{"id":309,"date":"2026-05-08T23:18:43","date_gmt":"2026-05-08T23:18:43","guid":{"rendered":"https:\/\/jua.ai\/articles\/detailed-hourly-weather-forecast\/"},"modified":"2026-07-04T05:04:43","modified_gmt":"2026-07-04T05:04:43","slug":"detailed-hourly-weather-forecast","status":"publish","type":"post","link":"https:\/\/jua.ai\/articles\/detailed-hourly-weather-forecast\/","title":{"rendered":"Detailed Hourly Forecasts That Match How Traders Trade"},"content":{"rendered":"<p><em>Written by: Olivier Lam, Physical AI Team, Jua.ai AG | Last updated: June 26, 2026<\/em><\/p>\n<h2 id=\"key-takeaways\">Key Takeaways for Energy Desks<\/h2>\n<ul>\n<li>Energy trading P&amp;L depends on forecast freshness, hourly detail, and probabilistic coverage, not just access to raw weather data.<\/li>\n<li>Traditional NWP runs only 2\u20134 times per day, which leaves six-hour gaps that expose traders to unpriced intraday weather shifts.<\/li>\n<li>EPT-2 outperforms ECMWF HRES on 10 m wind, 100 m wind, 2 m temperature, and surface solar radiation across every lead time in the 0\u2013240 hour range.<\/li>\n<li>Physics-constrained foundation models like EPT-2 enable the 24-times-daily refresh cadence that matches intraday trading horizons at ~5 km resolution while cutting compute costs by four orders of magnitude versus traditional NWP.<\/li>\n<li><a href=\"https:\/\/meetings-eu1.hubspot.com\/guett\/energy-trading?uuid=d780665f-ff71-439c-addf-c80e49af0627\" target=\"_blank\"><strong>Schedule a benchmark call<\/strong><\/a> to see how EPT-2 performs against your current provider and close the gap between physical reality and your trading screen.<\/li>\n<\/ul>\n<h2>Where Traditional NWP Leaves Traders Flying Blind<\/h2>\n<p>The two supercomputers that operate global NWP, one at ECMWF and one at NOAA, can run their full algorithm twice a day. With smaller supplementary runs, the energy industry receives roughly four global forecasts per 24-hour period. A single traditional 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 structure has capped update frequency at two to four runs per day for forty years.<\/p>\n<p><a href=\"https:\/\/www.bloomberg.com\/news\/articles\/2026-03-26\/energy-traders-turn-to-ai-to-forecast-the-weather-forecast?embedded-checkout=true\" target=\"_blank\">ECMWF&#8217;s two-week outlook is the definitive reference point for traders repricing risk around heating demand, renewable output, and system tightness<\/a>, yet between the four daily runs traders stare at stale numbers. When a model revises its output mid-cycle, the trader usually discovers the change because someone else has already traded on it.<\/p>\n<p>The workflow built on top of those forecasts compounds the problem. A typical morning begins at 6 a.m. The desk downloads overnight ECMWF and GFS runs as raw grib files (binary gridded format used by meteorological agencies), processes them through an in-house pipeline, cross-references an internal meteorology team or a consultancy, and stitches together spreadsheets, terminal screens, and vendor dashboards. By the time a coherent view of the day exists, the market has already moved. <a href=\"https:\/\/www.bloomberg.com\/news\/articles\/2026-03-26\/energy-traders-turn-to-ai-to-forecast-the-weather-forecast?embedded-checkout=true\" target=\"_blank\">Traders are now turning to AI and machine-learning tools specifically to anticipate forecast revisions<\/a>, which signals that the existing update cadence is structurally insufficient for intraday positioning.<\/p>\n<h2>Update Cadence That Matches Day-Ahead and Intraday Markets<\/h2>\n<p>That structural insufficiency becomes clear when you line up forecast updates against actual market mechanics. Day-ahead power markets in Europe clear the evening before delivery. Intraday markets trade continuously up to minutes before gate closure. A forecast that refreshes four times per day leaves gaps of six hours or more between updates. During those gaps, wind ramps, convective precipitation events, and solar irradiance shifts can move generation output by hundreds of megawatts without triggering any alert.<\/p>\n<p>Point-solution SaaS vendors that resell processed NWP outputs do not solve this problem because they inherit the same four-runs-per-day ceiling from the underlying models they redistribute. That limitation alone would be manageable if they preserved the full probabilistic information, but they typically strip out ensemble outputs (probabilistic forecasts from multiple model runs used to quantify uncertainty) and benchmarking tooling in the process. AI weather research outputs from labs, including Microsoft Aurora and Google DeepMind GraphCast, face a different constraint. Their models could theoretically update more frequently, yet they are typically updated four times per day in research mode, without a productised operational refresh schedule.