Evaluating the consistency of forest disturbance datasets in continental USA

Abstract. Forests play a crucial role in the Earth System, providing essential ecosystem services and sustaining biological diversity. However, forest ecosystems are increasingly impacted by disturbances, which are often integral to their dynamics but have been exacerbated by climate change. Despite the growing concern about these trends, the lack of consistent and temporally continuous data on forest disturbances at large spatial scales hinders our ability to accurately characterize changes in disturbance regimes and respond to these changes. In this study, we evaluate the consistency in spatial distribution and extent, disturbance timing and causal agent (when available), of five forest disturbance datasets available for the conterminous United States, to identify advantages as well potential shortcomings and inaccuracies of different mapping approaches. Consistency refers to the extent to which different forest disturbance datasets report similar timings and causal agents for overlapping disturbance events, reflecting their level of agreement. Specifically, we compare data from the Forest Inventory and Assessment (FIA), the Insect Disease Survey (IDS), both regularly conducted by the U.S. Department of Agriculture (USDA), the literature survey by the International Tree Mortality Network (ITMN) and two satellite-based datasets, the Global Forest Change (GFC) and North American Forest Dynamics Forest Loss Attribution (NAFD). All datasets report disturbance timing with a temporal granularity of one year, FIA and ITMN are point-based, and IDS, GFC and NAFD are spatially explicit. FIA, IDS, ITMN and NAFD report on disturbance agent, with different classification groupings. We find a moderate spatial agreement between the spatially explicit datasets and the point-based ones, with IDS, GFC and NAFD overlapping with 24 %, 58 % and 42 % of FIA disturbed patches, and on average 35 % of the ITMN reported mortality events. The datasets show similar trends in total disturbed extent over conterminous USA (CONUS) for the common period of 2001–2010, but with more pronounced differences at smaller scales, and when accounting for disturbance agents. The datasets agree well in disturbance timing: the mean difference is less than one year, while the variability in differences ranges from about 1 to 4 years. For FIA, we find better agreement with other datasets when the disturbance timing coincides with the inventory year, compared to disturbances reported as occurring in years between inventories. The satellite-based datasets tend to show an earlier detection of disturbance events, compared to the other datasets, possibly due to the inconsistent revisiting times of the inventory datasets (FIA and IDS). Our results show that although the datasets exhibit reasonably good agreement in disturbance timing, their spatial correspondence is considerably lower. Furthermore, the datasets show low agreement in terms of disturbance agent, which results from differences in grouping but also potentially on the methodology used to report causes. Our findings thus underscore the importance of careful data quality assessment and consideration of their inherent uncertainty when using single forest disturbance datasets for further applications. Specifically, for smaller scales and for disturbance agent attribution, we recommend careful comparison of more than a single dataset. Our study further highlights the need for improved data integration to advance the understanding of changes in forest disturbance regimes and their drivers.