Modeling spatially biased citizen science effort through the eBird database
Modeling spatially biased citizen science effort through the eBird database
复制标题
DOI：
10.1007/s10651-021-00508-1
复制
发表时间：
2021-06
影响因子：
3.8
通讯作者：
Becky Tang;J. Clark;A. Gelfand
中科院分区：
文献类型：
--
作者：
Becky Tang;J. Clark;A. Gelfand
公民科学数据库作为生态信息来源的重要性与日俱增，但不同地点的努力差异是此类数据所固有的。预计会出现空间偏差数据，即在研究区域内未统一采样的数据。偏差的进一步引入是不同地点采样活动水平的可变性。这激励了我们的工作：通过访问位置的空间数据集和这些位置的采样活动，我们提出了一种基于模型的方法来评估这些位置的工作量。调整访问地点和工作量方面的潜在空间偏差对于开发可靠的物种分布模型（SDM）至关重要。使用 eBird（致力于鸟类区系的全球公民科学数据库）以及宾夕法尼亚州和德国的说明性区域的数据，我们对观测位置和观测活动的空间依赖性进行了建模。我们采用点过程模型来解释空间中观测到的位置，拟合地统计模型来解释位置处的观测工作，并探索优先采样的潜在存在性，即两个过程之间的依赖性。总而言之，我们结合了有关位置和活动的信息，提供了更丰富的采样工作概念。由于 SDM 经常因其预测能力而被使用，因此我们方法的一个重要优势是能够预测未观察位置和区域内的工作量。通过这种方式，我们可以适应点参考数据之间的偏差，例如所需的面尺度密度。我们简要说明了我们提出的方法如何应用于 SDM，并通过模型的努力证明了预测的改进。
Citizen science databases are increasing in importance as sources of ecological information, but variability in effort across locations is inherent to such data. Spatially biased data—data not sampled uniformly across the study region—is expected. A further introduction of bias is variability in the level of sampling activity across locations. This motivates our work: with a spatial dataset of visited locations and sampling activity at those locations, we propose a model-based approach for assessing effort at these locations. Adjusting for potential spatial bias both in terms of sites visited and in terms of effort is crucial for developing reliable species distribution models (SDMs). Using data from eBird, a global citizen science database dedicated to avifauna, and illustrative regions in Pennsylvania and Germany, we model spatial dependence in both the observation locations and observed activity. We employ point process models to explain the observed locations in space, fit a geostatistical model to explain observation effort at locations, and explore the potential existence of preferential sampling, i.e., dependence between the two processes. Altogether, we offer a richer notion of sampling effort, combining information about location and activity. As SDMs are often used for their predictive capabilities, an important advantage of our approach is the ability to predict effort at unobserved locations and over regions. In this way, we can accommodate misalignment between point-referenced data and say, desired areal scale density. We briefly illustrate how our proposed methods can be applied to SDMs, with demonstrated improvement in prediction from models incorporating effort.