Mapping microhabitat structure and connectivity on a tropical inselberg using UAV remote sensing

Abstract

In the past decade, unoccupied aerial vehicles (UAVs or drones) have emerged as powerful tools for ecologists, and the quality and diversity of information they can reconstruct is increasing. Rocky outcrops or inselbergs are complex three-dimensional (3D) ecosystems with several spatial microhabitats that are difficult to characterize using ground-based methods. UAV-mounted cameras and photogrammetric software can be used to obtain 3D models of whole inselbergs with a spatial resolution of up to 4cm and small areas with a spatial resolution of up to 8mm. The shape and volume of eroded depressions and channels can be reconstructed. This allows simulation of the flow of rainwater that creates local differences in hydrological conditions and connectivity among microhabitats. By capturing the near-infrared (NIR) light spectrum, we mapped proxies of photosynthetic activity. This revealed that the microphytic crusts of tropical inselbergs can have higher values of potential photosynthetic activity than the vegetation on the rock. Overall, we show that in systems where the major ecological gradients depend on the 3D structure of the landscape, drone imaging can help to reconstruct spatial variation in microhabitat structure, including proxies for habitat quality and connectivity.

Keywords

Guiana Shield complex habitats photogrammetry NDVI 3D modeling

I Introduction

In 1849, Alexander von Humboldt wrote about the granite outcrops on the banks of the Orinoco River during one of his travels through South America. He was fascinated by the structure and the striking black color of the rocks, which he misinterpreted as manganese oxides (Rascher et al., 2003). Inselbergs, German word for “island mountains” (Bornhardt, 1900), are found in all major biomes worldwide. They are resistant rocks such as granite or gneiss that have weathered slower than the surrounding landscape, and they are known as mosaics of mosses, lichens, crusts of algae and cyanobacteria, and mat-forming plants that grow directly on the rock (Porembski and Barthlott, 2000).

While a relatively large body of literature describes the fauna and flora of inselbergs (Merlijn et al., 2010; Porembski and Barthlott, 1996, 2000; Sarthou et al., 2017), ecological dynamics of inselberg ecosystems are not well understood. Large portions of rocky outcrops are dry environments with xerophytic plants, while other parts of the same rock where intercepted rainfall accumulates in eroded depressions can form very moist environments with specific vegetation and fauna (De Paula et al., 2016). During the rainy season, these eroded basins can periodically hold standing water (Buschke al., 2013; Withers, 2000) and house a wide diversity of aquatic animals as well as some aquatic macrophytes (Merlijn et al., 2010). Due to differences in shape and catchment, some pools may hold water almost year-round while others can hold water for just a few days without rain (Vanschoenwinkel et al., 2009). Besides depressions, the eroded channels in the rock also affect the spatial distribution of biodiversity, sometimes giving rise to seep or ephemeral flush vegetation with lichens, mosses, and carnivorous plants that grow where streams of water trickle down the slope (Porembski and Barthlott, 1996). Overall, the micro-topography of inselbergs results in a wide range of environmental conditions that can be larger than those found in the surrounding landscape matrix (Main, 2000). However, quantifying these micro-geographic features is not straightforward. To reconstruct the shape of pools, researchers have relied on rough approximations, assuming simple geometries such as cylinders or half-ellipsoids (Altermatt and Ebert, 2010; Tuytens et al., 2014). Although simplification is often appropriate, the shape of pools determines the hydrology of aquatic habitats and affects species distributions and diversity patterns (Vanschoenwinkel et al., 2009, 2013), and can often be quite irregular.

The three-dimensional (3D) structure of the rock also regulates the flow of intercepted rainwater across the inselberg, so eroded channels on inselbergs act as corridors between eroded basins that can contain vegetation patches or pools, facilitating dispersal (Castillo-Escrivà, 2017; Heino, 2017; Vanschoenwinkel et al., 2008). This exchange can increase or decrease local diversity in rock pools depending on whether incoming organisms have negative effects on resident species (Vanschoenwinkel et al., 2013). Flowing water may also negatively impact populations by eroding local reserves of plant seeds, dormant propagules, and invertebrate animals, and removing them from the inselberg (Tuytens et al., 2014). Based on growth patterns of microphytic communities (lichens, algae, and cyanobacteria) in such channels, it is often possible to infer the presence of hydrological connections (Vanschoenwinkel et al., 2008). However, on inselbergs with small-scale topographic differences, it is extremely difficult to assess the shape of the rock, as well as the shape, position, and catchments of the individual basins, unless investigators implement ground-based survey techniques using theodolites or total stations. Nevertheless, these procedures are very expensive and labor-intensive (Kindermann, 2016), so creating a high-definition photogrammetric mosaic and 3D models of an inselberg would require weeks of fieldwork.

As a result, there is a need for time- and cost-effective methods to map the structure of inselbergs. The spatial resolution of aerial pictures and satellite images typically does not allow for the detection of landscape elements smaller than 30cm. One possible solution is to use sensors mounted on UAVs in conjunction with photogrammetry that allows the creation of 3D surface models in almost any desirable spectral range with high spatial resolution and accuracy (Colomina and Molina, 2014). It is also possible to explore the near-infrared NIR spectrum to map the spatial distribution of potential photosynthetic activity (Strong et al., 2017). Moreover, UAV-based reconstructions of the physical habitat structure of the rock could provide essential information about the abundance and quality of different microhabitats, but the feasibility of this approach remains to be tested. Given the high levels of endemism found in inselberg plants and animals (Merlijn et al., 2010; Porembski et al., 1998), and the fact that their populations strongly rely on the presence of microtopographic features that form diverse microhabitats, there is a need to accurately quantify microhabitat structure.