<\/p>\n<p>The compute economics of physics-constrained foundation models remove that ceiling. A single EPT-2 inference runs 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 a traditional NWP run. EPT-2 RR, Jua&#8217;s rapid-refresh variant, updates up to 24 times per day. EPT-2 HRRR delivers the same cadence at up to 5 km native resolution over Europe. Actual-generation power forecasts on the Jua platform refresh every 15 minutes, so the trading screen tracks the physical system rather than lagging it.<\/p>\n<h2>Hub-Height Wind and Other Variables That Move P&amp;L<\/h2>\n<p>Consumer weather apps report wind at 10 meters above ground, which is the standard meteorological surface observation height. Wind turbines operate at hub heights between 80 m and 160 m for onshore installations and above 100 m for most offshore platforms. Wind speed increases non-linearly with height because of the atmospheric boundary layer. A forecast that is accurate at 10 m can be materially wrong at 100 m, and the error compounds across a multi-gigawatt portfolio.<\/p>\n<p><a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">EPT-2 outperforms ECMWF HRES on 100 m wind across the full 0\u2013240 hour lead-time range<\/a>, which is the variable most directly relevant to wind-turbine power output. Jua for Energy&#8217;s weather forecast surface covers wind at 11 height levels from 10 m to 200 m, surface solar radiation (SSRD), precipitation timing, cloud cover, temperature at multiple levels, and pressure, for 25 variables in total. 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 6-hour increments. Aurora and most AI peers use a fixed 6-hour roll-forward that compounds error at each step. EPT-2 avoids that roll-forward error entirely.<\/p>\n<p>Precipitation timing shapes hydro dispatch and gas demand. Solar radiation shapes daytime power balance and solar-asset positioning. Fixed-timestep models that interpolate between 6-hour outputs introduce artifacts at the sub-6-hour scale. Those artifacts remain invisible in daily RMSE (root mean square error) statistics but matter directly for intraday trades.<\/p>\n<h2>Why a Foundation-Model-Plus-Agent Platform Solves These Gaps<\/h2>\n<p>Solving these technical limitations, including update frequency, hub-height accuracy, and sub-6-hour artifacts, requires a fundamentally different approach to weather forecasting. Jua is a foundation model and agent company in the same category that OpenAI and Anthropic compete in, applied to the physical economy rather than the digital one. EPT is a general spatiotemporal transformer foundation model that learns the governing physics of complex systems, such as conservation of mass, momentum, and energy, directly from observational data. Athena is an AI agent that plans, reasons, and calls tools to turn natural-language objectives into deliverables. The relationship mirrors the one Anthropic has to Claude Code, a horizontal AI platform with a flagship vertical product. Jua for Energy is that vertical product.<\/p>\n<p>A hypothetical intraday workflow shows the difference in practice. At 6 a.m., instead of downloading grib files, a trader opens the Jua platform. The Day-Ahead briefing, auto-generated on the overnight model run, shows model consensus across 25+ models, what moved since the previous run (model delta), whether models are converging or diverging as lead time shortens, and the price implications already written in. At 10 a.m., a divergence alert fires because EPT-2 and ECMWF ENS disagree on 100 m wind over northern Germany for the evening peak. The trader types into Athena: <em>&#8220;What is the 100 m wind forecast spread across models for northern Germany tonight?&#8221;<\/em> Athena returns the comparison widget in approximately 90 seconds. The trader acts. The market re-prices an hour later.<\/p>\n<p><a href=\"https:\/\/nebius.com\/customer-stories\/jua\" target=\"_blank\">Athena turns raw physics predictions from EPT-2 into trading decisions by reading market context and modeling participant behavior<\/a>, which converts model skill into realised P&amp;L. <a href=\"https:\/\/nebius.com\/customer-stories\/jua\" target=\"_blank\">A 1 GW wind portfolio that gains four percentage points of forecast accuracy saves approximately \u20ac1.5 M per year<\/a>, and that saving scales to hundreds of millions for large portfolios.<\/p>\n<p><a href=\"https:\/\/meetings-eu1.hubspot.com\/guett\/energy-trading?uuid=d780665f-ff71-439c-addf-c80e49af0627\" target=\"_blank\"><strong>Run a live comparison<\/strong><\/a> on your own region and variables against 25+ models on the Jua platform.