|SAHA, DEBASISH - PENNSYLVANIA STATE UNIVERSITY|
|KEMANIAN, ARMEN - PENNSYLVANIA STATE UNIVERSITY|
|RAU, BENJAMIN - U.S. FOREST SERVICE (FS)|
|MONTES, FELIPE - PENNSYLVANIA STATE UNIVERSITY|
Submitted to: Atmospheric Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/30/2017
Publication Date: 1/31/2017
Citation: Saha, D., Kemanian, A., Rau, B.M., Adler, P.R., Montes, F. 2017. Designing efficient nitrous oxide sampling strategies in agroecosystems using simulation models. Atmospheric Environment. 155:189-198.
Interpretive Summary: The large uncertainties associated with measurement of nitrous oxide emissions can be improved by increasing the frequency of measurements, but this also increases measurement costs. Using a model, we evaluated two strategies to determine the time and frequency of nitrous oxide measurement, 1) measurement over a range of fixed time intervals and 2) using a list of rules to describe the time and frequency of measurement. We found that although the accuracy of the nitrous oxide estimate increased as sample frequency increased, using rules to determine the time of sampling reduced the frequency of sampling to achieve the same measurement accuracy. These results demonstrate that it can be more efficient to measure nitrous oxide emissions based on predefined rules or events rather than a fixed frequency throughout the year.
Technical Abstract: Cumulative nitrous oxide (N2O) emissions calculated from discrete chamber-based flux measurements have unknown uncertainty. This study used an agroecosystems simulation model to design sampling strategies that yield accurate cumulative N2O flux estimates with a known uncertainty level. Daily soil N2O flux was simulated for Ames, IA (corn-soybean rotation, anhydrous ammonia fertilizer injected in fall), College Station, TX (corn-vetch rotation, corn fertilized twice), Fort Collins, CO (irrigated corn, fertilized), and Pullman, WA (winter wheat, fertilized in fall and spring). These fluxes were used as daily measurements surrogates. The flux data was sampled using either a fixed interval (1 to 32 days) or a rule-based (decision tree-based) sampling method. Two types of rule-based regression trees were constructed and used to predict sampling events. One used high-input (HI, including soil inorganic nitrogen) and the other low-input (LI, without soil inorganic nitrogen) N2O flux predictor variables as identified by Random Forest. The uncertainty of the annual N2O flux estimation increases along with the fixed interval length. A 4- and 8-day interval sampling is required at College Station and Ames, respectively, to yield ±20% accuracy in the flux estimate, while a 12-day interval renders the same accuracy at Fort Collins and Pullman. Both HI and LI rule-based method provided the same accuracy as that of fixed interval with 60% reduction in sampling numbers. The improvement in efficiency is higher in sites with greater flux variability.