The main goal of the study was to demonstrate that the imagery generated using an off-the-shelf drone combined with a custom data collection and processing pipeline can lead to a high-spatial-resolution quantification of microhabitat structure and connectivity of a complex ecosystem. We use this method to test two important unanswered questions for this ecosystem. From earlier work, it was known that rock pools could interact with neighbors via temporary overflows (Hulsmans et al., 2008). However, to what extent this flow might connect entire clusters of pools is unknown. This is because the catchments and associated outlets (i.e. pour points) cannot be determined without high-spatial-resolution digital terrain models. We hypothesize that water flow on inselbergs between pools may be associated with nested catchments and proper dendritic networks. Second, it is known that the surface of inselberg rocks is characterized by many different lichens and cyanobacteria. However, it is unknown which parts of the inselbergs are photosynthetically active. Here we test the hypothesis that in the dry season when temperatures are high, most active chlorophyll will be present in the black cyanobacterial crust on the inselberg and not in other microhabitats such as dry pool beds, gullies, or patches of terrestrial vegetation.

We investigated the extent to which a relatively simple UAV can help to reconstruct geographic variation in the microhabitat structure of a Neotropical inselberg. For this we developed a custom methodology (including survey techniques on the field and data processing steps) to create a detailed 3D model of the inselberg topography, using the images from a simple RGB (red, green, and blue) on-board camera. In addition to the overall 3D model, we quantified the shape and size of microhabitats and we reconstructed their hydrological connectivity. Finally, we used a camera with red (660nm) and near-infrared (850nm) light sensors (NIR+R) to measure the spatial distribution of potential photosynthetic activity in different microhabitats on the inselberg.

II Methods

2.1 Study area

The Colombian Guiana Shield is a granite craton in the geochronological province Ventuari-Tapajós (Díaz-Merlano et al., 2015). This region is dominated by lowland plains and parts of the granitic base rock are exposed as inselbergs with elevations between 40 and 700 meters above sea level (m.a.s.l.). These rocks belong to the igneous complex of Parguaza granite of Precambrian age from 1950 to 1800 million years old (Bonilla-P et al., 2013). However, it is unknown how long ago they became exposed as inselbergs. Currently, the inselbergs are surrounded by savannas, riparian forests, rivers, and lagoons.

For this study, we selected one shield inselberg (Main, 2000) known as Karikari rock (6.0946°N, 67.4857°W) located in the Bojonawi nature reserve. Karikari has approximately 22 hectares and reaches a maximum altitude of 70 m.a.s.l. For the habitat reconstructions in this study we focused on a representative section of the rock of 1.7 hectares and five eroded depressions that capture a range in different shapes (Figure 1).

Figure 1.

Geographic location of Karikari rock and orthomosaic generated for the complete Karikari rock. The pink outline shows our study area with the five pools inside. The blue dots represent the pour points (outlets) of each pool.

2.2 Data collection

UAVs were originally developed for military use; however, in recent decades, different types have been tested as tools for scientific purposes. When appropriate sensors are mounted on UAVs, they can provide high-spatial-resolution imagery that is excellent for quantifying the spatial habitat heterogeneity of a site. Currently, UAVs are applied in diverse fields including crop science, precision agriculture, wildlife research, forestry, aerial photography and videography, mapping and surveying, casualties’ detection and medical assessment, asset inspection, payload carrying, carbon loss estimates, and land management (Anderson et al., 2019; Baena et al., 2017; Basso and de Freitas, 2020; Boon et al., 2016; Huang et al., 2020; Scholefield et al., 2019; Tang and Shao, 2015). UAVs can be classified based on their size and flight endurance (Watts et al., 2012), or on their aerodynamic features (Nex and Remondino, 2014). Multi-rotors, the most widely used category, take off and land vertically, so they are ideal for surveying small areas with obstacles or irregular terrain.

We used a multi-rotor DJI Phantom 4 (DJI, Shenzhen, China) with an on-board GPS receiver and automatic pilot. A tablet with an Android Operating System was connected to the UAV’s remote control. We used the Pix4D Capture application (Pix4D S.A., Lausanne, Switzerland) on the tablet to monitor the drone status and trajectory and to create and upload the flight plan. Two sensors were used on-board: an RGB camera and a NIR+R camera (attributes of the cameras used for the data collection are available in Table S1 from the online supplementary document). The RGB camera (DJI FC330) is an integral part of the UAV and it is mounted on a gimbal for image stabilization. All images taken with this camera are geo-tagged, thanks to the on-board GPS. The infrared sensor we used is a MAPIR Survey 2 – NDVI (Normalized Difference Vegetation Index) Red+NIR (Peau Productions Inc., San Diego, CA, USA), which is sensitive to the near-infrared spectrum (810–890 nm, peak at 850 nm) and red spectrum (630–690 nm, peak at 660 nm). The NIR+R camera was placed pointing toward the nadir, beneath the UAV’s frame, with a mount strapped to the landing gear.

The main guiding principle for extracting 3D features from images using photogrammetry is to make sure that each point or zone of interest is included in overlapping pictures with different perspectives while maintaining a constant distance from the object of interest. The UAV can accomplish this by moving over its target area while maintaining enough overlap and side-lap between images. We performed two procedures for collecting images. The first procedure was aimed at capturing the entire inselberg (Dataset 1). For this we programmed the UAV to fly over the rock at a constant height of 40m above ground level following a grid-like trajectory, which is useful for covering large areas (Figure S4), while taking RGB images at a 10° off-nadir angle, which improves the visibility of steep surfaces (Ajayi et al., 2017; Rossi et al., 2017), resulting in 80% side overlap and 72% front overlap. We performed a test after setting the flight speed to make sure the images were in focus. Meanwhile, the NIR+R camera was set to acquire images every three seconds. The second procedure was aimed at capturing a more detailed model of five pools (Dataset 2 to 6). For each of these pools, we programmed two circular paths orbiting at a height between 5m and 24m, and with radii from 5m to 31m, with the target pool at the center and the camera pointing toward it at off-nadir angles between 20° and 48° (Figure S4). A circular path was chosen because it is the most efficient way of fulfilling the guiding principle mentioned above when the object of interest is small and has steep surfaces, as is the case for deep and narrow pools such as P2. The off-nadir angles were selected as a necessity for pointing the camera at the pool at the center. Radii, and therefore the number of pictures evenly distributed on each path, were proportional to the size of each pool. An average of 103 pictures was captured for each pool. Details of the datasets of collected images taken of specific pools and of the whole Karikari rock with RGB and NIR+R sensors are available in Table S2.

