Regional wind-farm icing-event production-loss mapping from satellite climatology
Blade icing is the largest single cause of cold-climate wind-farm curtailment, yet its spatial pattern is highly localised. Combining MODIS/VIIRS cloud-phase data with ERA5 wet-bulb fields produces grid-scale icing-climate indices that point records cannot.
Sensors
- MODIS Terra / Aqua: Cloud-phase discrimination (ice vs. liquid water) from the 3.7 µm and 11 µm band combination at 1 km spatial resolution; four overpasses per day per location from the two platforms combined. The MYD06/MOD06 cloud-property products carry cloud-top temperature and thermodynamic phase flags used to identify supercooled liquid water layers, the primary source of in-cloud icing. Archive extends to 1999 (Terra) and 2002 (Aqua).
- VIIRS NOAA-20 (and Suomi-NPP): Successor to MODIS with comparable cloud-phase retrieval capability; 750 m resolution for most bands, roughly one overpass per platform per day at mid-latitudes. VIIRS extends the cloud-phase record forward from 2012 (Suomi-NPP) and provides cross-calibration continuity with MODIS for multi-decadal icing-frequency analysis.
- ERA5 reanalysis (ECMWF): Hourly gridded fields at approximately 31 km horizontal resolution, covering 1940 to near-present. Wet-bulb temperature, 2 m temperature, relative humidity, precipitation type and wind speed at multiple pressure levels are all available. ERA5 is the backbone for identifying hours when wet-bulb temperature falls in the icing-risk window (typically 0 °C to roughly −8 °C for freezing rain and wet snow; colder for in-cloud rime). Uncertainty in complex terrain is real: ERA5 smooths topography, so sub-grid orographic corrections are necessary.
- MERRA-2 reanalysis (NASA GMAO): Hourly reanalysis at 0.5° × 0.625° resolution from 1980 to present. Useful as an independent cross-check on ERA5 icing-hour counts and for sites where ERA5 and MERRA-2 diverge, which itself signals orographic uncertainty worth flagging to a client. MERRA-2 liquid-water path and ice-water path fields supplement the satellite cloud-phase retrievals.
- MODIS Terra / Aqua MOD10 / MYD10 snow-cover products: Daily 500 m binary and fractional snow-cover maps. Used here not to detect icing directly but to validate the seasonal onset and retreat of cold conditions at candidate sites, and to identify persistent snow-on-ground periods that correlate with elevated rime-icing frequency in the literature.
Why a met mast cannot answer the regional question
A single meteorological mast records temperature, humidity and occasionally icing-sensor data at one point. Cold-climate wind development rarely involves a single point. A developer assessing a 200 km ridge system, or a portfolio manager reviewing assets across three countries, needs to know which sub-regions accumulate the most icing hours per year, how that pattern shifts between La Niña and El Niño winters, and where the uncertainty is large enough to affect a financing decision.
Ground station networks in cold, mountainous or boreal regions are sparse almost by definition. The terrain that makes a site windy also makes it expensive to instrument. Satellite climatology fills that gap not by replacing a met mast at the chosen site, but by providing the spatial context that tells you which sites are worth masting in the first place, and what the inter-annual variance of icing risk looks like across a 20-year energy-yield period.
What the cloud-phase signal actually measures, and what it does not
The physical basis is straightforward. Supercooled liquid water droplets, the primary agent of in-cloud rime icing on turbine blades, emit and reflect radiation differently from ice crystals at the 3.7 µm shortwave-infrared band. MODIS MOD06 and its VIIRS equivalent exploit this contrast to assign a thermodynamic phase to each cloudy pixel. A persistent record of supercooled liquid water cloud at or below ridge-top elevation, combined with ERA5 wet-bulb temperatures in the icing window, constitutes the core icing-climate index.
