Predicting terrain-induced wind turbulence for smokejumper parachute operations

Perdigão field observations

We evaluate modelled mean winds and TKE against data collected from the Perdigão field campaign, which took place in Portugal during 2017 (Mann et al. 2017; Fernando et al. 2019; WindsP Web Application 2020). The field site is characterised by two parallel ridges oriented northwest to southeast, forming a ridge–valley–ridge system (Fig. 1). The ridges are approximately 4 km long with slopes of ~35%. The ridgetops rise 200–250 m above the valley floor, with a ridge-to-ridge distance of 1.4 km. The surrounding terrain to the southwest and northeast is relatively flat compared with the ridge–valley–ridge system. The site is covered by patches of pine and eucalyptus trees with grass and shrub vegetation in between. Vasiljević et al. (2017) provide an in-depth description of the terrain and vegetation including photos of the site. There is a 2 MW wind turbine with a hub height of 78 m and rotor diameter of 82 m located on the southwest ridge near tower tse04 (Fig. 1). We describe possible effects of the turbine on the observation network and implications for interpretation of results in the Results and Discussion sections. Interestingly, Tomaszewski et al. (2018) investigated how turbulence induced by a wind turbine could pose a hazard to small aircraft. Although not the focus of this study, it is worth noting that wind turbines and other large bluff bodies also have the potential to produce turbulence hazards to aviation resources.

Fig. 1.

Perdigão field site and tower locations used in this study. Tower rsw06 is the calibration tower. Solid black lines indicate Transect A and Transect B, which approximately intersect the towers. The mean wind direction is from the southwest and aligned with the transects. The inset depicts the broader area; the field site is indicated with a red star.

We focus on a meteorological period with steady southwest winds approximately perpendicular to the ridges. Wind speed, wind direction and TKE were measured along two transects in the streamwise direction spanning the ridge–valley–ridge system (Fig. 1). We use sonic anemometer data from three 100-m towers (tse04, tse09 and tse13) with measurements 10, 20, 30, 40, 60, 80 and 100 m above the ground and six 60-m towers (rsw06, tnw07, tnw10, rsw03, tse06 and tse11) with measurements at 10, 20, 30, 40 and 60 m. Tower tnw07 had additional measurements at 4, 6, 8 and 12 m. The quality-controlled, tilt-corrected 5-min and high-rate observed data were obtained from the NCAR Earth Observing Laboratory (Oncley 2021). The raw data were sampled at 20 Hz (Oncley 2019). We investigate the time period 22:00 to 22:30 UTC on 4 May 2017 as the evaluation case. Palma et al. (2020) calculated a bulk Richardson number of −0.03 for this time period, which indicates near-neutral atmospheric conditions. These conditions were prone to flow separation over the southwest ridge and formation of a recirculation zone between the two ridges (Menke et al. 2019). We choose this case to evaluate model performance under a double-ridge system exposed to a cross-flow under neutral atmospheric stability conditions and expect that findings will extend to other similar cases.

Model description and configuration

We use the WindNinja conservation of mass and momentum model described in Wagenbrenner et al. (2019) for this work. This model estimates a solution to the steady-state, incompressible Reynolds-averaged Navier-Stokes (RANS) equations.

NASA Shuttle Radar Tomography Mission (SRTM) (Earth Resources Observation and Science Center (EROS) 2017) data at 30-m resolution are used for the digital elevation model (DEM). The computational domain extends ~3 km beyond the study area in all directions and is constructed using the ‘fine’ mesh choice in WindNinja. This choice allocates 100,000 cells for the mesh using the meshing approach described in Wagenbrenner et al. (2019), which affords higher mesh resolution in the layers near the surface and coarser mesh resolution higher in the domain (Fig. 2). The near-ground horizontal and vertical cell resolution is 24 m. The domain height is 1597 m above ground level (AGL). The ‘brush’ option is used to set uniform vegetation throughout the domain corresponding to a roughness length of 0.43 m, zero-plane displacement height of 1.8 m and vegetation height of 2.3 m.

