REDWOODS: AO Demonstrator & Redwood Ecosystems

• REDWOODS is an advanced adaptive optics module integrated into ShaneAO for high-contrast exoplanet imaging via second-stage wavefront control.
• It employs innovative methods such as the Wynne corrector and multi-WFS SCAO to mitigate atmospheric and chromatic errors for diffraction-limited sensing.
• In ecology, redwoods denote complex forest ecosystems used as benchmarks for precise 3D reconstructions, DBH estimation, and biomass inference.

REDWOODS most specifically denotes “Real time Exoplanet Direct imaging via Wavefront control Of Optical DefectS,” a new adaptive optics module being added as a sub-bench to the ShaneAO system on the Shane 3-m telescope at Lick Observatory. In the technical literature represented here, the term also appears in a separate ecological sense, where redwoods denote coast redwood and mixed-evergreen redwood forests used as demanding benchmarks for 3D reconstruction, diameter estimation, biomass inference, and geo-localised monitoring. The astronomical and ecological usages are distinct, but both center on high-fidelity measurement in optically difficult environments (Gerard et al., 14 Aug 2025, Korycki et al., 2024).

1. REDWOODS as a second-stage adaptive-optics demonstrator

REDWOODS is funded by NSF, LLNL’s LDRD program, and UC Observatories. It is described as an engineering demonstrator: its primary goal is not routine science operations, but on-sky testing of advanced high-contrast imaging technologies developed in LLNL’s High Contrast Testbed (HCT). Those technologies are aimed at closing the roughly $\gtrsim 100\times$ performance gap identified for ground-based and future extremely large telescopes to reach habitable-zone exoplanets (Gerard et al., 14 Aug 2025).

Within ShaneAO, REDWOODS occupies an intermediate role between the existing first-stage AO system and LLNL’s off-sky laboratory development stack. ShaneAO already provides first-stage AO correction using a conventional high-order deformable mirror driven by a Shack–Hartmann wavefront sensor. HCT is the laboratory facility in which new AO and high-contrast imaging methods are developed and characterized off-sky. REDWOODS deploys HCT concepts on-sky inside ShaneAO as a second-stage, diffraction-limited AO and focal-plane wavefront control module (Gerard et al., 14 Aug 2025).

Its stated scientific aims are improving on-sky correction of residual atmospheric speckles after the first AO stage, enabling higher contrast and smaller inner working angles for exoplanet and disk imaging, and demonstrating control schemes and optical technologies directly relevant for ELT instrumentation and future space missions. That framing places REDWOODS less as a finished facility instrument than as a pathfinder for extreme-AO architectures (Gerard et al., 14 Aug 2025).

2. Optical layout, sensing channels, and broadband SCC operation

The REDWOODS opto-mechanical layout begins with a periscope fold mirror inside ShaneAO that taps the corrected beam and routes it through relay optics that maintain diffraction-limited performance from the first AO stage. From there, the design provides switchable paths to a FAST/SCC focal-plane wavefront sensing mode in the near-IR, a three-sided reflective pyramid WFS mode, a second-stage DM or use of the common ShaneAO DM depending on implementation, and a dedicated real-time controller implementing multi-wavefront-sensor single conjugate AO (Gerard et al., 14 Aug 2025).

A central optical element is the Wynne corrector. In this design, the Wynne corrector is a refractive optical system inserted upstream of a reflective off-axis parabola that changes beam size linearly with wavelength on the OAP. The concept is based on Wynne’s 1979 bandwidth-extension idea for speckle interferometry, repurposed here for broadband self-coherent camera operation. The HCT design uses two triplet lens groups in materials whose refractive-index curves cross at a chosen center wavelength, with $\lambda_0 = 480\,\text{nm}$ in the laboratory design and $\lambda_0 \approx 1.05\,\mu\text{m}$ for REDWOODS. The design is optimized so that chromatic scaling of the pupil compensates the chromatic phase behavior in the SCC, allowing effective SCC fringes over a fractional bandwidth $\Delta\lambda/\lambda_0 \sim 30\%$ in the lab and $\sim 20\%$ for REDWOODS (Gerard et al., 14 Aug 2025).

