Skip to main content
Dryad logo

Hierarchical multi-grain models improve descriptions of species’ environmental associations, distribution, and abundance

Citation

Mertes, Katherine; Jarzyna, Marta; Jetz, Walter (2020), Hierarchical multi-grain models improve descriptions of species’ environmental associations, distribution, and abundance, Dryad, Dataset, https://doi.org/10.5061/dryad.cz8w9gj0f

Abstract

The characterization of species’ environmental niches and spatial distribution predictions based on them are now central to much of ecology and conservation, but implicitly requires decisions about the appropriate spatial scale (i.e. grain) of analysis. Ecological theory and empirical evidence suggest that range-resident species respond to their environment at two characteristic, hierarchical spatial grains: (i) response grain, the (relatively fine) grain at which an individual uses environmental resources, and (ii) occupancy grain, the (relatively coarse) grain equivalent to a typical home range. We use a multi-grain (MG) occupancy model, aided by fine-grain remotely sensed imagery, to simultaneously estimate species-environment associations at both grains, conduct grain optimization to measure response grain, and apply this analysis framework to an example species: a medium-sized bird (Tockus deckeni) in a heterogeneous East African landscape. Based on home range analysis of movement data, we calculate an occupancy grain of 1km for T. deckeni. Using a grain optimization procedure across 32 grains from 10m to 500m, we identify 60m as the most strongly supported response grain for a suite of environmental variables, slightly coarser than opportunistic behavioral observations would have suggested. Validation confirms that the accuracy of the optimized MG occupancy model substantially exceeds that of equivalent single-grain (SG) occupancy models. We further use a simulation approach to assess the potential impacts of accounting for the multi-scale structure of species’ environmental requirements on estimates of population size. We find that the more strongly supported MG approach consistently predicts a minimum population sizes in the study landscape that is much lower than that provided by the SG model. This suggests that SG approaches commonly used in conservation applications could lead to overly optimistic abundance and population estimates and that the MG approach may be more appropriate for supporting species conservation goals. More generally, we conclude that multi-grain approaches of the sort presented, and increasingly enabled by growing high-resolution remotely sensed data, hold great promise for offering a more mechanistic framework for assessing the appropriate grain(s) for population monitoring and management and enable more reliable estimates of abundances and species’ distributions.

Methods

During April 2014 – December 2014, we conducted surveys for Tockus deckeni (von Der Decken's hornbills) within 75, 1km x 1km sampling sites ("1km grid cells") at Mpala Research Centre ("MRC"; Laikipia, Kenya). Each 1km grid cell was surveyed approximately bimonthly during morning (06:30-10:30) and afternoon (15:30-18:30) periods. Each indvidual survey searched at least 10% of the 1km grid cell's area and lasted at least 30 minutes. Because we searched an entire 1km grid cell across multiple surveys, individual survey routes were not identical. Any adult T. deckeni encountered during a survey was followed for up to 15 consecutive locations. We recorded each bird location using distance-and-bearing methods: distance measured by rangefinder (uncertainty ± 1m), and bearing measured by compass (uncertainty ± 1°). 

We then converted these survey data into series of detections and non-detections in a grid of 10m x 10m cells laid over MRC. To determine the area searched effectively during an individual survey, we calculated the median distance at which T. deckeni were first detected during the study period (44m) and buffered each survey route by this distance. All observations outside this searched area (i.e., more than 44m from the survey route) were discarded. Since accounting for detection probability requires repeated visits to the same location, we also discarded any 10m grid cells surveyed less than three times during the study period. Finally, to avoid false negatives in non-detection data, we established a conservative minimum distance (500m) between detections and non-detections in searched areas where T. deckeni were detected. For the minimum survey duration (30 minutes), we estimated the median distance traveled by T. deckeni (287m). For a survey in which T. deckeni was detected, only 10m grid cells that were both within the searched area and at least 500m from any detection were considered non-detections. For a survey in which no T. deckeni were detected, all 10m grid cells within the searched area were considered non-detections.

Usage Notes

The data are presented as a matrix in which every row represents an indiviudal 10m x 10m grid cell in the study area.

Columns 2:21 contain detection event for an individual survey: either 1 (detection of adult T. deckeni), 0 (non-detection of adult T. deckeni), or NA (no survey performed).

Columns 22:41 contain dates for each detection event. Each 10m cell has a varying number (3-20) survey occasions and an equivalent number of detection events. (The value NA was used to fill empty survey events and dates.)

The accompanying RData file "grids_1km_10m" provides the spatial coordinates for 1km x 1km and 10x 10m grids over the study area.

Funding

National Science Foundation, Award: BCS-1333424

National Aeronautics and Space Administration, Award: NNX13AP11H