Wetlands Water Level Measurements From the New Generation of Satellite Laser Altimeters: Systematic Spatial‐Temporal Evaluation of ICESat‐2 and GEDI Missions Over the South Florida Everglades

The ICESat‐2 and GEDI missions were launched in 2018, becoming the new generation of space‐borne laser altimeters. These missions provide unprecedented global geodetic elevations, opening great opportunities for water level monitoring. The potential of these altimeters has been demonstrated in open‐water environments such as lakes, rivers, and reservoirs. However, detailed evaluations in vegetated environments, such as wetlands, floodplains, and other areas not constrained by water canal networks, are essential for continued improvement and further hydrological application. We developed a systematic accuracy assessment of ICESat‐2 ATL08, and GEDI L2A products to monitor spatial‐temporal water level and depth dynamics over the South Florida Everglades wetlands. The evaluation was performed on data acquired between 2020 and 2021, using gauge‐based water level and depth estimates as references. The results showed an RMSE of 0.17 m (water level) and 0.15 m (water depth) for ICESat‐2 and 0.75 m (water level) and 0.37 m (water depth) for GEDI. The analysis suggested that nighttime acquisitions were more accurate for both missions than daytime ones. The low‐power beams achieved slightly higher accuracies than those of the high‐power beams over the evaluated wetlands. Water level retrieval was more problematic in densely vegetated areas; however, we derived a correction model based on the leaf area index that improved the accuracy by up to 75% for water depth retrievals from GEDI. Furthermore, the analysis provides new insights to understand the potential of the altimeters in monitoring the spatial‐temporal dynamics of water levels in the evaluated wetlands.


Introduction
Wetland ecosystems contribute in different ways to sustain several human and natural systems.For example, the wetlands play a crucial role in water supply, groundwater recharge, floods, droughts regulation, and pollutants and nutrients removal (Acreman & Bullock, 2003;Cheng et al., 2020;Narayan et al., 2017;Quin et al., 2015).Wetlands worldwide also contribute to global carbon and water storage (Abril & Borges, 2019;Qiu et al., 2021;Wada et al., 2017), thus being also important for climate regulation.It is estimated that approximately 33% of the global wetland areas have been lost to date (Hu et al., 2017).Different factors related to human activities and global environmental changes have contributed to the global wetland area decline (Åhlén et al., 2021;Bogardi et al., 2012;Fluet-Chouinard et al., 2023;Van Asselen et al., 2013), which impact both regional and global ecological balances (Fluet-Chouinard et al., 2023).
Consistent and accurate hydrological monitoring of wetlands is essential to understanding the status of the ecosystems and designing strategies for management, conservation, and restoration (Tiner et al., 2015).In situ water level measurements from gauges provide valuable hydrological data with high temporal resolution.However, the gauges only allow a point-wise perspective of the water surface, which is not spatially comprehensive, particularly in vast wetland areas.For instance, gauge networks typically do not cover palustrine environments like marshes, sloughs, swamps, and floodplains on a global scale, resulting in a lack of critical hydrological information over such environments (Alsdorf et al., 2007;Kim et al., 2009).These wetland types, which are the focus of the present study, are highly dependent on spatial-temporal hydrological patterns.For example, different factors such as the duration and depth of flooding significantly influence the vegetation distribution, nutrient accumulation, and the ecosystem's biota (Todd et al., 2010;Watts et al., 2010).Space-based water level monitoring of wetlands has been demonstrated to be an essential complement to gauge networks and, in some cases, is the only source of information in non-instrumented areas (Kim et al., 2009;McCabe et al., 2017;Palomino-Ángel et al., 2019Palomino-Ángel et al., , 2022)).The increasing availability of data from new satellite missions also poses an opportunity to advance space-based hydrological applications, providing tools for calibrating and validating hydrological models (Fang et al., 2019;Tan et al., 2017).
Satellite altimetry uses radar or laser altimeter sensors on a nadir geometry to accurately determine the vertical position (elevation) of the Earth's surface with respect to a reference surface (Bufton, 1989;Crétaux & Birkett, 2006;Fu & Cazenave, 2001;Zhou et al., 2017).Radar altimeters were designed to monitor ocean water levels (Ablain et al., 2015;Benveniste, 2011;Chelton et al., 2001;Woodworth & Menéndez, 2015).However, their applications have expanded in recent decades, demonstrating the usefulness in monitoring water levels in inland water bodies such as rivers and lakes (Alsdorf et al., 2007;Birkett, 1998;Crétaux & Birkett, 2006;Dettmering et al., 2016).Wetland water level monitoring is a challenge using satellite radar altimetry, considering the limited spatial resolution of most radar altimetry missions and the complex nature of the signal reflected from the surface (Guo et al., 2009;H. Wang et al., 2019).Nevertheless, Lee et al. (2009) have demonstrated their potential for monitoring water levels in wetlands with different vegetation cover types.The authors used TOPEX/ Poseidon data over swamp forests, saline marshes, and brackish marshes in Louisiana wetlands, finding errors of up to 0.46 m over swamps and 0.18 m over marshes.Other applications have used ENVISAT as reference data to calibrate interferometric synthetic aperture radar observations and generate absolute water level change maps over Louisiana wetland swamps forest (Kim et al., 2009).
Space-borne laser altimetry is a newer technology than radar altimetry and, hence, was less explored for hydrological monitoring of inland water bodies and wetlands.The first earth observation system using a laser pulse to detect the Earth's surface was the Geoscience Laser Altimeter System (GLAS) on board the Ice, Cloud, and land Elevation Satellite (ICESat) that operated from 2003 to 2009.Although the mission's primary goal was to measure ice sheet mass balance, the data have allowed water level monitoring in lakes (X.Wang et al., 2013) and rivers (Baghdadi et al., 2011), with errors ranging from 0.13 m for lakes to 1.14 m for rivers.In 2018, two new satellite laser altimeters were launched into orbit: The Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2; Markus et al., 2017) and the Global Ecosystem Dynamics Investigation (GEDI) high-resolution laser ranging (Dubayah et al., 2020), becoming the new generation of space-borne laser altimeters.ICESat-2 has been evaluated over lakes and reservoirs worldwide (Ryan et al., 2020;Xu et al., 2021), reporting a root-mean-square error (RMSE) between 0.08 and 0.28 m. C. Yuan et al. (2020) assessed water level measurement accuracy from ICESat-2 over 30 reservoirs in China using gauge data and reported an RMSE of 0.06 m.Xiang et al. (2021) evaluated the lake water level retrievals from ICESat-2 and GEDI over the Great Lakes in North America using in situ gauging stations and reported an RMSE of 0.06 and 0.28 m, respectively.The same authors validated the accuracy of ICESat-2 and GEDI for water level retrievals over the main channel of the lower Mississippi River and reported an RMSE of 0.12 and 0.40 m, respectively (Xiang et al., 2021).Zhang et al. ( 2022) also compared ICESat-2 and GEDI for water level retrievals using gauge data over four lakes in China and reported RMSE varying from 0.04 m to 0.13 for ICESat-2 and RMSE from 0.35 to 0.46 m for GEDI.Table S1 in Supporting Information S1 summarizes the presented accuracy assessments.
All the mentioned studies were focused on evaluating the capabilities of ICESa-2 and GEDI for water level retrievals in lakes, reservoirs, and rivers.However, their specific performance and strengths in vegetated wetland environments have not been thoroughly evaluated and discussed.Rosenqvist et al. (2020) compared the water level obtained by ICESat-2 during the dry and wet seasons along a transect traversing the seasonally inundated Varzea floodplain forest and the Solimões river.The authors found that ICESat-2-derived water levels varied according to gauge data for two observation dates (one in the dry and one in the wet seasons), but no accuracy analysis was provided.
This study aimed to provide a systematic assessment of the performance and accuracy of the ICESat-2 and GEDI water level measurements over wetlands.Our pioneering study, which is part of the hydrogeodesy special issue (https://agupubs.onlinelibrary.wiley.com/hub/journal/19447973/homepage/call-for-papers/si-2023-000544),covers both the technical aspects of geodetic measurement methodologies and their applications to wetland hydrological monitoring.We conducted the analysis over South Florida's Everglades considering its dense network of gauge stations, which was ideal for ground truthing.Additionally, the variety of wetland types (with different palustrine environments such as sloughs, marshes, and swamps) and hydrological regimes (controlled, natural flow) allowed us to evaluate the missions over different conditions.We assessed the accuracy of the ICESat-2 ATL08 and GEDI L2A products in the entire Everglades region and analyzed the potential of the data for monitoring the spatial-temporal dynamics of water level.Moreover, we evaluated the effect of the acquisition times, vegetation types and density, and beam modes of the sensor on the water level measurement accuracy.Finally, we proposed a correction model for the altimeter observations based on the error analysis.

