Terra Daily — May 28, 2026

Research Worth Reading

• HybridOM: Hybrid Physics-Based and Data-Driven Global Ocean Modeling with Efficient Spatial Downscaling — Introduces HybridOM, a framework that fuses physics-based numerical solvers with data-driven deep learning for global ocean modeling, balancing computational efficiency with physical consistency. For ML engineers interested in climate, this is a concrete example of how hybrid modeling addresses a core tension in Earth-system simulation: maintaining long-term stability while leveraging learned representations.

• ACE2-NEMO: Coupling an ML Atmospheric Emulator to a Full-Depth Dynamical Ocean Model — Couples a machine-learned atmospheric emulator to a full-depth dynamical ocean model to study how fast atmospheric variability shapes slow climate variability and sensitivity. Directly relevant to engineers working on Earth-system model coupling and understanding emergent properties that neither component model captures alone.

• Economic Nonlinear Model Predictive Control for Microgrids with Generator Up and Downtime Constraints — Develops an economic NMPC scheme for microgrid optimal power flow that includes discrete generator runtime and startup cost constraints, tackling the mixed-integer nonlinear programming problems that arise in real distributed energy systems. Systems engineers will recognize the challenge of embedding combinatorial logic into continuous optimization loops.

• Skillful High-Resolution Weather Forecasting Independent of Physical Models — Presents an ML-based weather prediction system that operates without traditional numerical physical models, delivering forecasts faster and with higher skill. Demonstrates end-to-end learned forecasting at operational resolution — a useful reference point for engineers building or evaluating next-gen forecasting stacks.

• Day-Ahead Electricity Price Forecasting Using a Multivariate Group Lasso Method — Applies multivariate group Lasso to forecast day-ahead electricity prices using CAISO data, capturing complex temporal group effects in price signals. Engineers working on energy market prediction can use this structured regularization approach to handle high-dimensional feature sets without overfitting.

Technology & Innovation

• Q&A: Can China Turn Hydrogen Into Its Next Clean-Energy Industry? — China has designated hydrogen as a key “future industry” in its energy transition strategy. The Q&A covers electrolysis efficiency, infrastructure scaling, and cost competitiveness with fossil fuels — a useful snapshot of where the engineering bottlenecks actually are in the hydrogen value chain.

• As Geothermal Networks Grow, So Does the Call for a New Utility Model — Eversource’s neighborhood-scale geothermal network in Framingham, Massachusetts serves ~140 customers through shared ground-source heat pump infrastructure, representing a first-of-its-kind utility-owned district thermal system. Highlights emerging business and regulatory models for scaling geothermal beyond single-building installations.

• Africa’s Electric Motorcycle Manufacturers Get Ready for the Next Phase — A decade-long overview of Africa’s electric motorcycle sector, from early pilots to commercial rollouts, with an addressable market of over 30 million motorcycles. Useful context for engineers evaluating two-wheeler electrification opportunities in emerging markets.

• Driving Electric: Designing EV Carshare to Expand Access to Clean Transportation — RMI explores EV carshare models that bundle vehicle costs, insurance, and maintenance to expand access to clean transportation. Covers design considerations for cities looking to complement existing transit with short-term EV access programs — relevant for systems thinkers working on mobility-as-a-service platforms.

Open Source Projects

• Mechanical Engineer II — Cala Systems — Cala Systems is hiring a Mechanical Engineer II to design and prototype systems for energy storage or thermal management. A direct entry point for mechanical engineers looking to apply hardware design skills to sustainable energy applications.

Today’s Synthesis

Several pieces today point toward a converging opportunity at the intersection of ML-driven simulation and distributed energy infrastructure. HybridOM and ACE2-NEMO both demonstrate that hybrid physics-ML frameworks are maturing beyond toy problems — they’re being coupled to full-depth Earth-system models to capture emergent climate variability. For ML engineers, the transferable skill is building learned emulators that respect conservation laws and interface cleanly with legacy simulation code. Meanwhile, Economic Nonlinear Model Predictive Control for Microgrids shows that the downstream control layer for distributed energy systems is grappling with its own hard problems: mixed-integer nonlinear optimization under combinatorial constraints like generator uptime and startup costs. And Day-Ahead Electricity Price Forecasting illustrates how structured ML methods can extract actionable signals from high-dimensional market data. The concrete idea: teams building digital twins for microgrids or regional energy systems need both capabilities — fast surrogate models for weather forecasting, and robust MPC layers that can act on those predictions while respecting discrete operational constraints. Engineers who can bridge learned simulation with real-time optimal control are positioned to work on the software backbone of grid-interactive buildings, community microgrids, and virtual power plants.