An effective assessment protocol for continuous geospatial datasets of forest characteristics using USFS Forest Inventory and Analysis (FIA) data

Geospatial datasets of forest characteristics are modeled representations of real populations on the ground. The continuous spatial character of such datasets provides an incredible source of information at the landscape level for ecosystem research, policy analysis, and planning applications, all of which are critical for addressing current challenges related to climate change, urbanization pressures, and data requirements for monitoring carbon sequestration. However, the effectiveness of these applications is dependent upon the accuracy of the geospatial input datasets. A comprehensive set of robust measures is necessary to provide sufficient information to effectively assess the accuracy of these modeled geospatial datasets being produced. Yet challenges in the availability of reference data, in the appropriateness of assessment methods to dataset use, and in the completeness of assessment methods available have continued to hamper the timely and consistent application of map assessments. In this study we present a suite of assessments that can be used to characterize the accuracy of geospatial datasets of modeled continuous variables-an increasingly common format for modeling such attributes as proportion or probability of forestland as well as more traditionally continuous attributes such as leaf area index and forest biomass. It is a comparative accuracy assessment, in which each modeled dataset is compared to a set of reference data, recognizing both the potential for error in reference data, and probable differences in spatial support between the datasets. When used together, this proposed suite of assessments provides essential information on the type, magnitude, frequency and location of errors in each dataset. The assessments presented depend upon reference data with large sample sizes. The U.S. Forest Service (USFS) Forest Inventory and Analysis (FIA) database is introduced as an available reference dataset of sufficient sampling intensity to take full advantage of these assessments and facilitate their prompt application after modeled datasets are developed. We illustrate the application of this suite of assessments with two modeled datasets of forest biomass, in Minnesota and New York. The information provided by this suite of assessments substantially improves a user's ability to apply modeled geospatial datasets effectively and to assess the relative strengths and weaknesses of multiple datasets depicting the same forest characteristic.