The honest limits matter here. MODIS at 1 km and VIIRS at 750 m cannot resolve individual turbine positions; they provide a grid-cell average that may straddle a ridge crest and its sheltered lee slope simultaneously. Four overpasses per day miss the diurnal cycle of fog formation and dissipation, which can be pronounced in complex terrain. Thick ice cloud above a supercooled layer can obscure the lower layer entirely, causing undercount. And ERA5's 31 km grid smooths the very orographic features, valley cold pools and summit exposure, that drive the spatial variability a developer most wants to resolve. These are not reasons to abandon the method; they are the uncertainty bands that must accompany any icing-frequency map delivered to a client.
Building the icing-climate index: method in plain language
The derivation proceeds in three steps. First, ERA5 hourly wet-bulb temperature fields are screened to identify hours within the icing-susceptible range at each grid cell. Freezing precipitation (freezing rain, freezing drizzle) typically occurs between 0 °C and −5 °C wet-bulb; in-cloud rime accumulates over a wider range down to roughly −20 °C, though accretion rates fall sharply below −10 °C. Each hour meeting the temperature criterion is flagged as a candidate icing hour.
Second, MODIS and VIIRS cloud-phase composites are used to weight those candidate hours. A candidate hour at a grid cell that also shows persistent supercooled liquid water cloud in the satellite record receives a higher icing-probability score than one where the cloud phase is predominantly ice or where the sky is clear. This step is what separates the combined approach from a pure reanalysis method: ERA5 alone cannot reliably distinguish supercooled liquid from ice-phase cloud.
Third, the annual and seasonal icing-hour totals are computed over the available multi-decadal archive, producing a distribution rather than a single number. The inter-annual standard deviation of icing hours is often as important to an energy-yield model as the mean, because a P90 production estimate must account for the probability of an anomalously severe icing winter. The resulting index is delivered as a gridded dataset at the ERA5 native resolution, with optional downscaling to finer grids using terrain-following interpolation where digital elevation data supports it.
Translating icing hours into production-loss estimates
Icing-hour counts are not directly production losses. The conversion requires turbine-specific power curves, hub-height wind speeds during icing events, and assumptions about operator response: whether the turbine is curtailed automatically, allowed to run with degraded output, or shut down for safety. Published studies, including work in the journal Remote Sensing and IEA Wind Task 19 reports, have used icing-hour indices calibrated against SCADA production data from Nordic and Canadian fleets to derive regional loss factors. Typical reported production losses in high-icing Scandinavian sites range from 1% to 17% of annual energy production depending on site elevation and latitude, with the highest losses at exposed sub-alpine ridges above 600 m.
At the portfolio screening stage, the satellite-derived icing index feeds a simplified loss model: icing hours multiplied by a curtailment probability and an average power loss fraction. The output is an expected annual energy loss expressed in MWh per MW installed, with a confidence interval reflecting both the meteorological uncertainty and the scatter in published loss-factor relationships. This is enough to rank candidate sites and to flag which ones require a dedicated icing-measurement campaign before financial close.
Archive depth and inter-annual variability: the case for 20-plus years
A single winter of met-mast icing data tells you almost nothing about the P90 year. The MODIS archive runs from 2000 to present, giving over two decades of cloud-phase observations. ERA5 extends to 1940, though the pre-satellite era reanalysis is less constrained by observations. The practical sweet spot is the post-2000 period where MODIS, ERA5 and MERRA-2 overlap, providing roughly 24 years of consistent data at the time of writing.
That span is long enough to capture several positive and negative Arctic Oscillation winters, which are the dominant driver of cold-climate icing variability across northern Europe and Canada. It is not long enough to characterise multi-decadal trends in icing frequency with high confidence, particularly given ongoing Arctic warming. Any deliverable should state the archive period explicitly and note that the trend direction in icing frequency is site-dependent: some high-latitude sites are seeing fewer severe icing seasons; others are seeing more freezing-rain events as the temperature distribution shifts.
Satellize runs this analysis on the open MODIS, VIIRS and ERA5 archives as part of its satellite-data analytics service, with the Kingdom of Tonga crop-estimation programme demonstrating the same principle of deriving actionable indices from multi-source open data.