Fig. 2.

Eicks Fire terrain and jump operation points of reference. Jump Spot indicates the intended landing site. Landing Site indicates the actual landing site where the accident occurred. Transects A, B and C indicate transects sampled for wind and turbulence analysis from simulations. The mean wind direction is from the southwest, aligned with the transects. The inset depicts the broader area; the site of the jump operation is indicated with a red star.

The domain-average initialisation method in WindNinja is used. This sets the initial flow field and the boundary conditions (Wagenbrenner et al. 2019) using a logarithmic wind profile. This profile is defined by an input 10-m wind speed, wind direction and roughness parameters. The 10-m wind speed and direction are set to 2.9 m s⁻¹ and 229° to match the calibration tower rsw06 (Fig. 1).

A sensitivity analysis was performed to investigate model sensitivity to the chosen ground roughness, mesh resolution, turbulence model and numerical discretisation schemes by independently varying each of these parameters. Results of the sensitivity analysis are presented in Appendix 1.

Model evaluation methods

We compare modelled vs observed wind speed, wind direction and TKE at the sensor locations shown in Fig. 1. Model results are discussed qualitatively, including a description of the presence and structure of key flow features throughout the domain. Model performance is also quantified in terms of the root mean square error (RMSE), mean bias error (MBE) and mean absolute percentage error (MAPE):

where φ is the observed value, φ′ is the difference between the predicted and observed and N is the number of observations. A circular mean is used for wind direction calculations.

The RANS simulation includes a transport equation for TKE and outputs TKE directly as detailed in Wagenbrenner et al. (2019). TKE is the mean kinetic energy per unit mass contained in the eddies of a turbulent flow. The measured TKE, k, is calculated from the observed 20 Hz sonic anemometer data as one half of the sum of the variances, σ², of the velocity fluctuations:

where u′, v′, w′ are the instantaneous fluctuations about the mean and the overbar indicates the mean (e.g. u′ = u − u̅). The modelled and observed turbulence is also investigated in terms of the root-mean-square velocity fluctuations and the turbulence energy spectral density. The root-mean-square velocity fluctuations, U_rms are calculated as:

We anticipate that eddies of different sizes, time scales and strengths will impact parachute flight control differently. As a first investigation into this, we examine the turbulence energy spectral density of the sonic anemometer data computed using a Fast Fourier Transform. These transformed data are plotted as the spectral density vs the associated time scale on a log-log plot to depict the energy spectrum and highlight key features of the turbulent flow. For example, the peak is associated with production of turbulence and the largest eddy sizes in the flow. We compare the energy spectral density signatures at different locations in the terrain.

Eicks fire case study

We present modelled winds and turbulence for an operational jump conducted by smokejumpers in southwest New Mexico on the afternoon of 24 May 2021 as a case study to demonstrate the potential utility of high-resolution turbulence predictions for smokejumping operations. The jump operation took place in the lee of a tall ridge exposed to a strong southwesterly wind, approximating near-neutral atmospheric conditions. The jump ended with a hard landing that resulted in a smokejumper fatality. The details of the case are published in the Eicks Fire Smokejumper Fatality Learning Review Technical Report (Wildland Fire Lessons Learned Center - Fire Smokejumper Fatality: Learning Review – Technical Report; https://lessons.fs2c.usda.gov/incident/eicks-fire-smokejumping-accident-fatality-2021) and indicate that turbulence generated from upwind terrain features likely played a role in the hard landing.

The Eicks Fire was located in the Animas Mountains of southwest New Mexico (Fig. 2). This is a high-elevation desert with rugged terrain and peaks rising over 1000 m above the valley floor. The Eicks Fire was ~20 ha in size at the time of the jump. The designated jump spot was south of the fire on a grassy slope with a west aspect (Fig. 2). Streamers released by the smokejumpers indicated west winds of ~6 m s⁻¹ in the jump spot and the High-Resolution Rapid Refresh (HRRR) forecast modelled westerly 10-m winds of 6.2 m s⁻¹ from 261° at the time of the jump.

