The Department of Energy’s Atmospheric Radiation Measurement (ARM) user facility—jointly operated by nine national laboratories, including Lawrence Livermore—was established in 1989 with the aim of collecting continuous data to help researchers better understand atmospheric processes and improve how these processes are represented in Earth system models (ESMs). ARM now operates over 450 instruments across three fixed ground-based observatories as well as various mobile facility sites. Researchers routinely use ARM measurements to better understand clouds, aerosols, precipitation, solar radiation, and their interactions and to improve their representations in Earth system models.
Over the years, ARM has developed data products specifically designed for linking global models to ARM measurements, thereby supporting model evaluation and improvement. Incorporating ARM data into ESMs is challenging primarily because of the difference in temporal and spatial scales. ESMs normally have temporal resolutions of ~30 minutes and spatial resolutions of tens to hundreds of kilometers. ARM usually makes measurements at a much higher temporal frequency via a limited number of surface stations. The LLNL ARM data infrastructure team has played a leading role in developing novel data analysis algorithms and tools to bridge the scale gaps between Earth system models and ARM field observations.
Two of ARM’s long-standing products for modelers are ARM Best Estimate (ARMBE) and large-scale model forcing data sets based on the constrained variational analysis approach (VARANAL). ARMBE merges quality-controlled cloud, radiation, and atmospheric quantities that are often used in climate model evaluations into yearly data sets tailored for climate modelers with hourly temporal resolution, similar to the frequency of climate model output. ARMBE data can be used for both process studies and statistical evaluation of models at ARM sites. ARMBE-type data products effectively address a key difficulty for climate modelers—identifying the best estimate from multiple data streams that were obtained from various instruments and reviewed with different levels of data quality assurance or derived using different algorithms.
The VARANAL large-scale forcing data provide the critical initial and boundary conditions to drive single-column and cloud-resolving models and large-eddy simulations, and they are constrained by surface and top-of-atmosphere observations. VARANAL is created by combining ARM measurements and weather forecast model reanalysis data. It has become the most widely used forcing data for cloud modeling studies in the field since it was first released in late 1990s.
Over the last several years, the LLNL team has developed diagnostic tools and data sets to further facilitate use of ARM data in climate model evaluation and improve model‒observation comparison. The ARM Diagnostics for Climate Model Evaluation (ADCME) package combines ARM data sets with an open-source Python-based toolkit, which creates diagnostic plots and tables to compare model simulations with ARM data. ADCME uses data products such as ARMBE, VARANAL, and ARM Cloud Retrieval Ensemble Data (ACRED). The package also provides process-oriented metrics for evaluating convection onset in ESMs. In 2021, ADCME was integrated into the Energy Exascale Earth System Model.
What’s next
The LLNL ARM team is exploring the use of instrument simulators as another avenue for modelers interested in using ARM data to evaluate clouds simulated in ESMs. By generating “pseudo-instrument observations,” which emulate an actual instrument’s limitations, this approach supports apple-to-apple comparisons between modelled clouds and ARM detailed observations of clouds. As ESMs continue increasing model resolution, developing data and tool products for supporting evaluation and development of kilometer-scale models has become a priority for the LLNL ARM team for the coming years.