<\/p>\n<h2>Side-by-Side: Frequency, Ensembles, Hub Height, and Workflow<\/h2>\n<table>\n<thead>\n<tr>\n<th>Capability<\/th>\n<th>NWP Incumbents (ECMWF HRES \/ ENS)<\/th>\n<th>AI Research Outputs (Aurora \/ GraphCast)<\/th>\n<th>Jua for Energy (EPT-2 \/ EPT-2e \/ Athena)<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><strong>Update Frequency<\/strong><\/td>\n<td><a href=\"https:\/\/www.bloomberg.com\/news\/articles\/2026-03-26\/energy-traders-turn-to-ai-to-forecast-the-weather-forecast?embedded-checkout=true\" target=\"_blank\">2\u20134\u00d7 per day<\/a>, hard HPC compute ceiling at ~8,400 kWh and \u20ac1,000\u2013\u20ac20,000 per run<\/td>\n<td>Typically 4\u00d7 per day in research mode, no productised operational schedule<\/td>\n<td><a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">Up to 24\u00d7 per day (EPT-2 RR)<\/a>, 15-min refresh for actual-generation power forecasts, ~0.25 kWh and $0.20\u2013$15 per inference on a single GPU<\/td>\n<\/tr>\n<tr>\n<td><strong>Ensemble Skill<\/strong><\/td>\n<td>ECMWF ENS, 50-member gold standard for probabilistic NWP, CRPS (continuous ranked probability score) benchmark<\/td>\n<td>No productised ensemble equivalent from Aurora or GraphCast<\/td>\n<td><a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">EPT-2e beats the 50-member ECMWF ENS mean on both RMSE and CRPS at virtually every lead time<\/a>, 4 runs per day<\/td>\n<\/tr>\n<tr>\n<td><strong>Hub-Height Variables<\/strong><\/td>\n<td>Wind at standard pressure levels, 10 m surface wind, no native hub-height product<\/td>\n<td>Aurora, 10 m wind and no SSRD output, fixed 6-hour roll-forward that compounds error at sub-6-hour scale<\/td>\n<td><a href=\"https:\/\/nebius.com\/customer-stories\/jua\" target=\"_blank\">Wind at 11 height levels from 10 m to 200 m<\/a>, SSRD, precipitation, 25 variables total, native any-\u0394t (no roll-forward error), up to 5 km native resolution<\/td>\n<\/tr>\n<tr>\n<td><strong>Workflow Integration<\/strong><\/td>\n<td>Raw grib files via ECMWF MARS, member access, no productised cross-vendor benchmarking or agent layer<\/td>\n<td>Research code or limited API, no agent layer, no benchmarking surface, pipeline must be built by the customer<\/td>\n<td><a href=\"https:\/\/nebius.com\/customer-stories\/jua\" target=\"_blank\">REST API + Apache Arrow + <code>pip install jua<\/code> SDK<\/a>, Athena natural-language agent (~90 s per query), 25+ models on one platform, auto-generated briefings, divergence and correction alerts<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h2>FAQ: Detailed Hourly Forecasts for Trading Teams<\/h2>\n<h3>What is a detailed hourly weather forecast and why does it matter for energy trading?<\/h3>\n<p>A detailed hourly weather forecast provides atmospheric variable predictions, such as wind speed at multiple heights, surface solar radiation, precipitation, and temperature, at hourly or sub-hourly intervals rather than the 6-hour or 12-hour intervals typical of standard NWP outputs. For energy trading, hourly granularity matters because European intraday power markets trade continuously up to gate closure, and renewable generation can shift by hundreds of megawatts within a single hour due to wind ramps or passing cloud cover. A forecast that resolves only at 6-hour intervals cannot support intraday positioning decisions with the precision those markets require. Hub-height wind variables, such as wind at 80 m, 100 m, or 160 m, are specifically necessary because turbine power output is a cubic function of wind speed, and a 10% error at hub height translates directly into a material generation forecast error across a multi-gigawatt portfolio.<\/p>\n<h3>How does EPT-2 compare to ECMWF HRES for hourly wind and solar forecasts?<\/h3>\n<p>EPT-2, documented in the peer-reviewed technical report arXiv:2507.09703, demonstrates the performance advantage outlined in the key takeaways when evaluated against more than 10,000 real ground stations using the open-source StationBench methodology with no post-processing or station fine-tuning. EPT-2e, the ensemble variant, beats the 50-member ECMWF ENS mean on both RMSE and CRPS at virtually every lead time. ECMWF HRES has held the benchmark for forty years, and EPT-2 now surpasses it on the four variables that most directly drive energy P&amp;L. Jua for Energy does not replace ECMWF, because serious customers keep their ECMWF subscription and run Jua for Energy alongside it. Jua for Energy instead displaces the manual plumbing around the incumbent feed.<\/p>\n<h3>What is StationBench and why does it matter for evaluating hourly forecast accuracy?<\/h3>\n<p>StationBench is Jua&#8217;s open-source benchmarking methodology that evaluates forecast accuracy against more than 10,000 real ground-truth observation stations globally, without post-processing or station-specific fine-tuning. Most vendor accuracy claims are evaluated on gridded reanalysis data, which is a smoothed historical reconstruction, rather than on actual station observations. Station-based evaluation is more demanding because it tests the model against the physical reality at specific points, not against a smoothed average. StationBench is the evaluation surface used to validate EPT-2&#8217;s outperformance of ECMWF HRES and Microsoft Aurora, and it is available as an open-source tool so customers and meteorologists can run their own independent evaluations rather than relying on vendor-provided graphics.