A WindNinja simulation was initialised with a domain-average wind of 6.2 m s⁻¹ from 261° at 10 m AGL to assess local winds and turbulence during jump operations.

Perdigão mean wind and TKE

The modelled time-averaged wind field contains several notable flow features, including speed-up over the first ridge and flow separation over portions of both ridges with recirculation zones downwind (Figs 3 and 4). Flow speed-up occurs over sections of the second ridge that are not impacted by flow separation from the first ridge. For example, speed-up is visible over the south end of the second ridge (Fig. 3). This is likely due to the broken terrain that comprises the southern end of the first ridge, which allows flow to channel through small gaps in the terrain rather than being forced up and over the ridge. Instead of separating over the first ridge, the flow maintains its momentum as it is funnelled through the gaps in its approach to the second ridge. There is also flow channelling around the northern end of the ridges and through terrain gaps in the southern end of the ridges, visible particularly in the lower 40 m AGL plot (Fig. 3a). Flow blocking upwind of the first ridge is visible in both the 40 m and the 100 m AGL slices (Fig. 3).

Fig. 3.

Top view of the modelled flow field along terrain-following slices at (a) 40 m AGL, and (b) 100 m AGL.

Fig. 4.

Side view of the modelled flow field through (a) Transect A, and (b) Transect B. Tower labels indicate maximum sensor height for that tower.

Transects A and B pass through the central portion of the ridge–valley-ridge system characterised by nearly parallel ridges of similar height without gaps in the terrain (Fig. 1). The transects pass through an area of flow separation induced by the first ridge (Figs 3 and 4). The horizontal extent of the wake region that forms behind the first ridge is approximately two times the height of the ridge and extends above and downwind of the second ridge (Fig. 4). These modelled flow separation characteristics qualitatively match the lidar observations reported in Menke et al. (2019). There is flow recirculation between the ridges along both transects (Fig. 4). Two recirculation regions can be seen along both transects, one centred immediately downwind of the upwind ridge and the other centred closer to the second ridge and slightly lower in elevation (Fig. 4). The height of the modelled recirculation region is approximately equal to the height of the first ridge and the maximum strength of the modelled reverse flow is approximately 1 m s⁻¹ (Fig. 5). This closely matches the observed recirculation patterns reported in Menke et al. (2019) from scanning lidars during near-neutral stability southwesterly flow conditions (Fig. 6).

Fig. 5.

Modelled flow aligned with the input wind direction (229°) along (a) Transect A, and (b) Transect B. Negative values indicate reversed flow (u, wind speed).

Fig. 6.

Average line of sight (LOS) velocities observed by WindScanner lidars during southwesterly flow conditions in the Perdigão field experiment. Positive LOS velocities indicate southwesterly flow. Transect 1 and 2 correspond to Transect A and B, respectively, in this paper. Reproduced from Menke et al. (2019).

The modelled TKE field at 40 m and 100 m AGL is shown in Fig. 7. There is higher TKE at 100 m AGL compared with 40 m AGL (Fig. 7). The sensors exposed to the highest TKE in these 40 m and 100 m AGL slices are located on the second ridgetop (tnw10, tse13 and tse11; Fig. 7). This is a region of strong horizontal and vertical shear (Figs 3 and 4). The wake region from the first ridge tapers down over the second ridge and the areas with strongest TKE are located near the edges of the wake over the second ridge where horizontal shear is high (Fig. 3). There is strong vertical shear because the flow is passing over the second ridge (Fig. 4). Enhanced TKE can also be seen between the two ridges along the streamlines immediately east of tse06 (Fig. 7), which is near the edge of the wake region generated by the first ridge and a region of strong horizontal shear (Fig. 3b).

Fig. 7.

Top view of the modelled TKE field along terrain-following slices at (a) 40 m AGL, and (b) 100 m AGL.