A recommended practice for capturing images for photogrammetry is the distributed placement of discernible objects or markers at the scene, such as bright sharp corners or spots. These are known as as Ground Control Points (GCPs) or tie points, and they are useful for improving the quality of the resulting model. For dataset 1, collected over the entire Karikari rock, we did not place such markers, but relied on natural features on the rock surface. For datasets 2 to 6, collected from individual pools, we used two types of markers to establish discernible features: target boards and a 3D axis frame (Figure S1).

2.3 3D mapping of topography using RGB imagery

Transforming images to a 3D model requires photogrammetry – a technique that allows the obtainment of geometric information of an object from images. It uses structure from motion (SfM) algorithms to reconstruct 3D surfaces based on (a) multiple images from different viewpoints, (b) intrinsic camera parameters (focal length, sensor size, principal point, and lens distortion parameters) and (c) orientation and location from which the camera captured the photographs and/or location of discernible features in the scene that are manually marked on the photographs. SfM algorithms work by finding common feature points in the images and running optimization routines capable of calculating accurate positions of the cameras in space, and thus of the features visible in the images (Anderson et al., 2019; Dandois and Ellis, 2010). Furthermore, images capturing light from diverse spectral bands (infrared, visible, ultraviolet, etc.) can be used. When discernible features in the images are manually marked, they are called manual tie points and they help the SfM algorithm produce better results. If the precise spatial coordinates of these manual tie points are known, they are called Ground Control Points. Their task is to orient the resulting 3D model according to a desired coordinate system. When the camera locations are known or at least three GCPs are known, the model can be correctly oriented according to the coordinate system. We used Agisoft Photoscan Professional Edition 1.4.0 (Agisoft LLC11, St. Petersburg, Russia) for all photogrammetry processes.

Spatial calibration of the camera is often necessary to obtain reliable photogrammetric models. The goal of this calibration is to find an accurate measurement of the camera’s intrinsic parameters. Photoscan uses a pinhole camera model and the Brown-Conrady model (Brown, 1966; Fryer and Brown, 1986) for the non-linear distortion of the lens. The necessary data for the spatial calibration was gathered in a separate location, a flat grass field with easy access. We took a set of 429 aerial RGB images from different perspectives, at 40m height (the same altitude of the grid-like flight plan used for surveying the entire inselberg). We placed 20 GCPs in the field with high-precision local coordinates, so that the GCPs evenly covered the entirety of the sensor frame. Each of the GCPs was made of textile with a 15cm diameter black circle and a 5cm diameter white circle in the middle serving as the target. A preliminary photo alignment process was performed (point correlation and triangulation) and all GCPs were marked in the images. Then we employed the camera optimization function in Photoscan to calculate the geometric intrinsic parameters associated with the tangential and radial distortion of the images due to the camera lens. The resulting distortion plot and residuals plot are available in Figure S8.

We used the MAPIR Camera Control (MCC) software utility to correct the NIR images as much as possible with the hard-coded values supplied by MAPIR (Calibrating Images in MAPIR Camera Control Application Guide, Peau Productions Inc., San Diego, CA, USA) taken during a clear sunny day with a camera of the same model as ours. This permits feature comparison within the same images or between images acquired at similar lighting conditions.

2.3.1 Complete Karikari rock

After loading the RGB images from the grid-like trajectory flights over the whole inselberg using Photoscan, all the photogrammetric processing was performed to obtain an orthomosaic (Figure 1) and a dense point cloud, as is shown in the flow chart of the methodological framework of the online supplementary document (Figure S2). An orthomosaic is a georeferenced image obtained from a collection of smaller images where the geometric distortion has been corrected and orthorectified, so it can be used to measure true horizontal distances. Since we did not use GCPs on this site, only manual tie points were placed on the model to ensure a good alignment, and the model orientation, position, and scaling were derived from the coordinates of all the geo-tagged images. Moreover, the intrinsic camera parameters obtained from the calibration were imposed on the photogrammetric model of the inselberg to avoid distortion in the surface reconstruction (Griffiths and Burningham, 2019). We processed this model with and without the intrinsic parameters’ constraint to measure their impact in avoiding model deformations caused by the algorithm trying to optimize for camera alignment (Figure S6). The original point cloud was classified (terrain and non-terrain classes) and rasterized using Global Mapper 19 (Blue Marble Geographics, Hallowell, U.S.A.) to create a digital terrain model (DTM). DTMs are similar to digital surface models, but they only contain information from the earth’s surface as they remove any non-terrain elements captured by images (e.g. vegetation, water, etc.). From this DTM we calculated the drainage network of the whole rock. Also, we located the pour points (the point on the surface at which water flows out of a pool) of all pools on the DTM to obtain the drainage basin polygons for each pool.

2.3.2 Individual pools

After loading the RGB images from each orbital photo dataset in Photoscan, and running an initial low-quality alignment step, tie points were placed manually on the target boards and the four ends of the 3D frame, serving as GCPs necessary to set the orientation of the model. Each end of the 3D frame was marked in at least 32 images, evenly distributed around the pool. This number was chosen to allow for an optimal photogrammetric adjustment; however, often the additional benefit of marking more than 15 evenly distributed images per tie point diminishes quickly. Then we continued to perform a final photo alignment, obtaining a dense point cloud and a detailed orthomosaic, as shown in the flow chart of the methodological framework of the online supplementary document (Figure S3).

