 0.2 and r²= 0.99. The spans of smoothing describe local sensitivity in iterations of LOWESS models (Supporting Information).

Figure 3

Results of classification tree analysis of the binary probability of the non-native species presence in Kruger National Park (KNP). Probability of presence was determined on the basis of water runoff in the park (mean annual runoff from the surrounding watershed; million m³) and road density adjacent to the park (density of major roads within 10-km radius outside the KNP boundary) (Supporting Information) (%, percentage of cases for each class; bars, representation of the percentage of absent [gray] and present [black]). The splitting variable name and split criterion are given for each node. Vertical depth of each node is proportional to its improvement value. Overall misclassification of the optimal tree was 14.0%, sensitivity 0.92 and specificity 0.81 (Supporting Information).