AbstractsBiology & Animal Science

Agricultural census data have been identified to possess the potential to provide constraints on carbon uptake by croplands at the regional scale. In this study, we build on previous efforts and further assess this potential quantitatively by comparing 1) fractional cropland coverage in southwestern Ontario, Canada derived from agricultural statistics against three different remotely sensed land cover maps; and 2) carbon uptakes determined from agricultural data with simulations generated by a satellite data-driven biospheric model. In addition, we assimilated the census data-derived carbon uptakes with modeled estimates in a Bayesian inverse approach to determine if and by how much constrain the crop data can provide, as exhibited by uncertainty reductions. Uncertainties in census data-derived gross primary production (GPP) estimates are carefully quantified using a Monte Carlo simulation. In general, results from the fractional cropland coverage comparison indicate significant value of the agricultural census data by revealing biases in the spatial distribution of croplands, as found in all three of the satellite land cover products. However, we find that the carbon uptake values derived from crop harvested records are still subject to significant uncertainties that have been underestimated or neglected altogether in past studies. The Monte Carlo simulation suggests that the largest source of uncertainty can be traced to errors in the growth efficiency, followed by harvest production records, and then the harvest index. As result, attention must be paid to such errors when using the agricultural census data for carbon accounting purposes or to provide constraints to simulations of crop carbon uptake.