A solar-powered desalination plant in NEOM cannot prove its carbon benefit. A UAE aviation fuel producer cannot certify its low-carbon product to European buyers. An Egyptian cement exporter now must reckon with the EU's Carbon Border Adjustment Mechanism, which entered its financially binding phase this January, and its exposure has less to do with how clean its process is than with whether it can supply verified, region-specific data to prove it. An exporter pushed onto a conservative default value instead of its real, better number is effectively paying for being unmeasured, not for being dirty.
Across the Arab world, governments are investing billions in renewable energy, hydrogen, desalination, and sustainable infrastructure. Yet nearly every environmental assessment evaluating that investment relies on data developed somewhere else, for different grids, different processes, a different era of industrial change. Call it what it is: not a missing-data problem, but a mismatched-speed problem. The global standard for producing environmental data was built for economies that change slowly, and nothing about the Arab world right now is slow.
I did not arrive at this conclusion from the outside. I ran into it directly, in the middle of my own research, while trying to solve what I thought was a completely different problem.
Where the Wall Was
As part of my PhD dissertation at KAUST, I built the e-Hydrogen Cost Optimizer, a Python framework that minimizes the Levelized Cost of Hydrogen for Saudi locations including Duba, Yanbu, and Jubail, combining hourly wind and solar profiles, electrolyzer costs, and storage configurations into cost estimates ranging from $2.67 to $5.22 per kilogram. The model worked. I could optimize the hydrogen system down to the hour and the riyal. Then I tried to answer one simple-sounding question: what is the carbon intensity of the electricity grid in Yanbu? Three databases gave me three different answers. None of them had measured Yanbu; each had substituted a regional or global average and called it close enough. That moment changed the direction of my research. I realized optimization was no longer the bottleneck. Data was. A cost curve is only as trustworthy as the environmental numbers feeding it, and in the Arab world those numbers are almost always borrowed, from Europe, the US, averaged across continents, or already out of date by the time they're published. The major environmental databases the field relies on, chiefly ecoinvent and its peers, were built to maximize global coverage and scientific consistency, not to track newly industrializing regions as they change. Analysts across the region routinely fall back on "Rest of World" or global proxy numbers never designed to represent local technologies or supply chains, and the resulting gap can be large enough to swamp the actual environmental difference an analysis was supposed to detect. This isn't for lack of trying. Ecoinvent and other providers have added regional datasets, and Gulf sustainability initiatives have started collecting facility-level emissions data of their own. But both approaches inherit the same process: a multi-year manual study cycle, built expert-by-expert. That cycle can eventually produce a good dataset for a grid or a plant that holds still long enough to study. It cannot keep pace with a region where new renewable capacity, new industrial capacity, and new grid mixes are coming online faster than the study cycle can complete even once.
From One Tool to a Research Agenda
My optimizer already sits inside a life-cycle assessment (LCA) workflow built on Brightway, linking cost optimization to environmental accounting instead of treating them as separate exercises. That integration exposed the Yanbu problem in the first place, and convinced me the fix isn't a better background database. It's a different way of producing environmental data altogether. What should exist, but doesn't, is a real-time, open, federated framework for generating environmental data directly from industrial sources, designed for the speed of the region's transformation rather than retrofitted from a slower one. A better static database wouldn't solve this; it would just be a slower-moving target with a regional label on it.
I think of my optimizer as the first working piece of that system: it already shows that cost optimization and environmental accounting can share one pipeline instead of living in separate spreadsheets. The research agenda ahead of me is building the three things that pipeline still needs, in order. First, automated inventory construction from operational data. Most facilities already generate enormous volumes of sensor data, but a life-cycle inventory isn't a data feed, it's an allocation. A shared utility stream serving multiple product lines must be split between them by some defensible rule, and today that split is a modeler's judgment call, made once, by hand. The research question is whether it can become a repeatable, auditable function instead of a one-time decision buried in a spreadsheet. Second, once inventories update continuously, their uncertainty must update with them. Classical LCA assumes the background system holds still between revisions. The Arab world's grids don't hold still, and because inventories are hierarchical, an error in one upstream number ripples through every downstream product built on it. The research question is whether a dataset can carry an honest, continuously revised margin of error, tight enough to be useful, honest enough not to quietly overstate its own confidence. Third, connecting facilities without forcing them to centralize raw data. Facilities should hold their own validated data locally, answerable to outside verifiers, rather than handing raw numbers to a central repository. The hard part is that verification and confidentiality pull in opposite directions: a verifier needs enough to trust a number without seeing the production volumes and process details that make the operation commercially sensitive. The research question is finding the minimum evidence sufficient for real trust, not importing a privacy protocol built for an unrelated domain.
