Fig. 1
figure 1

An overview of the study area highlighting the Grand Canyon National Park (black outline) and the Kaibab Plateau at elevations > 2130 m (7000 ft, green shape). High-angle faulting on the plateau is shown with ball and bar symbology pointing in the direction of the downthrown block, adapted from Billingsley 20001 (https://pubs.usgs.gov/imap/i-2688/), USGS public domain. Maps created in Esri ArcGIS Pro 3.5.0 (https://www.esri.com).

Full size image

Karst aquifers support millions of people worldwide2 and are critical for ecosystems and biodiversity3,4. However, the complexity of karst aquifers, coupled with their ability to rapidly transmit groundwater over large distances, makes them both vulnerable to contamination and challenging to model5,6,7,8. Traditionally, maps and surveys of solutional cave networks have been used to infer hydrogeologic properties of karst aquifers9,10,11, but their accuracy and resolution can limit their utility in predicting groundwater flow12,13. Using Grand Canyon National Park’s (GRCA) spring-based water supply as a case study, we evaluate if cave lidar can be an effective tool for characterizing regional flow paths within a remote karst aquifer at high spatial resolutions.

In the arid southwestern United States, many tribal nations, municipalities, and federal land management agencies are entirely reliant on karst aquifers for water14,15. Currently, Grand Canyon National Park’s water supply, which supports over 6 million visitors annually, is sourced from a single cave spring within the North Rim of the canyon16. While the recharge area of this spring is not well defined, previous hydrologic studies on North Rim springs within the park have come to three major conclusions: (1) Discharge from North Rim springs is dominated by snowpack on the Kaibab Plateau [Fig. 1]. (2) Groundwater flow is largely conduit-based and efficient, sinking hundreds of meters and traveling up to 20 km to canyon springs in days to weeks, creating intrinsic vulnerability17. (3) Flow paths from sinks to springs are highly complex18,19,20. Given the proximity of these sinkholes to infrastructure and GRCA’s dependence on karst-fed springs to support both ecosystems and operations, additional efforts to characterize this aquifer at higher resolutions are needed.

The rugged topography of the Grand Canyon remains an obstacle for comprehensive hydrologic study. The incision of the Colorado River exposes a multi-layered karst aquifer composed of alternating Paleozoic shales, siltstones, sandstones, and carbonates21. The largest springs (by discharge) for the Grand Canyon region emerge from the Redwall and Muav limestones, which lie over 700 m beneath the surface of the Kaibab Plateau22. No public well logs on the plateau reach this depth, and lineament analysis and electromagnetic resistivity studies are limited in modeling general properties of this aquifer23.

Fortunately, the exposure of these layers by the erosion of the canyon provides a rare window into the inner structures of this aquifer. Over 80 km of cave passages within the Redwall and Muav limestones have been traditionally surveyed in the past two decades24. Caves form in carbonate rock when groundwater maintains its chemical aggressiveness along a flow path for long distances25. Cave maps from these traditional surveys suggest major groundwater flow paths develop along structural weaknesses within the rock mass beneath the Kaibab Plateau’s surface26,27. While utilizing maps to infer hydrologic properties of an aquifer is well documented9,28,29 their two-dimensional nature and inherent subjectivity complicates analysis.

Terrestrial lidar, which measures the return time of light pulses to precisely characterize a three-dimensional space or object, has a demonstrated record of success in measuring cave environments30,31,32. In addition, extracting geologic features from lidar point clouds has been well established in the literature33,34,35. Our study objectives were to: (1) characterize caves within the Grand Canyon’s North Rim with a handheld mobile laser scanner (MLS) to better understand geologic controls on groundwater flow within the rock mass directly, and (2) compare the accuracy of this methodology with other independent surveys, including cave survey and radiolocation control points.

