Maintaining biodiversity is achieved by understanding the spatial distribution of suitable conditions to sustain viable populations to prioritize conservation efforts. Land managers are limited by available spatial information on rare plant species and can benefit from the spatial and temporal scale of herbaria data held in open access data aggregators. In many cases, herbaria data are the only spatial data available. Unmeasurable error and bias in museum specimen collections can lead to uncertain predictions and negatively impact setting policy on the ground. The development of cross-validation methods have increased the modeling power and spatial inference of species distribution models but limitations and error in primary data hinders predictions of the current and future population trends. Population extents of 59 plant species mapped in a uniform and defined manner by the Colorado Natural Heritage Program was used to understand the impact of unmeasurable spatial bias, geo-referencing errors, and sparse data that comprise herbaria datasets. These mapped populations were used as the true distribution of species to examine the spatial error of herbarium records over time and across habitat specificity of the species. Species distribution models for recent herbarium records were compared to mapped populations to understand how multiple sources of unmeasurable error and bias impact the resulting species distribution inference.
Results/Conclusions
Contrary to assumed patterns of biases and lack of precision, older specimens are no more likely to have spatial error than collections made after the adoption of hand held global positioning system usage. Error is likely introduced due to an overreliance and confidence in the stated geographic coordinates. Spatial error is more likely in species with a smaller range and is far larger than the spatial scale needed to be informative for environmental and topographic predictor variables. Bias and error in herbaria data decrease the model performance metrics. When bias and error cannot be measured, appropriate modeling techniques cannot be used to address the problems. Despite spatial error in natural history collections, herbaria are critical to species distribution model directed conservation. To make the best use of what is often the only existing data, we must understand how these disparate sources of error and bias alter the resulting model predictions.