Smart Simon M, Stevens Carly J, Tomlinson Sam J, Maskell Lindsay C, Henrys Peter A
Centre for Ecology & Hydrology Lancaster, Lancaster, United Kingdom.
Lancaster Environment Centre, Lancaster University, Lancaster, United Kingdom.
PeerJ. 2021 Jan 12;9:e10632. doi: 10.7717/peerj.10632. eCollection 2021.
Estimation of the impacts of atmospheric nitrogen (N) deposition on ecosystems and biodiversity is a research imperative. Analyses of large-scale spatial gradients, where an observed response is correlated with measured or modelled deposition, have been an important source of evidence. A number of problems beset this approach. For example, if responses are spatially aggregated then treating each location as statistically independent can lead to biased confidence intervals and a greater probably of false positive results. Using methods that account for residual spatial autocorrelation, Pescott & Jitlal (2020) re-analysed two large-scale spatial gradient datasets from Britain where modelled N deposition at 5 × 5 km resolution had been previously correlated with species richness in small quadrats. They found that N deposition effects were weaker than previously demonstrated leading them to conclude that ". We use a simulation study to show that their conclusion is unreliable despite them recognising that an influential fraction of the residual spatially structured variation could itself be attributable to N deposition. This arises because the covariate used was modelled N deposition at 5 × 5 km resolution leaving open the possibility that measured or modelled N deposition at finer resolutions could explain more variance in the response. Explicitly treating this as spatially auto-correlated error ignores this possibility and leads directly to their unreliable conclusion. We further demonstrate the plausibility of this scenario by showing that significant variation in N deposition at the 1 km square resolution is indeed averaged at 5 × 5 km resolution. Further analyses are required to explore whether estimation of the size of the N deposition effect on plant species richness and other measures of biodiversity is indeed dependent on the accuracy and hence measurement error of the N deposition covariate. Until then the conclusions of Pescott & Jitlal (2020) should be considered premature.
评估大气氮（N）沉降对生态系统和生物多样性的影响是一项紧迫的研究任务。对大规模空间梯度进行分析，即观察到的响应与测量或模拟的沉降相关，一直是重要的证据来源。这种方法存在一些问题。例如，如果响应在空间上是聚集的，那么将每个位置视为统计独立会导致置信区间有偏差，并且出现假阳性结果的可能性更大。使用考虑残余空间自相关的方法，佩斯科特和吉特拉尔（2020年）重新分析了来自英国的两个大规模空间梯度数据集，其中先前已将5×5公里分辨率下模拟的氮沉降与小样方中的物种丰富度相关联。他们发现氮沉降的影响比先前证明的要弱，从而得出结论“……”。我们通过模拟研究表明，尽管他们认识到残余空间结构变异中有影响的一部分本身可能归因于氮沉降，但他们的结论是不可靠的。这是因为所使用的协变量是5×5公里分辨率下模拟的氮沉降，这使得更精细分辨率下测量或模拟的氮沉降有可能解释响应中更多的方差。将此明确视为空间自相关误差忽略了这种可能性，并直接导致了他们不可靠的结论。我们通过表明1平方公里分辨率下氮沉降的显著变化在5×5公里分辨率下确实被平均化，进一步证明了这种情况的合理性。需要进一步分析来探讨氮沉降对植物物种丰富度和生物多样性其他指标影响大小的估计是否确实取决于氮沉降协变量的准确性以及因此的测量误差。在此之前，佩斯科特和吉特拉尔（2020年）的结论应被视为不成熟的。