The vertical slices (Fig. 8) show a region of enhanced TKE between the two ridges just above the recirculation zone that extends vertically and downwind just over the second ridge, following the mean flow. A similar pattern in TKE is shown in both transects, although Transect A has slightly higher TKE values.

Fig. 8.

Side view of the modelled TKE field through (a) Transect A, and (b) Transect B. Tower labels indicate maximum sensor height for that tower.

The modelled mean wind speed profile approximates the observed profile at the calibration tower, rsw06, with the modelled mean speed falling within the range of the reported 5-min mean speeds for four out of the five sensor heights (Fig. 9). The modelled speeds fall within the range of the 5-min observations for the other tower locations on the first ridge except for the lowest three heights on tower tse04 and the lowest height on tower rsw03 (Fig. 9). The model slightly underpredicts at the lowest three heights at tse04 and slightly overpredicts at the lowest height at rsw03 (Fig. 9). The modelled speed profiles slightly underpredict in the valley, but still fall within the range of the 5-min averages observed at most sensors on the valley towers (Fig. 9; tse09, tse06, tse11, tnw07). The modelled speed profile is underpredicted at the second ridge towers (Fig. 9). This could indicate that the modelled wake region from the first ridge is propagating too far downwind or perhaps is too wide as it passes over the second ridge. The wake region narrows as it approaches the second ridge, with towers tnw10 and tse13 just inside the wake region (Fig. 3). If the modelled wake region were slightly narrower over the second ridge, the wind speeds would match the observed profiles more closely.

Fig. 9.

Modelled vs observed mean wind speed at each tower. Grey circles indicate 5-min mean observations. Black triangles indicate 30-min mean observations. The top row of panels are the southwest ridge towers; the middle row of panels are the valley towers; the bottom row of panels are the northeast ridge towers.

The modelled wind direction profile matches the observed wind direction profile at the calibration tower, rsw06 (Fig. 10). The modelled direction profiles closely match the observed profiles at the other two tower locations on the first ridge (Fig. 10). The wind direction observed at the first ridge is ~230°, with little vertical directional shear (Fig. 10). The observed 5-min wind directions in the valley exhibit substantial variability and the 30-min averages indicate vertical directional shear, consistent with the presence of a recirculation region (Fig. 10). The modelled direction profiles match the observed valley profiles well with all of the modelled profiles falling within the range of observed 5-min mean observations (Fig. 10). The observed direction profiles have much less variability and directional shear at towers tse13 and tnw10 on the second ridge, which suggests that these towers are outside of the turbulent recirculation region (Fig. 10). The modelled direction profiles match the observed profiles well, with all modelled directions falling within the range of the observed 5-min average directions (Fig. 10).

Fig. 10.

Modelled vs observed mean wind direction at each tower. Grey circles indicate 5-min mean observations. Black triangles indicate 30-min mean observations. The top row of panels are the southwest ridge towers; the middle row of panels are the valley towers; the bottom row of panels are the northeast ridge towers.

The modelled TKE is within the range of the observed 5-min mean TKE at the lower two heights of the calibration tower, rsw06 (Fig. 11). The simulation slightly overpredicts TKE at the upper three heights at rsw06 (Fig. 11). Slight overpredictions in TKE at the upper tower heights can also be seen for the towers tse04 and rsw03 on the first ridge (Fig. 11). The observed mean and variability in TKE was higher at the valley towers (Fig. 11). The modelled TKE matches the observed profiles well, with the modelled TKE falling within the range of the reported 5-min means, except for at tower tse11, where TKE is slightly underpredicted at all heights (Fig. 11). The shape of the modelled mean TKE profile closely matches the observed profiles, with TKE increasing with height (Fig. 11). The highest observed TKE values were measured at the second ridge (Fig. 11). There is also substantial variability in the observed TKE on the second ridge at tower tnw10 (Fig. 11). The simulation underpredicts TKE on the second ridge, especially at tnw10; however, the shape of the modelled TKE profiles matches the observed profiles with increasing TKE up to ~30 m and then steady or slightly decreasing TKE from 30 to 100 m (Fig. 11).

