{"id":262,"date":"2026-05-08T15:05:39","date_gmt":"2026-05-08T15:05:39","guid":{"rendered":"https:\/\/jua.sites.aigrowthagent.co\/best-ai-for-weather-forecasting\/"},"modified":"2026-07-12T05:00:28","modified_gmt":"2026-07-12T05:00:28","slug":"best-ai-for-weather-forecasting","status":"publish","type":"post","link":"https:\/\/jua.ai\/articles\/best-ai-for-weather-forecasting\/","title":{"rendered":"AI for Weather Forecasting: Physics-Constrained vs NWP"},"content":{"rendered":"<p><em>Written by: Olivier Lam, Physical AI Team, Jua.ai AG | Last updated: July 11, 2026<\/em><\/p>\n<h2 id=\"key-takeaways\">Key Takeaways for Energy Professionals<\/h2>\n<ul>\n<li>Traditional NWP models are limited to 2\u20134 daily runs because of high computational costs, which creates a structural forecasting lag for energy professionals.<\/li>\n<li>First-generation pattern-based AI models improve speed but lack physics constraints, so they underperform on extreme weather events that drive energy trading P&amp;L.<\/li>\n<li>Physics-constrained models like Jua\u2019s EPT family learn conservation laws from data and deliver higher accuracy on wind, temperature, and solar variables at a fraction of NWP cost.<\/li>\n<li>Operational advantages include up to 24 daily updates, productised ensembles, native any-\u0394t forecasting, and API\/SDK integration that cuts engineering overhead.<\/li>\n<li><a href=\"https:\/\/meetings-eu1.hubspot.com\/guett\/energy-trading?uuid=d780665f-ff71-439c-addf-c80e49af0627\" target=\"_blank\">Book a demo with Jua<\/a> to run live benchmarks on your region and variables in under five minutes.<\/li>\n<\/ul>\n<h2>How AI Weather Models Fit Into Today\u2019s Forecasting Landscape<\/h2>\n<p>Numerical weather prediction has been the industry standard for forty years. ECMWF HRES and NOAA GFS decompose the atmosphere into three-dimensional grid cells and solve differential equations inside each one. The method works. A single NWP simulation consumes approximately 8,400 kWh of compute and costs \u20ac1,000\u2013\u20ac20,000 to run on high-performance computing infrastructure. That cost ceiling caps update frequency at two to four runs per day, which has been a hard constraint for the energy industry for decades.<\/p>\n<p>The first generation of AI weather models changed the speed and cost profile. <a href=\"https:\/\/deepmind.google\/discover\/blog\/graphcast-ai-model-for-faster-and-more-accurate-global-weather-forecasting\/\" target=\"_blank\" rel=\"noindex nofollow\">Google DeepMind&#8217;s GraphCast<\/a>, published in Science in 2023, beat ECMWF HRES on 90% of 1,380 verification targets across medium-range forecasts. <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/aurora-a-foundation-model-of-the-atmosphere\/\" target=\"_blank\" rel=\"noindex nofollow\">Microsoft Aurora<\/a>, a 1.3-billion-parameter foundation model published in Nature in 2025, <a href=\"https:\/\/news.microsoft.com\/source\/features\/ai\/microsofts-aurora-ai-foundation-model-goes-beyond-weather-forecasting\/\" target=\"_blank\" rel=\"noindex nofollow\">beat existing numerical and AI models across 91% of medium-range weather forecasting targets when fine-tuned<\/a>. ECMWF&#8217;s own AIFS followed. These models showed that data-driven approaches could match or exceed NWP on standard verification metrics at a fraction of the inference cost.<\/p>\n<p>The limitation of this first generation is structural. GraphCast, Aurora, and AIFS are pattern-based models that learn statistical associations from historical data without explicitly encoding the conservation laws that constrain what the atmosphere can physically do. A 2026 study published in Science Advances found that AI models including GraphCast, Pangu-Weather, and FuXi systematically underestimate the intensity and frequency of record-breaking extreme events, with the underestimation error growing as events exceed the range of historical training data. For energy professionals, this translates directly to imbalance costs and missed trading edges when a wind ramp is not predicted or a cold snap is underestimated.<\/p>\n<p>The second generation, physics-constrained foundation models, addresses this gap by learning conservation laws from observational data rather than imposing them as hard symbolic constraints. Jua&#8217;s EPT family sits in this category and targets energy-relevant variables directly.