ASPIRE: ARM-driven Simulation for Physical Insight Research Engine

Nearly all of the water available for U.S. energy production comes from precipitation governed by

atmospheric processes spanning microscopic cloud microphysics, turbulent mixing, and large-scale storm

dynamics. Accurate prediction of seasonal and year-to-year water availability is essential for hydropower

management, while microphysical processes such as freezing rain and wet snow directly threaten energy

infrastructure reliability through atmospheric icing. Advancing our understanding of these processes —

and translating that understanding into actionable prediction — requires deep, process-level insight into

individual precipitation events, not just statistical associations derived from surface observations alone.

The DOE's Atmospheric Radiation Measurement (ARM) user facility provides the richest observational

resource available for this purpose, continuously capturing the evolving atmospheric state across dozens of

simultaneous data streams. The scientific power of ARM lies in this multi-stream simultaneity: when

independent data streams converge on a consistent physical explanation, they enable genuine causal

understanding rather than statistical inference alone. Yet ARM datasets are significantly underutilized —

meaningful scientific analysis currently requires labor-intensive workflows that produce results only years

after data collection, and opportunities for near-real-time insight to guide data collection during active

events are routinely missed.

ASPIRE is designed to close this gap. Its vision is a future in which scientists have real-time situational

awareness during ARM field campaigns, can diagnose the microphysical and turbulent processes

controlling surface precipitation as events unfold, and can systematically identify where current process

models require improvement — accelerating both the science of cloud microphysics and the applied

prediction of U.S. water for energy.

AI is essential to this vision for three reasons. First, the volume and diversity of ARM data streams —

spanning ground-, air-, and space-based sensors across radar, lidar, radiometer, and in-situ modalities —

exceed what any human analyst can synthesize in real time; agentic AI enables automated multi-stream

fusion and consistency analysis that elevates output from pattern recognition to physically grounded causal

inference. Second, the iterative coupling of observational analysis with process model simulation —

running, interpreting, and refining ensembles of atmospheric simulations in response to what observations

reveal — is computationally and logistically intractable without autonomous agent coordination; ASPIRE's

agentic reflexion loop compresses this cycle from weeks to hours. Third, AI-enabled sensitivity analysis

and uncertainty quantification allow ASPIRE to distinguish parametric uncertainty from structural model

deficiency, directly identifying where new physics representations or enhanced observations are most

needed and providing a principled basis for testing next-generation ML-based microphysics emulators.

Together, these capabilities transform ARM data from an underutilized archive into a continuously

actionable scientific resource for U.S. energy applications.