None of this is guaranteed to work cleanly. EU verifiers currently trust a known methodology; a novel, continuously-updating pipeline will need to earn that trust, and some facilities will be reluctant to instrument their operations without a clear incentive. That adoption gap is itself worth researching, the first pilot should treat institutional buy-in as seriously as the technical pipeline.
How I Would Start
I wouldn't try to model an entire country. I'd start with a single desalination facility, most likely an MSF or RO plant, because desalination is one of the region's largest energy consumers and one of the least accurately represented processes in existing life-cycle inventories, ecoinvent proxies for it are generic and rarely reflect Gulf-specific feed salinity, intake temperature, or the local grid's marginal generation mix.
Months one and two: instrument the boundary. Pull historical and live tags from the plant's SCADA and energy-management systems over OPC-UA or Modbus, specific energy consumption per cubic meter of permeate, brine rejection rate and salinity, thermal input for any MSF trains, membrane flux decline for RO, and map each tag to a foreground unit process in a Brightway-based LCA model, the same environment my optimizer's cost engine already exchanges data with. This is where allocation gets decided explicitly rather than inherited from a template: energy shared between product water and any co-generated steam or power is split by exergy content rather than mass or volume, since exergy allocation is the rule that tracks thermodynamic quality correctly across different outputs. Every emission factor carries its allocation method as metadata, not a hidden assumption, so a verifier can trace a number back to the rule that produced it.
Months three and four: run the automated pipeline in parallel with the facility's existing manual LCA or CBAM declaration, using the manual study as ground truth for validation. Quantify the delta process-by-process, whether it traces to allocation choice, missing foreground flows, or a stale background proxy. Wrap each output in a Bayesian credible interval, updating the posterior on each emission factor as new SCADA batches arrive rather than re-running a static Monte Carlo, so the interval itself gets tighter, not just the mean.
Months five and six: publish the pipeline, the allocation methodology, and validation results as open source, structured so a second facility in a different sector, a power generator or cement kiln, where allocation looks different (fuel-switching ratios and clinker substitution rates instead of exergy splits), can adopt the same scaffolding without starting from zero. Any query a verifier runs against the published dataset returns the value plus its current credible interval and allocation provenance, never a bare number. This starts with tools that already exist and exchange data with each other, not from nothing. What's missing is the layer that replaces a static background dataset with a foreground inventory the facility generates about itself, continuously, with allocation and uncertainty made explicit, and that layer is what this research agenda is built to add.
Why Here, Why Now
Most discussions of this problem default to "build an Arab version of ecoinvent." That's the wrong target. Another static database, however, regionally specific, would be out of date again within a few years, for the same reason the current ones are out of date now. Consumer electronics have a term for a product designed to stop working, or stop being good enough, on a schedule: planned obsolescence. It's usually an indictment, a phone battery engineered to degrade, a printer counting down to a forced replacement. But strip out the profit motive and the underlying engineering idea is sound: if a product's usefulness is going to expire, the responsible move is to design that expiration on purpose, rather than let it fail silently and surprise the user. Environmental datasets today get the worst version of this without any planning. They go obsolete just as fast, sometimes faster, than a rapidly industrializing grid can outrun them, but nobody designed for that. There's no expiration date printed on an ecoinvent proxy, no flag telling an analyst the number she's using went stale eighteen months ago. The obsolescence is real; it's just unplanned, and unplanned obsolescence in a carbon figure doesn't cost a customer a repair bill, it costs an exporter a shipment held at the border while someone reruns the math. The system I'm proposing takes that engineering discipline and applies it honestly, in the open, to environmental data itself. Every dataset it produces would carry an explicit validity window and scheduled revalidation, the way a well-run piece of infrastructure carries a maintenance interval instead of running until it breaks. A number wouldn't quietly go stale; it would expire on a known clock, trigger its own successor, and hand the analyst a fresh estimate with fresh uncertainty attached. That's the real innovation here, not another database, but one built for its own obsolescence by design.
This reframes the region's thin historical environmental record as an opening rather than a deficit. Europe and the US have decades of legacy environmental infrastructure built around the slow-revision model, and a lot invested in it working as it is. The Arab world has comparatively little of that to defend, and more reason to want something faster: it is the region building the plants this data needs to describe, in real time, right now. That's a genuine first-mover advantage, not a consolation prize. Saudi Arabia, with its industrial buildout, digital infrastructure, and explicit sustainability mandate, is close to an ideal place to prove it, with lessons that could extend across the region.
I began my journey believing the hardest problem in sustainable energy was optimizing future systems. I now think the harder problem is making sure those optimizations are built on environmental data that deserves to be trusted. I'm currently scoping the pilot described above, starting conversations with a Saudi industrial partner about instrumenting one facility, and mapping which of its existing sensor infrastructure could feed a first-generation pipeline without requiring new hardware. My optimizer solved one part of this puzzle. This research agenda is the next one, and it's the one I intend to build.