Functionally, the Wynne corrector produces a wavelength-dependent magnification $M(\lambda)$ such that the effective pupil size on the OAP varies approximately linearly with $\lambda$. HCT tests showed that without the Wynne corrector, focal-plane speckles and DM print-through are chromatically smeared in broadband. With an on-axis Wynne corrector, DM features remain sharply defined across the band, but the SCC Lyot-stop pinhole is off-axis by 3.4 mm, and the Wynne system introduces significant chromatic dispersion for off-axis beams. Translating the Wynne assembly off-axis improves SCC fringe visibility while relaxing wavefront errors in the coronagraphic pupil, indicating a trade-off between pinhole position and chromatic dispersion. For REDWOODS, this informs alignment strategy: the corrector must be aligned with the off-axis SCC reference pinhole in mind so that broadband fringe quality is preserved without degrading Strehl (Gerard et al., 14 Aug 2025).

On the wavefront-sensing side, the REDWOODS configuration includes a FAST/SCC mode at $\lambda_0 = 1.05\,\mu\text{m}$ with Wynne-enabled $\Delta\lambda/\lambda_0 \sim 20\%$, a fully-reflective three-sided near-IR pyramid WFS with $\Delta\lambda/\lambda_0 \sim 50\%$, and kHz-speed SHWFS control using reduced intensities instead of slopes. Together these define a mixed focal-plane and pupil-plane sensing architecture rather than a single-sensor AO loop (Gerard et al., 14 Aug 2025).

3. Multi-WFS SCAO control and loop calibration

REDWOODS adopts a multi-wavefront sensor single conjugate AO architecture in which a fast WFS and a slow WFS share a common DM. In the stated conceptual model, the fast arm may be an SCC or pyramid sensor running at $\lambda_0 = 480\,\text{nm}$0100 Hz to kHz rates with diffraction-limited sensitivity to small residual speckles, while the slow arm may be the existing SHWFS or a lower-rate focal-plane sensor handling larger errors and long-term drifts. Because both arms are noisy and partially non-common-path, the controller must combine multiple measurements into one DM command stream (Gerard et al., 14 Aug 2025).

The measurement model is written as

$\lambda_0 = 480\,\text{nm}$1

where $\lambda_0 = 480\,\text{nm}$2 is the common-path disturbance phase, $\lambda_0 = 480\,\text{nm}$3 and $\lambda_0 = 480\,\text{nm}$4 are temporal non-common-path errors, and $\lambda_0 = 480\,\text{nm}$5 and $\lambda_0 = 480\,\text{nm}$6 are noise terms. A generic multi-WFS reconstructor then estimates DM commands through

$\lambda_0 = 480\,\text{nm}$7

with the reconstructors derived by least-squares or minimum-variance optimization. In the regularized pseudo-inverse form,

$\lambda_0 = 480\,\text{nm}$8

and similarly for $\lambda_0 = 480\,\text{nm}$9 (Gerard et al., 14 Aug 2025).

The key control result reported from HCT analysis is that adding a high-pass filter to the integral controller on the fast WFS significantly reduces transfer of temporal non-common-path error between arms. Without this high-pass filter, strong NCP at $\lambda_0 \approx 1.05\,\mu\text{m}$0 can push the system outside the desired dynamic range; the analysis uses a requirement of total phase error $\lambda_0 \approx 1.05\,\mu\text{m}$1. In time-domain form, the fast arm is represented as

$\lambda_0 \approx 1.05\,\mu\text{m}$2

while the slow arm uses a conventional integral law,

$\lambda_0 \approx 1.05\,\mu\text{m}$3

In simplified frequency-domain form, the residual transfer for a common-path disturbance is written as

$\lambda_0 \approx 1.05\,\mu\text{m}$4

This formulation makes explicit that REDWOODS is designed to distribute authority spectrally across sensing channels rather than assigning a fixed modal partition to each sensor (Gerard et al., 14 Aug 2025).

HCT closed-loop measurements further showed that at 100 Hz the SHWFS error transfer functions, expressed in a zonal actuator basis, match the model well, while FAST/SCC Fourier-mode ETFs show strong variations in optical gain $\lambda_0 \approx 1.05\,\mu\text{m}$5 as a function of spatial mode. Some modes show overshoot and non-standard behavior, so optical gains must be carefully calibrated and divided out in the command matrix to achieve uniform temporal response. These calibration results are directly relevant to REDWOODS, where SCC or pyramid sensing is embedded in a multi-WFS RTC (Gerard et al., 14 Aug 2025).