To get rid of non-terrain elements (such as vegetation and noise), we classified the point cloud into terrain and non-terrain classes, using automatic and manual editing tools in Global Mapper 19. For the point clouds of pools P2, P5A, and P5B (Table 1), the removal of zones with water or dense vegetation created empty regions, especially in concave areas where a simple linear interpolation would not correctly represent its shape. In order to approximate the missing bottom surface of those pools, we used AutoCAD Civil 3D (2018, Autodesk Inc.) to draw consecutive cross-section profiles following the general pattern of the basin and, in conjunction with the terrain cloud point, generate a complete DTM for the pool.

Table 1.

Attributes of reconstructed rock pools basins. The maximum depth, area, and volume were obtained using the RGB orthomosaic and DTMs of each pool. The catchment area was calculated using a flow accumulation operation from the DTM data.

Pool	Maximum depth (cm)	Maximum area (m²)	Maximum volume (L)	Catchment area (m³)	Type of curves (area–volume)
P1	2.99	2.42	50.44	11.30	Concave–convex
P2	13.24	3.16	220.84	41.52	Concave–convex
P3	8.32	22.36	1053.96	132.74	Concave–convex
P4	15.88	51.98	2778.43	6016	Convex–convex
P5A	30.70	7.08	690.97	10,591	Convex–convex
P5B	7.81	3.89	109.71	10,591	Convex–convex

RGB: red, blue, green; DTM: digital terrain model

Based on each pool’s DTM we wanted to calculate a relationship between water depth, surface area, and volume storage for each pool (Figure 2). In order to do that, we used the Fill tool from ArcMap to obtain the pool’s outline at full capacity, the pool’s maximum depth at full capacity, and the DTM clipped by the pool’s outline at full capacity. Finally, the storage capacity extension for ArcMap was employed to calculate the depth–area–volume curve. Based on modeled pool shapes we classified our pools in “Pits gnamma,” “Pans gnamma,” and “Armchair gnamma” according to Campbell (1997), Timms (2013), Aguilera et al. (2014), and Timms and Rankin (2016).

Figure 2.

3D reconstructions of six rock pool basins.

To analyze the shape and volume of pool P5 (dataset ID 6) we decided to subdivide it into two pools, P5A and P5B, because they have a soil bank with two big trees in the middle. These two pools connect after heavy rains, and P5A and P5B differ in the amount of water they need to overflow, so P5B (which is downstream from P5A) overflows earlier than P5A. However, ecologically speaking, P5 (dataset ID 6) is just one ecosystem, because when it is full of water organisms can occupy and interact in the whole pool.

2.4 Reconstructing the catchments and hydrological connectivity of aquatic habitats

Based on the DTM from Dataset ID 1, we reconstructed the hydrological connectivity of the whole rock surface by identifying the drainage network (Figure 3) on the rock that can link up rock pools during rains and mediate dispersal. We used the watershed tool in Global Mapper 19 to generate the drainage networks. This tool fills depressions in the terrain data and applies the D-8 algorithm (Jenson and Domingue, 1988) to estimate the flow direction and flow accumulation at each raster cell for calculating the drainage network (Tarboton et al., 1991). The value of each cell in the flow accumulation raster is a proxy for the peak volume flow rate relative to other cells during a rain event when all rain becomes runoff without any interception, evapotranspiration, or loss to groundwater (ArcGIS 10.6 documentation). This watershed tool was applied to each pool by choosing its particular pour point and letting the software calculate its corresponding catchment area and stream network draining to it. To visualize the general drainage network for Karikari rock in a way that matched the spatial resolution, only cells with accumulated flow from an area greater than 9m² were shown as part of a stream.

Figure 3.

Watershed model of the catchments of five pools (P1: pink; P2: green; P3 orange; P4: yellow and modeling the shape and volume of pools P5: white) and the eroded channels through which, based on the digital elevation model, most water would flow (blue lines). For interpretation of the references to colours in this figure legend, refer to the online version of this article.

2.5 Reconstructing spatial variation in potential photosynthetic activity

Using NIR+R cameras via NDVI analysis Equation 1 (Rouse et al., 1974), it is possible to detect proxies of photosynthetic activity, monitor eutrophication and to estimate the spatial extent and biomass of different aquatic algae species in freshwater and marine ecosystems (Pettorelli et al., 2011). It assumes that healthy photosynthetic cells contain active chlorophyll, which absorbs light in the visible range and reflects in the NIR range. Higher levels of NDVI indicate the presence of healthy photosynthetic pigments, both in plants and microphytic communities (Gamon et al., 1995; Malahlela et al., 2018), and NDVI values can also increase concerning moisture conditions (Karnieli et al., 1999). To compare differences in potential photosynthetic activity between different microhabitat types (rock, dry pools, grass, and evergreen trees), we randomly selected 20 replicates in 20×20cm areas within each habitat type present in Figure 4. Then we used Kruskal–Wallis tests and associated pairwise Wilcoxon rank-sum tests, performed in R (R Development Core Team, 2008), to test for differences in NDVI scores between different microhabitat types (Figure 5).

Equation (1) NDVI: normalized difference index.

NIR: near-infrared reflectance band (850 nm). R: red reflectance band (660 nm).

N D V I = (N I R - R) / (N I R + R)

Figure 4.

Micro-geographic variation in potential photosynthetic activity (NDVI) on a section of Karikari rock characterized by many different microhabitats.

Figure 5.

Sampled points were randomly selected to complete 20 replicates in 20×20 cm areas within each habitat type class present in Figure 4.

III Results

3.1 3D mapping of the complete inselberg using RGB imagery

We found that the unconstrained model resulted in an erroneous convex surface (see the comparison map in the online supplementary document Figure S6). However, the imposition of the parameter’s constraint resolved the issue and produced a very good representation of the terrain while preserving a small re-projection error of 0.648 pixels.

