Identification of Vegetation Surfaces and Volumes by Height Levels in Reservoir Deltas Using UAS Techniques—Case Study at Gilău Reservoir, Transylvania, Romania

Abstract

The hydrophilic vegetation from reservoir deltas sustains rapid expansions in surface and important increases in vegetal mass against a background of a significant influx of alluvium and nutrients from watercourses. It contributes to reservoir water quality degradation and reservoir silting due to organic residues. In this paper, we propose an evaluation method of two-dimensional and three-dimensional parameters (surfaces and volumes of vegetation), using the combined photogrammetric techniques from the UAS category. Raster and vector data—high-resolution orthophotoplan (2D), point cloud (pseudo-LIDAR) (3D), points that defined the topographic surface (DTM—Digital Terrain Model (3D) and DSM—Digital Surface Model (3D))—were the basis for the realization of grid products (a DTM and DSM, respectively). After the successive completion of the operations within the adopted workflow (data acquisition, processing, post-processing, and their integration into GIS), after the grid analysis, the two proposed variables (topics) of this research, respectively, the surface of vegetation and its volume, resulted. The data acquisition area (deriving grids with a centimeter resolution) under the conditions of some areas being inaccessible using classical topometric or bathymetric means (low depth, the presence of organic mud and aquatic vegetation, etc.) has an important role in the reservoirs’ depth dynamics and reservoir usage. After performing the calculations in the abovementioned direction, we arrived at results of practical and scientific interest: Cut Volume = 196,000.3 m³, Cut 2D Surface Area = 63,549 m², Fill Volume = 16.59998 m³, Fill 2D Surface Area = 879.43 m², Total Volume Between Surfaces = 196,016.9 m³. We specify that this approach does not aim to study the vegetation’s diversity but to determine its dimensional components (surface and volume), whose organic residues participate in mitigating the reservoir functions (water supply, hydropower production, flash flood attenuation capacity, etc.).

1. Introduction

UAS techniques are implemented in more and more strategic and other areas of the civil sector, from the supervision of natural and anthropogenic risk events to territorial management and the delivery of products for commercial purposes, the latter being a new trend in drone utilization [1,2,3,4,5,6,7].

UAS technologies have been confirmed useful in the contemporary topometric context of field research. Thus, the use of this technology brings, on the one hand, extra accuracy, and, on the other hand, allows access in areas difficult to access (or areas inaccessible to classical topometric technology) [5,8,9,10,11,12,13].

Smaller aircrafts (UAVs) help to conduct field research, in the areas hard to reach for classic aircrafts, as they ensure independence, ease of operation, and very good accuracy (if properly equipped and calibrated), allowing a significant reduction in working time [5,10,14,15,16].

Numerous studies involving UAS techniques have been dedicated to natural hazards and risks in general, and hydrological ones in particular, as these phenomena need to be monitored and managed under the existing conditions of amplified climate change effects [17,18,19,20,21,22,23,24].

Performing several flights during flood periods facilitates good monitoring of flooded areas, sustaining optimal decisions in managing the post-flood situation [25,26,27,28].

Other profile studies are focused on various hydraulic models built from cartographic resources or medium- or high-resolution aerial images [9,29,30,31].

The implementation of UAS techniques is increasingly developing in the fields of resource prospecting, construction, technological management, and occupational safety due to the provided opportunities [32,33,34].

On the other hand, numerous applications of light flight techniques are implemented in forestry and agriculture, vegetation monitoring [35,36,37,38,39,40,41,42], as well as in the evaluation of forests’ status, implying the use of high-performance electro-optical sensors for near-infrared images [43].

The studies that have involved satellite and UAS technology have effectively focused on environmental components, including plant and wildlife habitats, wetlands, etc., have mainly been conducted in recent years, being targeted on current directions of interest [44,45,46,47,48,49].

The topometric products derived from the use of this technology can be classified (in our case) into two broad categories [50,51]:

-

Products providing 2D planimetric information (x, y) (example: orthophotoplan, etc.).

-

Products providing 3D altimetric information (x, y, z) (DEM, DSM, 3D model, etc.).

