Terra Daily — June 3, 2026

Research Worth Reading

• Samudra 2: Scaling Ocean Emulators across Resolutions — Introduces Samudra 2, a neural emulator for ocean general circulation models that achieves fine spatial resolution with multi-year autoregressive rollouts, enabling orders-of-magnitude speedups over traditional OGCMs for climate science applications.

• Scalable Uncertainty Quantification for Extreme Weather Forecasting via Empirical Neural Tangent Kernels — Proposes NTK-UQ, a Neural Tangent Kernel-based method for uncertainty quantification in deep learning weather models, addressing the critical gap of deterministic forecasts lacking confidence estimates for extreme weather events.

• Surrogate Modeling of Interconnector Flows: A Machine Learning Alternative to Full-Scale Power System Simulations with Application to Cross-Border Electricity Exchange — Presents a machine learning surrogate modeling approach to replace computationally expensive full-scale power system simulations for cross-border electricity exchange analysis, improving tractability in high-renewable-penetration scenarios.

• Enhancing Collective Self-Consumption through Water Storage Heater Flexibility — Explores demand-side flexibility via water storage heaters to align consumption with renewable generation in Renewable Energy Communities, improving collective self-consumption and supporting net-zero transitions.

• Impedance Modeling and Stability Analysis of Droop-Controlled Inverter Under Unbalanced Power Grid Operating Conditions — Addresses oscillatory stability risks in grid-tied inverters under unbalanced grid conditions, providing improved impedance modeling for renewable energy integration into power systems.

• Unstable Poles Arising in AC Power Grid Subsystem Representations — Analyzes small-signal stability in AC power grids using subsystem interconnections, focusing on PQ and other models to derive scalable stability certificates for renewable-rich grids.

Technology & Innovation

• China and Britain Hit Nuclear Milestones — Reports on recent nuclear energy milestones achieved by China and the UK, relevant to low-carbon baseload power generation and the evolving global nuclear landscape.

Open Source Projects

• In Massachusetts, parked EVs will start feeding the grid this summer — Electric school buses in Acton and Boxborough, Massachusetts will use their ~200 kWh batteries for vehicle-to-grid (V2G) services during summer months, charging overnight when grid power is cleanest and feeding energy back during peak demand. This is a real-world V2G deployment case study demonstrating distributed storage at the grid edge.

Policy & Regulation

• DOE bars homes from using rebates to ditch fossil-fueled heating — The Department of Energy has ruled that federal energy efficiency rebate programs will no longer cover switching from fossil-fueled heating to electric alternatives, directly impacting home electrification adoption. This policy shift has significant implications for heat pump deployment and residential decarbonization timelines.

Community Finds

• The Electricity Economy Is Having Its Moment — Examines the convergence of surging electricity demand—driven in part by data centers and AI—with the energy transition, highlighting grid stress, supply constraints, and the economic forces reshaping the U.S. power sector.

Today’s Synthesis

If you’re an ML engineer eyeing climate work, there’s a sweet spot forming at the intersection of extreme weather forecasting and grid stability . Today’s NTK-UQ paper tackles uncertainty quantification for weather models, while the impedance modeling work addresses inverter stability under real-world unbalanced grid conditions—two sides of the same coin. As renewable penetration grows, grid operators need probabilistic weather inputs paired with fast stability analysis, not just point forecasts. The V2G deployments in Massachusetts show distributed storage is already at the grid edge, but orchestrating millions of assets requires ML surrogates that can run in real time. Your systems skills transfer directly here: building differentiable pipelines that couple weather uncertainty, grid physics, and distributed control. Start by prototyping a surrogate that maps forecast confidence intervals to droop-control setpoints, then validate against the impedance models in the unbalanced-grid paper. The bottleneck isn’t data—it’s deployable, physics-informed ML that operators actually trust.