A Science plus Technology Vision for WMO - Advancing Weather, Climate with Data Science

Implement a data-centric governance model for weather and climate analytics today. This mission centers on standardized metadata, robust data quality checks, and open, secure data sharing across agencies and borders. Target: ingest 2 petabytes of new observations each day from satellites, aircraft, ground stations, ocean sensors, and radar networks, with end-to-end latency under 5 minutes for critical datasets. Establish a centralized data broker, a common catalogue, and a transparent access framework that serves researchers, operators, and decision-makers.

See also: 141 Moral Issues in Capital Management.

Create a modular data fusion stack that combines live observations with model outputs. Use a shared data model and FAIR principles to ensure data are Findable, Accessible, Interoperable, and Reusable. Implement a metadata registry with machine-actionable descriptors and data lineage tracking. Deploy scalable microservices that ingest, validate, and publish data to a 24/7 compute grid.

Invest in capacity building and talent retention: train 250 data scientists and 180 meteorologists per year; fund 5 joint data-science projects between WMO members annually; ensure at least 30% participation from developing regions within partner programs.

Open-source tools and reproducible workflows become the baseline: adopt xarray, Dask, Apache Arrow, and Jupyter-based dashboards; provide templates for reproducible notebooks; maintain a shared Git repository with CI pipelines. Establish 3 priority software packages per year to accelerate forecast verification, anomaly detection, and climate scenario analysis.

Operational metrics define success: latency for real-time surface observations under 3 minutes; data quality targets at 95% accuracy for critical nowcasts; 90% of core datasets published with machine-readable metadata; and a 20% annual increase in data-powered decision support use by member countries.

Which Research Priorities Foster Weather with Climate Forecast Enhancement?

This mission centers on strengthening data and model integration to boost weather and climate forecast accuracy. By combining radar, satellite, surface, and in-situ observations with coupled models, we can shorten update cycles and reduce bias across scales.

Advance data assimilation methods and ensemble streams to deliver probabilistic forecasts that support risk-based decisions. Build streamlined pipelines that ingest observations within hours and propagate uncertainties through forecasts for days to seasons.

Invest in robust observation networks, computing capacity, and cross-domain linkages to ensure consistent initial conditions. Leverage machine learning for post-processing, bias correction, and rapid anomaly detection while preserving physical constraints.

This mission requires strong collaboration across agencies, regions, and research communities to share data, benchmarks, and software, accelerating their adoption into forecast centers.

Key Research Areas

Priority	Rationale	Actions	KPIs
Integrated data assimilation across atmosphere-ocean-land-cryosphere	Aligns initial conditions across domains to reduce cross-domain biases.	Implement coupled EnKF/4D-Var, unify QC, share pilot datasets across centers.	RMSE reduction 15-25% for 3–7 day forecasts; update latency < 3 hours for core products.
Observational network optimization and real-time ingestion	Maximizes impactful observations within assimilation windows.	Prioritize high-impact satellites, expand radiosonde and surface networks in underserved regions, automate QC.	Coverage improvement 20–40% in target regions; ingestion latency under 1 hour.
Climate-weather coupling for seasonal forecasts	Improves initialization for climate-scale forecasts and boundary conditions.	Develop seamless coupling between seasonal climate models and daily-scale weather models; cross-validate biases.	Skill gain in 2–6 month forecasts; probabilistic calibration metrics improved by 10–20%.
Uncertainty quantification and ML-enhanced post-processing	Delivers reliable probability forecasts and actionable risk metrics.	Use ML to learn residual biases, calibrate ensembles, quantify uncertainty; ensure physics constraints.	Reliability metrics improved; under-dispersion reduction; user trust indicators.

Implementation Path and Collaboration

See also: Innovative Insurance Solutions in a Changing Climate.

Establish a joint R&D agenda with clear milestones over multi-year cycles and a governance structure to coordinate data sharing and software stewardship. Build capacity through training programs, shared datasets, and open-source tooling accessible to national meteorological services.

How Can Worldwide Observation Networks Be Upgraded for Real-Time Data?

Deploy modular, interoperable sensor grids with edge processing to deliver near-real-time data streams. Each node fuses measurements locally for rapid quality checks and forwards vetted data to central hubs, reducing backhaul load and enabling faster alerts.

Increase coverage with a mix of low-cost surface sensors and enhanced fixed stations to close observation gaps. Set a target that 90% of critical observations reach the data center within five minutes of collection, and plan the addition of 100 new micro-sensors per 1000 km² in high-priority regions over five years. Include portable profilers for rapid seasonal campaigns.

Establish multi-layer downlink using regional ground stations, satellite links, and dense relay nodes to shrink latency across regions. Prioritize storm-prone areas and coastal zones where rapid updates save lives and property.

Adopt open standards and shared data formats such as NetCDF, CF conventions, SensorML, and the WMO Information System. Create a common API for data ingestion and push data into regional data cubes with clear provenance and accurate timestamps to ensure traceability.