The main data outputs of the photogrammetric processes are summarized in Table 2. Karikari rock (Dataset 1) was reconstructed with a point cloud density up to 990 points/m², which resulted in a raster digital model with a spatial resolution up to 3.9cm/pixel (Figure 6) and an orthomosaic with a spatial resolution of up to 1.9 cm/pixel (Figure 1). This digital model results from the interpolation and rasterization of the dense 3D point cloud and includes vegetation. The products generated for each dataset are exportable to be easily viewed by amateur users in Google Earth and similar programs.

Table 2.

Overview of the spatial resolution of reconstructed 3D point clouds (using unoccupied aerial vehicles photogrammetry) for the entire inselberg Karikari rock (Dataset 1) and five individual eroded rock pool basins (Dataset 2–6).

Dataset ID	Points (million)	Cloud density (points/m²)	Digital model pixel size (mm)	Orthomosaic pixel size (mm)
1	209.5	990	38.6	19.3
2	16.7	29,394	7.1	3.5
3	12.0	59,105	5.3	2.6
4	12.6	25,176	7.8	3.9
5	8.7	19,868	8.5	4.3
6	9.6	19,238	9.4	4.7

Figure 6.

Reconstructed digital elevation model (DEM) for the complete Karikari rock.

In both the orbital-based and grid-based UAV datasets, small erosion channels, colored signs of water flow, and type of coverage (dry vegetation, green vegetation, black rock crust) are clearly visible and delimited. On the other hand, alternative data sources available to our study area, such as satellite imagery, are unfit to represent concave surfaces with a diameter less than 1.5m and discern small patches (less than 30×30cm) of low vegetation from rock surfaces (Figure S7). Therefore, the localization of features of interest, e.g. rock basins, eroded channels, and vegetation patches, is limited, ruling out the possibility of using satellite imagery for studying microhabitats such as the selected pools.

3.2 Modeling the shape and volume of individual pools

Using UAV survey imagery, we reconstructed the models of individual pools with an average point cloud density of 3 points/cm², resulting in DTMs with an average spatial resolution of 7.6mm/pixel and orthomosaics with an average spatial resolution of 3.8mm/pixel (Table 2).

3D reconstructions of the pools revealed two typical shapes according to the nomenclature used by Timms and Rankin (2016): bucket-shaped pools, called “pan gnamma” (Figure 2a, b, and c) and deeper pools with rounded bottoms, called “pit gnamma” (Figure 2d, e, and f). Moreover, it was possible to calculate the relationship between water depth, water surface area, and volume for each pool (Figure 2m to r). These depth–area–volume curves are represented by two functions: the blue line shows the area of the surface of the water for a given water depth in the pool (as one would see when the pool is being filled as it rains). It is calculated based on the summed area of all DTM pixels whose height is below the given water level. The red line shows the volume contained in the pool for a given water depth. The shape of both curves depends on the specific morphology of each pool.

Calculated curves revealed two tendencies: the first one shows a concave function for surface area vs. depth and a convex function for volume vs. depth. This tendency is found in bucket-shaped pools (P1, P2, and P3). It implies that the initial slope at the bottom of the pool is very weak because they have flat bottoms. Therefore, initially, the area will increase rapidly, and afterward, when the water reaches the steep pool sides, the surface area of the pool no longer changes much. As it is possible to corroborate with the cross-section profiles (Figure 2g to i), these pools can be classified as typical pan gnamma, as those modeled by Vanschoenwinkel and colleagues in 2009. The second tendency was for pools with convex functions for both the surface area vs. depth and volume vs. depth. This tendency was found in P4 and P5 (Figure 2d to f). If we look at the cross-section profile for these pools (Figure 2j to l), it is shown they have a more gradual change in the slope of their sides, and their profiles resemble more a half ellipsoid, which is congruent with the pit gnamma morphology (Aguilera et al., 2014). However, P4 and P5B have a peculiar depth profile that is asymmetric, with different slopes on both sides of the pool. Such a pool-type morphology is called an armchair hollow or armchair gnamma (Timms, 2013; Timms and Rankin, 2016).

3.3 Reconstructing catchments and hydrological connectivity of aquatic habitats

The calculated drainage network (Figure 3) was coherent with the eroded channels and colored signs of water flow seen on the rock in the orthomosaic, and this match constitutes a good rule of thumb to check the coherence of the generated elevation models (Figure 6). Also, a graphical model of a nested set of catchments of different pools was generated. Catchment sizes for the studied pools varied between a minimum area of 11.30 m² for P1 and a maximum of 10,591 m² for P5 (Table 1). The watershed model showed that four out of five selected pools are part of the same watershed. Moreover, these four pools are situated sequentially along the same drainage system. Two micro-basins, P2 and P3, are nested in the catchment of P4, which in turn is part of the catchment of P5. P5 represents the outflow (pour point) of the entire drainage system. As such, overflowing water can pass from one pool to another before flowing from the rock to the surrounding ecosystems. The drainage basin of P1, however, was isolated from the other pools.

A similar drainage network (at the same level of detail determined by the threshold drainage area of 9m²) would be achieved if a 50cm/pixel satellite elevation dataset were used. However, the corresponding 30cm/pixel satellite imagery would be insufficient to discern eroded channels and small features that are crucial for checking the coherence of such calculated drainage networks.

3.4 Reconstructing spatial variation in potential photosynthetic activity

Figure 4 shows a capture of the potential photosynthetic activity derived from NIR+R images via NDVI analysis that is representative of the whole inselberg. The analysis revealed an absence of detectable photosynthetic pigments in almost all vegetation patches on the inselberg, except for small areas of grass and a few remaining green leaves present in dry trees. The upper margin of Figure 4a and b shows a section of evergreen trees in the adjoining riparian forest with the higher NDVI scores. However, even though the study took place in the dry season, we found NDVI levels corresponding with green pigments and potential photosynthetic activity in microphytic communities crusted in the bare granite rock (areas that appear darker gray in the RGB panel and darker green in the NDVI panel from Figure 4).

