Independent Wind Engineering

Recover wind measurement data. Reduce project uncertainty.

Independent wind engineering for developers, investors and asset owners. We recover compromised LiDAR and SODAR campaigns, independently assess energy yield, and identify the causes of wind farm underperformance. Technical conclusions supported by validated models, quantified uncertainty and transparent reporting.

Book a diagnostic Services
3D mesoscale wind resource simulation over terrain
Wind resource modeling Mesoscale flow downscaled to CFD over terrain
RSD Rectification Lead practice
Independent EYA Review Lender & IE scrutiny
Virtual Met Mast Mesoscale, long-term series
Operating Assets Diagnostics & monthly monitoring
How we work
Wind resource assessment is ultimately an exercise in risk management. Technical decisions must be transparent, justified and defensible.

Every assumption—from measurement strategy and long-term correction to wake modeling and uncertainty analysis—affects project value and investment decisions. Our role is to ensure those assumptions are transparent, technically justified and appropriate for the project context.

We do not rely on a single methodology simply because it is common practice. Instead, we evaluate alternative approaches, quantify their impact on results and document the basis for key decisions. The outcome is a defensible technical position that can withstand review by lenders, investors and independent engineers.

See how we engage
Services

Led by measurement campaign recovery.

RSD rectification is where we are most distinct. We reconstruct and rectify SODAR and LiDAR campaigns in the environments that compromise them most — complex terrain, coastal and humid sites, cold climates — using our own mesoscale virtual met mast, physics-informed machine learning and advanced data analysis, so that a fragmented record becomes a usable one. Wherever we use machine learning, it is trained on high-resolution physics and validated against measurements.

RSD reconstruction workflow: fragmented LiDAR/SODAR record, reconstruction with a virtual met mast, continuous time series with uncertainty, and consistency checks against a concurrent met mast (shear profile, Weibull, wind rose, scatter)
Illustrative example — figures shown are not from a client project.
Lead practice · RSD Rectification & Virtual Met Mast

Recovering value from compromised measurement campaigns.

Remote sensing campaigns can return data with reduced availability, sectoral gaps, or noise contamination — particularly in complex terrain, near forests or infrastructure, and in high-precipitation environments. Where the validated subset of the record is sufficient to anchor a calibration, the campaign can often be recovered through numerical reconstruction rather than redeployed. Using mesoscale modeling calibrated against the reliable hours of the existing record, combined with machine-learning reconstruction that incorporates sectoral and stability dependence, we synthesize a continuous hub-height time series at the sensor location — a virtual met mast. The reconstructed series is cross-validated against retained measurements, and uncertainty is quantified separately for measured and reconstructed portions. The deliverable is a documented record suitable for IE or lender review.

  • Diagnostic of RSD data quality & recoverability
  • WRF mesoscale calibration on trustworthy hours
  • Learned reconstruction model (sectoral, stability-aware), trained on trustworthy hours
  • Virtual met mast time-series synthesis
  • Cross-validation on held-out measured hours
  • Uncertainty quantification for reconstructed data
  • Long-term extension via MCP
  • Documentation suitable for IE / lender review
LIVE / HUB-HEIGHT FLOW · 300 M
ERA5 · 31 KMWRF · 3 KM NESTLONG-TERM SERIES20+ yrs
Mesoscale Virtual Met Mast

A long-term, site-resolved wind series at hub height.

Our own mesoscale modelling chain produces a long-term wind time series at hub height for any site, resolving the terrain, coastal and boundary-layer effects that global datasets miss. It serves as an independent reference for long-term correction, as the fallback for RSD reconstruction, and as the resource baseline for operating-asset analytics. Results are validated against measurements and calibrated to the client's own data where available.

SimulationMesoscale model

WRF nested to refined resolution, finer where terrain demands it, run for a full year at the site.

Long-term extensionStatistical model

Trained on the simulated year to extend the series over decades of reanalysis, preserving the wind distribution and sector climatology.

ValidationMeasurements

Calibrated and validated against masts and LiDARs; error statistics are reported with every deliverable.