<\/p>\n<h2>How Physics-Constrained Models Like EPT-2 Work<\/h2>\n<p>Equation-based NWP encodes physics explicitly. The governing partial differential equations of fluid dynamics are discretised and solved numerically at each grid point. The physics is correct by construction, but computational cost scales with resolution and domain size, so update frequency remains bounded by HPC infrastructure.<\/p>\n<p>Pattern-based AI models replace the dynamical core with a learned operator, typically a graph neural network or transformer, trained to map atmospheric state at time <em>t<\/em> to state at time <em>t + \u0394t<\/em>. Most state-of-the-art models implement this as a fixed-interval roll-forward. Aurora and most peers are trained on a fixed 6-hour grid and roll forward in 6-hour steps, which compounds error at longer lead times. Pure AI weather models often show larger imbalances and error accumulation over extended forecasts because they are not explicitly constrained by physical equations.<\/p>\n<p>Physics-constrained foundation models occupy a third category and are designed to avoid this error-compounding problem. The EPT family is a general spatiotemporal transformer foundation model that learns the governing physics of complex systems, including the conservation laws that constrain mass, momentum, and energy, directly from observational data in a latent representation that is integrated forward in time. EPT-2 produces forecasts at native any-\u0394t, which means it is trained to predict at arbitrary time steps rather than rolling forward in fixed increments. EPT-2 does not roll. This architectural choice removes the fixed-interval roll-forward and eliminates the error-compounding that affects fixed-interval peers at longer lead times, as documented in <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>.<\/p>\n<h2>EPT System Components That Matter for Energy<\/h2>\n<p>The system components relevant to energy professionals are:<\/p>\n<ul>\n<li><strong>Deterministic flagship:<\/strong> EPT-2, global, 20-day horizon, four runs per day, benchmarked against more than 10,000 ground stations via StationBench.<\/li>\n<li><strong>Ensemble variant:<\/strong> EPT-2e, 60-day horizon, updated 4\u00d7 daily, beats the 50-member ECMWF ENS mean on both RMSE (root mean square error) and CRPS (continuous ranked probability score) at virtually every lead time per <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>.<\/li>\n<li><strong>Rapid refresh:<\/strong> EPT-2 RR, up to 24 runs per day, which reduces the staleness problem between standard NWP cycles.<\/li>\n<li><strong>High-resolution:<\/strong> EPT-2 HRRR, native 5 km resolution over Europe.<\/li>\n<li><strong>Agent layer:<\/strong> Athena, an AI agent that turns natural-language objectives into briefings, benchmarks, backtests, and custom widgets, with typical queries resolving in approximately 90 seconds.<\/li>\n<\/ul>\n<p>Data pipelines ingest more than 5 petabytes of weather and climate data from over 120 sources, including geostationary and polar-orbiting satellites, surface station networks such as SYNOP, METAR, and proprietary feeds, national radar networks, ocean buoys, ERA5 reanalysis, and operational ECMWF HRES initial-condition fields. The StationBench evaluation methodology, which is open-source, uses more than 10,000 real ground stations and applies no post-processing or station fine-tuning, and it provides the external validation anchor for all accuracy claims.<\/p>\n<p><a href=\"https:\/\/meetings-eu1.hubspot.com\/guett\/energy-trading?uuid=d780665f-ff71-439c-addf-c80e49af0627\" target=\"_blank\">Compare EPT-2&#8217;s system components against your current setup in a live demo.<\/a><\/p>\n<h2>Strategic Trade-offs for Energy Trading and Utilities<\/h2>\n<p><strong>Accuracy versus speed.<\/strong> NWP incumbents such as ECMWF HRES at 9 km resolution remain the universal benchmark for deterministic accuracy, particularly on extreme events. The 2026 Science Advances study confirms that for severe, record-breaking events, physics-based HRES consistently outperforms pattern-based AI models. EPT-2 outperforms ECMWF HRES on every lead time and on 10 m wind, 100 m wind, 2 m temperature, and surface solar radiation across the full 0\u2013240 hour range, per <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>, while running on a single GPU in minutes at approximately 0.25 kWh and $0.20\u2013$15 per simulation, which is roughly four orders of magnitude cheaper than the equivalent NWP run. Pattern-based AI peers achieve competitive RMSE on standard variables but underperform on extremes and lack physics constraints.