Results

Lidar metrics

We find that mobile laser scanners (MLS) are an excellent tool for efficiently generating accurate three-dimensional models of multi-kilometer cave systems at centimeter-scale resolution. Three perennially discharging remote caves in the North Rim of the Grand Canyon were scanned, including GRCA’s current water source: Roaring Springs Cave. Over 10.2 km of passages were documented in 25 days. The average point density of these point clouds is 1,615 points/m2 and features as small as 10 cm in diameter can be resolved. Average registration root mean squared error (RMSE) between scans was less than 10 mm, with a minimum overlap between scans of 20.1% [Table 1]. Maximum depths from cave entrances exceeded 1 km horizontally and 400 m vertically beneath the North Rim. We estimate the total compounding error of this methodology to be 0.1%, or a drift of +/− 1 m per 1 km in the X and Y dimensions using a shallower cave system as a proxy (Lava River Cave). Lidar point clouds for all cave systems were georeferenced using 3–5 control points with an average RMSE of 6.8 mm and 10.2 mm in the X and Y, respectively [Table 1]. To our knowledge, this dataset is currently the highest resolution and most extensive three-dimensional characterization of caves within the Redwall and Muav limestones.

Table 1 Lidar point cloud scan registration metrics for each cave system.
Full size table

Relationship between flow paths and geologic structures

Our lidar point cloud data show that present-day groundwater flow within the Muav and Redwall limestones exploits sub-vertical joint sets and bedding plane dip within the rock mass. From both the lidar data and traditional surveys, it is evident that collapse from solutional enlargement exposes these features as planar surfaces across all three cave systems [Fig. 2]. For both caves in the Muav limestone (Roaring Springs and Muav West), exposed bedding planes remain consistent over their overall extents, and joint sets are clearly visible between parallel passages [Fig. 3]. The lidar-documented consistency in bedding planes and parallel passages suggest flow paths in the Muav limestone preferentially enlarge joints to more closely align with the dip direction of bedding planes. While bedding and joint control is also prevalent in the Redwall East cave passages, distinct passage morphologies are also identifiable in the lidar data.

Fig. 2
figure 2

A comparison of in-cave photos (left panels a, c) and a lidar-derived model (right panels b, d) of Muav Cave West showcasing planar features exposed by collapse. Joints (top panels a, b) and bedding planes (bottom panels c, d) orientations control groundwater flow direction. The cave model consists of 750 million polygons, with an average edge length of 2 cm. This model was generated using Poisson Reconstruction36. Water surface on the bottom left panel is manually added as a featureless plane.

Full size image
Fig. 3
figure 3

Top-down view of cave lidar data highlighting straight walls (a) and parallel passages (b, c) created by solutional enlargement of sub-vertical joint sets. Arrows correspond to approximate cave stream flow direction (blue) and bedding dip direction (black) averaged over each cave. The consistency of these features across kilometers of passages suggests development of flow paths forms along joint sets within both the Redwall and Muav limestones. Scale bar applies to all caves. 3 cartography created by A. Mildice

Full size image

Comparison with independent surveys

We compared lidar and traditional cave survey data products. During lidar scanning efforts of the Muav West cave, an experienced survey team simultaneously mapped the cave using traditional survey methods37,38,39. While the scan team covered twice the cave distance in the same amount of time, the survey team was faster when documenting more hazardous passages, including low crawls partially filled with water. The lidar-derived cave data had 98% spatial agreement and overlap with the traditional survey-based map when superimposed in plan view [Fig. 4]. Differences between datasets occurred in isolated areas but, surprisingly, the two surveys did not diverge significantly as distance from the entrance increased. Most differences can be attributed to where the cave sketcher had to subjectively define what was a wall or a ceiling in two dimensions. However areas that were difficult to survey, such as passages that required swimming and setting survey stations on mud covered walls, also resulted in discrepancies.

Fig. 4
figure 4

A top-down view comparing traditional cave survey with lidar-derived model of the Muav West Cave. While both surveys were done independently, the two datasets are remarkably consistent over the extent of the cave. Additional panels of zoomed-in views of the entrance (a), midpoint (b), and end (c) of the cave provided to show greater details. Cartography by A. Mildice.

Full size image

We quantified the compounding errors (drift) of our scanning methodology in a more accessible and smaller cave system outside of the Grand Canyon, Lava River Cave near Flagstaff, Arizona. This proxy cave system was similar to Grand Canyon caves in length from the entrance (1 + km) and passage dimensions but was shallow enough to establish radiolocation control points from the surface to the cave. Total compounding error of the lidar derived data was measured by determining the average distance between surface control points and their subterranean counterparts in the X and Y dimensions. Pairwise point-to-point tensions were calculated using CloudCompare’s Align tool40. Our results indicate that the error in the X and Y dimensions increased as distance from the entrance control points increased. At 500 m from the entrance the RMSE was 0.5 m of drift, increasing to 0.87 m at 1.07 km of distance from the entrance for a compounding error of ~0.1%.