Satellite Laser Altimetry Background
Satellite laser altimeters are active remote sensing instruments that use laser pulses to obtain the distance from the instrument to the ground surface by measuring the time the transmitted signal takes to travel from the instrument to the ground and back (Bufton, 1989;Zhou et al., 2017).A differentiating aspect among laser altimeters is their return pulse detection mechanism.A full waveform lidar system uses a linear detection mode that converts the return pulse into an analog waveform, capturing the entire temporal profile of the reflected laser energy (Hofton et al., 2019).In contrast, photon-counting lidar systems use detectors that are sensitive at the single photon level and record the arrival time associated with a single photon (Hofton et al., 2019;Neuenschwander et al., 2021;Zhou et al., 2017).In this study, we used data measured by both laser detection mechanisms, as the GEDI mission is a waveform system, and the ICESat-2 is a photon-counting system.
When altimeters observe land surfaces, different aspects of the surface and the operating characteristics of the satellite systems impact the signals received at the sensor.These aspects encompass solar background noise, scattering and attenuation in the atmosphere, laser energy, and surface reflectance (Blair et al., 1999;Hofton et al., 2019;Neuenschwander et al., 2021;Zhou et al., 2017).The vegetation types also influence the returning pulse, leading to a distorted returning waveform (for full-waveform systems) or fewer returned photons in photon counting systems.This feature poses a challenge for processing algorithms to identify layers in the vertical profile (i.e., canopy height and ground height), especially in densely vegetated areas (Hofton et al., 2019;Neuenschwander et al., 2021).This study used the terrain surface height derived from the altimeters to indicate water level height in wetlands.

Study Area
The Everglades region is located in southern Florida, and it extends over an area of 9,150 km 2 from the margin of Florida Bay in the south to the Everglades Agricultural Area (EAA) in the north (Figure 1).The area supports a diverse mosaic of different wetlands, including freshwater marshes, swamps, sloughs, and wet prairies (Figure 1a).The area also presents diverse vegetation communities where the sawgrass (especially Cladium jamaicense) is the most abundant, interspersed with patches of shrubs with a mix of swamp and bayhead shrub species, and trees with a mix of swamp, hammock, and bayhead tree species.Other graminoid marsh and graminoid prairie species are also abundant in the area.The northwestern part of the region has an important presence of cypress forests (David, 1996;Gunderson, 1994;Heffernan et al., 2009;Ruiz et al., 2019).
The region presents a subtropical climate with annual cycles of wet and dry seasons characterized by fluctuations in precipitation and evapotranspiration, which are the main natural drivers of the hydrology in the area (Todd et al., 2010).The approximate annual accumulated precipitation over the region is 1,400 mm/year, with a peak during the wet season from June to October and a lower precipitation rate during the dry season from November to May (Sinha et al., 2018).The area is characterized by a flat topography with an average slope of 0.002% ( Bourgeau-Chavez et al., 2005) and elevations below 5 m that present an altitudinal gradient increasing from south to north (Figure 1b).

Water Resources Research
The wetlands in the area are divided by a network of channels and levees forming units with different hydrological patterns where some areas have managed surface water flow, and others remain with a natural surface water flow.The natural flow areas correspond to the Everglades National Park (ENP) and the Big Cypress National Preserve (BCNP) in Figure 1a, and the managed areas correspond to the Water Conservation Areas (WCAs in Figure 1a).The inundation patterns over the natural flow areas are driven mainly by the regional precipitation regimes, generating higher water levels during the wet season and lower water levels in the dry season.In contrast, the network of channels and levees impacts the hydrological connectivity in the managed water flow areas (Palaseanu & Pearlstine, 2008), where differences of up to 1 m in water level are common between the hydrologic areas (e.g., Figures 1c and 1d).

Data Sets and Data Preprocessing
To evaluate the potential of the altimeters in estimating water levels and depths over the studied wetlands, we used five data types: (a) Along-track heights from the laser altimetry product ICESat-2 ATL08; (b) Along-track heights from the laser altimetry product GEDI L2A; (c) Ground-based water level and depth surface generated from the Everglades Depth Estimation Network (EDEN); (d) Digital Elevation Model (DEM) of the Everglades (Figure 1b); and (e) Two vegetation data types generated from optical remote-sensing observations: land cover maps derived from the Copernicus Land Cover (ANC18) product and the Leaf Area Index (LAI) derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) MCD15A2H product.We used data acquired during the period 2020-2021 (2 years), to have sufficient observations of the variability of the wet and dry seasons within the study area.The data types and processing are summarized in Figure 2 and described below.

