Harnessing iNaturalist to quantify hotspots of urban biodiversity: the Los Angeles case study
Harnessing iNaturalist to quantify hotspots of urban biodiversity: the Los Angeles case study
复制标题
利用iNaturalist量化城市生物多样性热点：洛杉矶案例研究
DOI：
10.3389/fevo.2023.983371
复制
发表时间：
2023-06
期刊：
影响因子：
--
通讯作者：
Joscha Beninde;Tatum W. Delaney;Germar Gonzalez;H. B. Shaffer
中科院分区：
文献类型：
--
作者：
Joscha Beninde;Tatum W. Delaney;Germar Gonzalez;H. B. Shaffer
保护规划的一个主要目标是优先保护和管理拥有最大生物多样性的地区。然而，这种空间优先排序往往受到有限的数据可用性的影响，导致决策受到少数标志性或濒危物种的驱动，对共同发生的分类群的好处不确定。我们认为，基于野外观测的多物种栖息地偏好应该指导保护规划，以优化尽可能多的物种的长期持久性。方法利用生境适宜性建模技术和来自iNaturalist社区科学平台的数据，提出了一种开发空间明确的生境适宜性模型的策略，从而实现更好的基于地点的保护优先级。以大洛杉矶地区为例，利用Maxent和Random Forests建立了1200个陆生物种的适宜性模型，这些物种至少有25个发生记录，分别来自植物（45.5%）、节肢动物（27.45%）、脊椎动物（22.2%）、真菌（3.2%）、软体动物（1.3%）和其他分类类群（< 0.3%）。该建模策略进一步比较了iNaturalist数据集固有的空间稀疏和分类偏差文件修正，在荒地和城市子区域联合和单独建模物种，并使用零模型和未用于训练模型的物种和事件的“测试”数据集验证了模型的性能。结果所有物种组合的生境适宜性平均模型在不同模型设置下相似，但随机森林模型在模型评价中获得最高的AUCROC和AUCPRG中值。不同分类类群对城市化梯度的响应差异不大，而本地和非本地物种在大多数城市和大多数野生生境中表现出截然不同的模式，并且在城市-野生界面附近的平均生境适宜性均达到峰值。我们的建模框架完全基于开源软件，我们的代码供进一步使用。鉴于iNaturalist等平台提供的城市生物多样性数据越来越多，该建模框架可以很容易地应用于其他地区。以这种方式量化大量具有代表性的本地物种的栖息地适宜性，为进一步的生态研究和保护决策提供了清晰的数据驱动基础，最大限度地提高了当前和未来保护工作的影响。
Introduction A major goal for conservation planning is the prioritized protection and management of areas that harbor maximal biodiversity. However, such spatial prioritization often suffers from limited data availability, resulting in decisions driven by a handful of iconic or endangered species, with uncertain benefits for co-occurring taxa. We argue that multi-species habitat preferences based on field observations should guide conservation planning to optimize the long-term persistence of as many species as possible. Methods Using habitat suitability modeling techniques and data from the community-science platform iNaturalist, we provide a strategy to develop spatially explicit models of habitat suitability that enable better informed, place-based conservation prioritization. Our case study in Greater Los Angeles used Maxent and Random Forests to generate suitability models for 1,200 terrestrial species with at least 25 occurrence records, drawn from plants (45.5%), arthropods (27.45%), vertebrates (22.2%), fungi (3.2%), molluscs (1.3%), and other taxonomic groups (< 0.3%). This modeling strategy further compared spatial thinning and taxonomic bias file corrections to account for the biases inherent to the iNaturalist dataset, modeling species jointly and separately in wildland and urban sub-regions and validated model performance using null models and a “test” dataset of species and occurrences that were not used to train models. Results Mean models of habitat suitability of all species combined were similar across model settings, but the mean Random Forest model received the highest median AUCROC and AUCPRG scores in model evaluation. Taxonomic groups showed relatively modest differences in their response to the urbanization gradient, while native and non-native species showed contrasting patterns in the most urban and the most wildland habitats and both peaked in mean habitat suitability near the urban-wildland interface. Discussion Our modeling framework is based entirely on open-source software and our code is provided for further use. Given the increasing availability of urban biodiversity data via platforms such as iNaturalist, this modeling framework can easily be applied to other regions. Quantifying habitat suitability for a large, representative subset of the locally occurring pool of species in this way provides a clear, data-driven basis for further ecological research and conservation decision-making, maximizing the impact of current and future conservation efforts.