Discussion

We demonstrate that mobile lidar scanners (MLS) provide an accurate and efficient method for characterizing remote cave systems in three-dimensions and enables high-resolution documentation of features directly within a karst aquifer. The methodology and equipment outlined achieved compounding errors of 0.1% over 1 km from established control points. In addition, the mobile lidar was able to scan in difficult areas, including partially flooded crawls, climbs, and rappels. Contiguous models of multi-kilometer cave systems enable relationships between passages to be quantified, which we use to better understand how geologic structures influence modern-day aquifer flow paths. While traditional cave survey equipment is currently lighter and more durable, MLS provides more objective and higher resolution data for geologic characterization compared to traditional cave maps.

We find MLS point clouds are valuable for quantifying orientations of geologic structures that control aquifer flow paths, particularly when highlighting relationships between structures across cave passages. Traditional cave maps, especially their two-dimensional nature and focus as a navigational tool limit their utility in characterizing ceilings and walls, where important geologic structures are exposed. In addition, hand-sketched maps are more subjective and rely on the experience of the cartographers to determine what features to prioritize on the map. In this study, we document handheld lidar scanners to be faster than sketch-based traditional surveys for data acquisition, but suboptimal on climbs and partially flooded passages with less than 30 cm of airspace. While this study is focused on bedding planes, joints and passage morphologies, features with dimensions > 10 cm are visible.

Lidar data application: Understanding aquifer flow paths

Given these encouraging results, we compare our lidar-derived findings to previous predictions and past studies on groundwater flow within the Kaibab plateau. Lidar data from all three GRCA caves show flow paths within the Redwall and Muav limestones developing along subvertical joint sets along bedding plane dip [Fig. 3]. This control on passage morphology is consistently observed in the lidar data, despite the three caves being located on opposite sides of the North Rim over 30 km apart. This pattern supports the longstanding hypothesis that groundwater flow, karst and cave development in the Redwall-Muav carbonates are concentrated along sub-parallel joint sets related to regional faults26,27. Given the increased densities of sinkholes near normal fault complexes17, our results indicate modern day flow paths within the Kaibab Plateau are, at least in part, a product of a tectonic extensional regime.

One notable difference between the Redwall and the Muav cave passage morphologies was the evidence of phreatic (below water table) cave development in the Redwall East cave. Hypogene speleogenesis can occur when upwelling “deep” groundwater mixes with epigenetic (surface) water41,42 and is characterized by the presence of distinct formations. Hypogene formations, such as bubble trails, blind cupolas, and three-dimensional maze passages were present in sections above the stream passage of the Redwall East lidar-derived model [Fig. 5]. These features are consistent with stable isotope results from nearby springs, which suggest that deeper upwelling groundwater is a partial contributor to snowmelt-dominated discharge20,43,44,45. Given the age of speleothems within other Redwall caves on the North Rim of the Grand Canyon46 the presence of these structures above the stream channel suggests present-day groundwater flow may utilize paleo-aquifer flow paths.

Fig. 5
figure 5

Three-dimensional renders of a section of Redwall East Cave point cloud enable visualization of passage morphologies that are traditionally difficult to portray in two dimensions. Viewed in profile (a), distinctive conical ceiling features and vertical connectivity between passages are apparent. Looking from within the model (b) the geometrically complex network of passages is accurately modelled.

Full size image

Conclusion

This study demonstrates that mobile lidar scanners are viable tools for efficiently documenting hazardous GPS-denied environments. We digitized over 10 km of remote cave passages beneath the North Rim of the Grand Canyon at previously unattainable resolutions. Our methodology enabled successful scanning through cave passages partially filled with water, crawlways less than 0.5 m in diameter, and on vertical cliffs using technical rope expertise. While adoption of lidar for cave documentation is not widespread, our results show mobile lidar data provides a detailed contiguous three-dimensional representation of caves that are less subjective than traditional surveys. This method allows for accurate identification and extraction of structural features and bedding planes that are related to groundwater flow and cave development. We achieved compounding errors of 0.1%, or a drift of 1 m per 1 km of cave passage, when comparing our methodology to control data. Validating previous predictions, we find cave development within the Redwall and Muav limestones utilize subvertical joint sets and bedding dip direction in three caves across the eastern, southern, and western flanks of the North Rim. This detailed characterization of the Redwall and Muav limestones shows consistent flow path trends, presenting mechanisms governing Kaibab plateau groundwater-flow systems.