<\/p>\n<p><strong>Generality versus specialization.<\/strong> Aurora extends predictions to air quality, ocean waves, and tropical cyclones, which shows the breadth possible with large foundation models. EPT is a general physics foundation model, and the same architecture that learns atmospheric dynamics already predicts plasma behaviour inside a tokamak. The data and the fine-tune change from one physical system to the next. For energy professionals, the relevant specialization is the Jua for Energy product surface, which covers 25 variables including wind at 11 height levels from 10 m to 200 m, surface solar radiation, and power forecasts for solar, wind onshore, wind offshore, load, and residual load across five countries.<\/p>\n<p><strong>Cost versus performance.<\/strong> EPT-2 was trained on 8 \u00d7 H100 GPUs over 10 days. Microsoft Aurora required 32 \u00d7 A100 GPUs over 18 days. At inference, EPT-2 runs approximately 25% faster than Aurora. For quant teams evaluating build-versus-buy, the relevant comparison is total integration cost rather than inference cost alone. Raw AI research subscriptions require teams to build ingestion pipelines, ensemble logic, benchmarking harnesses, and hindcast access, which consumes engineering capacity that could be spent on alpha research. The Python SDK installs with <code>pip install jua<\/code>, and the REST API exposes more than 25 models through a single schema with Apache Arrow support for large payloads.<\/p>\n<p><a href=\"https:\/\/meetings-eu1.hubspot.com\/guett\/energy-trading?uuid=d780665f-ff71-439c-addf-c80e49af0627\" target=\"_blank\">See how EPT-2&#8217;s cost-performance profile compares to your current provider.<\/a><\/p>\n<h2>Implementing Jua for Energy in Production<\/h2>\n<p>Benchmarking against ground stations is the correct evaluation methodology for energy applications. <a href=\"https:\/\/weatherbench2.readthedocs.io\/\" target=\"_blank\" rel=\"noindex nofollow\">WeatherBench 2<\/a> provides the community&#8217;s standard independent evaluation framework using ERA5 reanalysis as ground truth. StationBench extends this to real observational data using the ground-station network described earlier, which is the relevant test for energy professionals whose P&amp;L depends on local wind and temperature accuracy rather than gridded reanalysis skill. The table below compares EPT-2, ECMWF HRES, and Aurora on the four variables most relevant to energy trading, per <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a> and the StationBench methodology.<\/p>\n<table>\n<thead>\n<tr>\n<th>Variable<\/th>\n<th>EPT-2 (0\u2013240 h)<\/th>\n<th>ECMWF HRES (0\u2013240 h)<\/th>\n<th>Aurora (0\u2013240 h)<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>10 m wind speed (RMSE)<\/td>\n<td><a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">Outperforms HRES at every lead time<\/a><\/td>\n<td>Benchmark reference<\/td>\n<td><a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">Loses to EPT-2 across full range<\/a><\/td>\n<\/tr>\n<tr>\n<td>100 m wind speed (RMSE)<\/td>\n<td><a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">Outperforms HRES at every lead time<\/a><\/td>\n<td>Benchmark reference<\/td>\n<td><a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">Loses to EPT-2 across full range<\/a><\/td>\n<\/tr>\n<tr>\n<td>2 m temperature (RMSE)<\/td>\n<td><a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">Outperforms HRES at every lead time<\/a><\/td>\n<td>Benchmark reference<\/td>\n<td><a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">Loses to EPT-2 up to ~130 h lead time<\/a><\/td>\n<\/tr>\n<tr>\n<td>Surface solar radiation (RMSE)<\/td>\n<td><a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">Outperforms HRES at every lead time<\/a><\/td>\n<td>Benchmark reference<\/td>\n<td>No SSRD output published<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Hindcast validation is the second requirement for production deployment because different stakeholders need historical data for different purposes. Quant teams need years of historical forecast data to backtest systematic strategies, while utilities need hindcast parity testing before replacing a pipeline component. To serve both use cases, hindcast data is available across multiple Jua and third-party models on the Jua platform. Backtests run in approximately five minutes via Athena or directly through the SDK for programmatic access.