Next Article in Journal
Integration and Detection of a Moving Target with Multiple Beams Based on Multi-Scale Sliding Windowed Phase Difference and Spatial Projection
Previous Article in Journal
Assessment of Variability and Attribution of Drought Based on GRACE in China from Three Perspectives: Water Storage Component, Climate Change, Water Balance
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Satellite Imagery-Estimated Intertidal Seaweed Biomass Using UAV as an Intermediary

1
College of Marine Ecology and Environment, Shanghai Ocean University, Shanghai 201306, China
2
Key Laboratory of Marine Environmental Survey Technology and Application, Ministry of Natural Resources, Guangzhou 510310, China
3
Key Laboratory of Tropical Marine Ecosystem and Bioresource, MNR, Fourth Institute of Oceanography, Ministry of Natural Resources, Beihai 536007, China
4
Key Laboratory of Marine Ecological Monitoring and Restoration Technologies, MNR, East China Sea Environmental Monitoring Center, Shanghai 201206, China
*
Author to whom correspondence should be addressed.
These authors contributed equally to this work.
Remote Sens. 2023, 15(18), 4428; https://doi.org/10.3390/rs15184428
Submission received: 15 August 2023 / Revised: 31 August 2023 / Accepted: 5 September 2023 / Published: 8 September 2023

Abstract

:
The aim of this study was to use unmanned aerial vehicles (UAVs) as a supplement to satellite remote sensing to accurately assess benthic seaweed biomass in intertidal zones, in order to improve inversion accuracy results and investigate the spatial distribution patterns of seaweed. By adopting non-multicollinearity vegetation indices (feature sets) from PlanetScope and Sentinel-2, and using benthic seaweed biomass inverted from multispectral UAV imagery as the label set for satellite pixel biomass values, machine learning methods (Gradient boosting decision tree, GBDT) can effectively improve the accuracy of biomass estimation results for Ulva pertusa and Sargassum thunbergii species (Ulva pertusa, R S e n t i n e l 2 2 = 0.74, R P l a n e t S c o p e 2 = 0.8; Sargassum thunbergii, R S e n t i n e l 2 2 = 0.88, R P l a n e t S c o p e 2 = 0.69). The average biomasses of Ulva pertusa and Sargassum thunbergii in the intertidal zone of Gouqi Island are 456.84 g/m2 and 2606.60 g/m2, respectively, and the total resources are 3.5 × 108 g and 1.4 × 109 g, respectively. In addition, based on the hyperspectral data, it was revealed that a major source of error is the patchy distribution of seaweed.