ICESAt-2 ATL08 Product
The ICESat-2 mission was launched on September 2018, carrying onboard the ATLAS instrument (Neuenschwander & Pitts, 2019).The instrument operates at 532 nm in the green range of the electromagnetic spectrum and has a laser pulse rate of 10 kHz that generates a 13 m diameter footprint on the ground.The pulses from ATLAS are split into six individual beams arranged in three pairs (Figure 3a).Each pair of beams, separated by 3.3 km in the across-track direction, consists of a high energy (100 mJ) and a low energy (25 mJ) beam (i.e., strong and weak beams) (Neuenschwander et al., 2021).The strong and weak beams of each pair are separated by 90 m.As ICESat-2 moves along its orbit, the six beams describe six tracks on the Earth's surface repeated every 91 days.
The ICESat-2 data are acquired, processed, and archived by the NASA National Snow and Ice Data Center, Boulder, Colorado, USA.The processing includes the generation of several higher-level data products for land ice, sea ice, vegetation and land, ocean elevation, and inland water.We used the ALT08 land/water vegetation elevation product which provides estimates of terrain and canopy height in the along-track direction of the satellite.
The ATL08 terrain and canopy heights are provided at a fixed step size of 100 m along the ground track, referred to as a segment.Each segment has a width of 13 m, considering the diameter of the ATLAS instrument footprint, thus forming rectangular segments of 100 × 13 m 2 .We used the terrain surface height as an indicator for water level height in wetlands.Among the variables of the ATL08 product, we used the best-fit terrain height (h_te_best_fit), which is the best-fit terrain elevation at the mid-point location of each 100 m segment.The midsegment terrain elevation is determined by selecting the best of three fits (linear, third-order, and fourth-order polynomials) to the terrain photons and interpolating the elevation at the mid-point location of the 100 m segment (Neuenschwander et al., 2021).
We downloaded all the ATL08 data acquired by ICESat-2 between 2020 and 2021 within the study area using the NASA Earth Data Search (https://search.earthdata.nasa.gov/search?q=ATL08%20V006).A total of nine ground tracks of the mission crosses the study area with each track revisiting it every 91-day (Figure S1a in Supporting Information S1).An operational off-nadir pointing at different angles of the ATLAS instrument generates ground tracks that shift several kilometers each revisit (Markus et al., 2017).The varying positions of the repeated ICESat-2 tracks imply that laser altimetry measurements are not conducted in the same exact location, unlike most altimetry missions that acquire data along fixed ground tracks (Figure S1b in Supporting Information S1).
The ATL08 product is distributed in HDF5 format, and the heights are referenced to the WGS84 ellipsoid (ITRF2014 reference frame).To compare with the other data sets, a first step in data preprocessing included height transformation to the NAVD88 datum.We performed the transformation using the Vertical Datum transformation tool (VDatum) developed by NOAA (available at https://vdatum.noaa.gov/welcome.html).Additionally, an important challenge to the ATL08 product is the effect of solar background noise, affecting the accuracy in identifying ground photons used to determine the terrain heights (Liu et al., 2021).A higher number of ground-reflected photons implies a better ability for the product to estimate the terrain height (Neuenschwander et al., 2021).Based on the satellite velocity and the expected number of reflected photons for land surfaces, at least 50 reflected photons per segment are used to accurately represent the surface (Neuenschwander et al., 2021).Taking the criterion into account, the second step in data preprocessing was the removal of observations with less than 50 ground-labeled photons per segment.Finally, the third step of the data preprocessing consisted of detecting and removing outliers.Considering the flat topography of the study area, with heights not exceeding 5 m, outlier removal was performed using an interquartile range (IQR) method (Yuan et al., 2020).Outliers were detected by the 1.5 times the difference between the third and first quartiles of the data set: IQR = q 0.75 q 0.25 (1) Where q 0.25 and q 0.75 are the first and third quartiles of measured altimeter terrain height (ATH), and ATH outlier are the detected outliers which are beyond the threshold (Equation 2).

GEDI L2A Product
The GEDI mission was launched in December 2018 and operated onboard the International Space Station (ISS) until March 2023.NASA plans to reinstall the sensor in 2024 to resume measurement operations.GEDI is a full waveform, multibeam laser altimeter.The instrument emits three laser beams simultaneously; one of them is split into two beams (coverage beam), and the other two lasers remain at full power (full beam).Each of these four beams is then dithered to produce eight ground tracks that are spaced approximately 600 m apart on the Earth's surface in the across-track direction (Hofton et al., 2019;Liu et al., 2021).Each beam transect has 30 m diameter footprints spaced every 60 m along the ISS's track (Figure 3b).
The GEDI L2A Geolocated Elevation and Height Metrics product provides estimates of terrain elevation, canopy height, and relative height metrics.The GEDI product estimates the different heights by interpreting the waveform of the laser pulse as it interacts with the reflecting surfaces (Hofton et al., 2019).The waveform can have a simple (single mode) form or a complex and multimodal form with each mode representing energy reflected from different vertical layers within the laser footprint.Simple waveforms are common in oceans, open water bodies or bare-ground regions, whereas complex waveforms are common in rough terrain or vegetated areas.The first mode in a complex waveform corresponds to the surfaces having the highest elevation within a footprint (i.e., top of the canopy), and the last mode of the waveform corresponds to the surface having the lowest elevation (i.e., terrain-water surface) (Hofton et al., 2019).These reflecting surfaces are reported in the GEDI L2A product as the elev_highestreturn and the elev_lowestmode variables, respectively.We used the elev_lowest_mode, which represents the elevation of the center of the lowest mode of the waveform relative to the reference ellipsoid (Hofton et al., 2019).
We downloaded all the GEDI L2A data acquired between 2020 and 2021 within the study area using the NASA Earth Data Search (https://search.earthdata.nasa.gov/search?q=C1908348134-LPDAAC_ECS#).Considering that the sensor operates onboard the ISS, the ground tracks are limited by its orbit and do not have periodical observations.During the studied period, the sensor crossed the study area a total of 88 times at different swaths and locations.The product is distributed in HDF5 format, and the heights are referenced to the WGS84 datum.
The preprocessing of the GEDI L2A product included the transformation of the reported heights to the NAVD88 datum to compare with other data sets.Additionally, the product provides a quality flag (quality_flag) that was used to remove erroneous or lower-quality observations.Finally, in the same way as for the ATL08 product, outlier removal was performed using an interquartile range (IQR) method (Equations 1 and 2).

EDEN Water Surface Products
To spatially continuous interpolations using the Radial Basis Function multiquadric interpolation of EDEN gauge network data (Haider et al., 2020).Daily median water levels from each gauge are used to generate the surface water level product.Water depth is calculated by subtracting the EDEN ground digital elevation model from the predicted water level.The continuous water surfaces are generated on the 400 m EDEN grid and are recorded as elevation relative to the North American Vertical Datum of 1988 (NAVD 88).The products report a RMSE of 4.78 cm for the entire EDEN region (Haider et al., 2020) and are available at https://sofia.usgs.gov/eden/models/watersurfacemod_download.php.
We used water surface products Version 3 (Haider et al., 2020) and not the measured levels at gauges.Although gauges measurements are more accurate, the information is available in a finite number of gauge stations (277 within the study area), which does not necessarily coincide with the altimetry data location.The continuous water level and depth surfaces provide values throughout the Everglades and, hence, enabled us to conduct a systematic comparison between the space-borne altimetry and ground-based observations.

