Statistical Downscaling of NEON Precipitation Water Isotope Datasets for Daily Time Series Generation

Abstract:

The National Ecological Observatory Network (NEON) provides open access data products including sub-daily precipitation amounts and biweekly precipitation stable water isotope (ẟ²H, ẟ¹⁸O) concentrations at 37 field sites across the United States. Unfortunately, the relatively infrequent biweekly sampling intervals of the precipitation isotope concentrations complicate their application in driving many high-resolution hydrometeorological models, which often require daily to sub-daily forcing datasets. Here, we present an application of a statistical method to temporally downscale precipitation isotope datasets across the NEON network. Our resulting workflow was automated and can be reproduced with python and R scripts in a published Jupyter Notebook, allowing the user to temporally downscale available stable water isotope data at any or all of the NEON sites for specific time intervals. First, the notebook used the “neonUtilities” R package to download and organize the precipitation amount and stable water isotope datasets. Next, python scripts calculated biweekly sums of the precipitation amount time series to temporally correspond with the precipitation stable water isotope sampling frequency. Then, the seasonal time series component of each isotope time series was quantified and removed. Site-specific daily time series statistics were approximated from the biweekly deseasonalized series and used to generate isotope values conditioned on daily precipitation amounts. Lastly, the seasonal time series component was added and the final synthetic ẟ²H and ẟ¹⁸O time series captured precipitation amount effects, seasonality, and site-specific statistical relationships. Ensemble sets can be created by specifying the number of synthetic time series generated at each site and the Jupyter Notebook is easily adaptable for user-specific research questions. The new stable water isotope time series can be incorporated as an environmental tracer in hydrologic models (e.g. evapotranspiration partitioning, subsurface processes), as ensembles for model selection, and with numerous model and parameter sensitivity and uncertainty analyses.