Fig. 11.

Modelled vs observed mean TKE at each tower. Grey circles indicate 5-min mean observations. Black triangles indicate 30-min mean observations. The top row of panels are the southwest ridge towers; the middle row of panels are the valley towers; the bottom row of panels are the northeast ridge towers.

As mentioned in the Methods, there was a wind turbine located on the southwest ridge near tse04. Although the turbine undoubtedly impacted near-turbine flow and turbulence downwind, we do not think it is likely that the turbine substantially affected observations at the sonic anemometers or lidar scan used in this work. We believe that the mean flow and turbulence signatures described above are induced by the terrain and only minimally impacted by the turbine. This is in line with findings reported in Wildmann et al. (2019) that showed that the turbine effect on the flow only propagates downwind ~200–400 m. Fernando et al. (2019) reported measurements with and without the turbine operating and found that the turbine only affected flow up to ~300 m downwind. Additionally, modelling by Wenz et al. (2022) found that the turbine had little effect on the flow beyond 500 m downwind or below the ridge top. They reported a well-defined wake up to 340 m downwind of the turbine that rapidly decayed beyond this distance. The turbine wake simulated by Wenz et al. (2022) under neutral atmospheric stability did not follow the terrain, but drifted upward, which compares well with the findings reported in Wildmann et al. (2019).

Characterising turbulence at Perdigão

We anticipate that different temporal and spatial scales of turbulence will impact parachute flight control differently. For example, a wind gust of very short duration, say less than 1 s, is not likely to cause loss of control as the short impulse would not change the momentum of the parachuter significantly. Similarly, a very long timescale fluctuation, say of 20-min duration, is also not likely to cause flight control difficulty as the wind changes are slow enough that the operator could adjust flight to compensate. We anticipate that turbulence of the order of seconds to minutes may be the most important timescales for parachute flight. Our analysis in previous sections focused on TKE because it is directly computed during our RANS simulation process, which uses a k-epsilon type turbulence closure, but this is a simple first-order description of turbulence. Here, we investigate more detailed turbulence characteristics and examine other potentially useful turbulence metrics for smokejumper operations.

Turbulence energy spectra

We computed the turbulence energy spectral density at two tower locations representing a location of relatively high TKE and low TKE to compare the scales of turbulence. For this purpose, 40-m data from tower rsw06 (top of the first ridge) and tower tnw10 (top of second ridge in wake from first ridge) were analysed (Fig. 12). There is a slight increase in the peak energy for tnw10 at a time scale of ~600 s compared with rsw06. The energy cascade in the inertial subrange from ~100 to 0.1 s follows the expected −5/3 power law at both locations. Eddies with time scales of ~5–10 s are anticipated to be of importance for parachute functionality, given that smokejumper flying time is ~30–60 s. As the energy spectra continuously span the temporal range relevant to smokejumpers at a known rate and scales with the peak energy, a single metric can sufficiently describe the turbulence under these atmospheric conditions.

Fig. 12.

Turbulence energy spectra computed from sonic anemometer data 40 m AGL for rsw06 and tnw10 in the Perdigão domain. Red lines indicate power law regressions in the inertial subrange for each dataset.