<\/p>\n<p>Documentation requirements for regulated utilities include peer-reviewed technical reports. EPT-2 is documented in <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>, and EPT-1.5 in <a href=\"https:\/\/arxiv.org\/abs\/2410.15076\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2410.15076<\/a>. Both reports are available for internal risk and regulatory review. The key metrics for evaluation are RMSE for deterministic accuracy, CRPS for probabilistic skill, and energy score for multivariate ensemble assessment.<\/p>\n<p><a href=\"https:\/\/meetings-eu1.hubspot.com\/guett\/energy-trading?uuid=d780665f-ff71-439c-addf-c80e49af0627\" target=\"_blank\">Start integrating Jua forecasts with a demo of the API and SDK.<\/a><\/p>\n<h2>Readiness and Opportunity Assessment for Your Team<\/h2>\n<p>The following checklist covers the four dimensions relevant to evaluating Jua for Energy for production deployment.<\/p>\n<p><strong>Technical readiness:<\/strong><\/p>\n<ul>\n<li>Python environment available. <code>pip install jua<\/code> installs the SDK from PyPI in seconds.<\/li>\n<li>REST API access. <code>POST \/v1\/forecast\/data<\/code> and related endpoints are documented at <code>query.jua.ai\/docs<\/code>, with Apache Arrow support for large payloads.<\/li>\n<li>Hindcast access available for backtesting across multiple Jua and third-party models.<\/li>\n<li>ENTSO-E integration for direct grid data, including actual generation, capacity, and PSR classifications for European power markets.<\/li>\n<\/ul>\n<p><strong>Operational needs:<\/strong><\/p>\n<ul>\n<li>Intraday horizons covered by EPT-2 RR, which delivers up to 24 runs per day, while actual-generation power forecasts refresh every 15 minutes.<\/li>\n<li>Day-ahead and multi-day horizons covered by the EPT-2 deterministic flagship, which runs four times per day with a 20-day horizon, and by the EPT-2e ensemble, which extends to 60 days.<\/li>\n<li>Dissemination latency improved by a typical Jua run that completes approximately 2.5 hours ahead of competing operational runs at the same cycle.<\/li>\n<\/ul>\n<p><strong>Organizational factors:<\/strong><\/p>\n<ul>\n<li>Procurement timeline aligned with your governance. Physical trading houses have closed within two weeks of a live benchmark, while regulated utilities typically run longer evaluation cycles anchored to peer-reviewed documentation.<\/li>\n<li>Internal champion identified. Meteorologists who run the live benchmark on their own region and variable often become internal champions when the numbers speak, because the benchmark returns a head-to-head comparison in seconds.<\/li>\n<\/ul>\n<p><strong>Strategic goals:<\/strong><\/p>\n<ul>\n<li>Replacing manual morning prep. Day-Ahead and Intraday briefings auto-refresh on every new model run and cover model consensus, model delta, convergence tracking, and price implications.<\/li>\n<li>Consolidating fragmented pipelines. This automation is possible because more than 25 models, including 10 proprietary AI models from the EPT family and 15 third-party NWP and AI models, run on a single platform with one schema and one API, which removes the integration overhead that makes manual prep necessary today.<\/li>\n<li>Quantified economics. The time savings and operational efficiency translate directly to P&amp;L impact. A 1 GW wind portfolio that gains four percentage points of forecast accuracy saves approximately \u20ac1.5 M per year, and a 1 GW solar portfolio at the same accuracy gain saves approximately \u20ac3 M per year.<\/li>\n<\/ul>\n<p><a href=\"https:\/\/meetings-eu1.hubspot.com\/guett\/energy-trading?uuid=d780665f-ff71-439c-addf-c80e49af0627\" target=\"_blank\">Test your readiness with a live benchmark on your portfolio&#8217;s region and variables.<\/a><\/p>\n<h2>Common Pitfalls When Evaluating AI Weather Models<\/h2>\n<p><strong>Relying on vendor-provided graphics instead of independent benchmarks.<\/strong> Many AI weather vendors publish accuracy claims illustrated with cherry-picked maps or single-event case studies. The correct evaluation methodology is a head-to-head benchmark on the evaluator&#8217;s own region and variable, against a known reference model, using an independent ground-truth dataset. The StationBench methodology described above makes EPT-2 claims auditable. Meteorologists evaluating any AI weather model should demand access to the benchmarking surface before accepting vendor accuracy claims.