A met mast usually measures for one to three years, but a wind farm is financed on 20 to 25 years of expected production. Long-term correction, often called MCP (measure-correlate-predict), uses a long reference record to estimate what the short onsite measurement would look like over a typical long period.
At most Indian wind sites, the south-west monsoon between roughly May and September produces the majority of the year's energy. That seasonality is the main reason long-term correction in India needs more care than a standard workflow gives it.
Why the monsoon complicates MCP
- One season dominates. If the measurement includes an unusually strong or weak monsoon, the annual mean is pulled with it, and a short campaign may contain only one monsoon.
- Correlation can mislead. A reference can show a high annual correlation simply because it captures the monsoon-to-dry-season swing, while tracking day-to-day variation within the monsoon poorly.
- Direction changes by season. Wind direction shifts between the monsoon and the rest of the year, so a single directional relationship between site and reference may not hold year-round.
- Data gaps cluster. Lightning, cyclones, and power outages tend to hit during particular seasons, so missing data is rarely random.
Choosing a reference: ERA5, MERRA-2, or local data
ERA5 and MERRA-2 are global reanalysis datasets that provide decades of consistent hourly data and are the usual starting points. Neither is right by default. The choice should be based on how well each reproduces the onsite record at the time scales that matter: hourly and daily correlation, seasonal bias, and the diurnal pattern.
Local sources such as NIWE data, nearby operational wind farm SCADA, or IMD stations can add value as cross-checks, but they need their own consistency checks because instrument changes and site changes over the years can introduce false trends.
Practical checks that make the result defensible
- Compare correlation and residuals for monsoon and non-monsoon months separately, not only annually.
- Test more than one MCP method, such as linear regression, variance ratio, and matrix methods, and explain why the selected one fits the site.
- Check the reference for long-term trends or step changes before treating its long-term mean as stable.
- Use concurrent periods that cover complete seasons, so no month is over- or under-represented.
- Report how sensitive the long-term mean is to the choice of reference period.
How this shows up in P90
Long-term correction uncertainty and inter-annual variability are usually among the largest items in an Indian wind uncertainty budget. A reference that fits the monsoon well, and a campaign that covers more than one monsoon, are the two most effective ways to narrow the gap between P50 and P90. That directly improves the amount of debt the project can support.
When a lender's technical advisor reviews the assessment, the questions they ask most often are about reference selection and the seasonal fit. Documenting those choices clearly is usually quicker than defending them afterwards.