Graphical Abstract

1. Introduction

Benthic seaweed is an important primary producer in offshore ecosystems [1], (and its biomass serves as a critical indicator for the research being conducted on seaweed beds [2]. It is essential for assessing ecological value, seasonal succession, and carbon storage [3,4,5]. Seaweed productivity is closely linked to climate variables [6]. However, in relation to the background of global warming, ocean acidification, and other environmental conditions, there has been evidence of a marked yearly decrease in seaweed resources along coasts worldwide [7,8,9]. Biomass indicators are thus crucial references for the sustainable development, utilization, protection, and restoration of seaweed resources. Nonetheless, quadrat sampling the method of benthic seaweed biomass research conducted in intertidal zones at present can only reveal detailed information about seaweed abundance in a small area [10,11,12], making it difficult to perform large-scale surveys effectively. Furthermore, the difficulty in the research of genuinely characterizing seaweed biomass due to tidal influence and patches distributions exists as another challenge for performing accurate seaweed biomass assessments in intertidal zones. Therefore, it is vital to accurately assess seaweed biomass over extended periods of time and across large areas to successfully evaluate the ecological values, restoration efforts, and carbon sequestration occurring in offshore ecosystems.
Remote sensing observations offer several advantages over in situ surveys, such as a broad observation range, time and cost-effectiveness. The satellite remote sensing method is especially useful for obtaining a large amount of datasets over long time spans across large areas, with applications in monitoring the vegetation, including forests [13], grasslands [14] and giant kelp habitats [15]. These features share a high degree of distribution and homogeneity characteristics. In contrast, the satellite remote sensing method is rarely employed in the research to study vegetation that has a complex composition and presents spatial heterogeneity, such as benthic seaweed growing in intertidal zones. By capturing multiple bands and spectral information concerning ground objects, even minor details can be reflected by the acquired image data [16,17]. While the resolution of satellite remote sensing technology may limit an accurate assessment of seaweed parameters, such as abundance, biomass information is considerably contributed to a few constructive species [18], giving satellite remote sensing techniques a central role in biomass estimation practices. To investigate narrow and complex intertidal zones, researchers have turned to unmanned aerial vehicles (UAVs) for classification mapping tasks [19] and the retrieval of biomass from dominant species [20]. UAV remote sensing technology is a powerfully complement satellite remote sensing technology [21,22,23], and efforts have been made in the field to integrate both technologies [24,25,26] with successful inversion results.
When selecting an inversion model to perform a biomass assessment, the most commonly used models in the field include linear or non-linear regression equations obtained by regression fitting of single or multiple vegetation indices or band reflectances with biomass and other parameters. These models are known as Single Linear Regression Models (SLRMs) or Multiple Linear Regression Models (MLRMs) [27,28]. While these models are straightforward to use, their accuracy is limited and they rely on stacking data to produce good results. In recent years, the machine learning method has emerged as a new programming paradigm that has been widely used in the research to perform biomass assessments [29,30,31].
Due to the factors of environmental and biological stress, intertidal seaweed often exhibits a patchy distribution pattern [32,33]. This type of distribution pattern can cause the seaweed to blend with other land-cover types in satellite images at insufficient spatial resolutions. However, at present, limited research exists on the spectral mixing of benthic seaweeds. Only a few studies have been conducted on floating seaweed [34,35].
Overall, evaluating the biomass of seaweed in a narrow intertidal zone using satellite remote sensing technology is a highly important and challenging task, with limited prior research being available in the literature [17,36,37]. Directly inferring the seaweed biomass in this study area through combining sampling data with satellite imagery resulted in significant errors. As such, in this study, we attempted to use the biomass information obtained via multispectral UAV as the ground truth for satellite image raster, and we conducted a reverse assessment of seaweed biomass in an intertidal zone. By incorporating hyperspectral data into our assessment, we identified the main sources of errors. Twenty spectral indices were established based on Sentinel-2’s R, G, B, edge2, 8a bands and PlanetScope’s R, G, B, edge, NIR, which constituted a label set for Halodule sarmentosa and Ulva linza, species present in the intertidal zone. GBDT was also employed to assess seaweed biomass in the intertidal zone of Gouqi Island, Zhejiang Province, China. Hybrid spectral analysis was conducted by integrating high-resolution hyperspectral data for exploring the mixing of seaweed with seascapes such as seawater, rocks, and beaches to investigate the factors associated with areas of high seaweed biomasses and their patchy spatial distribution characteristics. Lastly, based on the patchy distributions of seaweed, conversion equations and patch-based interpolation methods for estimating the seaweed coverage–biomass relationship at varying spatial scales were computed. This study included seven survey stages as shown in Figure 1: (Section 2.2) data collection; (Section 2.3) data processing; (Section 3.1) biomass analysis; (Section 3.2) statistical analysis; (Section 3.3) relationship between Vis; (Section 3.4) biomass inversion; (Section 3.5) mixed spectral analysis; (Section 3.6) biomass retrieval; (Section 4.1) distribution analysis; (Section 4.2) carbon sequestration and (Section 4.3) spatial scale analysis (Figure 1).

2. Materials and Methods

2.1. Study Area

The study was conducted in the intertidal zone of Gouqi Island, Zhejiang Province, China. Gouqi Island (30°43′1.64″N, 122°46′3.25″E) is situated in the south of Ma’an Islands, Shengsi County, Zhejiang Province, China. Gouqi Island has unique natural landforms of islands and reefs, and has the characteristics of intertidal biodiversity. Nearshore sediments are mainly composed of rocks and reefs, where many seaweed grow, live and breed, forming a complex nearshore marine ecosystem [38] (Figure 2).
In this study, Ulva pertusa and Sargassum thunbergii, which were mainly distributed in the intertidal zone, were selected as the research objects. In order to highlight the actual growth and supply values of seaweed on Gouqi Island, the near-shore area of Houtou Bay, which is located at a considerable distance from human impact and represents the investigation site, was selected as the UAV aerial photography and field sample collection area.