Digital Elevation Model (DEM)
We used ground elevation data from the EDEN DEM (available at https://sofia.usgs.gov/eden/models/groundelevmod.php) to calculate the altimeter-derived water depth.Due to Florida's flat topography and tectonic stability, no measurable topographic changes are expected to occur during time scales of tens of years.Thus, the altimetry-based measurements represent water level changes.The DEM was developed using the High Accuracy Elevation Database (HAED) for the Everglades, a field-collected database of sub-water terrain surface elevation.
The HAED database combines ground survey data with data collected using the USGS's Airborne Height Finder (AHF) system (Jones et al., 2012), which adapted a helicopter-based survey system to differential GPS technology.The database comprises approximately 54,000 terrain elevation points at an approximate spacing of 400 m across the Everglades region.The EDEN DEM derived from the HAED database was produced using a multistage iterative approach yielding sub-water level elevation values with centimeter accuracy (Jones et al., 2012).

Vegetation Data
We introduced ancillary data to assess the effect of vegetation cover type and density on the accuracy of the water level retrievals.ICESat-2 ATL08 product provides land cover information for each segment using the Copernicus Land Cover (ANC18) product at the 100 m resolution (Buchhorn et al., 2020).The land covers were grouped into five descriptive categories of vegetation structure, considering the types of vegetation present in the study area from the ANC18 product.The evaluated land cover categories are open water, herbaceous vegetation, shrubland vegetation, open canopy forest, and closed canopy forest.
The Leaf Area Index (LAI) was used as an indicator of vegetation density.We derived the LAI data from the MCD15A2H Version 6.1 Moderate Resolution Imaging Spectroradiometer (MODIS) Level 4, Combined Fraction of Photosynthetically Active Radiation, and Leaf Area Index product.The product is an 8-day composite data set with a 500 m pixel size (Knyazikhin et al., 1999).The main advantage of MODIS products compared to other satellites (e.g., Landsat) is the higher temporal coverage of 8 days repeat observations.The higher temporal resolution comes with a lower spatial resolution (500 m), which corresponds well with the EDEN water surface resolution of 400 m.

Methodology
The data sources have different spatial resolutions, varying from individual footprints with a 30 m diameter to grided products with a cell size of 500 m.Thus, it is necessary to homogenize the resolution of all the products to perform the accuracy evaluation of the altimeters.We used the spatial resolution of the EDEN water surface products as the spatial resolution of analysis.The EDEN region is covered by a total of 57,071 cells of 400 m × 400 m, with water level and depth data on a daily scale.The Copernicus Land Cover (ANC18) product has an original spatial resolution of 100 m.All 16 ANC18 pixels within a single EDEN cell were used to determine the land cover of the EDEN cell by assigning the class of the majority of ANC18 pixels within that cell.The MCD15A2H product from which the LAI was extracted has a spatial resolution of 500 m.In this case, each EDEN pixel was assigned the LAI value of the centroid of the cell.The same procedure was applied for each 8day composite of MCD15A2H during the evaluated period 2020-2021.
The space-borne water levels for each EDEN cell derived from the altimeter observations (h_te_best_fit for ICESat-2 ATL08 and elev_lowest_mode for GEDI L2A), were obtained by averaging the measurement values that fall within each EDEN cell for a particular observation day.The GEDI L2A product has 30 m diameter footprints spaced every 60 m along its track.Thus, the number of GEDI L2A data footprints that fall within an EDEN cell varied from 1 to 10, depending on the track location within a cell.For ICESat-2, there were segments of 100 m along the ground track.Because the ICESat-2 track azimuth is almost N-S, the number of segments that fall within an EDEN cell varied from 1 to 4. Figure S2 (in Supporting Information S1) presents example observations from both altimeters over nine cells of the EDEN grid.Cell number 2 in Figure S2 in Supporting Information S1 had 10 GEDI L2A footprint samples from the full power beam to calculate the average water level, while cell number 7 had 2 GEDI L2A footprint samples from the coverage beam.The observation geometry of ICESat-2 allowed obtaining, in some cases, observations of the weak and the strong beam on the same EDEN cell (e.g., cells 2, 5, and 8 in Figure S2 in Supporting Information S1).In such an event, the average water level was calculated for each beam separately in order to compare values for both beams.The altimetry-based water depth was calculated by subtracting the EDEN ground digital elevation model from the altimetry water level for each EDEN cell.

Accuracy Assessment
Based on water level and depth reference values extracted from EDEN water surface products, we calculated statistical values to measure the accuracy of ICESat-2 and GEDI water levels and depths.First, we evaluated the accuracy of estimating water levels using the EDEN water level product.The EDEN product reports the water level even when it drops below the ground level.In such circumstances, altimeters are expected to measure only the ground level.Therefore, we identified the cases where the EDEN product reported water levels below the ground level by subtracting the DEM from the EDEN product.A negative value indicates that the EDEN measurement is below the DEM.In such cases, we assigned the DEM value to the EDEN cell for comparison with the altimeters.Second, we evaluated the accuracy of altimeters for tracking water depth using the EDEN water depth product.All water depths equal to zero in the EDEN water depth product were excluded, which allowed for evaluating the performance of altimeters to track only changes in the water surface.
Considering the daily temporal resolution of the EDEN water surface products, it was possible to compare the altimeter measurements with ground truth data from the exact day of the acquisition.The statistical analysis included the mean bias (MB), mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (CC).The MB is defined as: where x is the reference value derived from the EDEN water surface products, y is the water level derived from ICESat-2 or GEDI data, and n is the number of EDEN cells used in the analysis.The mean absolute error (MAE) is: The root mean square error (RMSE) is: The correlation coefficient (CC) is: Water Resources Research 10.1029/2023WR035422 PALOMINO-ÁNGEL ET AL.
The statistical metrics were calculated for all the available observations of each altimetry mission considering different acquisition times (day vs. night), vegetation (land cover and leaf area index), and different beam modes (strong vs. weak for ICESat-2 and full vs. coverage for GEDI).Finally, we proposed a correction model for the altimeter observations based on the error analysis using the leaf area index.

Spatial Consistency Analysis
After the accuracy assessment, we evaluated the ability of the products to represent the spatial variation of the water depth in the study area.For this purpose, we used sample transects from both altimeters crossing the EDEN region on different dates (representing the wet and dry seasons in the area), covering different land covers and LAI.We compared the altimeters transects with the reference EDEN transects, analyzing the errors for the different beam modes (strong vs. weak for ICESat-2 and full vs. coverage for GEDI) and applying the proposed correction model.