Automate quality control at the edge with lightweight checks and machine-learning assisted anomaly detection. Use automated QC flags, metadata validation, and cross-verification with neighboring sensors to reduce false alarms and improve confidence in alerts.

Governance must align with the mission: secure cross-border data-sharing agreements, stable funding for maintenance, and robust cybersecurity with access controls and audit trails. Establish a two-tier model in which regional nodes manage local fusion and a global center coordinates standards, benchmarks, and progress metrics.

See also: Manifesto 2024.

Implementation steps include gap analysis, pilot networks in diverse climates, phased scale-up, and continuous monitoring of latency, uptime, and data quality. Publish dashboards for stakeholders and build capacity through training programs tied to the data pipeline.

Which Data Standards with Interoperability Underpin the Global Vision?

Adopt a core stack: CF-compliant NetCDF-4 for gridded fields, BUFR and GRIB for observations and forecasts, and metadata expressed in ISO 19115/19139 profiles. Publish data with persistent identifiers and clear licenses, and expose access through OGC API standards to enable seamless cross-system use.

This mission relies on open, well-documented standards that teams can adopt incrementally. Build a shared metadata model anchored in the WMO Core Metadata Profile and ISO 19115, with a controlled vocabulary for variables, units, and provenance. Attach data quality flags and lineage details in machine-readable form to support automated discovery, citation, and reproducibility.

Open interfaces prove interoperability: implement OGC API - Features and OGC API - Coverages, plus WMS/WMTS for visual access when needed. Provide data in multiple encodings (NetCDF/CF, JSON-LD for metadata) and ensure consistent spatial reference systems and temporal axes across datasets.

Governance and stewardship drive consistency: define licensing, access rules, and versioning; track data lineage; require metadata completeness at intake; maintain history and change logs. Use DOIs for dataset release and assign stable identifiers to forecast products and observation streams.

Implementation plan includes multi-institution pilots, with targets: 80% of new data streams CF-compliant within two years; 90% of forecast and observation feeds accessible through API within 15 minutes of generation; API uptime goal of 99.5%, and documentation published in a central repository with sample queries.

How Can AI, Data Assimilation, plus HPC Transform Predictive Skill?

Adopt a three-layer stack now: AI surrogates for fast subgrid physics, AI-informed data assimilation to tighten analyses, and HPC-enabled workflows to scale ensembles. This mission aligns science with operations and yields faster turnaround and stronger forecast confidence.

Implementation outline

• AI surrogates: replace costly subgrid components such as radiative transfer and cloud microphysics with physics-informed networks that produce outputs at 1/2 to 1/5 of the compute cost. Validate against full-physics runs across 5 representative cases and keep error bounds within 0.5–1.0 K for near-surface temperature and within 0.5–1.0 m/s for key wind bands.
• Data assimilation enhancements: fuse AI-predicted covariances with an ensemble-variational framework; allow adaptive inflation tuned by online performance; keep observation impact high while controlling spurious signals. Target 10–15% RMSE reduction for 24–72 h forecasts in pilot regions.
• HPC workflows: containerize components, parallelize assimilation loops with MPI and multi-threading, and minimize I/O stalls through staged data movement. Run ensembles of 32–64 members on clusters with tens of thousands of cores; aim for end-to-end 48-h forecast generation times under 2–3 hours on peak runs.

Concrete results to expect

• Production environment: AI surrogates cut per-kernel time by 40–60%; overall ensemble loop time drops 20–35%; data assimilation cycles complete within the forecast window.
• Reliability: AI-guided covariances reduce ensemble spread misalignment by 15–25%, improving calibration metrics for key fields.
• Implementation readiness: deploy continuous integration for model code, ensure reproducible experiments via versioned datasets, and maintain an auditable trail for forecasts and post-processing.

Next steps include a 6-month pilot in one region, expanding to adjacent areas as results prove robust, and establishing governance with clear data access, reproducibility, and audit trails for forecasts and post-processing.

Which Governance, Access, and Collaboration Models Guide Open Data?

Adopt a three-layer governance model aligned to the mission: a Steering Board, a Data Steward Network, and an Access & Compliance Office. The Steering Board sets policy, prioritizes datasets, and approves licensing. The Data Steward Network handles metadata, quality checks, and dataset lifecycle for each item. The Access & Compliance Office manages licensing, user authentication, and audit trails.

Licensing should default to CC0 for public data and CC BY 4.0 for data requiring attribution, with explicit terms for derivative works. Each dataset carries a machine-readable license in the metadata. Implement a data catalog with DCAT-AP or DataCite DOIs. Maintain a central API gateway to serve data through REST endpoints with rate limits and usage logs. Include privacy and safety constraints, ensuring sensitive information is protected.