4. Three-sided reflective pyramid sensing, limiting magnitude, and mode tradeoffs

A second strand of REDWOODS development concerns the three-sided reflective pyramid wavefront sensor. The instrument is modeled for the Shane 3-m telescope with a WFS wavelength centered at $\lambda_0 \approx 1.05\,\mu\text{m}$6 and a bandpass from $\lambda_0 \approx 1.05\,\mu\text{m}$7 to $\lambda_0 \approx 1.05\,\mu\text{m}$8, set by the ShaneAO SHWFS dichroic on the short-wavelength side and the REDWOODS camera quantum efficiency on the long-wavelength side. Unlike a four-sided transmissive pyramid, the REDWOODS mask is a 3-sided reflective pyramid, yielding three main pupil images and emphasizing achromatic, low-light sensitivity over a broad near-IR range (Hurtado et al., 19 Aug 2025).

For small aberrations, the reconstruction is treated linearly through an interaction matrix $\lambda_0 \approx 1.05\,\mu\text{m}$9 relating detector signals $\Delta\lambda/\lambda_0 \sim 30\%$0 to modal coefficients $\Delta\lambda/\lambda_0 \sim 30\%$1,

$\Delta\lambda/\lambda_0 \sim 30\%$2

The reported linear regime extends to approximately $\Delta\lambda/\lambda_0 \sim 30\%$3 rad peak-to-valley. Limiting-magnitude simulations adopt $\Delta\lambda/\lambda_0 \sim 30\%$4, $\Delta\lambda/\lambda_0 \sim 30\%$5, $\Delta\lambda/\lambda_0 \sim 30\%$6, $\Delta\lambda/\lambda_0 \sim 30\%$7, and $\Delta\lambda/\lambda_0 \sim 30\%$8, with photon counts computed from a Vega-spectrum model for integration times of $\Delta\lambda/\lambda_0 \sim 30\%$9 and $\sim 20\%$0. The coefficient-space signal-to-noise ratio is defined as

$\sim 20\%$1

and the adopted criterion for reliable DM commands is $\sim 20\%$2 (Hurtado et al., 19 Aug 2025).

Instrument	Fourier	Zernike
3 m, 30 e$\sim 20\%$3 RON, 1 ms	$\sim 20\%$4	$\sim 20\%$5
3 m, 30 e$\sim 20\%$6 RON, 10 ms	$\sim 20\%$7	$\sim 20\%$8
10 m, 1 e$\sim 20\%$9 RON, 1 ms	$M(\lambda)$0	$M(\lambda)$1
10 m, 1 e$M(\lambda)$2 RON, 10 ms	$M(\lambda)$3	$M(\lambda)$4

These simulations indicate that, for the actual REDWOODS configuration, high-order Fourier modes require very bright guide stars, while low-order Zernike modes remain measurable to around $M(\lambda)$5 at $M(\lambda)$6. The paper characterizes this limiting magnitude as “relatively poor,” and therefore treats throughput modeling, modal completeness, and pixel-selection masks as open engineering issues rather than settled performance limits (Hurtado et al., 19 Aug 2025).

The study also compares two pyramid masks. The shallow apex mask produces a footprint about $M(\lambda)$7 smaller, has higher sensitivity, and is more exposed to aliasing. The steep apex mask produces a larger footprint and reveals a central pupil image corresponding to spatially filtered super-Nyquist high spatial frequencies, thereby acting as an optical anti-alias filter. Despite differences in ROI and frame-rate ceiling, the measured total computational latencies are close—574 $M(\lambda)$8 for the steeper apex mask and 589 $M(\lambda)$9 for the shallower apex mask—and the derived $\lambda$0 dB closed-loop bandwidths are similarly close, $\lambda$1 and $\lambda$2, respectively. The resulting difference in bandwidth error is only about $\lambda$3 rad, whereas aliasing differences are much larger. Operationally, the tradeoff is therefore driven primarily by aliasing versus sensitivity, not by temporal bandwidth (Hurtado et al., 19 Aug 2025).

Laboratory validation used a 4f optical bench and imaged the shallow-apex REDWOODS mask. The best pupil image showed the expected three-ring structure but also unexpected ringing that persisted after careful alignment. The paper attributes this to a likely mismatch between idealized mask models and actual fabricated surface properties, motivating metrology-driven simulation updates (Hurtado et al., 19 Aug 2025).