<\/p>\n<h3>Can Jua for Energy integrate with existing trading pipelines and internal models?<\/h3>\n<p>Jua for Energy exposes a REST API with Apache Arrow payload support and a Python SDK installable via <code>pip install jua<\/code> from PyPI. The API provides access to 25+ models, including 10 proprietary AI models from the EPT family plus 15 third-party NWP and AI models such as ECMWF HRES, ECMWF ENS, ECMWF AIFS, NOAA GFS, DWD ICON, Microsoft Aurora, and GFS GraphCast, under a unified schema. Hindcast data is available across multiple Jua and third-party models for backtesting systematic strategies. ENTSO-E grid data integrates directly for European power-market data. Quant developers at capital-markets funds typically stand up the integration in days rather than the quarter it takes to build equivalent pipelines from raw AI-weather research subscriptions. The developer dashboard is at developer.jua.ai and documentation at docs.jua.ai.<\/p>\n<h3>Who uses Jua for Energy and what types of organizations benefit most?<\/h3>\n<p>Jua for Energy is used by regulated utilities, physical trading houses, and capital-markets and quantitative funds across five continents. Current customers include Axpo, TotalEnergies, Statkraft, EnBW, EDF, and Hydro-Qu\u00e9bec. Regulated utilities use the full platform surface, including maps, weather forecasts, automatic briefings, divergence and correction alerts, and power forecasts, across meteorology, dispatch, and trading teams. Physical trading houses prioritize API-first access and ensemble outputs for programmatic integration with internal risk engines. Quantitative funds use the Python SDK and hindcast data to backtest systematic weather-signal strategies. The common thread across all three archetypes is the need for forecast accuracy that is independently verifiable, a refresh cadence that matches intraday trade horizons, and workflow tooling that eliminates the manual morning preparation routine.<\/p>\n<h2>Conclusion: Move Before the Weather and the Market Do<\/h2>\n<p>The structural gap in energy-trading weather forecasting is not a data problem because ECMWF HRES has provided high-quality global forecasts for forty years. The gap instead sits in frequency, granularity, and workflow. Four daily NWP runs leave six-hour blind spots between updates. Consumer apps and point-solution SaaS vendors inherit that ceiling. AI weather research outputs from large labs close part of the accuracy gap but deliver raw files without ensembles, benchmarking, or an agent layer.<\/p>\n<p>Jua for Energy closes the remaining gap with EPT-2, a physics-constrained foundation model that <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">outperforms ECMWF HRES on the four variables that drive energy P&amp;L across all lead times<\/a>, refreshing up to 24 times per day at up to 5 km native resolution. Athena, the AI agent, turns a natural-language question into a briefing, a benchmark, or a backtest in approximately 90 seconds. <a href=\"https:\/\/nebius.com\/customer-stories\/jua\" target=\"_blank\">Jua serves major utilities across four continents, with sales cycles compressed to as little as two weeks<\/a> from live benchmark to procurement.<\/p>\n<p>The savings scale quickly. The four-percentage-point accuracy gain mentioned earlier translates to \u20ac1.5 M annually for a 1 GW wind portfolio and \u20ac3 M for an equivalent solar portfolio. The live benchmark takes five minutes. The market does not wait.<\/p>\n<p><a href=\"https:\/\/meetings-eu1.hubspot.com\/guett\/energy-trading?uuid=d780665f-ff71-439c-addf-c80e49af0627\" target=\"_blank\"><strong>Book a demo<\/strong><\/a> and see EPT-2 benchmarked against your current forecast provider on your own region and variables before the next trade window opens.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Jua delivers 24x-daily hourly weather forecasts at ~5 km resolution \u2014 giving energy desks the intraday accuracy to trade with confidence.<\/p>\n","protected":false},"author":103,"featured_media":308,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"inline_featured_image":false,"footnotes":""},"categories":[11],"tags":[],"class_list":["post-309","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\/309","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=309"}],"version-history":[{"count":2,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/posts\/309\/revisions"}],"predecessor-version":[{"id":732,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/posts\/309\/revisions\/732"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/media\/308"}],"wp:attachment":[{"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/media?parent=309"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/categories?post=309"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/tags?post=309"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}