  • One simulated year at 10-minute resolution; extended over decades at hourly resolution by a statistical model that preserves the wind-speed distribution, sector climatology and seasonal and diurnal structure of the simulated year
  • Validation report against the client's measurements: bias, correlation at 10-minute, daily and monthly scales, distribution and extreme-value comparison — including where the model underperforms
  • Sector statistics, Weibull parameters and WAsP-ready tables per hub height
  • Calibration against the client's mast or RSD record (MCP) with documented skill
  • Resource maps at hub height for site and layout screening
  • Direct input to EYA review, RSD reconstruction and operating-asset normalisation

Validation results are reported with each deliverable, including where the model underperforms.

Supporting work around the three practices

Scoped, self-contained pieces of work that feed into a measurement campaign, a yield assessment, or an operating-asset review. Each is delivered as a documented dataset or memo.

Early stage

Site Screening

Mesoscale resource screening and comparative benchmarking for the go/no-go decision, before any instrument is deployed.

Early stage

Measurement Campaign Design

Instrument selection, mast and remote-sensing placement, and data-coverage analysis designed for later bankability.

Measurement

Mesoscale Wind Data

Site-specific WRF hindcasts delivered as hub-height time series with a methodology note.

Measurement

LiDAR & SODAR Flow-Curvature Correction

Terrain-induced flow distortion correction for remote-sensing devices in complex terrain.

Assessment

Wind Data Analysis & Long-term Correction

Quality control, multi-method MCP with inter-method variance, and long-term extension against multiple reanalysis datasets.

Operation

SCADA Performance Review

Availability, outage classification, power-curve health and resource-normalized performance from exported 10-minute SCADA data.

FEATURED / FINANCIAL CLOSE
P50 YIELD 3,142 MWh
Energy Yield Assessment & Independent Review

Energy yield assessment for financial close and IE review.

An Energy Yield Assessment prepared for financial close produces the P50, P75 and P90 figures that lenders and investment committees rely on. Whether an assessment is accepted is decided by the lender's independent engineer, not by the label on the report — so we build ours to survive that review: every assumption documented with its source, long-term correction run with several methods and the spread between them reported, uncertainty propagated term by term across measurement, long-term, wake and method, and validation of any modelled series against the site's own measurements included in the report. Where a third-party EYA already exists, we act as the independent technical reviewer and deliver a structured memo for lender or investment-committee use.

  • WRF mesoscale modeling (200 m – 1 km)
  • Met mast & LiDAR data quality control
  • Multi-method MCP regression with inter-method variance
  • Long-term correction using multiple reanalysis datasets
  • Wake loss modeling & layout sensitivity
  • P50 / P75 / P90 with explicit uncertainty propagation, model-derived where ensembles are available
  • Independent third-party EYA review
  • Documentation built for lender & IE acceptance
FEATURED / OPERATING ASSETS
MONTHLY FLEET SCORE · WT03 −3.9 % · since April
Operating-Asset Performance Analytics & Monitoring

Turning operating data into defensible performance evidence — once, or every month.

For owners, lenders and acquirers of operating wind assets, the question is rarely whether production is below expectation, but why. We ingest exported 10-minute SCADA data, apply quality control and outage classification, and separate the contributions of availability, curtailment, power-curve degradation and the wind resource itself. Because the resource term is modeled with the same mesoscale and long-term correction methods used in our yield work, the result is a performance assessment that can be compared directly against the original EYA. Delivered as a documented report and a reusable dataset, and suited to acquisition due diligence, warranty discussions and independent owner reviews.

Continuous independent monitoring

The same analysis can run every month instead of once. Our monitoring platform scores each turbine against the best-behaved machines on site, cross-checks every nacelle sensor against its neighbours, and flags the performance signatures that a conventional OEM report scores as "available": a pitch or controller offset that only bites in partial load, a yaw reference several degrees off, a drifting anemometer, a curtailment window wider than the permit, a power curve degrading quarter by quarter. Each flag arrives with its evidence and a confidence, in a one-page monthly note written by an analyst, not a dashboard alert — so that the next deviation is caught in its second week, not its eighth month. The OEM keeps its service contract; you gain an independent second reading of the same data.