2.2. Data Collection

Multispectral images were collected using a DJI M300 RTK quadcopter UAV (Dà-Jiāng Innovations Science and Technology Co., Ltd., Shenzhen, China) equipped with an MS600pro multispectral camera (Dà-Jiāng Innovations Science and Technology Co., Ltd., Shenzhen, China). UAV aerial photography was conducted between 11:00 and 02:00 (UTC/GMT+08:00) from 25 to 30 June 2021. Prior to each flight, the sensor underwent calibration using an Ag-60 calibration gray plate and at least 5 ground control points (GCPs) were established for each route. UAV aerial photography was conducted at a height of 80 m with a flight speed of 5 m/s, a lateral overlap rate of 70%, and a heading overlap rate of 80%. The shore-land outward expansion distance was set at 10 m, while the sea outward expansion distance was set at 20 m according to Chen et al. [30]. To ensure that the imaging time for Sentinel-2B and PlanetScope coincided with the UAV’s lowest tide time, the relevant data were selected from the images captured at 2:00 p.m. on 29 June, 2021 for Sentinel-2B (S2B_MSIL2A, “https://scihub.copernicus.eu/ (accessed on 21 April 2023)”) and 23 June, 2021 for PlanetScope Scene “https://www.planet.com/ (accessed on 21 April 2023)”. Figure 3 displays the band information collected for the multispectral sensors used in the UAV and satellite imaging tasks, as well as the first principal component spectral curves for Ulva pertusa and Sargassum thunbergii.
Following the UAV flight, field sampling was conducted by collecting 57 samples of Ulva pertusa and Sargassum thunbergii in quadrats measuring 0.25 × 0.25 m2. After placing the quadrat, we recorded the longitude and latitude coordinates and took photos to record. We collected all target species in the quadrat. We took them back to the laboratory and used an electronic balance to measure the wet weight of specific seaweed and calculate its biomass. The wet weight of each sample was recorded, as well as the biomass, longitude and latitude of the collected Ulva pertusa and Sargassum thunbergii at each sampling point.
To collect the spectral data of the dominant intertidal seaweed species, an ASD Field Spec Handheld device was employed (Field Spec Handheld, Analytical Spectral Devices, Inc., Boulder, CO, USA). This device has a wavelength observation range of 325–1075 nm and is capable of observing both visible and near-infrared bands commonly used in vegetation research. It features a spectral sampling interval of 1 nm, a spectral resolution of 3 nm, and a field of view of 25°. Spectrometer optimization was performed every 10–15 min with dark-current acquisition occurring every 5 min. Prior to the spectral measurement, calibration against a reference whiteboard was conducted to obtain a reflectivity value of 1. The spectrometer was then directed towards the target feature for spectral data collection in real time, averaging 5 sets of data at 10 s intervals across 3 measurement points covering the lower, middle, and upper sections of the thallus [20]. The hyperspectral information of common intertidal zone surface features, such as seawater, reefs, and beaches, was collected. The dataset contained 290 spectral reflectance curves from the island.