The Kruskal–Wallis test revealed significant differences in the NDVI values from four microhabitats: rock, dry pools, grass, and trees p < 0.0001 (Figure 5). According to the Wilcoxon pairwise test, dry pools and grass had similar values, while rock with crusted microphytic communities and evergreen trees had significantly higher NDVI values. Also, we detected high NDVI scores associated with crusted microorganisms inhabiting wet channels. The dry channels had biological soil crusts observed in the field. Dry rock pools had lower NDVI scores compared with wet channels and crusted microphytic communities in the rock.

IV Discussion

The developed methodology allowed us to test two important hypotheses about this ecosystem: that habitat patches occur in dendritic, hydrologically connected networks with nested catchments, and that the photosynthetic activity of micro-vegetation crusted in the bare rock can be higher than that of other vegetation associated with the rock. The study serves as a proof of principle, illustrating that a wealth of important habitat information relevant for the study of ecological and evolutionary dynamics in this ecosystem can be surveyed in a time- and cost-effective way with a minimal impact on the ecosystem. We also demonstrated the ability to create complete high-definition photogrammetric 3D maps of inselbergs, as well as to reconstruct the eroded channels that transport water and depressions that can store water. Such information is nearly impossible to map and quantify using only ground-based techniques or satellite data. Our results showed that water flow on inselbergs between pools is associated with nested catchments and proper dendritic networks. Several studies have shown that hydrochoric dispersal mediated by eroded channels that connect rock pools can affect the genetic structure of populations (Hulsmans et al., 2008) as well as the composition of communities (Vanschoenwinkel et al., 2008). However, these studies considered the simple case of neighboring pools linking up by simple connections on a relatively flat rock (Tuytens et al., 2014). The current study allowed us to define nested networks of catchments, which is a prerequisite to accurately modeling the volumes of water, not just those transported among pools but also those that end up in vegetation patches. The reconstructed drainage systems in this paper will be helpful for defining moisture-based microhabitats depending on how abundantly and frequently each area of the inselberg receives flowing water, which could, in turn, explain the spatial distribution of different organisms.

The 3D reconstructions of pool basins generated in this study can be used to build more accurate hydrological models that can predict flooding cycles and reconstruct long-term variations in the hydrological disturbance in this system. However, current models use buckets or half-ellipsoids as approximations of rockpool morphometry. We propose the use of such approximations combined with photogrammetric techniques to characterize pool types. This will allow not only quantification of the size of catchments but also calculation of the area–depth–volume functions and the drainage basin area of each pool. Both analyses can lead to more accurate predictions of inundation patterns, which in turn can be used to estimate the extinction risk of populations that exist in these habitats (Pinceel et al., 2017). Since hydrology determines the selection regime in temporary ponds, more accurate models combined with time-series data will help us to better explain both ecological and evolutionary dynamics. For instance, hydrological disturbance regime was an important driver of diversity patterns in rock pools, both within clusters of pools (Vanschoenwinkel et al., 2013) and across climate gradients (Brendonck et al., 2015). Rockpool hydrology was also linked to evolutionary bet-hedging strategies, with populations experiencing harsher conditions with short inundations showing adaptive life-history traits that help to cope with these conditions (Pinceel et al., 2017).

On inselbergs, there is usually little vegetation that can interfere with the creation of general terrain models. Some emerging aquatic plants, for instance, could be easily removed after the photogrammetric process, as shown in this study. However, we acknowledge that generating a DTM is challenging when vegetation is dense, especially for achieving a correct assessment of the shape of the pools’ basins. If basins are filled with dense vegetation or if the water level is too high, the shape of the basin can only be estimated using tie points from the visible edges of the basin.

On sites with more dense vegetation, it is recommended to use either ground-based survey techniques for small areas or multi-return light detection and ranging (LiDAR) for larger areas. The light pulses generated by the LiDAR can still penetrate between branches and leaves. Nonetheless, LiDAR equipment is much more expensive and would require more expensive, heavier drones to operate (Dandois and Ellis, 2010). See the online supplementary material (page 6) for further discussion of the advantages and disadvantages of this methodology.

We have shown that by adding a simple multispectral camera to an affordable off-the-shelf UAV it is possible to reconstruct fine-scale variation in potential photosynthetic activity. To our knowledge, this is the first study that quantitatively confirms the idea that the black biological soil crust on tropical inselbergs is photosynthetically active (Büdel et al., 2000). What is more, at the very least in the dry season, during which this survey took place, microphytic communities were photosynthetically active, while pools revealed very low signals of photosynthetic activity, although some had a wet crust at the bottom (P3) or at least a few centimeters of water (P5). The lowest levels of reflectance were detected on the river’s water and dry vegetation. We also detected that the presence of these communities growing epilithically and endolithically on the bare rock of inselbergs have different degrees of association with humidity on the rock surface and in the drainage channels (Figure 4). To validate our NDVI results it is still important to perform additional ground-truthing and to identify and quantify the algae communities present. Even though satellite sensors are the most common source for collecting NDVI information, using UAVs for this purpose is particularly relevant for fine-scale analysis in ecosystems like inselbergs, where fluxes of energy, carbon, and nutrients are still poorly understood. Given the affordable and repeatable nature of UAV, these surveys facilitate the assessment of rapid temporal changes in habitats and potential for seasonal changes to be recorded and monitored, and we anticipate that UAVs will become a standard tool used by biologists for targeted fine-scale biomonitoring.

Taken together, our findings demonstrate the potential for UAVs to monitor ecological habitat conditions in ecosystems that show the strong small-scale geographical variation that would be impossible to map using the ground-level methods or satellite imagery currently available. Therefore, we are confident the developed methodological framework and data processing pipeline are general and can support high-spatial-resolution spatial and temporal ecological monitoring studies in other ecosystems. The confirmed idea that inselberg microhabitats such as rock pools, which are classical models in ecology and evolutionary biology (Brendonck et al., 2010), can be connected via complex dendritic network challenges ecologists to revisit the processes that are responsible for the persistence of biodiversity in this ecosystem.