5. Deformable-mirror hardware, integration status, and on-sky program

The HCT hardware roadmap feeding REDWOODS includes a Boston Micromachines 492-actuator high-order deformable mirror with selectable electronic modes. One mode provides $\lambda$4 inter-actuator stroke with 16-bit electronics, yielding $\lambda$5 stroke resolution for space-style ultra-fine control. Another provides $\lambda$6 stroke with 14-bit electronics for ground-based AO. A custom 3:1 multiplexer permits software-only switching between stroke/resolution profiles. For REDWOODS, the stated relevance is high-order, high-precision DM control compatible with second-stage diffraction-limited sensing and a high-frame-rate RTC (Gerard et al., 14 Aug 2025).

On the sensing side, the planned deployed modes are a Wynne-enabled FAST/SCC channel at $\lambda$7 with $\lambda$8, a fully-reflective three-sided near-IR pyramid WFS with $\lambda$9, and kHz-speed SHWFS control using reduced intensities instead of slopes. Explicit contrast and inner-working-angle numbers are not given in the proceeding, but the intended capabilities are improved suppression of residual speckles for exoplanets and faint companions, improved sensitivity to circumstellar disk substructure at small angles, and on-sky demonstration of multi-WFS SCAO logic, Wynne-enabled broadband SCC, reduced-intensity SHWFS processing, and a compact reflective near-IR pyramid WFS (Gerard et al., 14 Aug 2025).

The integration schedule reported in the proceedings is specific. A pre-fabrication review was completed in April 2024. REDWOODS then entered the integration and testing phase, with installation at ShaneAO/Lick expected to be completed in fall 2024 and internal source testing plus on-sky testing planned through September 2026. Compared with existing ShaneAO operation, REDWOODS adds a second stage of diffraction-limited sensing and control, multiple wavefront sensors optimally combined, and advanced coronagraphic/focal-plane wavefront sensing methods (Gerard et al., 14 Aug 2025).

6. Redwoods as quantitative ecological observatories

In ecological remote sensing, redwoods refer to coast redwood and mixed-evergreen redwood forests whose size, bark morphology, canopy layering, and carbon density make them technically demanding measurement targets. A central example is the UC Santa Cruz Forest Ecology Research Plot, a ForestGEO site in mixed-evergreen and redwood forest in the Santa Cruz Mountains with 16 ha area, $\lambda_0 = 1.05\,\mu\text{m}$0 mapped stems, and repeated censuses on a 5-year cycle. In this context, tree diameter at breast height is identified as a “primary data point used in ecological monitoring and carbon accounting efforts,” and the redwood case is emphasized because small errors in diameter can translate into large errors in volume and biomass (Korycki et al., 2024).

The paper “NeRF-Accelerated Ecological Monitoring in Mixed-Evergreen Redwood Forest” compares mobile laser scanning and NeRF reconstructions for DBH estimation. MLS is acquired with a Unitree B1 quadruped carrying an Ouster OS0-128 3D LiDAR, IMU, stereo vision, GNSS+RTK, and ROS2/LIO-SAM. NeRF reconstruction uses an iPhone 14, the NeRFCapture app, ARKit metric camera poses, and Nerfacto in the nerfstudio framework. Two redwood subplots are studied: Dataset A, $\lambda_0 = 1.05\,\mu\text{m}$1 with 11 coast redwood stems, and Dataset B, $\lambda_0 = 1.05\,\mu\text{m}$2 with 6 redwoods and 3 Douglas-firs. NeRF point clouds are reported as 3–4× denser than MLS point clouds, with fieldwork durations of 5 and 8 minutes for NeRF, 30 and 40 minutes for SLAM, and 45 and 52 minutes for manual DBH, respectively (Korycki et al., 2024).

The principal algorithmic contribution is convex-hull DBH modeling, introduced because RANSAC cylinders systematically underfit redwood trunks with deeply furrowed bark and irregular cross-sections. Using 20 cm vertical slices, DBSCAN filtering, 2D convex hulls, and a maximum-over-models rule for partial trunks, the study reports an overall RMSE of $\lambda_0 = 1.05\,\mu\text{m}$3 cm for NeRF + convex hull, consistently outperforming cylinder-based approaches by a factor of 3–4×. In the same study, SLAM + convex hull is reported at approximately $\lambda_0 = 1.05\,\mu\text{m}$4 cm RMSE over both datasets, so NeRF slightly outperforms MLS in DBH estimation while being faster and less hardware-intensive (Korycki et al., 2024).