Methods

Three caves connected to perennial springs were selected for scanning. Connectivity between sinks of the Kaibab Plateau and selected caves was established through previous tracer studies and discharge measurements from the Grand Canyon National Park Science and Resource Management Division17,23.

In-cave scanning methods

Methodology and equipment for data collection and processing was identical for all caves. A simultaneous localization and mapping (SLAM) near-infrared (905 nm) handheld mobile lidar scanner (MLS) (GeoSLAM Zeb Horizon, Ltd. Nottingham, UK) was used to generate three-dimensional point clouds of all passages. The GeoSLAM scanner has 6 mm relative accuracy, as reported by the vendor, and records 300,000 points a second at ranges up to 100 m. All scans were completed as a loop, with the scanner starting and stopping at the same exact position to minimize error and data gaps using a tripod mounted frame. Scan duration was limited to < 25 min and laser returns over 30 m from the scanner were discarded to maintain precision47. To mitigate the chance of data corruption, scans requiring the scanner to be transported through low airspaces, or up and down cliffs were done twice for redundancy. Scans in crawls less than 0.5 m in diameter were completed with the scanner pointed towards previously scanned passage to allow the SLAM algorithm to match features more easily. Transporting the scanner across hazardous areas, such as deep water and climbs, was accomplished by handing off the scanner to a securely positioned assistant. Consequently, extraneous points in the lidar point cloud were removed manually. A minimum overlap of 10 m between scans was maintained to facilitate registrations between adjacent scans.

Georeferencing and data processing

All cave point clouds in this study were georeferenced using 3–5 control points recorded from each cave’s entrance scan. If no established control points were available in near the entrance, a Leica GG04 + GNSS antenna with Precise Point Positioning was used to record latitude, longitude (WGS84) and elevation above the geoid (EGM 2008). Geodetic coordinates were projected into grid and then ground coordinate reference systems (CRS) using a combined scale factor (CSF) from a central point with the lowest RMSE. CSF was calculated using the National Oceanic and Atmospheric Administration’s (NOAA) and National Geodetic Survey’s (NGS) “NCAT” and “vDatum” tool48.

For the proxy cave system (Lava River Cave), additional at-depth control points were recorded at the midpoint and end of the proxy cave using longwave radiolocation antennae. A 25 cm diameter radio transmitter was levelled over each control point in the cave, emitting a quasi-static magnetic field at 3946 Hz49. A 42 cm radio antenna was used to locate the null, or center of this field on the surface, where a GPS coordinate was recorded.

Lidar scans were exported from the lidar unit with FARO Connect (ver. 2024.4.0) using the “standard” preset with the “transient” filter enabled with no subsampling. Critically, scans were exported using a filetype that enabled “normals” to be saved for each point to facilitate converting the clouds to meshes. A continuous end-to-end model for each cave was generated by manually registering scans together using FARO SCENE (ver. 2018.0.0648). Registration error between scans was determined using FARO SCENE’s cloud-to-cloud registration at default settings for each cave system. The resulting cave clouds were spatially subsampled to a minimum distance of 1 mm between points.

Traditional cave survey methods

A traditional cave survey was conducted during scanning efforts for Muav West Cave in October 2022, with an additional trip in October 2023. Data collection methods were based on established cave survey techniques37,38,39. The survey team used three modified Leica e7400x distometers (“Disto X2’s”) to measure distance, inclination, and azimuth between stations. Instrument calibration was conducted in the field before survey, attaining an average error of < 0.4° for each DistoX2 with a standard deviation of < 0.3°. Cross checking of measurements by independently backsighting each survey shot ensured azimuth and inclination measurements were within 2° of agreement. Survey shot distance measurements were only recorded if their independently measured back-sight agreement was < 6.1 cm (0.2 ft). A 2D sketch of the cave was drawn around these measurements in plan view, with cross sections at each survey station. Measurements from surveys were digitized using COMPASS Cave Survey Software50. Sketches of the cave were combined to create a seamless map using Adobe Illustrator. Inventory of formations, sediments, biology, and human impacts was done for all survey stations51.