The accuracy of the final products will be determined, on the one hand, using the photometric equipment used for data acquisition (camera, sensor, etc.) and, on the other hand, by the optimal planning of the flight mission. Last but not least, it is very important to process the data knowingly (photogrammetric procedures and software used) [9,10,11,12,13,23,42,52,53].

Finally, the data integration into the GIS environment to capitalize the research results, followed by spatial analysis, represents the last research step.

The confluence, or the discharge areas in general and the deltaic areas in particular, have some topometric, geological, and geomorphological specifications:

-

They are hard-to-reach areas (sometimes inaccessible);

-

They are very dynamic areas, as from the point of view of the substrate they are the result of the continuous sedimentation–erosion game.

Their monitoring is more difficult precisely because of this dynamism but also owing to the difficult direct access caused by the presence of swampy terrain and land in formation and consolidation. Under these circumstances, UAS techniques become extremely useful and efficient [54,55].

Also, the huge amounts of seasonal vegetation, which rapidly and consistently grow on the substrate of recent sediments and unconsolidated slime, loaded with nutrients, annually increase the amount of organic mud, which contributes to the deterioration of the water quality, accelerating the silting of the lake sector [56,57].

Based on these considerations, this article proposes a method of precise determination of the two- and three-dimensional parameters (areas and volumes of vegetation) using combined photogrammetric techniques such as UAS.

Our methodological approach starts from a very simple logic. Once the resulting DTM and DSM grids are integrated into a GIS environment, they can be compared. Therefore, from the grid representing the DSM, we subtracted the grid representing the DTM. Consequently, the current volume of vegetation is obtained since only vegetation is present in the area of interest (AOI), without any other anthropogenic interference. We underline that this approach does not aim to study the diversity of the vegetation but to determine its dimensional metric components (surface and volume), whose organic residues participate in mitigating the reservoir functions (water supply, hydro-power production, flash flood attenuation capacity, etc.). We mention that, in our opinion, the contribution of this approach is substantial in areas inaccessible using classical topometric means.

The Study Area

The Somesul Rece Delta has developed at the confluence of the two arranged watercourses (Someșul Cald and Someșul Rece), which form the Someșul Mic River at the exit of the Apuseni Mountains (Transylvania, Romania), under the effect of the combined alluvial transport of rivers and of the Gilău reservoir silting (Figure 1) [8,9,57].

Starting from the year 1972, the year of it being put into use, the Gilău reservoir was used as the main surface source for the water supply of almost an entire county (Cluj County—currently 6.674 km², representing 2.8% of Romanian territory), resulting in its strategic importance for the mentioned area [57].

2. Materials and Methods

For the thematic framing of the study, specialized literature was used, respectively, the Web of Science and ScienceDirect databases, and also the national bibliographic support provided by the Babeș-Bolyai University, e.g., studies, maps, and other materials, etc.

For general cartographic support, the Digital Elevation Model of Europe (EU-DEM), 2017, and Topographic Map of Romania, 1:25,000, 1978–1982, were used, together with our own observations of the hydrographic basin afferent to the study area.

For the processing of the general and thematic information, software purchased by Babeș-Bolyai University and the Faculty of Geography, as well as software related to the used equipment and proprietary or free/open source software, was used: Microsoft Excel 2016, ESRI ArcGIS 10.8.1, DJI GO 4, Pix4Dcapture, Agisoft Metashape, and Global Mapper.

For the purpose of this research (determinations of surfaces and volumes related to the vegetation developed in the study area), a methodology inspired by the principles of digital photogrammetry assisted using UAS was used. Therefore, the grid representing the DTM was subtracted from the grid representing the DSM. Consequently, the volume of vegetation was obtained, since only vegetation is present in the area of interest (AOI), without other anthropogenic interference.

A commercial Phantom 4 Pro drone, produced by the DJI manufacturer, was used to acquire the photogrammetric data (Figure 2).

This device belongs to the category of microdrones [57], and its characteristics can be observed on the manufacturer’s website [61,62].