  • SCADA export ingestion & quality control
  • Outage classification: auditable rule table first, learned model for ambiguous cases, every call carries its rule and confidence
  • Curtailment and grid-loss decomposition
  • Power-curve health and degradation tracking
  • Resource-normalized performance vs. the original EYA
  • Monthly fleet scoring and sensor cross-checks, per turbine, with evidence and confidence for each flag
  • Detection of pitch / yaw / anemometer / curtailment-configuration signatures from 10-minute SCADA
  • One-page monthly analyst note; quarterly review call
  • Acquisition due diligence & warranty support
  • Documented report and reusable dataset
From measurement to assessment: LiDAR and SODAR data, mesoscale and CFD modelling, reconstructed time series and uncertainty, energy yield and project value
Modeling & data analysis From measurement to assessment: remote-sensing and mast data, mesoscale and CFD modelling of complex terrain, reconstructed time series with quantified uncertainty, and the energy yield, losses and project value that follow from them.
Capabilities

The methods behind the three practices, and the bespoke work we take on when a problem does not fit a standard scope.

WRF virtual met mast
Our own automated WRF chain, driven by reanalysis and nested down to refined resolutions, produces a hub-height time series for the site and extends it over the long term with a learned statistical model. Delivered as time series, WAsP tables and sector statistics, validated against tall-tower measurements, and calibrated against the client's own measurements where available.
MCP & long-term correction
Multi-method MCP regression (linear, matrix, variance ratio), reanalysis-based long-term correction, and AI-enhanced approaches for short measurement records.
AI weather models
Where a client needs short-term production forecasts, we evaluate operational AI weather models (ECMWF AIFS, GraphCast-class) against the site's own production record before relying on them, and report the skill they actually achieved.
Physics-informed machine learning
Learned RSD reconstruction, statistical long-term extension of mesoscale runs, and classifier models for operating data. Always trained on physics-based simulations or measurements and tested against held-out data.
RSD reconstruction with a virtual met mast
Gap-filling, quality control and learned reconstruction of SODAR and LiDAR records. Where the instrument drops out, the gaps are filled from our own refined-resolution mesoscale virtual met mast rather than from coarse reanalysis, and a residual model learns the site-specific remainder — turning a fragmented campaign into a near-complete profile across the rotor layer.
Wake & layout modeling
Eddy-viscosity and engineering wake models, layout-sensitivity analysis, and validation against operational data where available.
Uncertainty quantification
Component-wise uncertainty propagation across measurement, long-term, modeling, methodological, and wake terms — producing P50/P75/P90 bands with traceable composition.
Cloud HPC infrastructure
AWS ParallelCluster and Azure CycleCloud for large WRF runs and CFD simulations, scaled to project scope without permanent infrastructure investment.
Extreme event analysis
Typhoon and extreme-wind statistics for site selection and design-basis verification — particularly relevant to onshore and offshore sites in monsoon and cyclone-prone regions of Asia-Pacific.
Operating-asset technical analytics
Our own monitoring pipeline: automated SCADA ingestion, outage classification, fleet-referenced power-curve scoring, sensor cross-checks and resource-normalized performance metrics — the same code for every asset, run once for a diagnostic or monthly for continuous monitoring, so results are comparable across a portfolio and over time.
Tools we work with
WRF, WindSim, WindPRO, WAsP, MASCOT, OpenFOAM, Python (xarray, MetPy, numpy, scikit-learn, PyTorch), NVIDIA Earth-2 / CorrDiff, ECMWF AIFS and GraphCast-class models, SQL, and GPU and HPC capacity on AWS and Azure.
Bespoke technical work
Met-ocean characterization, CFD and fluid–structure problems, and foundation or mast engineering questions that sit outside the three practices. Describe the problem by email and we will say plainly whether it is something we are well placed to take on.
Approach

Independent of project counterparties, vendors, and financial outcomes.