2.3. Data Processing

For UAV data processing methods, we used the reflectance values of the Ag-60 calibration plate prior to each aircraft takeoff for the spectral calibration. We selected at least 5 ground control points (GCPs) to perform geometric correction on the multispectral images, using a first-order polynomial for the geometric correction, where n = 1 in Equations (1) and (2):
x = i = 0 n j = 0 n i a i j X i Y j
y = i = 0 n j = 0 n i b i j X i Y j
( x , y ) represents the coordinates in the original image, while (X, Y) represents the ground coordinates that have been recorded. a i j and b i j represent the coefficients of a polynomial. Finally, the image is spliced to generate an orthophoto image, and then verified. After the verification step, the geometric correction error of the image was less than 0.5 pixels.
R M S = ( x r x i ) 2 + ( y r y i ) 2
x r and y r are the original coordinates in the input reference coordinate system, while x i and y i are the transformed coordinates. Based on Equation (4), the spatial resolution of the corrected pixel value is 0.058 m × 0.058 m. The grayscale values of the image were resampled using the bilinear interpolation method.
a G S D = f h
a: Pixel size; GSD: ground resolution; f: principal distance of lens; and h: flight altitude.
Based on Figure 2, it can be observed that the satellite spectral parameters of Ulva pertusa and Sargassum thunbergii species exhibit significant variations in the 750 nm red edge and the 850 nm near-infrared bands. Therefore, to calculate the VI of Sentinel-2, Band 8a was selected as the NIR band, while the red-edge band was selected as Edge 2. The Sentinel-2 data that has been downloaded is L2A data, which refers to the bottom-of-atmosphere reflectance data that has already undergone radiometric calibration and atmospheric correction processing. After resampling through SNAP (10 m). ENVI 5.1 (Exelis VIS, Herndon, VA, USA) is used to stack the R, G, B, edge2, NIR2 (8a) bands of the processed layer. The ground sampling distance between PlanetScope pixels varies with satellite altitude, but has an average of 3.7 m, and the “analytic ortho scene” product used in this article has been resampled to a uniform spatial resolution of 3 m. We used the level-3B analytic surface reflectance product, which has undergone orthorectification and radiometric correction. In this study, the vegetation index calculated from the G1 and G bands of PlanetScope showed a high level of consistency ( R d i f f e r e n t   V I s 2 = 0.98–0.99), therefore only the G band was considered in calculating the vegetation index, while the G1 band was not taken into account. A total of 20 VIs and their calculation formulas were selected (Appendix A). According to NDWI, we extracted the intertidal zone of Gouqi Island and removed any small fragments to obtain the intertidal zone area of Gouqi Island and its surrounding islands based on the PlanetScope data [39].
Different spectral bands and VIs may be associated with the same pigments, such as the interaction between near-infrared spectral bands and the internal structure of leaves, and the interaction between visible spectra (i.e., red and green) and chlorophyll, lutein and carotene. To prevent the over-fitting of the inversion model, the expansion coefficient was used to eliminate some model parameters with strong multicollinearity results. The calculation formula is as follows:
V I F = 1 1 R 2
The average biomass value at the grid points was computed from the average of biomass values for all quadrats in the Sentinel-2 (10 m) and PlanetScope (3 m) pixel data (Figure 4). We augmented this approach with supervised classification and GBDT techniques, as outlined in the research conducted by Chen et al. [30]. This enabled us to determine the biomass that corresponded with the pixel values of Sentinel-2 or PlanetScope imagery. The retrieved biomass in our investigation was acquired by dividing the total count of pixel values by the number of pixels (refer to Figure 4 for Ulva pertusa R2 = 0.80, RMSE = 93.07 g/m2, MAE = 38.64 g/m2; Sargassum thunbergii R2 = 0.73, RMSE = 796.12 g/m2, MAE = 329.18 g/m2).
We used the GBDT algorithm in machine learning to evaluate the biomass of seaweed in the intertidal zone of Gouqi Island. The specific calculation process was used as follows:
Input the training dataset T = { ( x 1 , y 1 ) , ( x 2 , y 2 ) , . . . , ( x i , y i ) } , x i X R n , y i Y R ; loss function L ( y , f ( x ) ) , and output regression tree f ~ ( x ) .
(i)
Initialization
f 0 ( x ) = a r g   min c i = 1 N L ( y i , c )
Assuming that the loss function is defined as the square loss, this choice is motivated by the fact that the square loss function is convex, a property that can be directly derived from c .
i = 1 N L ( y i , c ) c = i = 1 N ( 1 2 ( y i c ) 2 ) c = i = 1 N ( c y i )
c is the mean value of all training sample label values, i.e., biomass.
(ii)
Iterative training m = 1, 2,..., M trees.
For each sample (sampling sites) i = 1, 2,..., N. Calculate the negative gradient:
r m i = L ( y i , f ( x i ) ) ( x i ) f ( x ) = f m 1 ( x )
Compute the value of r m i and use it to update the biomass values for the corresponding samples. Then, use the updated dataset ( x i , r m i ), i = 1, 2,..., N as the training data to construct a new regression tree. Compute the corresponding leaf region, R m j , j = 1, 2,..., J, for each sample, where J is the number of leaf nodes in the regression tree. Calculate the optimal fitting value for each leaf region and use it to update the strong learner.
(iii)
Obtain the final learner GBDT
f ~ ( x ) = f M ( x ) = f 0 ( x ) + m = 1 M j = 1 J c m j I ( x R m j )
I is an indicator function, where I equals 1 if the condition holds and 0 otherwise.
Cross validation [40,41] was introduced as an effective method to perform the model evaluation and selection activities when the data were relatively small. Due to the time constraints resulting from the tidal activity, only a limited number of sampling points were available. Therefore, the LOOCV (Leave-One-Out Cross-Validation) method was used to partition the dataset into training and testing sets. The LOOCV method is commonly used in remote sensing inversion models [42,43]. If the size of a dataset is N, then for each iteration of LOOCV, one data sample is used for validation purposes, while the remaining N-1 samples are used for training purposes. This process is repeated N times, where, for each iteration, a different data sample is used for validation purposes, until all the samples have been validated once (Figure 5).
We used the VIs following screening as the feature set, the inversion biomass as the label set, and scikit-learn library (GradientBoostingRegister) in PyCharm2020.1 ×64 (JetBrains, Prague, Czech Republic) to build the GBDT model. To input the model parameters, we used the parameter search method of grid search. In this method, a two-dimensional parameter matrix in units of n_estimators and learning rate was traversed, with each point in the grid being evaluated as a candidate parameter set for the GBDT model. Once the optimal values of n_estimators and learning rate have been determined using the grid search method on the training set, these parameters can be used to calculate the performance on the test set.
The determination coefficient (R2), root mean square error (RMSE), and mean absolute error (MAE) were used as evaluation indicators for the prediction results of the model, as shown in Formulas (10)–(12). According to the area, the mixed spectral reflectance value (13) was calculated. This represents the spectral reflectance of objects within a specific area. The object’s area proportion was also taken into consideration.
R 2 = i = 1 n ( x i x ¯ ) 2 ( y i y ¯ ) 2 i = 1 n ( x i x ¯ ) 2 i = 1 n ( y i y ¯ ) 2
R M S E = i = 1 n ( y i y ¯ ) 2 n
M A E = i = 1 n | y i y ¯ | n
ρ A r e f = i = 1 n ( ρ A r e f i × R P C A 100 )