Access tiers: Public for forecast products and historical data; Research for verified researchers via two-factor authentication; Restricted for sensitive datasets under data sharing agreements. All access events logged for accountability. Fees: waive for non-profit research and education; charge modest fees for large commercial requests with annual caps.

Collaboration models: Create a common data governance charter, standard metadata, and API specs; adopt data exchange standards like DCAT-AP, OpenAPI, and SensorThings API for sensor data; use DataCite DOIs. Use cross-agency working groups with quarterly sprints; publish quarterly transparency reports on dataset counts, licenses, access counts, and incident logs.

Implementation blueprint

Launch a 90-day rollout: publish the governance charter, appoint data stewards, deploy license templates, and connect the catalog to the API gateway. Set measurable targets: access decisions within 2 business days, data quality score of at least 92% from automated checks, and catalog completeness of at least 85% across top 200 datasets.

Harvest feedback through regular sprints: collect use cases from forecasters and researchers, adjust tiers and licensing, and refine the catalog. Publish a quarterly transparency report detailing dataset counts, licenses, access counts, and incident logs to maintain trust.

How Can Capacity Building plus Knowledge Transfer Operate for Member Nations?

Launch three regional CBKT hubs anchored to a shared mission, delivering three streams: structured data-science training, applied forecast and climate-project work, and mentor-based coaching. Target 150 practitioners trained per hub each year, plus 25 regional mentors who support cohorts, peers, and national teams. Each hub connects to a central knowledge center hosting modular courses, hands-on exercises, reusable code, and multilingual materials that align with WMO data standards and interoperable systems.

Design knowledge transfer flows with fast feedback. Offer monthly micro-learning modules of 8–12 minutes, quarterly hands-on workshops in regional centers, and annual virtual bootcamps that bring together participants from multiple nations. Pair trainees with mentors from national meteorological services and international experts. Provide secondment opportunities lasting 3–6 months to work on national projects and share learnings back to the program. Materials align with WIS 2.0 interoperability and open data policies where authorized. Create a living playbook that captures successful pilots and supports scaling.

Secure a sustainable resource model by dedicating funds for delivery, repository development, and evaluation. Allocate roughly 60% to training delivery, 25% to the knowledge center and tooling, and 15% to monitoring and evaluation. Build partnerships with universities for credentialing and with industry to access real-world datasets. Translate materials into the main national languages. Provide cloud-based practice labs with secure data access to enable remote learning while protecting sensitive information.

Define practical metrics to track progress: number of staff trained, improvements in workflow efficiency, data-quality indicators, and end-user satisfaction. Publish a quarterly dashboard for member nations and conduct annual reviews with a diverse representation from regional groups. Use feedback to adapt curricula, update datasets, and refresh toolkits, ensuring alignment with national capacity plans and emergency response needs.

Implementation steps: establish a CBKT governance body within the WMO structure; roll out 3 regional hubs within 12 months; build a central repository of courses, datasets, and code; run two pilots per region in the first year; scale to all member nations within three years. Monitor milestones quarterly and adjust resource allocation to meet country-level demand. The result is a data-capable workforce ready to produce timely climate and weather services that support resilience.

What Metrics, Validation, and Evaluation Methods Track Progress Effectively?

Adopt a mission-aligned metric set with 8 to 12 indicators and publish a living dashboard within four weeks after each data release to keep teams focused and accountable.

Metrics to Track

Focus on three tiers: forecast accuracy, probabilistic skill, and data quality. For point forecasts, track MAE and RMSE across key variables (temperature, precipitation, wind) with regional stratification. For probabilistic forecasts, report CRPS and Brier scores, plus reliability curves to reveal miscalibration. Monitor data latency (time from observation to ingestion), data completeness (percentage of expected observations), and anomaly rate. Add governance metrics: model version count, documentation coverage, and reproducibility indicators (code availability, containerization, and version tagging). Set targets such as MAE < 1.5°C for daily temperature in mid-latitude regions, CRPS under 0.25 for precipitation probability, latency under 15 minutes for streaming feeds, and data completeness above 98% in critical stations. Review metrics monthly and compare against a climatology baseline to keep the mission transparent.

Validation and Evaluation Methods

Use rolling-origin validation with a 12-month forecast horizon and a 5-year data window to reflect seasonal cycles and climate trends. Apply spatio-temporal cross-validation to avoid overfitting by splitting regions and seasons. Do hindcasts for the past five to ten years and compare forecasts to observed outcomes. Conduct ablation studies to measure the impact of data sources, smoothing, and model components. Run calibration checks with reliability diagrams and PIT tests to ensure probabilistic outputs align with observed frequencies. Quantify uncertainty with prediction intervals and coverages (e.g., 90% intervals capture observed outcomes about 90% of the time). Track model drift by monitoring shifts in input and output distributions, updating the validation plan at least twice a year. Document the evaluation plan, publish code and data provenance where possible, and automate report generation to keep stakeholders informed about progress toward the mission goals.