Temporal Analysis
Finally, we evaluated the ability of the altimeters to measure temporal variations of the water depth within the study area.As the altimeters move along their orbits, the ground tracks cross the study area on different dates and locations, generating water level observations for the EDEN cells they cross on their path.Considering that the observation strategy of the two altimeters does not generate repeated ground tracks on the same location, we take advantage of the fact that different ground tracks from each altimeter intersect spatially, generating repeated observations over some EDEN cells.We calculated the number of valid observations from each altimeter for each EDEN cell during the study period (2020-2021).The analysis generated an observation frequency map showing the number of observations available from each altimeter for each EDEN cell over time.The map allowed the identification of the cells with the higher frequency of observations to evaluate the temporal variability of the water surface.We extracted the altimetry-based time series of water depth for each high-frequency observation cell, covering the different hydrological areas within the EDEN region.We compared the altimetry-based time series with the reference EDEN time series and analyzed the errors for different beam modes (strong vs. weak for ICESat-2 and full vs. coverage for GEDI).

Accuracy Assessment
We first evaluated the accuracy of the ICESat-2 ATL08 and GEDI L2A observations to estimate water level and water depth for the entire study area during the period 2020-2021 (Figure 4).There were 49,133 ICESat-2 ATL08 observations and 24,878 GEDI L2A observations available (after quality control) during the period.The observations were from eight ICESat-2 ground tracks (with a 91-day revisit time, completing 66 passes over the study area) and 73 GEDI ground tracks.The water levels ranged approximately between 0 and 5 m with respect to the North American Vertical Datum of 1988 (NAVD 88) (Figures 4a and 4b).When the topography was removed, water depth values were obtained, which ranged between 0 to approximately 2 m for the evaluated period (Figures 4c and 4d).We found that the altimetry water level products had an RMSE of 0.19 and 1.13 m and an MAE of 0.11 and 0.76 m for ICESat-2 and GEDI, respectively.For water depth estimation, the RMSE was 0.16 and 1.12 m, while the MAE was 0.09 and 0.79 m for ICESat-2 and GEDI, respectively.The mean bias (MB) indicates that the ICESat-2 water level and water depth were close to the reference data within 4-6 cm (MB of 0.06 and 0.04 m, respectively), whereas GEDI overestimated water level and depth by 70-80 cm (MB of 0.72 and 0.78 m, respectively).

Parameters Affecting Altimeters Water Level and Depth Retrievals
We analyzed the errors considering different beam modes (strong vs. weak for ICESat-2 and full vs. coverage for GEDI), acquisition times (day vs. night), and vegetation (land cover and leaf area index) (Figure 5).We found that the weak beam of ICESat-2 performed better than the strong beam for the studied wetlands, with up to 38% lower RMSE compared to strong beam observations under the same conditions (Figure 5).Regarding GEDI, we also found a slightly better performance for the coverage beam than the full beam, although not as marked as for ICESat-2.A summary of additional error metrics MB, MAE, and CC is presented in Figures S3 and S4 in Water Resources Research 10.1029/2023WR035422 Supporting Information S1.The following sections describe the parameters affecting altimeter water level and depth retrievals.

Acquisition Time
We separated the altimeters' observations considering the acquisition time (day and night).Better performance was demonstrated using only night observations for both sensors, decreasing the errors for both products (Figures 5a and 5d).The RMSE for ICESat-2 was up to 19% lower for nighttime observations compared to daytime observations.In the case of GEDI, the error was up to 35% lower using night observations.The lower errors were found for ICESat-2 using the weak beam observations acquired at night, with a RMSE of 0.13 m for water level and 0.11 m for water depth estimation (Figures 5a and 5d).Regarding GEDI, the lower errors were for the coverage beam observations acquired at night, with an RMSE of 0.86 m for water level and 0.89 m for water depth (Figures 5a and 5d).In terms of the number of observations, the weak (coverage) beams observations accounted for approximately 44% (29%) of the total sample size for ICESat-2 (GEDI).When it comes to the acquisition time of the observation, night valid samples accounted for approximately 62% and 36% for ICESat-2 and GEDI, respectively.Table S2 in Supporting Information S1 presents the number of valid observations for each scenario.

Land Cover
Additionally, we evaluated the effect of land cover on the estimation of water level and water depth for the two altimeters (Figures 5b and 5e).For the evaluation, we analyzed the five categories of vegetation in the study area.
For ICESat-2, the MB values suggest that the water level and depth estimation were very close to the reference EDEN data using weak beam observations.There was a trend of overestimation in forest land covers (MB of 0.28 m with the strong beam mode of ICESat-2).Both altimeters had the lower errors in herbaceous wetlands, with RMSE of 0.14 m and below for ICESat-2 using weak beam observations and with RMSE of approximately 1.00 m for GEDI using coverage beam observations.The most significant errors were in the closed forest for both products.

Leaf Area Index
We evaluated the effect of the LAI on the estimation of water level and depth (Figures 5c and 5f).Nearly 90% of the observations analyzed for both sensors presented an LAI of less than 2.0 m 2 /m 2 , where the most frequent range corresponded to 0.5-1.0m 2 /m 2 , with 47% of the observations for ICESat-2 and 45% of the observations for GEDI (Table S4 in Supporting Information S1).GEDI significantly overestimated water level and depth, especially in observations over dense land covers with higher LAI index, reaching MB values of up to 1.6 m (Figures S3c and  S4c in Supporting Information S1).The MB values for ICESat-2 suggest that the water level and depth estimation was very close to the reference EDEN.For both altimeters, there was a direct relationship between the LAI and the errors (i.e., the higher the LAI, the greater the error).In the case of GEDI, the RMSE reached a value of up to 2.00 m for observations with LAI greater than 5.0 m 2 /m 2 .For ICESat-2, the error was lower, close to 0.2 m in observations with LAI higher than 2.5 m 2 /m 2 , where it tended to stabilize.