3. Result

3.1. Relationship between Biomass and Satellite VIs

The correlation coefficient between the directly measured sample biomass and spectral coefficient was poor (R2 = 0.3), indicating that using this value alone for the satellite retrieval of the biomass may result in a high degree of randomness, resulting in poor regression evaluation results.
The VIs presented a better correlation with the biomass collected using the UAV. The biomass of each UAV ROI can be compared with the average biomass of all quadrats in a single pixel of Sentinel-2 or PlanetScope, as shown in Figure 6.
Within the range of the Sentinel-2 grid, (a) the overall biomass of Ulva pertusa collected in the quadrat was lower than the average biomass of Ulva pertusa in a 100 m2 area. (b) In the range of 2000–3000 g/m2, the sample data could well represent the biomass of Sargassum thunbergii in a 100 m2 spatial range. However, if the biomass is higher or lower than this value, the deviation of the sample biomass will be higher or lower than the average biomass of Ulva pertusa in a 100 m2 area.
Within the range of the PlanetScope grid, (c) the sampling data obtained for Ulva pertusa well represent the average biomass of Ulva pertusa within a 9 m2 area (R2 = 0.6). (d) In the range of 2000–3000 g/m2, the sample data can well represent the biomass of Sargassum thunbergii in a 9 m2 spatial range. However, when the biomass is higher or lower than this value, the deviation of the biomass is lower or higher than the average biomass of Sargassum thunbergii in the specified region.