Compared to other microdrones, the passive sensor (camera) of the Phantom 4 Pro has a mechanical trigger (shooter) and not an electronic one. This fact leads to the elimination of pillow-type distortions, which appear on the periphery of the captured scenes. Another important feature is the 1-inch sensor, which guarantees a real 20 MP resolution.

Calibration of the Phantom 4 Pro camera, regarding the specific coefficients, led to the following results (

Table 1. Calibration coefficients and correlation matrix.

P2	P1	K3	K2	K1	Cy	Cx	F	Error	Value
0.04	−0.14	0.17	−0.22	0.13	−0.2	−0.07	1	4.2	2433.9	F
−0.1	0.6	−0.03	0.03	0.08	−0.1	1		0.094	−36.81	Cx
0.63	−0.06	−0.03	0.05	−0.11	1			0.1	−21.06	Cy
−0.17	0.1	0.47	−0.54	1				0.00019	−0.002	K1
0.01	0.06	−0.98	1					0.00018	−0.005	K2
0	−0.07	1						0.00011	0.0018	K3
−0.08	1							7.10 × 10−6	−5 × 10−4	P1
1								8.30 × 10−6	−2 × 10−4	P2

).

Figure 3 shows that the effect of calibration is the relatively balanced distribution of errors on the surface of the captured image, which avoids their massive concentration in the marginal area, as well as in the ordinary photos.

Following the characteristics of the drone objective (Phantom 4 Pro type camera), we conditioned an 85% front lap and a 75% side lap, to ensure a sufficient density of information to obtain a high-accuracy model (Figure 4a,b).

Regarding the measurements and further data processing, the working methodology involved the following approach (

Table 2. The succession of operations within the adopted workflow.

Resulted Formats	Software	Process or Activity	Stage
	-	Equipment preparation	Data acquisition
	-	Checking the weather–climate context
	-	Flight corridor separation
	DJI GO 4	Sensor testing and verification
	DJI GO 4	Camera setup
	Pix4Dcapture	Mission route planning
	Pix4Dcapture	Actual flyby
Photograms (*.jpg 20MP) (5472/3078)	-	Data downloading
Validated photograms (*.jpg 20MP) (5472/3078)	-	Photogram filtering and validation
Validated photograms (*.jpg 20MP) (5472/3078)	-	Photogram filtering and validation	Data processing
*.las; *.laz	Agisoft Metashape	Simple point cloud
*.las; *.laz	Agisoft Metashape	Densified point cloud
*.las; *.laz	Global Mapper	Classification of points
*.tiff; *.jpg	Agisoft Metashape	Creating the orthophoto map
*.grd	Global Mapper	Creating the elevation models
*.collada; *.dae	Agisoft Metashape	Creating the 3D model (mesh)
Areas (m2); Volumes (m3)	Global Mapper	Grid operations	Post-processing and integration in GIS

).

Starting with the stage of mission preparation, the existence in the south–southeast part of the studied perimeter of a network of high-voltage power lines was taken into account, which conditioned the flight at a relative altitude of approximately 130 m. Thus, electromagnetic interference was avoided, as it can influence the RC equipment [57].

The accuracy of the inputs used in spatial analysis is very important. To check the reliability of the products used in the analysis, it was necessary to derive two additional products. Consequently, we used a synthetic NDVI (known as a “false NDVI”).

This indicator is based on the principle that it can be derived using RGB scenes. Compared to the classic NVDI, obtaining this indicator is much easier, as it does not require a multispectral camera. Some specialists say that the accuracy of the synthetic NDVI is not as high as the accuracy of the classic NDVI (which also includes the NIR band), but in terms of efficiency, this indicator is very appropriate.

The synthetic NDVI is based on the VARI (Visible Atmospherically Resistant Index) algorithm, expressed by Equation (1)

VARI = (Green − Red)/(Green + Red − Blue),

(1)

3. Results

After the flight, the data were downloaded to a computer and processed, obtaining several derived products.

Based on the acquired images in the visible spectrum, the raster data results were as follows:

-

High-resolution orthophoto plan (2D);

-

Point cloud (pseudo-LIDAR) (3D).

