Aligned configuration. The solver settings that can be matched across the three packages are matched: the wake deficit is scaled by the freestream speed, rotor averaging is linear, and FLORIS and PyWake sample the rotor with a nine-point cubature. The remaining differences are listed below.

Solver settings in this view

Setting	FLORIS	FOXES	PyWake
Deficit reference wind speed	Freestream (sosfs)	Freestream (ws_quadratic_amb_target)	Freestream (use_effective_ws=False)
Rotor grid	TurbineCubatureGrid, 9 points	grid16, 4 x 4 cell centers	CGIRotorAvg, 9 points
Power / thrust averaging	power linear-cubature, thrust linear-cubature	linear	linear

Differences that remain in this view:

FOXES: FOXES keeps its grid16 rotor model; the requested cubature with 9 points rotor grid is not applied.

← All comparisons

jimenez — four turbines

velocitydeflectionSoftware: FLORIS, FOXES, PyWake · Rotor diameter 240 m · Wind speed 9.8 m/s

View the one turbine scenario for this wake model →

Model parameters and implementation notes

The windIO inputs and their software-specific mappings are listed below. Matching literature-model names do not imply identical solver, rotor-averaging, superposition, turbulence, induction, or yaw treatments.

windIO case inputs

velocity: jensenalpha = 0.1
deflection: jimenezbeta = 0.1

FLORIS

velocity: jensenwe = 0.1
deflection: jimenezkd = 0.05

Uses sum-of-squares freestream deficit superposition (sosfs), the TurbineCubatureGrid, 9 points rotor grid with power linear-cubature, thrust linear-cubature averaging, and Crespo-Hernandez wake-added turbulence. Secondary steering, yaw-added recovery, transverse velocities, and active wake mixing are disabled.

Converts the windIO coefficient with kd = beta / 2 because FLORIS uses 1 + 2*kd*x/D where the other implementations use 1 + beta*x/D.

FOXES

velocity: JensenWakek = 0.1, induction = Betz
deflection: JimenezDeflectionbeta = 0.1, rotate = false

Uses Betz induction, quadratic wind-speed superposition with a freestream (ws_quadratic_amb_target) deficit reference, grid16, 4 x 4 cell centers rotor sampling, rotor-point partial wakes, and the rotor_wd wake frame. No separate wake-added turbulence model is configured.

FOXES keeps its grid16 rotor model; the requested cubature with 9 points rotor grid is not applied.

Uses JimenezDeflection with rotate=False, translating the wake path without rotating the wake cross-section.

PyWake

velocity: NOJDeficitk = 0.1
deflection: JimenezWakeDeflectionbeta = 0.1

Uses PropagateDownwind with squared-sum superposition, CGIRotorAvg, 9 points rotor averaging, freestream (use_effective_ws=false) deficit reference, and ct2a_mom1d induction. No separate wake-added turbulence model is configured.

Uses the lower-level NOJDeficit class rather than PyWake's Jensen_1983 literature preset.

Caveat: This is a coupled comparison: yaw-dependent thrust, velocity deficit, deflection, rotor averaging, and wake superposition can all contribute to differences. It does not isolate the deflection equation.

Layout & sample locations

Turbine positions (bold, rotated by yaw angle where applicable) and the cross-section sample planes (dashed) used by the plots below.

Streamwise velocity

Cross-sections

Horizontal contour

X-Y plane at hub height.

Vertical contour

X-Z plane at Y=0 (lateral centerline).

Cross-section contour

Y-Z plane at X=20D.

Effect of aligning the configuration

Streamwise RMSE for each software pair in both configurations of this case. Thedefault andaligned views differ only in the solver settings listed at the top of the page.

Software pair	Default RMSE (m/s)	Aligned RMSE (m/s)
FLORIS vs FOXES	0.2553	0.0517
FLORIS vs PyWake	0.4273	0.4155
FOXES vs PyWake	0.3588	0.4205

Comparison metrics

Pairwise error between software for each plotted quantity. Click a column to sort.

Software pair	Target	RMSE (m/s)	Max abs error (m/s)	Mean abs error (m/s)	NRMSE

How to interpret this comparison

Along the hub-height streamwise line the closest pair is FLORIS vs FOXES (0.052 m/s RMSE, 1.0% NRMSE) and the furthest is FOXES vs PyWake, at 0.42 m/s RMSE (8.0% NRMSE). For scale, the free-stream wind speed is 9.8 m/s.
For FOXES vs PyWake, cross-section RMSE is 0.17 m/s at 4D past first turbine, 0.017 m/s at 1D past last turbine, 0.020 m/s at 5D past last turbine, 0.006 m/s at 10D past last turbine. NRMSE is normalized by the range of the first-listed software's own values in that plot, so it can rise with distance even as RMSE falls, because the wake profile becomes shallower.
The horizontal contour shows how a small RMSE can hide a large maximum error: FLORIS vs PyWake has an RMSE of 0.18 m/s but a maximum of 5.20 m/s. Those maxima come from a small fraction of grid points, found at the rotor plane (X = 0) or at the sharp edge of a wake, where the packages disagree about whether a point is inside the wake. Use RMSE for overall agreement and the maximum to locate where to look.
Compared with the same scenario without yaw (Jensen), the largest streamwise RMSE between any pair goes from 0.41 to 0.42 m/s. A similar value means the separation comes mostly from the underlying velocity model; a clearly larger value points to the deflection and yaw-thrust treatment.
The Jensen top-hat deficit makes the contour maximum errors sensitive to which side of the wake edge a grid point falls on, so judge overall agreement by RMSE.
FLORIS maps the Jiménez coefficient as kd = beta / 2 (see the implementation notes above). Beyond that mapping, each package applies yaw to the thrust coefficient differently, which scales the deflection and is a leading candidate for the remaining separation.
At 4D past the first (yawed) turbine FLORIS and FOXES agree (0.037 m/s RMSE) while PyWake differs from both (0.16 and 0.17 m/s). Downstream of the last turbine all three agree to within 0.020 m/s in every cross-section, so with the configuration aligned the remaining separation is concentrated at the first yawed wake.
A value of 0.0000 means the two packages agree at the precision shown in the table.
These metrics measure how consistently the packages implement the same inputs. They are not errors against measurements, so close agreement does not show that a wake model is accurate, and differences should be read together with the implementation notes above.