3.2. Relationship between Different VIs and Biomass

Correlation and significance analyses was conducted on the 20 vegetation indices of Sentinel-2 (Figure 7a–d) and PlanetScope (Figure 7e–h), and the average biomass data obtained from the corresponding ground samples within each raster area, as well as the mean biomass derived from UAVs within each corresponding raster.
Within the pixel range of Sentinel-2, the number of significant differences between the average number of Ulva pertusa and 20 spectral parameters measured at the sample site, as well as the correlation value, was low. Except for the extremely significant correlation achieved with VARI (p < 0.01), the correlation of the other 19 spectral parameters was low (the average absolute value was 0.09). The average number of Sargassum thunbergii measured had no correlation with the 20 spectral indexes (the average absolute value was 0.03). The biomass measured by UAV presented a better correlation with 20 spectral indexes. There was a significant correlation between the UAV inversion biomass of Sargassum thunbergii and 20 VIs, and the correlation coefficient between Sargassum thunbergii and VIs was also improved. The change in the correlations for the UAV inversion biomass of Ulva pertusa and vegetation index is complex; however, the correlation coefficient improved in general.
Within the pixel range of PlanetScope, the absolute correlation values of the sampled and satellite data for the VIs mainly ranged from 0.3 to 0.55. The correlation values between the indices obtained from unmanned UAV remote sensing technology, which covered an average biomass within a grid cell, mainly ranged from 0.6 to 0.86. All indices showed significant correlations with the biomass values.

3.3. Relationship between VIs

According to the heat map of variance expansion coefficient (Figure 8), the VIF values of 15 spectral indexes in Sentinel-2 were greater than 10, and the VIF values of 15 spectral indexes in PlanetScope were greater than 10, that is, there were two pairs of multicollinearity outcomes between them.
To prevent the occurrence of overfitting in the inversion model, it was necessary to use a VI without multicollinearity as the characteristic parameter for obtaining seaweed biomass values. After screening, five VIs (Sentinel-2: NDVI, PVI, OSAVI, MSAVI, VARI; PlanetScope: RVI, PVI, MSAVI, CIred edge, VARI) with a VIF of less than 5 were selected as the characteristic variables to be used for biomass machine learning and training assessment models.

3.4. Satellite Remote Sensing Retrieval Based on UAV Biomass

The resulting GBDT model with optimal parameter values can then be applied to the new data to perform prediction or classification tasks. We employed LOOCV to partition the training and test sets and determine the minimum RMSE value for the biomass. The optimal parameters for the GBDT model of Ulva pertusa were determined as n_estimators = 117 and learning rate = 0.6, while the best parameters for the GBDT model of Sargassum thunbergii were determined as n_estimators = 102 and learning rate = 0.7. Based on the satellite and UAV data we obtained, the inversion results are presented in Figure 9.
The biomass value of Ulva pertusa was estimated through the inversion of Sentinel-2 and UAV imagery, with values ranging from 200–850 g/m2 and a concentration of approximately 590 g/m2. The actual average biomass value of Ulva pertusa within a single raster unit inverted by UAV fell in the range of 200–850 g/m2, centered around 590 g/m2. Sentinel-2 was used to estimate the biomass value of Sargassum thunbergii, with values ranging from 2000–3500 g/m2 and a concentration of approximately 2600 g/m2. The actual average biomass value of Sargassum thunbergii within a single raster unit inverted by UAV fell in the range of 2000–3500 g/m2, centered around 2600 g/m2.
The biomass value of Ulva pertusa was estimated through the inversion of PlanetScope and UAV imagery, with values ranging from 200–1200 g/m2 and a concentration of approximately 600 g/m2. The actual average biomass of Ulva pertusa within a single raster unit inverted by UAV fell in the range of 200–1200 g/m2, centered around 600 g/m2. The biomass of Sargassum thunbergii was estimated through the inversion of PlanetScope imagery, with values ranging from 0–6000 g/m2 and a concentration of approximately 2500 g/m2. The actual average biomass value of Sargassum thunbergii within a single raster unit inverted by UAV varied; however, over 70% of the data fell in the range of 1000–4000 g/m2.