Correction of Altimeter Observations
During the validation process, we found systematic errors in estimating the water levels and depths from the altimeters.We performed a correction that allowed us to adjust the altimeters' measurements and reduced the errors concerning the reference values.We used the relationship between the LAI and the errors observed to correct the observations of both altimeters.LAI is a parameter for which there are global products in different spatial and temporal resolutions.Thus, it has the potential to be used in other wetlands with similar vegetation types.For the correction, we first described the linear relationship between the EDEN and altimeter-derived water levels and depths as: where WL alt and WL EDEN are the altimeters and EDEN-derived water level and depth measurements, respectively, m is the slope, and b is the intercept.We set the intercept b to zero, considering that when WL EDEN is equal to zero, altimeters are expected to report water levels equal to zero (WL alt = 0).We determined the slope m between WL alt and WL EDEN within each LAI range defined in the validation process (Figures S5 to S12 in Supporting Information S1).We used the LAI ranges, considering that higher LAI values yield greater errors and thus change the slope between WL alt and WL EDEN within each LAI range.We found a strong relationship between the slope m and the LAI values, especially for the GEDI observations (Figures S13 and S14 in Supporting Information S1).The relationship was used to determine a model to estimate m using the LAI of the altimeter observations (Figures S13 and S14 in Supporting Information S1).We determined the model separately for the water level and depth observations of each altimeter beam mode.The estimated slope was termed m.Subsequently, the corrected water level (WL corrected ) can be calculated using the WL alt and the m as: The errors in the water level (depth) estimation from the altimeters significantly reduced after the correction (Figure 6).GEDI's errors in the water level (depth) went from an RMSE of 1.13 m (1.12 m) and an MAE of 0.76 m (0.79 m) before the correction to an RMSE of 0.75 m (0.37 m) and an MAE of 0.45 m (0.25 m) after the correction (Figures 4b,4d,6b and 6d).The tendency of GEDI to overestimate water level (depth) also reduced after the correction, going from an MB of 0.72 m ( 0.78 m) to an MB of 0.17 m ( 0.11 m).In the case of ICESat-2, the errors also decreased to a lower extent (Figures 4a, 4c, 6a and 6c).

Spatial Consistency Analysis
After the accuracy assessment, we evaluated the ability of the products to represent the spatial variation of water depth along the satellite's track.For this purpose, we used four sample transects crossing the EDEN region on different dates (Figure 7a).ICESat-2 reproduced well the spatial variation of the water depth along the transects (Figures 7b and 7c).The ICESat-2 weak beam presented an RMSE of 0.09 m, which was lower than the error obtained for the strong beam observations (RMSE of 0.14).The same trend was observed for the two ICESat-2 transects.There were variations in water depth between 0.0 and 0.5 m in both ICESat-2 transects, which were captured by the sensor.There was a tendency for higher errors in the shallower areas of the water, which is consistent with the results obtained in the accuracy evaluation (Figure 4c), where a higher dispersion of the data is evident in water depth close to zero.The transects for GEDI (Figures 7d-7g) presented errors of higher magnitude than ICESat-2, which is consistent with the accuracy evaluation results presented in Section 5.1.However, water depth variations were also captured by the sensor.The corrected transects (Figures 7e and 7g) presented RMSE of up to 30% lower for both full and coverage beams concerning the uncorrected data.Residuals of the observations are presented in Figure S15 in Supporting Information S1.

Temporal Analysis
We evaluated the ability of the altimeters to measure temporal variations of the water depth within the study area.First, we generated the observation frequency map showing the number of observations available from each altimeter for each EDEN cell (Figure 8).GEDI had 24,878 observations available (after quality control).The sensor operates onboard the ISS; thus, the ground tracks are limited by its orbit and do not have periodical

Water Resources Research
10.1029/2023WR035422 observations, leading to lower observation frequencies in the same location.The highest observation frequency for GEDI was four observations over the same EDEN cell in the 2 years (Figure 8a).ICESat-2 had 49,133 available observations (after quality control).In the 2 years, the highest observation frequency for ICESat-2 was nine observations over the same EDEN cell (Figure 8b).When both sensors were combined, the frequency of observation in the study area increased.The maximum observation frequency for the 2 years was 11 observations over the same EDEN cell (Figure 8c).
From the observation frequency map, we identified six sites with relatively high observation frequency (Figure 8c), covering the hydrological areas of EDEN.For each site, we generated the water depth time series from the altimeters and the EDEN water depth product (Figure 9).The altimeters allowed to reproduce the temporal variation of water depth with RMSE less than 0.08 m for ICESat-2 and 0.22 m for GEDI in the analyzed sites.A peak of maximum water depth is identified in the 2 years between October and December, followed by a low water level period with a minimum between March and May.The intensity of the water depth peaks was different for each EDEN hydrological area during the evaluated period, with higher variability in the WCA1, WCA2, and WCA3, reaching water depth above 1.0 m.Lower magnitudes of water depth variations were found in the natural areas of the ENP and BCPA for the evaluated period, with peaks not exceeding 0.8 m of maximum water depth.The combined observations of the sensors captured the temporal variations in the water depth; however, a higher frequency of observation is required to reproduce the magnitude of the water depth features in the study area.

Discussion
This study systematically assessed the performance and accuracy of the ICESat-2 ATL08 and GEDI L2A products for water level and depth measurements over South Florida's Everglades wetlands.We evaluated the effect of the acquisition times, vegetation types and density, and beam modes of the sensors on the water level measurement accuracy.We also proposed a correction model for the altimeter observations based on the LAI and analyzed the potential of the data for the spatial-temporal monitoring of water levels.The study provided new insights to understand better the performance of the new generation of satellite laser altimeter products, helped to improve understanding of the factors affecting the observations, and demonstrated their capability for monitoring spatial-temporal water level dynamics in wetlands.Water Resources Research 10.1029/2023WR035422

Accuracy of Laser Altimeters' Water Level Measurements in Wetlands
The accuracy assessment showed that the ICESat-2 ATL08 product performs better than the GEDI L2A product for detecting water levels and depths in the studied wetlands.The overall RMSE of all observations were 0.17 m (water level) and 0.15 m (water depth) for ICESat-2 and 0.75 m (water level) and 0.37 m (water depth) for GEDI using LAI-corrected observations (Figure 6).It is important to note that both altimeters use different measurement strategies, which may contribute to differences in the accuracy between the results obtained by the two missions.
The GEDI instrument utilizes a full waveform system with a 30 m diameter footprint every 60 m along its track.In contrast, the ICESat-2 mission uses a photon counting system with a high repetition rate, resulting in a 13 m diameter footprint every 0.7 m along its track.The high repetition rate of ICESat-2 offers more spatial details for estimating the ground-level parameters of the ATL08 product (Liu et al., 2021;Neuenschwander & Magruder, 2016), potentially resulting in lower errors.To our knowledge, no other studies have developed a comprehensive evaluation of the accuracy of these products for estimating water levels and depths in wetlands.
However, since we used the terrain surface height derived from the altimeters to indicate water level heights, we compared our results concerning other studies that validated the same products for terrain level estimation over land ecosystems and water level estimation over inland water bodies.
Most studies on land ecosystems validated terrain and canopy height retrievals.For the purpose of this discussion, we focused solely on the results of these studies for terrain height measurements.Similar to our results, the studies reported lower errors for ICESat-2 ATL08 than for GEDI L2A.For example, Liu et al. (2021) (n = 9,875).It is important to note that although we used the same products as the cited studies, the errors identified in our study are significantly lower than the reported errors for ground level over land ecosystems.The differences can be attributed to the vegetation characteristics of the studied areas, and to the fact that the cited studies assessed the accuracy of ground-level retrievals (not necessarily inundated with water).The ground can generate a distorted return signal due to the presence of elements (e.g., vegetation) or irregular topography.In the presence of water (i.e., inundating the ground), the more uniform surface of the water is expected to yield a strong signal return to the sensor (Neuenschwander et al., 2021;Neumann et al., 2021), increasing the number of returned photons for ICESat-2 or generating clearer waveform modes for GEDI, thus improving the ability to detect the lowest bound of the vertical profile (i.e., the water level).
Other studies have validated the products for estimating water levels in inland water bodies, such as lakes and reservoirs.To compare our validation concerning those studies, we analyzed our results in non-vegetated areas.
Relatively lower error values were found with respect to the vegetated areas, reaching RMSE of 0.10 m for ICESat-2 and 0.52 m for GEDI (Figure 5b), which coincides with the results reported by the previous studies.For instance, Zhang et al. (2022) reported errors of up to 0.13 and 0.46 m for ICESat-2 and GEDI, respectively, in different lakes in China.