<\/p>\n<p><strong>Ignoring physics constraints when evaluating AI models for extreme events.<\/strong> The extreme-event underestimation documented in the Science Advances study creates a specific risk for energy professionals. A pattern-based AI model may perform well on standard verification metrics while failing precisely when the forecast matters most, such as during a cold snap, a wind ramp, or a heat wave. Physics-constrained models that learn conservation laws from observational data are architecturally more robust to this failure mode. Evaluators should test explicitly on historical extreme events in their region, not only on average-condition RMSE.<\/p>\n<p><strong>Underestimating integration effort for raw AI research subscriptions.<\/strong> Subscribing to raw AI model outputs such as GraphCast, Aurora, or AIFS delivers model files without ensembles, hindcasts, productised refresh schedules, or workflow tooling. The quant team then builds the ingestion pipeline, ensemble logic, benchmarking harness, and hindcast access themselves. This work typically consumes a full engineering quarter. The correct comparison is total cost of integration, including the engineering time required to reach production. A productised platform with a documented SDK, Apache Arrow payload support, and hindcast access available on day one changes the integration calculus.<\/p>\n<h2>Frequently Asked Questions<\/h2>\n<h3>What is the fundamental difference between AI weather forecasting and traditional NWP?<\/h3>\n<p>Traditional NWP encodes the governing equations of atmospheric dynamics explicitly. The Navier-Stokes equations, thermodynamic relations, and radiative transfer schemes are discretised and solved numerically at each grid point. The physics is correct by construction, but the computational cost is severe, so a single simulation consumes approximately 8,400 kWh and costs \u20ac1,000\u2013\u20ac20,000 on HPC infrastructure, which limits update frequency to two to four runs per day. AI weather models replace the dynamical core with a learned operator trained on historical data. Pattern-based approaches learn statistical associations without encoding conservation laws, while physics-constrained approaches like EPT learn the governing physics directly from observational data in a latent representation. For energy professionals, the practical differences show up in accuracy on extremes, update frequency, and integration cost.<\/p>\n<h3>Can AI weather models be trusted for extreme events?<\/h3>\n<p>The structural limitation documented earlier, which is pattern-based models&#8217; inability to extrapolate beyond their training distribution, means they systematically underestimate extreme events. This behaviour is a documented structural limitation rather than a tuning issue. Physics-constrained models are architecturally more robust because EPT learns conservation laws from observational data, and its outputs respect the physical constraints that govern what the atmosphere can do. EPT-2 uses the StationBench validation approach detailed earlier in the article, with no post-processing or station fine-tuning, and the results are published in peer-reviewed technical reports on arXiv. For energy professionals, the practical recommendation is to evaluate any AI model explicitly on historical extreme events in the relevant region before deploying it in production.<\/p>\n<h3>Does Jua for Energy replace an ECMWF subscription?<\/h3>\n<p>Jua for Energy does not replace ECMWF. Most serious customers keep their ECMWF subscription and run Jua for Energy alongside it. ECMWF AIFS, which is ECMWF&#8217;s own AI model, runs on the Jua platform alongside EPT models. Jua for Energy displaces the plumbing around the ECMWF feed, including the in-house grib pipeline, the spreadsheet stitching, the manual benchmarking, and the morning-briefing analyst. The 7\u20139 a.m. manual prep routine compresses into a single workspace, refreshed up to 24 times per day, where every model, including ECMWF HRES, ENS, AIFS, GFS, Aurora, and EPT, appears on the same screen with one schema and one API.<\/p>\n<h3>How does EPT-2 differ from Aurora and GraphCast at the architecture level?