Mean velocity fluctuation

The mean velocity fluctuation can be calculated directly from the TKE, assuming isotropic turbulence (Eqn 5). This is he mean velocity fluctuation that can be expected about the mean wind speed over time. This could be a good metric to characterise turbulence for smokejumper operations because (1) it is intuitive as it is reported in speed units and has an interpretable meaning, and (2) it indicates an absolute force affecting a parachute. Eddy dissipation rate (EDR) is a commonly used turbulence metric in the aviation industry because it is proportional to the root-mean-square vertical acceleration experienced by an aircraft (Sharman et al. 2014); however, velocity fluctuations are also frequently used to assess turbulence hazards for simple quantification and communication. Guidelines issued by the National Weather Service delineate velocity fluctuations of 1.5–5.9, 6.0–10 and 11–15 m s⁻¹ into categories of light, moderate and severe turbulence for aircraft, respectively (National Weather Service 2024). The effect of each turbulence classification on a specific aircraft will depend on the size of the aircraft, with small aircraft being more susceptible to smaller velocity fluctuations than larger aircraft. One question is what magnitude of velocity fluctuation is sufficient to impact parachute functionality. There is not much guiding information in the literature on this topic. As the size and weight of a parachuter is much smaller than that of even a small aircraft, it is suspected that light turbulence would have a substantially larger impact on parachutes than on small aircraft. Thus, even velocity fluctuations of 1.5 m s⁻¹ could be potentially hazardous to smokejumpers, particularly if encountered near the ground. This will likely need to be assessed, along with other turbulence metrics, including EDR, via communication with smokejumpers regarding their in-flight experiences coupled with quantification of these turbulence metrics in the jump areas.

As turbulence levels can vary drastically with elevation, one efficient way to characterise turbulence for smokejumper operations may be to quantify the maximum value of a given turbulence metric within each column of air above the terrain spanning the jump profile. This will give smokejumpers a quick estimate of the spatial distribution of the most turbulent air in their operational area. The jumpers exit the aircraft at ~915 m (3000 feet) AGL and their parachutes are deployed by 460 m (1500 feet) AGL. The region between 460 m AGL and the ground is of most importance to the smokejumpers because this is the region they are flying their parachutes in. Arguably, regions closer to the ground are more critical than regions farther away from the ground, because the closer to the ground the smokejumper is, the less room for error they have in their manoeuvres. Fig. 13 shows the maximum mean velocity fluctuation within 460 m of the ground. The largest velocity fluctuations in the study area of the Perdigão domain are found on the upwind side of the second ridge, with mean velocity fluctuations of up to 1.1 m s⁻¹ (Fig. 13a).

Fig. 13.

Maximum mean velocity fluctuation below 914 m (1500 ft) AGL for the (a) Perdigão, and (b) Eicks Fire domains. Note the different colour scales in (a, b).

Eicks fire case

A WindNinja simulation was conducted to re-create the wind and turbulence conditions during the Eicks Fire accident. Fig. 14 shows terrain-following slices of the modelled flow field at 40 m and 100 m AGL. The modelled time-averaged wind field shows flow separation over the tall ridge upwind of the landing site, indicated by the recirculation zone in the top and side views (Figs 14 and 15).

Fig. 14.

Top view of the modelled flow field along terrain-following slices at (a) 40 m AGL, and (b) 100 m AGL for the Eicks Fire domain. Jump Spot indicates the intended jump spot.

Fig. 15.

Side view of the modelled flow field through (a) Transect A, (b) Transect B, and (c) Transect C for the Eicks Fire domain. Jump Spot indicates the intended jump spot.

Vertical profiles of the flow were extracted along three transects aligned with the mean wind direction. Transect A runs just south of the landing site, within the northern edge of the recirculation zone (Figs 14 and 15). Transect B runs through the landing site and Transect C runs through the intended jump spot (Figs 14 and 15). Transect A runs through a tall, sharp portion of the upwind ridge and flow recirculation can be seen behind the ridge, spanning a vertical distance from the valley floor behind the ridge to the top of the ridge, ~190 m (Fig. 15a). The landing site is located behind a smoother part of the ridge with a shallower distance to the lee-side valley bottom, on the north edge of the recirculation zone. Again, the recirculation zone spans the vertical distance from the lee-side valley floor to the ridgetop, ~75 m (Fig. 15b). The intended jump spot is located further up-drainage in a shallower region of the lee-side valley and just outside of the recirculation zone (Fig. 15c).

Top and side views of the modelled TKE fields are shown in Figs 16 and 17. There is higher TKE present throughout the 100 m height surface compared with the 40 m height surface (Fig. 16). The landing site is located in an area with relatively high TKE (TKE > 3 m² s⁻² at 100 m AGL; Fig. 14b). The intended jump spot is located in an area with lower TKE than the landing site (Fig. 16). The vertical slices show a peak region in modelled TKE immediately downwind of the tall ridge, starting at approximately the height of the tall ridge and spanning a vertical distance of ~190 m above this height (Fig. 17).