At larger scales, crown delineation from airborne data remains a companion problem. The MCRC graph-cut framework operates directly on the full 3D LiDAR point cloud, optionally fused with hyperspectral imagery through robust PCA, and the study explicitly notes its applicability to very large trees such as coast redwoods. Because MCRC assigns LiDAR returns to individual trees in 3D rather than segmenting only a canopy height model, it supports direct estimation of crown-level attributes in structurally complex forests (Lee et al., 2017).

7. Emerging generative and geo-localised reconstruction of redwood ecosystems

Several recent studies treat redwood forests as transfer domains or motivating applications for newer 3D reconstruction methods. The Open Forest Observatory is building a database of geospatial forest data and open-source methods for UAV-based forest mapping. Its current workflow relies on Structure-from-Motion, but the NeRF-based extension is motivated by higher quality, fewer artifacts, and better robustness to sparse views. The paper recommends 70–90% image overlap, combines overhead and oblique flights, and explicitly frames redwoods as the kind of tall, multi-layered stands for which moving from SfM to NeRF matters for fire management, restoration, and carbon accounting (Chanlatte et al., 16 Jun 2026).

ForestGen3D addresses a different sensing gap: the incompleteness of aerial LiDAR below dense canopies. It uses a conditional denoising diffusion probabilistic model to approximate

$\lambda_0 = 1.05\,\mu\text{m}$5

where $\lambda_0 = 1.05\,\mu\text{m}$6 is ALS and $\lambda_0 = 1.05\,\mu\text{m}$7 is TLS-like 3D structure, together with a convex-hull containment prior that keeps generated points spatially consistent with ALS observations. The framework is evaluated at tree, plot, and landscape scales and is reported to match TLS references in tree height, DBH, crown diameter, and crown volume. The paper presents this as directly relevant to redwood ecosystems, where ALS often captures canopy tops but poorly samples trunks, understory, and ladder fuels (Castorena et al., 19 Sep 2025).

A complementary line of work estimates biomass directly from point clouds. “Direct Estimation of Tree Volume and Aboveground Biomass Using Deep Regression with Synthetic Lidar Data” trains PointNet, PointNet++, DGCNN, and PointConv on synthetic lidar plots, then applies the trained regressors to real lidar. On synthetic data, the reported MAPE range is 1.69% to 8.11%; on real lidar, discrepancies relative to field measurements are 2% to 20%, whereas indirect methods based on tree segmentation and allometry, as well as FullCAM, show discrepancies from 27% to 85%. The implementation is on eucalyptus plantations, but the paper explicitly argues that the framework translates naturally to redwoods, where large-tree biomass often challenges allometric extrapolation (Pourdelan et al., 4 Mar 2026).

At the individual-plant scale, Sapling-NeRF fuses GNSS, LiDAR-based SLAM, and object-centric NeRF reconstruction into a three-level representation: coarse Earth-frame localisation, centimetre-accurate plot reconstruction, and dense sapling-scale modeling. Experiments in Wytham Woods and Evo show that stem height, branching patterns, and leaf-to-wood ratios can be captured with increased accuracy as compared to TLS, for saplings with heights between 0.5 m and 2 m in situ. The paper explicitly discusses translation of this pipeline to redwood ecosystems for regeneration monitoring under deep shade, complex understory, and long revisit intervals (Muñoz-Bañón et al., 26 Feb 2026).

Taken together, these studies indicate a bifurcated but technically coherent usage of the term. In astronomy, REDWOODS denotes a second-stage AO demonstrator for ShaneAO, integrating Wynne-enabled SCC, reflective pyramid sensing, multi-WFS SCAO control, and high-precision DM actuation. In forest science, redwoods denote structurally extreme ecosystems that serve as benchmarks for dense 3D reconstruction, DBH estimation, biomass inference, crown delineation, and repeatable geo-localised monitoring from canopy to sapling scale (Gerard et al., 14 Aug 2025, Korycki et al., 2024).