PRIMARY PRODUCTION OF AN ARCTIC WATERSHED: AN UNCERTAINTY ANALYSIS

We describe a scaling protocol that combines two hierarchically linked models with field surveys, spatially distributed weather data, and remotely sensed images to generate daily predictions of gross primary production (GPP) for a 9200-km² arctic watershed. A detailed process-based model of vegetation–atmosphere interactions, which has been tested in a variety of arctic ecosystems against independent hourly gas exchange data, forms the base of the hierarchy. This detailed model was used to construct a second and simpler, “big-leaf” model, which was calibrated for arctic conditions and which required many fewer parameters and input data. For landscape forcing data, we derived spatiotemporal data on weather conditions (maximum and minimum temperature, and irradiance) from weather stations throughout the watershed. Spatiotemporal descriptions of the biotic constraints on production, chiefly leaf area index (LAI) and total foliar nitrogen (N_f), were derived from field surveys, a land cover database, and normalized difference vegetation index (NDVI) data acquired from satellites. The scaling hierarchy avoided propagation of error via a compensation process, though the procedures involved still introduced uncertainty into daily GPP predictions averaging 16% of the growing season daily mean. The construction of the spatiotemporal temperature and irradiance fields introduced uncertainty of 1–2% at spatial and temporal resolutions of 1 km² and one day, respectively. The greatest uncertainty was introduced by assignment of LAI across the region, because of the highly heterogeneous landscape and the high sensitivity of production to changes in LAI at the low levels found in the Arctic. Uncertainty in vegetation properties introduced an uncertainty of ±15% in basin GPP predictions, a value commensurate with basin net ecosystem production (NEP). In conclusion, improved characterization of vegetation via remote sensing is required before any bottom-up approach to carbon budgeting can reduce uncertainty to a reasonable level.