Fig. 16.

Top view of the modelled TKE field along terrain-following slices at (a) 40 m AGL, and (b) 100 m AGL for the Eicks Fire domain. Jump Spot indicates the intended jump spot.

Fig. 17.

Side view of the modelled TKE field through (a) Transect A, (b) Transect B, and (c) Transect C for the Eicks Fire domain. Jump Spot indicates the intended jump spot.

A finger of relatively high TKE can be seen in the 100 m AGL slice (Fig. 16b) extending from immediately downwind of the tall ridge, through the landing site and to the east–northeast, approximately along Transect B. This finger delineates the boundary of the recirculation zone (Fig. 16b compared with Fig. 14b). The strong gradients in the flow along this boundary are the reason for the enhanced TKE in this region and this is something smokejumpers should look out for. Enhanced turbulence can be expected along the boundaries of lee-side recirculation zones. In this case, there is a sufficiently strong gradient between the landing site and the jump spot, such that turbulence conditions are substantially different between these two locations although the distance between them is small. In the absence of turbulence forecasts, smokejumpers could watch out for strong spatial gradients in mean wind forecasts or observations, as this likely translates to increased turbulence.

Another thing to note is how far downwind the wake and enhanced TKE persist in the domain. The wake from the tall ridge and the enhanced TKE associated with this wake extend horizontally from immediately downwind of the tall ridge to beyond the eastern edge of the domain (Figs 14 and 15), 2500 m or more than six times the ridge height. Smokejumpers should be aware that terrain effects can persist far downwind, and the next ridge or two downwind of a taller exposed ridge may be in a region of very high turbulence.

The maximum velocity fluctuation was ~1.6 and 1.2 m s⁻¹ in the column of air above the landing site and the intended jump spot, respectively (Fig. 13b). The velocity fluctuations at the landing site fall within the light turbulence rating for aircraft (National Weather Service 2024). As smokejumpers are substantially lighter than even a small aircraft, this light rating could translate to a substantial impact for their flight experience. Both TKE at 40 m and 100 m AGL and mean velocity fluctuation turbulence metrics indicate higher turbulence levesl at the landing site compared with the intended jump spot.

Implications for smokejumper jump spot assessments

WindNinja’s skill in simulating the distribution of turbulence under near-neutral atmospheric conditions at the Perdigão site suggests that it could be of use for smokejumper jump spot assessments under moderate to high wind conditions. As WindNinja is formulated for neutral atmospheric stability, model accuracy is expected to deteriorate under other stability regimes. Analyses of the energy spectral density reveal that a single metric is sufficient to characterise turbulence under these atmospheric conditions. TKE and the mean velocity fluctuation are the metrics presented in this work and are shown to correlate well with observations. It is unknown how either of these metrics correlate with parachute functionality or the ability of the smokejumper to fly. Future work is needed to assess how smokejumper in-flight experiences correlate with the various computed turbulence metrics. The EDR, which is the standard metric used in the aviation industry, should also be investigated. The turbulence metrics deemed most useful to smokejumpers could be incorporated into a turbulence forecast product customised for the smokejumper community, similar to what has been done by the National Weather Service for the aviation industry (National Weather Service Aviation Weather Center 2024). Because WindNinja is a fast-running model, with simulation times typically of the order of a few minutes, turbulence metrics could be generated in real time, for example, via a WindNinja web or mobile interface customised for the smokejumper community. Outputs such as the maximum mean velocity fluctuations or EDR in the jump column could be generated for viewing in the app. Another output that could be of use is the height AGL containing the maximum mean velocity fluctuation or EDR. Future work should evaluate relevant thresholds of EDR and mean velocity fluctuations based on smokejumper flight experiences. Specific turbulence metrics, thresholds for those metrics and display options will need to be selected based on testing and feedback from smokejumpers.