基于公民科学数据测算物种保护优先性的 方法优化研究——应用机器学习与智能优化 算法

Optimizing the Framework for SpeciesConservation Priorities Calculation Based onCitizen Science Data: Application of MachineLearning and Intelligent Optimization Algorithms

关键词:  风景园林  风景园林规划  生物多样性保护  物种 分布模型  人工智能  公民科学

Key words:  landscape architecture  landscape planning  biodiversity conservation  species distribution model  artificial intelligence  citizen science

摘要:

公民科学观测记录是生物多样性保护相关规划研究 和实践中常用的数据来源，但存在记录点代表性有限和热点聚 集等问题，充分了解其局限性及可能存在的偏差对于有效的保 护规划至关重要。选用广州市100条鸟类系统调查样线数据作 为基准，分析同时期公民科学数据的偏差情况及不同数据筛选 方法和优先性测算方法的改善效果。采用3种数据稀疏方式减 小公民科学数据热点聚集产生的影响，对鸟类记录点及其所处 环境进行机器学习并构建275种鸟类的分布模型，基于此测算 保护优先性。广州市公民科学观测记录数据热点聚集明显，对 其进行稀疏有助于减小物种分布模拟偏差，但与基于系统调查 数据得出的结果相比仍具有较大差距。对比传统丰富度方法与 智能优化算法的保护优先性测算结果显示，智能优化算法可以 更有效地识别丰富度不高但对特定物种更重要的区域，且对公 民科学数据采样不均问题带来的保护优先性测算结果偏差具有 良好的改善效果。因此，在利用公民科学观测记录进行规划研 究和实践时，宜采用多种方式进行数据筛选、物种分布模拟及 保护优先性测算，以取得更加可靠的结果。

Abstract:

Citizen science records are commonly used as data sources in planning research and practice. However, there are issues including limited representativeness of observation points and clustering of hotspots, leading to biases in the analysis based on them. It is important to fully understand their limitations and possible deviations for effective conservation planning. We used bird survey data from 100 transects in Guangzhou as a benchmark to assess how various data filtering and prioritization methods improves the results from concurrent citizen science data. Applied machine learning with a max entropy model to simulate 275 bird species' occurrence probabilities according to their records and environments, which then informed conservation priority calculations. The results revealed clustering in Guangzhou's citizen science data. Sparsity can alleviate deviation, yet it still lags significantly in representativeness compared to systematic survey-based results. Compared with the conservation priority estimation results of the traditional richness method and the intelligent optimization algorithm, it is found that the latter can identify habitats with low richness but more important for specific species effectively, and has an obvious alleviation on the bias of conservation priority estimation results caused by the uneven sampling of citizen science data. The results show that it is advisable to use a variety of methods to screen data, simulate species distribution, and estimate conservation priorities in order to obtain relatively reliable results when using citizen science records for planning research and practice.