Supplemental material

Supplementary_document - Mapping microhabitat structure and connectivity on a tropical inselberg using UAV remote sensing

Supplementary_document for Mapping microhabitat structure and connectivity on a tropical inselberg using UAV remote sensing by Ángela Aristizábal-Botero, David Páez-Pérez, Emilio Realpe and Bram Vanschoenwinkel in Progress in Physical Geography: Earth and Environment

Footnotes

Acknowledgements

We want to thank the Omacha Foundation () for providing access to Bojonawi Natural Reserve and for facilitating transportation.

Data availability statement

Some of the datasets generated and/or analyzed during the current study are included in this published article and its supplementary information files. Additional datasets are available from the corresponding author on a reasonable request.

Supplemental material

Supplemental material for this article is available online.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the University of Los Andes (Universidad de los Andes) and Colciencias (Departamento Administrativo de Ciencia, Tecnología e Innovación, Colombia) by grant number 647 for National Doctoral programs.

ORCID iDs

Ángela Aristizábal-Botero

David Páez-Pérez

References

Aguilera

Sato

Llambías

, et al. (2014) Erosion surface and granitic morphology in the Sierra de Lihuel Calel, Province of La Pampa, Argentina. In: Rabassa

Ollier

(eds) Gondwana Landscapes in Southern South America. Dordrecht: Springer, 393–422.

Ajayi

Salubi

Angbas

, et al. (2017) Generation of accurate digital elevation models from UAV acquired low percentage overlapping images. International Journal of Remote Sensing 38(8–10): 3113–3134.

Altermatt

Ebert

(2010) Populations in small, ephemeral habitat patches may drive dynamics in a Daphnia magna metapopulation. Ecology 91(10): 2975–2982.

Anderson

Westoby

James

(2019) Low-budget topographic surveying comes of age: Structure from motion photogrammetry in geography and the geosciences. Progress in Physical Geography: Earth and Environment 43(2): 163–173.

Baena

Moat

Whaley

, et al. (2017) Identifying species from the air: UAVs and the very high spatial resolution challenge for plant conservation. PLOS ONE 12(11): 1–21.

Basso

de Freitas

(2020) A UAV guidance system using crop row detection and line follower algorithms. Journal of Intelligent & Robotic Systems 97(3): 605–621.

Bonilla-P

Frantz

Charão-Marques

, et al. (2013) Petrografía, Geoquímica y Geocronología del Granito de Parguaza en Colombia. Boletín de Geología 35(2): 83–104.

Boon

Greenfield

Tesfamichael

(2016) Wetland assessment using unmanned aerial vehicle (UAV) photogrammetry. ISPRS The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences 41(1): 781–788.

Bornhardt

(1900) Zur Oberflächengestaltung und Geologie Deutsch-Ostafrikas. Dietrich Reimer Verlag 7(1).

10.

Brendonck

Jocque

Hulsmans

, et al. (2010) Pools “on the rocks”: Freshwater rock pools as model system in ecological and evolutionary research. Limnetica 29(1): 25–40.

11.

Brendonck

Jocqué

Tuytens

, et al. (2015) Hydrological stability drives both local and regional diversity patterns in rock pool metacommunities. Oikos 124(6): 741–749.

12.

Brown

(1966) Decentering distortion of lenses. Photogrammetric Engineering and Remote Sensing 32(3): 444–462.

13.

Büdel

Becker

Follmann

, et al. (2000) Algae, Fungi, and Lichens on Inselbergs. Berlin, Heidelberg: Springer.

14.

Buschke

Esterhuyse

Kemp

, et al. (2013) The dynamics of mountain rock pools – are aquatic and terrestrial habitats alternative stable states? Acta Oecologica 47(1): 24–29.

15.

Campbell

(1997) Granite landforms. Journal of the Royal Society of Western Australia 80(3): 101–112.

16.

Castillo-Escrivà

Aguilar-Alberola

Mesquita-Joanes

(2017) Spatial and environmental effects on a rock-pool metacommunity depend on landscape setting and dispersal mode. Freshwater Biology 62(6): 1004–1011.

17.

Colomina

Molina

(2014) Unmanned aerial systems for photogrammetry and remote sensing: A review. ISPRS Journal of Photogrammetry and Remote Sensing 92(1): 79–97.

18.

Dandois

Ellis

(2010) Remote sensing of vegetation structure using computer vision. Remote Sensing 2(4): 1157–1176.

19.

De Paula

LFA

Forzza

Neri

, et al. (2016) Sugar Loaf Land in South-eastern Brazil: A centre of diversity for mat-forming bromeliads on inselbergs. Botanical Journal Linnean Society 181(3): 459–476.

20.

Díaz-Merlano

Montes Arango

Marcés Guerrero

(2015) El Escudo Guayanés en Colombia. Un mundo perdido (1st Ed.). Colección Ecológica Banco de Occidente. iM Editores LTDA, Bogotá, Colombia. Available at: www.imeditores.com/banocc/guayanes/ (accessed 10 July 2019).

21.

Fryer

Brown

(1986) Lens distortion for close-range photogrammetry. Photogrammetric Engineering and Remote Sensing 52(1): 51–58.

22.

Gamon

Field

Goulden

, et al. (1995) Relationships between NDVI, canopy structure, and photosynthesis in three Californian vegetation types. Ecological Applications 5(1): 28–41.

23.

Griffiths

Burningham

(2019) Comparison of pre- and self-calibrated camera calibration models for UAS-derived nadir imagery for a SfM application. Progress in Physical Geography: Earth and Environment 43(2): 215–235.

24.

Heino

Alahuhta

Ala-Hulkko

, et al. (2017) Integrating dispersal proxies in ecological and environmental research in the freshwater realm. Environmental Reviews 25(3): 334–349.

25.

Huang

Zhang

, et al. (2020) Remote measurement of vital signs for unmanned search and rescue vehicles. In 2020 5th International Conference on Control and Robotics Engineering (ICCRE). Osaka, Japan: IEEE, pp. 164–168.

