ERA5 vs MERRA-2 for MCP in the Northeast of Brazil: comparative analysis

In the MCP (Measure-Correlate-Predict) methodology, the statistical relationship between the measurement campaign and a long-term reference series is what allows extrapolating a short period of data to the 20–25 year time horizon relevant to the project. The choice between ERA5 and MERRA-2 as this reference is not indifferent — and in the Brazilian Northeast, in particular, produces systematic differences that directly affect the estimated P50 and the uncertainty associated with P90.

This note compares the two reanalysis in terms of resolution, hourly correlation and bias by speed range in wind campaigns in the Northeast, and quantifies the impact of this choice on the final outcome of AEP.

What is at stake in the choice of reference

The quality of a MCP analysis depends on three factors in sequence: the correlation during the concurrent period, the representativeness of the long-term series and the adequacy of the method to the local wind regime. The reference reanalysis influences the three simultaneously.

A high r2 in the concurrent period does not guarantee a correct P50 — if the reanalysis systematically underestimates the wind in years of high wind activity and overestimate in years of low, extrapolation can be undetectable by the isolated correlation. The analysis of bias by speed range and by month is therefore as important as the global r2.

ERA5 and MERRA-2: comparative characteristics

ERA5, produced by the European ECMWF, and MERRA-2, produced by NASA, are the two global reanalyses most used in wind resource assessment. Their structural differences have direct implications for sites in the Northeast:

The horizontal resolution is the most striking factor for sites in complex terrain. ERA5, with ~31 km, better captures the coastal acceleration and the orographic effects of plateaus and ridges. In the Northeast, where many sites lie in plateaus raised a few kilometers from the coast, this difference of ~20 km between the two reanalysis is not negligible.

The context of the Brazilian Northeast

The wind regime in the Northeast is dominated by NE allys — steady, strongly seasonal and high-energy. The dry period (August–January) concentrates the most intense winds, with frequency of squid close to zero at altitude and high predictability of the vertical profile. The rainy period (February–July) brings local atmospheric instability, lower-intensity breezes and greater hourly variability — an analytically more demanding environment for any reanalysis.

Two specific phenomena of this region require attention in the selection of the reference MCP:

Low-level coastal jet (JCBN): The northeastern coast regularly experiences a jet with a maximum altitude of 200–600 m, more intense during the southern winter. Reanalyses with a resolution of 30–50 km do not capture the fine structure of this jet, resulting in underestimation of the peaks of velocity at shoreline sites — especially above 100 m of cube height.

Orographic effects of plateaus: Parks installed on plateaus in Ceará, Rio Grande do Norte and Bahia experience flow acceleration dependent on terrain geometry at a scale of 1–5 km, below the resolution of any reanalysis. Mesoscale models are necessary to capture this effect; the reanalysis serves only as a boundary condition for mesoscale downscaling.

Competition period: duration and quality

IEC 61400-12-1 does not prescribe minimum duration for the concurrent period MCP, but the industry's consensus points to well-established criteria:

• Minimum 12 months To cover the complete seasonality of the aliases
• 24–36 months To reduce the effect of meteorologically atypical years
• - r2 ≥ 0.80 In the competing period as acceptable quality threshold
• - r2 ≥ 0.85 As a threshold for bankability in due diligence reports

– Competitive periods below 12 months, or with monthly r2 consistently below 0.80, substantially increase the uncertainty associated with P90. In projects where P90 determines debt service coverage (DSCR), this uncertainty has a direct impact on the financing structure.

Correlation and bias by speed range

The bias analysis should not be limited to the global average error — the bias pattern per speed range is what reveals the nature of the problem. A reanalysis that consistently underestimates over 10 m/s affects exactly the upper tail of the distribution, which concentrates most of the energy.

The asymmetry of the bias of MERRA-2 has an important methodological consequence: linear correction of bias (mean adjustment and standard deviation) does not eliminate the problem, because the bias is not constant. The application of linear MCP with MERRA-2 without stratification by speed band tends to amplify the bias in the upper tail, exactly where the energy is.

Seasonal variation of the correlation

The temporal correlation between measurement and reanalysis is not stationary — it varies with the meteorological regime. In the Northeast, the transition between dry period (stationary alisis) and rainy period (convective instability) produces correlation degradation that is more pronounced in MERRA-2:

The seasonal degradation of MERRA-2 has consequences that go beyond the correlation: months with low r2 produce less reliable wind forecasts for the rainy period, which in the Northeast contributes 20–35% of the annual energy. If the MCP method does not stratify the seasonal correlation per season, this seasonal bias may result in AEP error of 3–5%.

Impact on estimated P50

The difference in bias between ERA5 and MERRA-2, amplified by the long-term extrapolation MCP, produces systematic differences in the final P50. The graph below shows the percentage variation in the P50 estimate of AEP for four reference scenarios, taking ERA5 with 3 years of concurrent period as the baseline:

The magnitude of these differences — 2 to 4% between scenarios with the same measurement campaign — is often underestimated in reports that document only the global r2 and omit the sensitivity to the reference analysis. In the due diligence process of international lenders, the absence of this analysis is one of the main sources of technical questioning.

Selection criteria

Renewene shall adopt the following decision hierarchy for sites in the Northeast:

1. ERA5 as the primary standard — greater resolution, coverage since 1940, greater acceptance in international reports (European lenders and multilateral banks).
2. MERRA-2 as a secondary validation reference — not as principal, but to quantify the uncertainty of reanalysis and identify divergences that require explanation.
3. – Weighted set ERA5 + MERRA-2 — when the difference between the estimates exceeds 2.0% after correction of biases per speed band; the weight is calculated proportional to the annual average r2.
4. Downscaling mesoscale mandatory — when the site is above 600 m high, less than 15 km from the coast with a pronounced topographic gradient, or in a steep valley where no reanalysis has sufficient resolution.

In any scenario, the report should explicitly document the reanalysis used, the full concurrent period with r2 monthly, the bias per speed band, the correction method applied and the uncertainty attributed to the choice of the reference as a separate component of the total uncertainty of AEP.