Every number we report comes with the assumption behind it and how much the result moves if that assumption is wrong. If a conclusion cannot survive a lender's engineer, it does not leave our desk.

01 / FIRST CONVERSATION

A scoping exchange

Describe the problem and the data you hold by email. We reply with our initial read on fit and scope, and set up a short call only where it would save both sides time.

02 / SCOPING

A short scoping document

If there is a fit, we send a short scoping document covering the question we are answering, the data required, the deliverable, the timeline, and the price.

03 / DELIVERY

Defined deliverables

Engagements conclude with a written technical report and a working artifact — a model, dashboard, tool, or dataset. Thirty days of follow-up is included by default. Retainer arrangements are available where they suit the work.

About

An independent wind engineering practice.

Spenta Power is an independent technical consultancy working in wind resource assessment, measurement, and the performance of operating wind assets. The practice is built around a single conviction: the technical assumptions behind a wind project — how the wind is measured, how records are corrected and extended, how uncertainty is quantified — determine its value, and they should be transparent, justified, and able to withstand independent review.

The work draws on more than a decade of experience across wind resource assessment, mesoscale and CFD modeling, remote-sensing measurement, and the performance analysis of operating renewable assets — including the design and operation of automated monitoring and reporting systems for utility-scale fleets. Machine learning is used where it makes a physically grounded method faster or more consistent, never as a substitute for one. We work as an independent party — for developers, investors, lenders, and the engineers who advise them — without ties to equipment suppliers or project outcomes.

The practice is based in Tokyo and works with clients internationally, across Europe and Asia-Pacific.

Engagements Worldwide
Contact info@spentapower.com
Insights

Notes on wind resource, measurement and operating-asset practice.

Method note · AI downscaling

Why AI downscaling only works when it is trained on resolved physics

A downscaling model cannot invent a ridge speed-up it has never seen. Trained on coarse data it returns smooth, plausible and wrong fields; trained on high-resolution WRF over the same terrain it reproduces the ridge, valley and coastal effects the physics resolved, for every site and every year, at a fraction of the compute. The expensive part is not the AI. It is the WRF archive it is trained on and the measurements it is validated against.

Mesoscale · Validation

WRF versus ERA5 over flat terrain: where mesoscale adds value, and where it doesn't

A full year of our 3 km WRF chain against a 200 m research tower on flat terrain tracks the tower closely at daily and monthly scales, but carries a modest positive speed bias near the surface that fades with height. On flat ground ERA5 alone does about as well, so WRF earns its cost in complex terrain and near coasts, not on plains. What matters for long-term correction is that the bias is stable across a decade of measurements: a stable bias calibrates out, a drifting one does not. Most of the remaining bias traced back to surface roughness — a land-cover question, not a boundary-layer-scheme question.

Measurement · SODAR reconstruction

Rebuilding a fragmented RSD campaign with a virtual met mast

Many sites have a SODAR or LiDAR record that is good when it works and missing when it matters. We fit a smooth profile across the measured heights each hour, fall back to our refined-resolution mesoscale virtual met mast when the instrument drops out, and let a residual model learn only what remains. Because the fallback already carries the site's terrain and coastal effects, the reconstruction stays physically consistent through the gaps, and the campaign's usable coverage becomes near-complete without extending the measurement period.

MCP · Long-term correction

The case for multi-method MCP in resource assessment

Each MCP method makes different assumptions about what is stable between the reference and target periods. The inter-method spread is typically 1.5 to 4 percent in long-term mean wind speed, which propagates to several percent in long-term energy. Carrying that spread as an explicit uncertainty term, rather than collapsing it through a single method choice, produces a more defensible result.

Onshore Wind · EYA Review

Common assumption risks in third-party EYAs

Reference correlations evaluated only on a single period, wake loss assumptions that may not match the as-built layout, loss factors carried forward from earlier projects without site-specific review, and uncertainty bands that do not fully capture complex-terrain variability. These are recurring areas worth checking in independent review.