26.

Hulsmans

Vanschoenwinkel

Pyke

, et al. (2008) Quantifying the hydroregime of a temporary pool habitat: A modelling approach for ephemeral rock pools in SE Botswana. Ecosystems 11(1): 89–100.

27.

Jenson

Domingue

(1988) Extracting topographic structure from digital elevation data for geographic system analysis. Photogrammetric Engineering and Remote Sensing 54(11): 1593–1600.

28.

Karnieli

Kidron

Glaesser

, et al. (1999) Spectral characteristics of cyanobacteria soil crust in semiarid environments. Remote Sensing of Environment 69(1): 67–75.

29.

Kindermann

(2016) From lake to sand — the archaeology of Farafra Oasis, Western Desert, Egypt. Azania: Archaeological Research in Africa 51(2): 282–284.

30.

Main

(2000) Habitat template for invertebrates on granite outcrops. Journal of the Royal Society of Western Australia 83(3): 139–147.

31.

Malahlela

Oliphant

Tsoeleng

, et al. (2018) Mapping chlorophyll-a concentrations in a cyanobacteria- and algae-impacted Vaal Dam using Landsat 8 OLI Data 8. South African Journal of Science 114(9–10): 1–9.

32.

Merlijn

Vanschoenwinkel

Brendonck

(2010) Freshwater rock pools: A review of habitat characteristics, faunal diversity and conservation value. Freshwater Biology 55(8): 1587–1602.

33.

Nex

Remondino

(2014) UAV for 3D mapping applications: A review Applied Geomatics 6(1): 1–5.

34.

Pettorelli

Ryan

Mueller

, et al. (2011) The normalized difference vegetation Index (NDVI): Unforeseen successes in animal ecology. Climate Research 46(1): 15–27.

35.

Pinceel

Vanschoenwinkel

Hawinkel

, et al. (2017) Aridity promotes bet hedging via delayed hatching: A case study with two temporary pond crustaceans along a latitudinal gradient. Oecologia 184(1): 161–170.

36.

Porembski

Barthlott

(1996) Plant Species Diversity of West African Inselbergs. Dordrecht, the Netherlands: Springer.

37.

Porembski

Barthlott

(2000) Inselbergs: Biotic Diversity of Isolated Rock Outcrops in Tropical and Temperate Regions. Berlin, Heidelberg: Springer-Verlag.

38.

Porembski

Martinelli

Ohlemüller

, et al. (1998) Diversity and ecology of saxicolous vegetation mats on inselbergs in the Brazilian Atlantic Rainforest. Divers Distributions 4(3): 107–119.

39.

R Development Core Team (2008) R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing.

40.

Rascher

Lakatos

Büdel

, et al. (2003) Photosynthetic field capacity of cyanobacteria of a tropical inselberg of the Guiana Highlands. European Journal of Phycology 38(3): 247–256.

41.

Rossi

Mancini

Dubbini

, et al. (2017) Combining nadir and oblique UAV imagery to reconstruct quarry topography: Methodology and feasibility analysis. European Journal of Remote Sensing 50(1): 211–221.

42.

Rouse

Jr Haas

Schell

, et al. (1974) Monitoring vegetation systems in the Great Plains with ERTS. NASA Special Publication 1(10-14): 309–317.

43.

Sarthou

Pavoine

Gasc

, et al. (2017) From inselberg to inselberg: Floristic patterns across scales in French Guiana (South America). Flora 229(1): 147–158.

44.

Scholefield

Morton

McShane

, et al. (2019) Estimating habitat extent and carbon loss from an eroded northern blanket bog using UAV derived imagery and topography. Progress in Physical Geography: Earth and Environment 43(2): 282–298.

45.

Strong

Burnside

Llewellyn

(2017) The potential of small-unmanned aircraft systems for the rapid detection of threatened unimproved grassland communities using an enhanced normalized difference vegetation index. PLOS ONE 12(10): 1–16.

46.

Tang

Shao

(2015) Drone remote sensing for forestry research and practices. Journal of Forestry Research 26(4): 791–797.

47.

Tarboton

Bras

Rodriguez-Iturbe

(1991) On the extraction of channel networks from digital elevation data. Hydrological Processes 5(1): 81–100.

48.

Timms

(2013) Geomorphology of pit gnammas in Southwestern Australia. Journal of the Royal Society of Western Australia 96(1): 7–16.

49.

Timms

Rankin

(2016) The geomorphology of gnammas (weathering pits) of Northwestern Eyre Peninsula, South Australia: Typology, influence of haloclasty and origins. Transactions of the Royal Society of South Australia 140(1): 28–45.

50.

Tuytens

Vanschoenwinkel

Waterkeyn

, et al. (2014) Predictions of climate change infer increased environmental harshness and altered connectivity in a cluster of temporary pools. Freshwater Biology 59(5): 955–968.

51.

Vanschoenwinkel

Buschke

Brendonck

(2013) Disturbance regime alters the impact of dispersal on alpha and beta diversity in a natural metacommunity. Ecology 94(11): 2547–2557.

52.

Vanschoenwinkel

Gielen

Vandewaerde

, et al. (2008) Relative importance of different dispersal vectors for small aquatic invertebrates in a rock pool metacommunity. Ecography 31(5): 567–577.

53.

Vanschoenwinkel

Hulsmans

De Roeck

, et al. (2009) Community structure in temporary freshwater pools: disentangling the effects of habitat size and hydroregime. Freshwater Biology 54(7): 1487–1500.

54.

Watts

Ambrosia

Hinkley

(2012) Unmanned aircraft systems in remote sensing and scientific research: classification and considerations of use. Remote Sensing 4(6): 1671–1692.

55.

Withers

(2000) Overview of granite outcrops in Western Australia. Journal of the Royal Society of Western Australia 83(3): 103–108.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.26 MB