Background: Under ongoing climate and land-use change, biodiversity is continuously decreasing and monitoring biodiversity is becoming increasingly important. National Forest Inventory(NFI) programmes provide valuable...Background: Under ongoing climate and land-use change, biodiversity is continuously decreasing and monitoring biodiversity is becoming increasingly important. National Forest Inventory(NFI) programmes provide valuable timeseries data on biodiversity and thus contribute to assessments of the state and trends in biodiversity, as well as ecosystem functioning. Data quality in this context is of paramount relevance, particularly for ensuring a meaningful interpretation of changes. The Swiss NFI revisits about 8%–10% of its sample plots regularly in repeat surveys to supervise the quality of fieldwork.Methods: We analysed the relevance of observer bias with equivalence tests, examined data quality objectives defined by the Swiss NFI instructors, and calculated the pseudo-turnover(PT) of species composition, that is, the percentage of species not observed by both teams. Three attributes of woody species richness from the latest Swiss NFI cycles(3 and 4) were analysed: occurrence of small tree and shrub species(1) on the sample plot and(2) at the forest edge, and(3) main shrub and trees species in the upper storey.Results: We found equivalent results between regular and repeat surveys for all attributes. Data quality, however,was significantly below expectations in all cases, that is, as much as 20%–30% below the expected data quality limit of 70%–80%(proportion of observations that should not deviate from a predefined threshold). PT values were about 10%–20%, and the PT of two out of three attributes decreased significantly in NFI4. This type of uncertainty –typically caused by a mixture of overlooking and misidentifying species – should be considered carefully when interpreting change figures on species richness estimates from NFI data.Conclusions: Our results provide important information on the data quality achieved in Swiss NFIs in terms of the reproducibility of the collected data. The three applied approaches proved to be effective for evaluating the quality of plot-level species richness and composition data in forest inventories and other biodiversity monitoring programmes. As such, they could also be recommended for assessing the quality of biodiversity indices derived from monitoring data.展开更多
Observer error,a type of nonsampling error,is pervasive in vegetation sampling and often of a consequential magnitude.Observer error rates should be reported along with published studies,although there currently exist...Observer error,a type of nonsampling error,is pervasive in vegetation sampling and often of a consequential magnitude.Observer error rates should be reported along with published studies,although there currently exists no standardized,easily comparable format.Here we describe five key metrics of observer error(i.e.imprecision between observers),how they are calculated,and how they can be reported and interpreted.Three metrics apply to species composition:pseudo-turnover,observer bias in species richness and underestimation of true species richness.Two metrics—cover agreement and observer bias in cover estimation—apply to categorical cover estimation.All metrics are simple to determine,could be calculated from virtually any multispecies sampling effort using two or more observers,and are easily compared with other studies.The metrics are all reported as percentages,allowing for relative comparisons among studies with greatly differing species diversities.We also describe how to decompose the amount of error in species composition and cover estimation into random and biased components.Such decomposition is useful in determining whether additional training may be necessary for some observers.Two of the five metrics—pseudo-turnover and cover agreement—have been quantified in previous studies,and we compile a list of published rates of pseudo-turnover within general habitat types,and published cover agreement categories,for comparison with future studies.Finally,we provide an example by calculating the observer error metrics for a real data set collected by three different observers.展开更多
基金funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 787638),granted to Catherine Graham。
文摘Background: Under ongoing climate and land-use change, biodiversity is continuously decreasing and monitoring biodiversity is becoming increasingly important. National Forest Inventory(NFI) programmes provide valuable timeseries data on biodiversity and thus contribute to assessments of the state and trends in biodiversity, as well as ecosystem functioning. Data quality in this context is of paramount relevance, particularly for ensuring a meaningful interpretation of changes. The Swiss NFI revisits about 8%–10% of its sample plots regularly in repeat surveys to supervise the quality of fieldwork.Methods: We analysed the relevance of observer bias with equivalence tests, examined data quality objectives defined by the Swiss NFI instructors, and calculated the pseudo-turnover(PT) of species composition, that is, the percentage of species not observed by both teams. Three attributes of woody species richness from the latest Swiss NFI cycles(3 and 4) were analysed: occurrence of small tree and shrub species(1) on the sample plot and(2) at the forest edge, and(3) main shrub and trees species in the upper storey.Results: We found equivalent results between regular and repeat surveys for all attributes. Data quality, however,was significantly below expectations in all cases, that is, as much as 20%–30% below the expected data quality limit of 70%–80%(proportion of observations that should not deviate from a predefined threshold). PT values were about 10%–20%, and the PT of two out of three attributes decreased significantly in NFI4. This type of uncertainty –typically caused by a mixture of overlooking and misidentifying species – should be considered carefully when interpreting change figures on species richness estimates from NFI data.Conclusions: Our results provide important information on the data quality achieved in Swiss NFIs in terms of the reproducibility of the collected data. The three applied approaches proved to be effective for evaluating the quality of plot-level species richness and composition data in forest inventories and other biodiversity monitoring programmes. As such, they could also be recommended for assessing the quality of biodiversity indices derived from monitoring data.
基金supported by the Inventory and Monitoring Program of the National Park Service。
文摘Observer error,a type of nonsampling error,is pervasive in vegetation sampling and often of a consequential magnitude.Observer error rates should be reported along with published studies,although there currently exists no standardized,easily comparable format.Here we describe five key metrics of observer error(i.e.imprecision between observers),how they are calculated,and how they can be reported and interpreted.Three metrics apply to species composition:pseudo-turnover,observer bias in species richness and underestimation of true species richness.Two metrics—cover agreement and observer bias in cover estimation—apply to categorical cover estimation.All metrics are simple to determine,could be calculated from virtually any multispecies sampling effort using two or more observers,and are easily compared with other studies.The metrics are all reported as percentages,allowing for relative comparisons among studies with greatly differing species diversities.We also describe how to decompose the amount of error in species composition and cover estimation into random and biased components.Such decomposition is useful in determining whether additional training may be necessary for some observers.Two of the five metrics—pseudo-turnover and cover agreement—have been quantified in previous studies,and we compile a list of published rates of pseudo-turnover within general habitat types,and published cover agreement categories,for comparison with future studies.Finally,we provide an example by calculating the observer error metrics for a real data set collected by three different observers.