Based on the point cloud classification, we were able to separate at least two major categories:

-

Points that define the topographic surface (DTM—Digital Terrain Model) (3D);

-

Points that define the surface occupied by the elements above the topographic one (including vegetation and anthropogenic objectives) (DSM—Digital Surface Model) (3D).

Based on these raster products derived from the point cloud, grid-type products can be generated, which are suitable for operations in a GIS environment.

Consequently, after we obtain the two grids (the DTM and DSM, respectively), the analysis of the grid will show the two dimensions: the planimetric surface occupied by vegetation and the volume, respectively.

The spatial distribution of the residues for the study area presents a favorable configuration for the development of advanced processing stages of the materials obtained after the overflight (Figure 5):

As a component of the results, the average values of the estimated errors are within reasonable limits (

Table 3. Average camera location error.

Total Error (m)	XY Error (m)	Z-Altitude Error (m)	Y-Latitude Error (m)	X-Longitude Error (m)
1.91775	1.2804	1.43	1.11	0.631441

).

During the processing of the photograms, the following conditions were rigorously adhered to:

-

Coplanarity condition;

-

Relative orientation;

-

Absolute orientation.

The simple point cloud was the first product derived from this processing. This product is the framework for subsequent planimetric and altimetric determinations and for the densified point cloud. Each point has Cartesian information (x, y, z) and color information (RGB) (Figure 6).

Based on the densified point cloud product, the DSM surface was generated using a micro-triangulation process.

To characterize the status of the vegetation in the studied area, we used two more raster products: the high-resolution orthophotoplan and the DTM (Digital Terrain Model).

Determining the DTM in this way involves returning to the densified point cloud and performing an automatic classification of the existing points.

As a component of the results, following this classification, we were able to separate the points that referred to constructions, land, high vegetation, low vegetation, and roads. The spatial distribution of these distinct points is shown in Figure 7.

Once the spatial configuration of each class is obtained, it is possible to obtain the DTM using micro-triangulation of the specific terrain points (Figure 8).

A high-resolution orthophotoplan of the studied area was derived as well (Figure 9).

Our approach started with the assumption that is possible to assess quantitatively the vegetation developed at the river mouth of the Someșul Rece, where the alluvial deposits have formed a real lake delta over time. For this purpose, because of the difficult conditions in the field (listed in the introductory part), we used UAS technology and the photogrammetric study type.

Similar approaches have been developed by some authors [63,64], but they have generally focused on obtaining photogrammetrically derived products (orthophotoplan, DSM, DTM), without exploiting these products in thematic applications (e.g., obtaining the NDVI, surface and volume of biomass, etc.).

As can be seen in Figure 10, the model’s reliability is low at the water surface (only a few percent) and of high and very high quality in the dry and semi-dry areas of the mouth.This was fully anticipated, knowing that the water surface is one of the most difficult things to model in 3D using photogrammetric techniques due to the high degree of reflectance and due to its excessive mobility (waves, currents, etc.).

Given these considerations, we presented several other conditions.

In the first phase, we selected the actual confluence area, which we delimited strictly on the boundary of the land surface to proceed with the grid operations. In this way, the analysis gained accuracy, and the inputs could be integrated into a GIS environment (Figure 11).

Because the distribution of vegetation is relatively easy to determine based on the high-resolution orthophotoplan and the synthetic NDVI, we also ensured a vertical “scan” of the vegetation. This was made possible by overlapping the previously determined elevation models (the DSM and DTM, respectively).

Subsequently, the elements could be highlighted by analyzing the overlapping profiles. Upon careful analysis, the height of the trees located in the profile alignment can be discerned.

Using extrapolation, we applied the same principle, but, this time, to the whole area of interest. In this sense, we compared the two surfaces: respectively, the DSM surface assimilated with the vegetation, and the DTM surface assimilated with the land (Figure 12).

After performing the calculations in the abovementioned direction, we arrived at results of practical and scientific interest (

Table 4. Dimensional elements of emerged vegetation at Gilău Lake delta topo surface.

Element	Dimension
Total Volume between Surfaces	196,016.9 m3
Total Surface	64,428.43 m2
Cut Volume	196,000.3 m3
Cut 2D Surface	63,549 m2
Fill Volume	16.59998 m3
Fill 2D Surface	879.43 m2

and Figure 13).