3.5. Spectral Analysis

The spectral reflectance curves of intertidal zone features where seaweed could grow were determined and transformed into multispectral spectral reflectance curves with the same bands as the satellite data through a spectral resampling method. The multispectral reflectance curves corresponding to the sampled points were extracted and recorded in the respective grid (Figure 10).
The spectral resampled data of reefs and beaches presented a similar trend to the multispectral reflectance curves obtained from the satellite data, while the seawater exhibited significant differences of 400–600 nm. To successfully calculate the mixed spectra values, the measured seaweed hyperspectral curves were combined with the spectra of from rocks, beaches, and seawater. The distribution of seaweed was determined using UAV surveys of seaweed-coverage areas, with any remaining areas replaced by spectra obtained from reefs, beaches, and seawater (Figure 11).
In Sentinel-2, the correlations between the mixed spectra of seaweed with seawater, beaches, and reefs, and satellite-extracted average spectral reflectance curves of seaweed were 0.97–0.98, 0.73–0.91, and 0.83–0.93, respectively. For PlanetScope, the correlations between the mixed spectra of seaweed with seawater, beaches, and reefs and satellite-extracted average spectral reflectance curves of seaweed were 0.96–0.97, 0.75–0.98, and 0.92–0.98, respectively.

3.6. Inversion of Seaweed Biomass in the Intertidal Zone of Gouqi Island

The intertidal seaweed biomass assessment model trained using UAV inversion dataset was used to quantitatively invert the seaweed biomass from the PlanetScope data, and the biomass of the intertidal seaweed Sargassum thunbergii and Ulva pertusa in the entire island of Gouqi Island was obtained (Figure 12).
The average biomass of Sargassum thunbergii in the intertidal zone of Gouqi Island is about 2606.60 g/m2. By counting the pixels of all Sargassum thunbergii distributions, the total value is 1.6 × 108 g/m2. Considering that the pixel size of the satellite data is 9 m2, the total resource of Sargassum thunbergii in the intertidal zone of Gouqi Island is about 1.4 × 109 g. Similarly, the average biomass of Ulva pertusa is about 456.84 g/m2, and the total biomass of all Ulva pertusa distribution pixels is 3.9 × 107 g/m2, with a total weight of about 3.5 × 108 g.
Sargassum thunbergii and Ulva pertusa are not distributed in beach areas throughout the island, but are widely distributed in reef areas. In the reef area, the biomass of Sargassum thunbergii is higher at low tide levels, but not distributed at high tide levels. Ulva pertusa has a higher biomass at mid to high tide levels than at low tide levels. In terms of their distribution positions, the tide level of the Sargassum thunbergii is lower than that of Ulva pertusa.

4. Discussion

The accuracy of the UAV and satellite biomass retrieval results was mainly affected by the errors evident in the mixed-pixel processing and model construction results. Random errors have the potential to accumulate into systematic errors that can significantly impact the final retrieval results. Compared to satellite remote sensing retrieval methods, the approach presented in this paper offered distinct advantages in biomass retrieval methods. It significantly improved the accuracy and efficiency of intertidal terrestrial biomass retrieval methods, while also reducing the need for extensive field samples. The use of drones introduced benefits, such as increased mobility, timeliness, and cost-effectiveness, combined with the high time-resolution and wide-coverage capabilities of the satellite data, making this method more applicable to the research.

4.1. Distribution of Seaweed Biomass in Intertidal Zones

According to the current situation on Gouqi Island, in areas with smaller currents (such as the northern mussel farm and the southeastern inner bay, less than 1 m/s) [44], the biomass of Sargassum thunbergii (2000–2500 g/m2) and Ulva pertusa (400–550 g/m2) is higher. In areas with large ocean currents (such as the southern open sea area and the eastern jet stream area, greater than 1 m/s horizontally) [44], the biomass of Sargassum thunbergii (0–2000 g/m2) and Ulva pertusa (0–450 g/m2) is relatively small. In the inversion results of satellite data with a spatial resolution of 3 m, both types of seaweed in the intertidal zone exhibit a certain aggregation model (SI).