<\/h3>\n<p>Three architectural differences matter for production use. First, EPT-2 produces forecasts at native any-\u0394t and is trained to predict at arbitrary time steps rather than rolling forward in fixed increments. Aurora and most peers roll forward in 6-hour steps, which compounds error at longer lead times. Second, EPT-2e is a productised ensemble with 10 members that beats the 50-member ECMWF ENS mean on both RMSE and CRPS at virtually every lead time. No AI peer has shipped a productised ensemble equivalent. Third, EPT is a general physics foundation model, and the same architecture learns atmospheric dynamics, plasma behaviour in a tokamak, and other physical systems. Aurora and GraphCast are research outputs from large companies&#8217; AI labs rather than foundation model platforms with agents on top of them. Jua&#8217;s category sits one level higher in terms of platform scope.<\/p>\n<h3>What does the evaluation process look like in practice?<\/h3>\n<p>The live benchmark is the standard evaluation trigger. A prospect selects a region and a variable that matters to their book, typically a wind-rich region of their home market, selects their current provider alongside EPT-2, and the Jua platform returns a head-to-head accuracy comparison in seconds. Backtests against years of historical forecasts run in approximately five minutes via Athena. Physical trading houses have closed procurement within two weeks of this benchmark. Regulated utilities run longer evaluation cycles anchored to the peer-reviewed technical reports on arXiv. Quant funds install the SDK, run their own backtests against historical forecast data, and decide on the basis of those results.<\/p>\n<h2>Conclusion and Next Steps<\/h2>\n<p>The shift from traditional NWP to physics-constrained AI foundation models expands the evaluation space rather than replacing one paradigm with another. ECMWF HRES remains the universal benchmark and the correct reference for extreme-event skill. Pattern-based AI models extend medium-range forecast skill by roughly 18 to 24 hours over traditional NWP on standard RMSE metrics but underperform on extremes and lack physics constraints. Physics-constrained foundation models like EPT-2 deliver the accuracy gains on energy-critical variables documented earlier while running at approximately four orders of magnitude lower cost per simulation.<\/p>\n<p>The four evaluation dimensions, which are model capability, operational usability, reliability, and integration fit, resolve to a clear production recommendation for energy professionals. A physics-constrained foundation model with a productised ensemble, up to 24 daily updates, and a documented SDK and API, benchmarked transparently against ground-truth observations, provides a robust foundation. Jua for Energy delivers all four, with EPT-2 and EPT-2e as the model layer, Athena as the agent layer, and a 25-model benchmarking surface that makes the comparison auditable in seconds.<\/p>\n<p>A 1 GW wind portfolio that gains four percentage points of forecast accuracy saves approximately \u20ac1.5 M per year. A 1 GW solar portfolio at the same accuracy gain saves approximately \u20ac3 M per year. The live benchmark is the fastest way to quantify what that means for a specific portfolio, region, and variable.<\/p>\n<p><a href=\"https:\/\/meetings-eu1.hubspot.com\/guett\/energy-trading?uuid=d780665f-ff71-439c-addf-c80e49af0627\" target=\"_blank\">Quantify the accuracy gain for your portfolio with a live benchmark.<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Jua&#8217;s physics-constrained AI outperforms NWP with 24 daily updates, better wind &amp; solar accuracy, and lower cost. Built for energy pros. Book a demo.<\/p>\n","protected":false},"author":103,"featured_media":261,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"inline_featured_image":false,"footnotes":""},"categories":[11],"tags":[],"class_list":["post-262","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-weather-forecasting"],"_links":{"self":[{"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/posts\/262","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/comments?post=262"}],"version-history":[{"count":2,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/posts\/262\/revisions"}],"predecessor-version":[{"id":790,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/posts\/262\/revisions\/790"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/media\/261"}],"wp:attachment":[{"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/media?parent=262"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/categories?post=262"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/tags?post=262"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}