4. Discussion

The vegetation of the deltas developed in reservoirs is of less interest, or even of negative interest, to institutional water administrators, as their primary interest is the good quality of water and the low silting rate of these water bodies, to capitalize for as long as possible and at a high efficiency on their capacity (useful volume) [57].

Or, as could be seen previously, these organic residues, together with their alluvium quantities, participate in mitigating reservoir functions (water supply, hydro-power production, flash flood attenuation capacity, etc.) [56,57]. However, the vegetation contributes to the retention of a great part of these significant quantities of alluvium transported by flash floods, and alluviums that contribute to the evolution/growth of the deltas.

On the other hand, the vegetation, especially when developed/tall, fully contributes to an increase in biodiversity in the delta area, favoring the development of more and more complex ecosystems and an increase in the occupied space. In the case of the Gilău reservoir delta, species unknown before began to appear before the development of the sedimentation space, even if temporarily (pelicans, swans, etc.), and seagulls, wild ducks, etc. have been registered for decades at the current site.

Compared to these previously mentioned elements, some authors have focused on studying several aspects related to wetlands, such as, for example, the soil conditions, as support for the evolution of the vegetation. Visible-near-infrared (Vis-NIR) spectroscopy has been considered an alternative to building spectral libraries, which couple soil data and Vis-NIR spectra using models. These models can provide optimal predictions for topsoil expansion [65].

Other authors have considered the accurate prediction of clay to be the basis for the assessment of soil quality because it governs the soil moisture and fertility dynamics, with an effect on the development of vegetation, including from wetlands, in the primary soils of the silt category. Research results have shown that γ-ray data were the most important for the prediction of topsoil, while in the subsoil, the slope was the most important [66].

Extensive wetlands play a crucial role in the global carbon cycle. Aboveground vegetation biomass (AGB) is a critical indicator of carbon storage in wetland ecosystems [67]. Some authors, using the normalized difference vegetation index (NDVI), climate data, and measured AGB data, have investigated the temporal and spatial variation in marsh AGB and its response to climate change, with the results demonstrating a good correlation between the AGB and annual maximum NDVI (NDVImax) [67].

Vegetation activity and phenology are significantly affected by climate change, and changes in vegetation activity and phenology can, in turn, affect local, regional, and global climate patterns. Using the observation minus reanalysis (OMR) method, some authors have investigated the possible effects of vegetation activity and changes in the growing season on the air temperature of the respective areas [68]. The results have showed that the average NDVI of certain vegetated areas increased significantly, and the growing season started earlier and ended later, resulting in its extension. The results suggest that climate-change-induced increases in vegetation activity and extended growing seasons may further exacerbate regional warming [68].

Wetlands not only affect local hydrology and ecosystems but also regulate human environmental conditions [69]. Some authors have demonstrated that wetland mapping, which is based on Google Earth Engine (GEE), feature optimization, and a random forest (RF) model (GFORF), provides new perspectives on studying them. The results obtained showed that the overall accuracy for wetland data was 82.73%, 83.16%, 82.47%, and 88.14% across the interval 1990–2010 [69]. As a result of improving the accuracy of wetland mapping, it is possible to identify the dynamics of wetlands, which justifies the implementation of ecological services and protection measures in this category of study [69].

Various studies have shown that forest biomass is a key biophysical parameter for studying climate change, ecological modeling, and forest management [70]. Compared with discrete return LiDAR data, full-waveform LiDAR data can provide more accurate and abundant information on the vertical structure of vegetation and thus have been increasingly applied to forest aboveground biomass (AGB) estimation [70].

Estimating the structural attributes of forests is crucial for their sustainable management and helps us to understand their contributions to global carbon storage [71]. Unmanned Aerial Vehicle Light Detecting and Ranging (UAV LiDAR) has become a promising technology and is being tested for use for forest management due to its ability to provide highly accurate estimates of three-dimensional (3D) structural information at a lower cost, greater flexibility, and finer resolution than airborne LiDAR [71].

