There are currently no guidelines in the System of Environmental-Economic Accounting Ecosystem Accounting (SEEA EA) for quantifying and disclosing uncertainty. However, without quantifying uncertainty, it is unclear whether or not accounting tables contain biased (erroneous) area estimates which do not reflect real land cover changes. We use Oslo municipality in Norway as a case study to illustrate best practices in quantifying unbiased area estimates using design-based statistical methods. As input for ecosystem extent accounts, we compared a custom Sentinel-2 land cover map with a globally available one called Dynamic World for 2015, 2018 and 2021. The design-based area estimation involved (i) generating a stratified probability sample of locations using the satellite-based maps to define strata, (ii) assigning ecosystem type labels to the samples using photointerpretation according to a response design protocol, and (iii) applying a stratified area estimator to produce 95% confidence intervals around opening, closing and change stocks in the extent accounting table. We found that pixel counting practices, currently adopted by the SEEA EA community, led to biased extent accounts, particularly for ecosystem conversions, with biases averaging 195% of the true change value derived from design-based methods. We found that the uncertainty inherent in state-of-the-art satellite-based maps exceeded the ability to detect real change in extent for some ecosystem types including water and bare/artificial surfaces. In general, uncertainty in extent accounts is higher for ecosystem type conversion classes compared to stable classes, and higher for 3-yr compared to 6-yr accounting periods. Custom, locally calibrated satellite-based maps of ecosystem extent changes were more accurate (81% overall accuracy) than globally available Dynamic World maps (75%). We suggest that rigorous accuracy assessment in SEEA EA will ensure that ecosystem extent (and consequently condition and service) accounts are credible. A standard for auditing uncertainty in ecosystem accounts is needed.

中文翻译：

---

生态系统核算的“不确定性审计”：基于卫星的生态系统范围在没有基于设计的面积估计和准确性评估的情况下存在偏差

目前环境经济核算体系生态系统核算（SEEA EA）中没有用于量化和披露不确定性的指南。然而，在没有量化不确定性的情况下，尚不清楚会计表格是否包含有偏差（错误）的面积估计，这些估计不反映真实的土地覆盖变化。我们以挪威奥斯陆市为例，说明使用基于设计的统计方法量化无偏面积估计的最佳实践。作为生态系统范围账户的输入，我们将定制的 Sentinel-2 土地覆盖图与全球可用的 2015 年、2018 年和 2021 年动态世界地图进行了比较。基于设计的面积估计涉及 (i) 使用以下方法生成位置的分层概率样本基于卫星的地图来定义地层，(ii) 根据响应设计协议使用照片解释为样本分配生态系统类型标签，以及 (iii) 应用分层面积估计器来生成围绕开仓、停仓和变化库存的 95% 置信区间在范围会计表中。我们发现，SEEA EA 社区目前采用的像素计数做法导致了范围计算的偏差，特别是对于生态系统转换而言，偏差平均为基于设计的方法得出的真实变化值的 195%。我们发现，最先进的卫星地图固有的不确定性超出了检测某些生态系统类型（包括水和裸露/人造表面）范围实际变化的能力。一般来说，与稳定类别相比，生态系统类型转换类别的范围核算不确定性更高，与 6 年核算期相比，3 年核算期的不确定性更高。定制的、本地校准的基于卫星的生态系统范围变化地图比全球可用的动态世界地图 (75%) 更准确（总体准确度为 81%）。我们建议，SEEA 环境评估中严格的准确性评估将确保生态系统范围（以及相应的条件和服务）账户的可信度。需要制定审计生态系统账户不确定性的标准。