Estimating National Trends in Soil Organic Carbon Stocks for US Agricultural Lands Using Large Datasets

Abstract:

Modeling soil processes across broad spatial scales to predict trends in soil C stocks is a challenging but manageable task using process-based models, large datasets and sufficient computing resources. Efforts to model soil processes and estimate trends in carbon stocks across broad scales began over two decades ago, and with the production of new and more comprehensive datasets as well as advances in computing technology, these analyses are becoming more tractable over time. We have used the DayCent ecosystem model to estimate soil C stock changes across US agricultural lands from 1990 to 2007, combining land survey data from about 400,000 locations with soil profile data from SSURGO, weather data from the NARR product, vegetation indices from MODIS EVI products, and management information from a variety of data sources. We assessed the accuracy and precision in the model estimates based on measurements of soil C stocks from long-term agricultural experiments. From 1990 to 2007, soil C stocks increased in US agricultural lands between 7 and 24 Tg C per year, but uncertainties generally exceeded ±50%.

Key challenges moving forward include continued production of large datasets and reducing uncertainty in the model-based estimates. Large datasets can be costly to produce and continually update, but are critical for evaluating broad scale patterns in soil C and other environmental variables, and support for the programs compiling these data is essential. The majority of the uncertainty can be traced back to limited measurements of soil C for evaluating the model and incomplete process understanding of soil C dynamics. Ongoing research along with increased monitoring of soil C through networks of measurement sites should help to fill this need.