Move it or lose it: Predicted effects of culverts and population density on Mojave desert tortoise (Gopherus agassizii) connectivity

Abstract

Roadways and railways can reduce wildlife movements across landscapes, negatively impacting population connectivity. Connectivity may be improved by structures that allow safe passage across linear barriers, but connectivity could be adversely influenced by low population densities. The Mojave desert tortoise is threatened by habitat loss, fragmentation, and population declines. The tortoise continues to decline as disturbance increases across the Mojave Desert in the southwestern United States. While underground crossing structures, like hydrological culverts, have begun receiving attention, population density has not been considered in tortoise connectivity. Our work asks a novel question: How do culverts and population density affect connectivity and potentially drive genetic and demographic patterns? To explore the role of culverts and population density, we used agent-based spatially explicit forward-in-time simulations of gene flow. We constructed resistance surfaces with a range of barriers to movement and representative of tortoise habitat with anthropogenic disturbance. We predicted connectivity under variable population densities. Simulations were run for 200 non-overlapping generations (3400 years) with 30 replicates using 20 microsatellite loci. We evaluated population genetic structure and diversity and found that culverts would not entirely negate the effects of linear barriers, but gene flow improved. Our results also indicated that density is important for connectivity. Low densities resulted in declines regardless of the landscape barrier scenario (> 75% population census size, > 97% effective population size). Results from our simulation using current anthropogenic disturbance predicted decreased population connectivity over time. Genetic and demographic effects were detectable within five generations (85 years) following disturbance with estimated losses in effective population size of 69%. The pronounced declines in effective population size indicate this could be a useful monitoring metric. We suggest management strategies that improve connectivity, such as roadside fencing tied to culverts, conservation areas in a connected network, and development restricted to disturbed areas.

Fig 1. Landscape resistance surfaces.

Three hypothetical landscapes (no barrier, absolute barrier that is one grid cell wide, barrier with culverts) and a real world study area in the Ivanpah Valley along the Nevada/California border were created. The Ivanpah Valley surface used the inverse of desert tortoise habitat suitability [62], with the addition of existing anthropogenic disturbance layers. Disturbance was applied using conversion factors to adjust cell values and increase resistance. Types of disturbance are reported in Table 2. The extent of each surface is 25 x 25 km with a resolution of 1 km².

Fig 2. Map of tortoise locations.

Individuals were from the Ivanpah Valley, along the Nevada/California border, and used as the original genetic data in forward-in-time simulations (n = 170). The state overview map has been adapted from the USGS National Boundary Dataset public domain and is for illustrative purposes only.

Fig 3. Predicted heterozygosity (Ho) and pairwise genetic differentiation (F_ST) over time.

Individuals were grouped relative to a linear barrier in three hypothetical landscapes (no barrier, absolute barrier, barrier with culverts) at three population densities (low, moderate, high) and a real world study area in the Ivanpah Valley (IV) along the Nevada/California border. In all hypothetical landscapes F_ST values for the scenario with culverts were intermediate, indicating that gene flow increased relative to an absolute barrier. At low densities, changes in Ho and F_ST likely reflected small population sizes that were increasingly isolated over time. Current estimates for IV likely indicate the time lag in detection, with Ho higher than predicted due to the scale and recentness of habitat loss and F_ST representing an existing signal of fragmentation from the interstate.

Fig 4. Predicted sPCA regression results.

The original genetic data and simulation results from a real world study area in the Ivanpah Valley along the Nevada/California border forecasting an increased signal of fragmentation with current levels of disturbance over time. Results are shown at discrete generations (5, 40, 200). Points represent genotypes and spatial locations. A correlation between synthetic variables (principal components) indicates spatial autocorrelation and is shown by the linear regression line. Note differences in x and y axes.