Parameters Affecting the Accuracy of Laser Altimeters' Water Level Measurement in Wetlands
We evaluated the influence of different parameters affecting the altimeters' observations.The acquisition time (day or night) was the first parameter to be assessed.We found that the observations at night provided lower errors in the estimation of the water level and depth for both sensors.Considering that the satellite laser altimeters are sensitive to solar background noise, especially for the photon-counting lidar system of ICESat-2, the night observations can yield lower errors (Hofton et al., 2019;Neuenschwander et al., 2021;Zhou et al., 2017).The same trend to lower errors using night acquisitions was reported in previous studies over lakes, rivers, and forests (Liu et al., 2021;Xiang et al., 2021;Zhang et al., 2022).We also analyzed the influence of the laser's intensity (strong vs. weak for ICESat-2 and full vs. coverage for GEDI) on the accuracy of the measurements.In the case of GEDI, we found a slightly better performance of the coverage beam (RMSE 0.73 m for all the LAI-corrected observations) than the full beam (RMSE of 0.76 m for all the LAI-corrected observations).For ICESat-2, we found a better performance of the weak beam than the strong beam (Figure 5).Previous studies have found a slightly better performance of the high-power beams of ICESat-2 and GEDI over lakes (Xiang et al., 2021) and forests (Liu et al., 2021).The difference in our results concerning previous studies may be attributed to the characteristics of the ecosystems.At the ground, photons are scattered once or many times by surface interactions, such as with vegetation, terrain, or water, and scatter in every direction, including back toward the sensor (Neuenschwander et al., 2021;Neumann et al., 2021).It is possible that the combination of the signal return from the bright water surface and that from the emerging vegetation may have influenced the relationship between the errors from the different beam modes.However, both pulses provide valuable data that can be used to describe spatial-temporal variations of water level, as shown in Figures 7 and 9.
Regarding vegetation type and density, we found that water level retrieval is more problematic in heavily vegetated areas (Figures 5b and 5c).Vegetation type influences the returning signal to the laser altimeters, leading to a distorted returning waveform for GEDI or fewer returned photons for ICESat-2.Densely vegetated areas pose a challenge for the processing algorithms to identify layers in the vertical profile (i.e., canopy height and ground height) (Hofton et al., 2019;Neuenschwander et al., 2021), thus increasing the errors in water level retrievals.We used the observed relationship between the leaf area index and the errors to derive a model that allowed correcting the observations of both altimeters.The model reduced the errors in the estimation of the water level and depth for both products.The improvement was more significant for GEDI, where the RMSE value of 1.13 m was reduced to 0.75 m (water level) and from an RMSE value of 1.12 to 0.37 m (water depth).Although the implemented correction significantly reduced the errors, there is still a trend to overestimate the water level using GEDI, which has also been found in previous studies (Zhang et al., 2022(Zhang et al., , 2023)).For ICESat-2, the model also reduced the RMSE but to a lesser extent, going from 0.19 to 0.17 m (water level) and from 0.16 to 0.15 m (water depth).Additional sources of uncertainty can contribute to the remaining errors in the observations of both altimeters.For instance, there are aspects related to the operational characteristics of the instrument (e.g., viewing angle, timing, determination of orbits, geolocation of footprints), processing algorithms settings, atmospheric effects (e.g., signal attenuation and atmospheric path delays), and wetland state factors, that can introduce additional uncertainty in the observations (Fayad et al., 2022;Hofton et al., 2019;Liu et al., 2021;Neuenschwander et al., 2021;Neumann et al., 2021;Zhang et al., 2022).Furthermore, the implemented validation methodology in the present study required aggregating the altimeter observations (Figure S2 in Supporting Information S1) to compare them with the EDEN water surface product.Every 400 m cell of the EDEN product may present interior variations of vegetation or topography, which could impact the individual observations of the altimeters (100 m segment for ICESat-2 and 30 m footprint for GEDI).
Previous studies have proposed correction models for observations from satellite laser altimeters based on other factors.For instance, Fayad et al. (2022) proposed correction models based on random forest regressors for GEDI water level retrievals in the Great Lakes in North America.Characterizing errors due to atmospheric factors, the instrument, and water surface factors was crucial to improving the observations.In forested areas, other studies have found different vegetation attributes related to the accuracy of ground-level retrievals from satellite laser altimeters.For example, attributes such as canopy height (Liu et al., 2021), canopy cover percentage (Liu et al., 2021;Neuenschwander et al., 2020), land cover and vegetation types (Liu et al., 2021) influenced the altimeters' accuracy.
Studies using airborne lidar on wetlands have also suggested that several vegetation attributes are associated with the accuracy of the observations.For instance, Medeiros et al. (2015) implemented a correction model based on biomass in coastal marshes, which improved ground-level retrievals from the airborne lidar by 38%.Other studies have used the Normalized Difference Vegetation Index (NDVI) (Buffington et al., 2016), species-specific bias based on vegetation cover maps (Hladik et al., 2013), and sensor signal parameters combined with wetland surface characteristics (Rogers et al., 2018;C. Wang et al., 2009).
Our proposed correction model is based on the leaf area index (LAI), a parameter for which global products are available at different spatial and temporal resolutions.Therefore, the model has the potential to be applied in other wetlands with similar vegetation conditions to those evaluated in the present study.We used LAI data available from the closest date to the altimeter observation during the 2-year evaluation period, which suggests the potential for applying the model under different vegetation conditions resulting from seasonal variations.However, the model does not incorporate additional sources of error, such as operational characteristics of the instrument, atmospheric effects, and other wetland state factors.Thus, we recommend that future studies integrate parameters accounting for these additional error sources.