Offshore Wind · Met-Ocean

Why offshore wind benefits from coupled modeling

The marine atmospheric boundary layer is more stability-dominated than its onshore counterpart, swell influences the local wind field through wave-induced stress, and floating LiDAR motion introduces a systematic bias. Coupled wind–wave modeling addresses these effects in ways that a marine-roughness adjustment to an onshore workflow does not.

Coastal · Extreme Events

Extreme event statistics under a non-stationary climate

Offshore wind and coastal infrastructure design has historically used extreme wind and wave statistics derived from stationary climate assumptions. Recent decades suggest the tails of these distributions are shifting, and site-specific extreme event modeling that incorporates climate scenarios is becoming part of the design conversation.

Operating Assets · Performance

Separating the wind from the turbine in underperformance reviews

A farm producing below its P50 may be suffering from availability losses, curtailment, power-curve degradation, or simply a weak wind year — and the remedies are entirely different. Normalizing production against an independently modeled resource, rather than against nacelle anemometry alone, is what makes the decomposition credible to a lender or a buyer.

A scoped first engagement SCOPED ENGAGEMENT

The Scoped Diagnostic.

Three entry points, one fixed engagement. Hand over an existing EYA, a compromised RSD measurement record, or a SCADA export from an operating asset; within fifteen working days you receive a structured technical diagnostic, a written memo and a 60-minute walk-through. Choose the track that matches your question.

Track A
EYA review
You hold a third-party energy yield assessment and need to know how much weight it will bear.
  • Which three assumptions move P50 and P90 the most, and by how much?
  • Is the long-term correction defensible with a different method and a different reference?
  • Do the wake and loss assumptions match the as-built layout?
Memo: ranked assumption risks with quantified sensitivity, and what to ask the original consultant.
Track B
RSD record
A LiDAR or SODAR campaign is fragmented, flagged, or stopped early, and the bank wants a usable record.
  • Which part of the record is trustworthy, and in which conditions did it fail?
  • Can the gaps be reconstructed to a standard an independent engineer will accept?
  • Reconstruct, extend, or redeploy — which costs less uncertainty?
Memo: data-quality regime, recoverability verdict and the recommended reconstruction pathway with its expected uncertainty.
Track C
Operating asset
A farm produces below budget and the OEM report says every turbine is available.
  • How much of the gap is wind, how much is downtime, curtailment or the turbines themselves?
  • Which turbines deviate from the fleet, since when, and with what signature?
  • Is a flagged underperformer real, or a drifting sensor?
Memo: loss decomposition per turbine, prioritised findings with evidence ready to forward to the OEM. Continues, if wanted, as monthly independent monitoring on the same method.
STEP 01
Handover & scoping
You share the EYA, the RSD record or the SCADA export under NDA. A short written kickoff — or a 30-minute call where preferred — confirms the diagnostic track, the principal questions and the delivery date.
STEP 02
Independent technical work
For EYA review, we re-run the critical analytical steps independently — MCP, long-term correction, wake assumptions, uncertainty propagation. For RSD diagnostic, we characterize the data quality regime, identify the validated subset, and run a feasibility pass on reconstruction. For operating assets, we quality-control the SCADA record, classify outages, and benchmark production against an independently modeled resource.
STEP 03
Findings memo & walk-through
A 10–15 page memo with the diagnostic findings for the chosen track, and a clear statement of what further work, if any, is warranted. Followed by a 60-minute walk-through with your technical team.
Duration
Typically 15 working days from complete data handover; confirmed per track at kickoff
Deliverable
Written memo + 60-minute walk-through
Scope
Fixed. Complex records confirmed at kickoff.
Fee
Fixed, quoted at kickoff from the track, the record length and the number of turbines or heights — agreed before any work starts. Monthly monitoring after Track C is priced per turbine and cancellable quarterly.
Request a diagnostic

Have a project we should look at?

Describe what you are working on, the data you hold, and what you would like assessed. We reply within one business day, and a first conversation is without obligation.

info@spentapower.com
Spenta Power 3-1-13 Kanamecho, Toshima-ku Tokyo 171-0043, Japan Tel +81 3-6825-1806 Mobile +81 80-9804-3090 JST (UTC+9)

Available for engagements worldwide