What this analysis cannot replace
Satellite climatology is a screening and ranking tool, not a bankable energy assessment. Lenders and turbine manufacturers will require site-level icing measurements, typically from heated icing sensors or ice-mass detectors on a met mast, before accepting a production estimate. The satellite-derived index sharpens the question: it tells you that Site A has roughly twice the expected icing-hour frequency of Site B, which justifies deploying the more expensive measurement campaign at Site A and adjusting the energy yield model accordingly.
The method also has nothing to say about icing events driven purely by fog at temperatures just below 0 °C in flat terrain, where the ERA5 temperature gradient is small and the cloud-phase signal is ambiguous. In those environments, a dense network of surface observations remains the primary source of truth. Satellite climatology is most valuable where the terrain is complex, the station network is thin, and the spatial variability of icing risk is highest. That is also, not coincidentally, where the best wind resources tend to be.
Typical figures
| Spatial resolution (cloud-phase input) | 1 km (MODIS MOD06/MYD06); 750 m (VIIRS) |
| Spatial resolution (ERA5 meteorological fields) | ~31 km native; downscaled to ~1 km with terrain-following interpolation where DEM supports it |
| Temporal resolution of icing-hour index | Hourly (ERA5 backbone); cloud-phase composited to daily and monthly |
| Satellite overpass frequency | 4 per day per location (MODIS Terra + Aqua combined); ~1 per day per platform (VIIRS) |
| Archive depth | MODIS: 2000 to present; ERA5: 1940 to present (post-2000 period recommended for combined analysis) |
| Key spectral bands (cloud-phase retrieval) | 3.7 µm (shortwave infrared) and 11 µm (thermal infrared) for liquid/ice discrimination |
| Icing-hour detection threshold | Wet-bulb temperature 0 °C to approximately −20 °C (accretion rate falls sharply below −10 °C); supercooled liquid water cloud flag required for in-cloud rime classification |
| Deliverable grid formats | GeoTIFF, NetCDF-4, CSV summary tables; QGIS/ArcGIS compatible |
| Latency (climatology product) | Retrospective analysis; typically delivered within 2 to 4 weeks of commission |
| Coverage | Global; most relevant for latitudes above 45°N/S and elevations above 400 m |
Analytics Satellize can run
| Annual icing-hour frequency map | ERA5 wet-bulb temperature screening combined with MODIS/VIIRS supercooled liquid water cloud frequency compositing | Gridded GeoTIFF showing mean annual icing hours per cell, with inter-annual standard deviation layer, covering the client's region of interest |
| Seasonal icing-risk profile | Monthly disaggregation of icing-hour index over the full archive period, stratified by icing type (freezing precipitation vs. in-cloud rime) using ERA5 precipitation-type flags | 12-month climatological chart per candidate site, plus regional heatmap showing peak icing month; delivered as PDF report and CSV |
| Site-ranking scorecard | Composite icing-severity score weighting mean icing hours, P90 icing-winter severity and elevation-corrected ERA5 wet-bulb exposure | Ranked table of candidate sites with icing-risk tier (low / moderate / high / severe) and recommended measurement-campaign priority |
| Expected annual production-loss estimate | Icing-hour index multiplied by published curtailment-probability and power-loss-fraction relationships from IEA Wind Task 19 literature; uncertainty propagated from both meteorological and loss-factor scatter | MWh/MW/year loss estimate with P50 and P90 bounds per candidate site, formatted for input to energy-yield models |
| Inter-annual variability and climate-index correlation | Pearson and Spearman correlation of annual icing-hour totals against Arctic Oscillation index and North Atlantic Oscillation index over the MODIS-era archive | Correlation coefficients and scatterplots per site; narrative section on whether a site's icing risk is predictable from seasonal climate forecasts |
| Icing-frequency trend analysis | Mann-Kendall trend test applied to annual icing-hour time series over the post-2000 archive; trend magnitude estimated by Theil-Sen slope | Trend significance and direction per grid cell, with honest statement of confidence limits given the 24-year record length; GeoTIFF trend layer |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.