Implications for Monitoring Water Level Dynamics in Wetlands
The spatial and temporal consistency of the altimeters' observations are also a critical factor for monitoring water level dynamics in wetlands.Our results demonstrated the ability of both sensors to reproduce the along-track water level profile for different transects in the study area.The transects crossed several hydrologic areas, and were taken in both dry and wet seasons (Figure 7).We found a better performance of ICESat-2 products with respect to GEDI products along the selected transects, with an RMSE of 0.08 m using the ICESat-2 weak beam and an RMSE of 0.20 m using the GEDI full beam for LAI-corrected observations.The spatial consistency of the data may contribute to study the spatial variation of the water surface, which is crucial to understanding water motion, storage, and water availability (Alsdorf et al., 2007).
To evaluate the effective temporal resolution of the altimeters, we used the number of valid observations within each EDEN cell during the study period (2020)(2021) to generate an observation frequency map for the Everglades region (Figure 8).Higher temporal resolution allows more detailed information on water level variations (Xiang et al., 2021).We found that ICESat-2 had the highest observation frequency, with nine valid observations over the same EDEN cell for the studied period, followed by GEDI, for which the highest observation frequency was four observations.When both sensors were combined, the observation frequency in the study area increased.The maximum observation frequency found for the 2 years was 11 observations over the same EDEN cell.The products captured the temporal variations in the water depth for the high-frequency observation cells (Figure 9) with an RMSE of 0.08 m for the weak beam of ICESat-2 and an RMSE of 0.22 m for the full beam of GEDI; however, a higher observation frequency is required to reproduce the magnitude of the water depth features in the study area.Note that the observation frequency map was estimated based on the EDEN grid cell size of 400 m.Therefore, in flat topography, coarser grid cell size may increase the observation frequency, indicating a good opportunity for more frequent measurements.However, even with higher temporal resolutions, ICESat-2 and GEDI are still not likely to fulfill some applications requiring water levels at short intervals (i.e., daily).It is recommended to explore integration with other data sources, such as radar altimeters (Lee et al., 2009) or interferometric synthetic aperture radar observations (Kim et al., 2009;Liao et al., 2020) from current and future missions.The upcoming data set from the new Surface Water and Ocean Topography (SWOT) mission, which was launched in December 2022, is expected to allow important hydrological observations over inland water bodies (Biancamaria et al., 2016).The potential of the mission to complement water level dynamics monitoring in wetlands remains to be determined.

Conclusions
The new generation of satellite laser altimeters provides unprecedented global geodetic elevations with dense observational coverage in space and time, opening up great opportunities for inland surface water monitoring.This study presented one of the first attempts to systematically assess the performance of the ICESat-2 ATL08 and GEDI L2A products for water level and depth measurements over wetlands and provided new insights into monitoring water level dynamics.The retrievals from ICESat-2 ATL08 and GEDI L2A products were validated against ground-based data from the EDEN's water surface products in the entire South Florida Everglades.The results showed that both products provide valuable information for monitoring water level dynamics in wetlands, especially after correcting systematic biases.The analysis suggested that: (a) for both products, the nighttime acquisitions are more accurate than those obtained during daytime, most likely because night observations are less affected by the solar background noise; (b) both beam modes of the sensors provide valuable data for tracking spatial-temporal variations in water level.The low-power beams (weak for ICEsat-2 and coverage for GEDI) achieved slightly higher water level accuracy than those of the high-power beams, most likely because of the resulting signal from the mixed wetland surface with water and emerging vegetation; (c) water level retrieval was more problematic in densely vegetated areas; however, statistical relationships between vegetation characteristics and errors can be determined to derive correction models for the altimetric products (such as the one proposed in this study); (d) the satellite laser altimetry products have a great potential for monitoring the spatial-temporal dynamics of water level and depth in wetlands, particularly with the combination of observations from the two missions.However, higher observation frequencies are required for some applications, and exploring the integration with other data sources is recommended.Our results support the idea that satellite laser altimetry may help advance our understanding of hydrological processes in wetlands worldwide, especially in remote and data-scarce regions. The

Figure 1 .
Figure 1.Location of the study area with hydrological units marked by black polygons.(a) Land cover (adjusted from the Statewide Land Use Land Cover map from the Florida Department of Environmental Protection available at https://geodata.dep.state.fl.us/datasets/FDEP::statewide-land-use-land-cover/about);(b) The Everglades Depth Estimation Network (EDEN) Digital Elevation Model (DEM); (c) An example water level map; and (d) An example water depth map for 11 October 2021 from the EDEN water surfaces products.The maps background is a true-color base map (Source: ESRI, DigitalGlobe, GeoEye, Earthstar Geographics, CNES/Airbus DS, USDA, USGS, AEX, Getmapping, AeroGRID, IGN, IGP, swisstopo, and the GIS User Community).

Figure 2 .
Figure 2. Data processing and accuracy assessment flowchart.

Figure 3 .
Figure 3. Illustration of the ground track configuration for (a) ICESat-2 and (b) GEDI missions.Dimensions in the figure are not shown to scale.The illustrations are based on schematics available in the ICESat-2 and GEDI Algorithm Theoretical Basis Documents(Hofton et al., 2019;Neuenschwander et al., 2021) andLiu et al. (2021).

Figure 5 .
Figure 5. Root Mean Squared Error (RMSE) for water level (top panels) and depth (bottom panels) estimations from ICESat-2 and GEDI.The subplots present the metrics discriminated for: (a), (d) the time of acquisition (Day and Night); (b), (e) the land cover (Permanent Water (PW), Herbaceous Wetland (HW), Shrubs (S), Open Forest (OP), Closed Forest (CF)); and (c), (f) the Leaf Area Index (represented by ranges of Leaf Area Index with 0.5 m 2 /m 2 intervals).
Neuenschwander et al. (2020) ICESat-2 ATL08 against GEDI L2A for terrain height retrievals at different sites around the U.S. mainland, Alaska, and Hawaii.The authors reported a RMSE of 2.24 m for ICESat-2 ATL08 (n = 32,666) and 4.03 m for GEDI L2A (n = 90,472).Neuenschwander et al. (2020)validated ICESat-2 ATL08 terrain heights in southern Finland and reported an RMSE of 0.73 m (n = 909,467).In the case of GEDI L2A, Guerra-Hernández and Pascual (2021) assessed the accuracy of terrain height retrievals in Spain and reported an RMSE of 4.48 m Copernicus Land Cover ANC18 product can be downloaded from the Copernicus global land cover viewer: [Tile W100N40, available at https://lcviewer.vito.be/2019].The Leaf Area Index data can be downloaded from the NASA Earth Data Search: [Product MCD15A2H Version 6.1, for the period January 2020-December 2021, available at https://search.earthdata.nasa.gov/search?q=C2222147000-LPCLOUD].