Spatial upscaling using thematic maps: An analysis of uncertainties in permafrost soil carbon estimates

Abstract

[1] Studies of periglacial regions confirm their importance in the global carbon (C) cycle, but estimates of ecosystem C storage or green-house gas fluxes from these remote areas are generally poorly constrained and quantitative estimates of upscaling uncertainties are lacking. In this study, a regional database describing soil organic carbon (SOC) storage in periglacial terrain (European Russian Arctic) was used to evaluate spatial upscaling from point measurements using thematic maps. The selection of classes for upscaling and the need for replication in soil sampling were statistically evaluated. Upscaling using a land cover classification and a soil map estimated SOC storage to 48.5 and 47.0 kg C m À2 , respectively with 95% confidence intervals (CI) within AE8%. When corrected for spatial errors in the LCC upscaling proxy, SOC was estimated to 46.5 kg C m À2 with a 95% CI reflecting propagated variance from both natural variability and spatial errors of AE11%. Artificially decreasing the size of the database used for upscaling showed that relatively stable results could be achieved with lower replication in some upscaling classes. Decreased spatial resolution for upscaling from 30 m to 1 km had little impact on SOC estimates in this region, but classification accuracy was dramatically reduced and land cover classes show different, sometimes nonlinear, responses to scale. The methods and recommendations presented here can provide guidelines for any future study where point observations of a variable are upscaled using remotely sensed thematic maps or classifications and potential applications for circum-arctic studies are discussed. For future upscaling studies at large geographic scales, a priori determination of sample sizes and tests to insure unimodal and statistically independent samples are recommended. If these prerequisites are not fulfilled, classes may be merged or subdivided prior to upscaling.

A new data set for estimating organic carbon storage to 3 m depth in soils of the northern circumpolar permafrost region

“…The Mollisols are also significant with 3.5 % coverage, but are not represented in the current data set because of a lack of data. In a regional study of SOC storage in periglacial terrain, Hugelius (2012) found that most permafrost-free mineral soil classes could be amalgamated into one class with very little effect on the overall upscaling results. Analogous tests of the NCSCDv2 show that there is no statistically significant difference in the 0-300 cm SOC storage (kg C m −2 ) for most of the permafrost-free mineral soil orders (t test, p > 0.05).…”

“…To accurately estimate SOC stocks of Histosols and Histels, the depth of surface Ohorizons is a key variable (Hugelius, 2012). When sampling soils in the field, researchers often strive to reach the bottom of deep organic soils in order to accurately describe the depth distribution and deep SOC storage of these deposits.…”

“…All calculations were based on the complete gap-filled data set as presented in the pedon data set spreadsheet. As there was insufficient representation of soil taxa for most individual regions of the NC-SCD (for discussion of sample sizes in thematic upscaling, see Hugelius, 2012), we grouped the data into the North American (including Greenland) and Eurasian sector. Table 1 provides a summary of the number of available pedons in upscaling soil classes from different geographic regions.…”

“…While it is recognized that the SOC pool stored in permafrost regions is very large and vulnerable to remobilization through permafrost thaw (Elberling et al, 2013), pool estimates are poorly constrained and quantitative error estimates are lacking (Hugelius, 2012). Tarnocai et al (2009) assigned qualitative levels of confidence for different components of the circumpolar SOC estimate.…”

Can C-band synthetic aperture radar be used to estimate soil organic carbon storage in tundra?

“…These means were subsequently weighted by the proportional representation of each thematic class in the study area in order to arrive at a weighted "landscape-level" mean for the entire study area. Statistical uncertainties in this type of approach are described in Hugelius (2012).…”