Individual tree detection based on a combination of Unmanned Aerial Vehicle Light Detecting and Ranging (UAV LiDAR) with Backpack LiDAR is another application of airborne technology, with high potential for accurate forest AGB estimation [55].

Other authors have tried to analyze the error range and resolution of drone images using a rotary wing, comparing them with the results of field measurements, and to analyze the stand patterns in the actual preparation of a vegetation map, comparing drone images with the aerial images provided by national specialized institutes [63]. The spatial analyses carried out concluded that the drone images made more accurate polygons than the 51 and 25 cm resolution images provided by the mentioned institutes [63].

The confluence of discharge areas in general, and the deltaic areas of the type related to the Gilău reservoir in particular, is hard to reach (sometimes inaccessible) and dynamic, as, from the point of view of the substrate, they are the result of a continuous sedimentation–erosion game. Monitoring them is more difficult precisely because of this dynamism but also owing to the difficult direct access caused by the presence of swampy terrain and land in formation and consolidation. It is due to these reasons that UAS techniques become extremely useful and efficient.

This type of area, located at the confluence of rivers and reservoirs, stores quantities of water that are difficult to quantitatively and qualitatively assess, recoverable for everyday use (water supply), hydropower, or involved in the process of the attenuation of flood waves (on this altimetric scale, the tranche of the characteristic volume of the same name is present).

Emerged and submerged seasonal vegetation, which grows rapidly and consistently on the substrate built from recent sediments, full of nutrients and unconsolidated mud, annually increases the amount of organic mud, which contributes to deteriorating the water quality and accelerating the silting of the lake sector.

The method proposed in this article is an obvious one in the low-cost category, with extremely easy applicability and which requires a relatively low workload. The technological flow of obtaining the finished product involves the automatic collection of aerial images, automatic downloading to the digital environment, processing using photogrammetric software, and obtaining derived products using the different models established in the specialized literature [8]:

-

3D reconstruction process based on UAV technology (drone) and the interpolation algorithm ‘‘Daisy’’ is cheap, relying on open-source solutions;

-

The accuracy of 3D reconstruction (5 cm) is much higher than traditional photogrammetric solutions;

-

The final product (DEM, DSM, orthophotoplan, false NDVI, vegetation grid, etc.) can be georeferenced and integrated into any GIS or CAD application;

-

This process allows an accurate qualitative and quantitative approach (distance, area, volume);

-

The procedure is of a non-invasive nature and is applicable in areas difficult to reach or inaccessible using traditional technology.

Some disadvantages of applying this technology are as follows:

-

Flights cannot be executed in conditions of winds over 60 km/h and in unfavorable light conditions; for this study, the flight was executed in calm weather conditions;

-

The flight autonomy is relatively low on the battery unit (under 40 min) to avoid system collapse and sustaining significant damage; instead, more batteries can be bought to change them every 30 min;

-

A low environment temperature is unfavorable, resulting in faster battery consumption;

-

Significant hardware resources are required in processing the data: in this case, a computer system including but not limited to an Intel i7/AMD Ryzen processor, 32 GB RAM, a video card of 16 GB, and a 2 TB SSD.

5. Conclusions

Following the proposed topic (determination of areas and volumes related to the vegetation developed in the study area), a methodology inspired by the principles of digital photogrammetry, assisted by UAV, was used. The succession of operations within the adopted workflow was successfully followed.

After the field stage and after the downloading and processing of data, several high-resolution derivative products were obtained, vector (dot) and raster (orthophotoplan, grid) products, subsequently integrated into the GIS environment.

As an additional element of validation and verification, an NDVI was derived to identify the status of vegetation and its distribution in the plan (2D distribution of information), followed by checking the reliability of the model resulting from the research (3D distribution of information).

The objective of the research was achieved by comparing, following the spatial analysis, the two surfaces—the surface assimilated to the vegetation (DSM) and the surface assimilated to the terrain (DTM), including at the level of transverse profiles over the delta area. The results reveal the values of the final parameters (volumes and areas of the emerged vegetation). Beyond these values, we expect the development of the methodology, which will make possible the calculation of the vegetal mass developed in this area based on its density.