4.2. Carbon Sequestration of Seaweed in the Intertidal Zone of Gouqi Island

According to the conversion ratio of wet to dry weight, the ratio of dry to wet weight of Sargassum thunbergii is about 0.57 [45], and the ratio of dry to wet weight of Ulva pertusa is about 0.19 [46]. Based on this, it can be concluded that the dry weight of Gouqi Island intertidal Sargassum thunbergii is about 8.0 × 108 g, and the dry weight of Ulva pertusa is about 6.6 × 107 g. The carbon sequestration of Sargassum thunbergii accounts for approximately 0.30 of the dry weight, while Ulva pertusa accounts for approximately 0.29 of the dry weight [47]. Therefore, the carbon sequestration capacity of Gouqi Island intertidal Sargassum thunbergii is 2.4 × 108 g, and the carbon sequestration capacity of Ulva pertusa is 1.9 × 107 g. The satellite imaging time is in summer, during the maturation and decline period of the intertidal seaweed on Gouqi Island. The biomass of intertidal seaweed at other times can be calculated in conjunction with satellite images and seaweed growth formulas from other periods.

4.3. Spatial Scale Differences in Seaweed Biomass in Intertidal Zones

The remote sensing inversion of VIs in the micro-scale region, similar to the method performed in this study [48] was conducted using UAV and satellite images, and was applied to different terrestrial vegetation types [16]. The methods included random forest regression and object-based image analysis [35], spectral unmixing [35,49], and the use of a vegetation index to upgrade the binary vegetation–vegetation classification from UAV to satellite images [50]. However, few studies on biomass information exist in the literature. For complex habitats with strong intertidal heterogeneity, the error was produced by the spectral mixing of local objects caused by an insufficient spatial resolution. In the mixed pixel, the intensity of light reflected from the seaweed was mainly created by the seaweed distribution area in the grid. The seaweed distribution area was proportional to the biomass, which is demonstrated as follows: the thickness of single-layer perforated Ulva pertusa was approximately 0.06 mm [20]. For commonly used marine survey quadrats (approximately 0.25 m), the coverage area was S ≈ volume (V), that is, m = ρ V ≈ ρ S. Suppose the biomass is B:
B = m D ρ S D = ρ C
where D is the area of the sample square (or satellite pixel) and C is the coverage of Ulva pertusa in the sample square (or pixel). The distribution area of Ulva pertusa in a specific quadrat (or pixel) was proportional to its biomass, that is, the spectrum of the image pixel could directly reflect the biomass of Ulva pertusa. We extracted the coverage of Ulva pertusa in 0.25, 0.5, 5.8 (UAV100 pixel values) and 10 m spatial resolution (0.25 and 0.5 m are in situ photography values; 3, 5.8 and 10 m are UAV data ROIs) through EVI (Figure 13).
The EVI classification thresholds of 0.25, 0.5, 5.8 and 10 m presented spatial resolutions of 2, 0.2, 0.03, and 0.015, respectively (Supplementary Materials for more). We counted the percentage of pixels for Ulva pertusa, which represents the coverage per unit area. We established the relationship between seaweed coverage and biomass for samples at 0.25, 0.5, 5.8, and 10 m samples (Figure 14). We provided the coverage biomass conversion equation at different spatial scales. The conversion equation is presented in Equation (14). D i is the square root of the area (the length of a side).
N t 1 N t 2 = ( L 1 L 2 ) 1 ( ρ 1 ρ 2 ) 1 2 ( D i 0.025 ) 1 10
We converted the conversion coefficients of 0.0625, 0.25, 9, 33.64, and 100 m2 to a unified spatial resolution of 0.025 m × 0.025 m for spatial resolutions of 1, 0.93303, 0.779977, 0.7302, and 0.6915, respectively (as shown in Figure 14b). Since the UAV pixel value was 0.058 × 0.058 m and the sample plot size used for sampling was 0.25 × 0.25 m, a unit sample plot of 4× the pixel size was used for biomass calculation during the inversion stage. Therefore, to calculate the biomass value (9, 33.64, and 100 m2) during the inversion step, we multiplied the value by 4 and then performed the calculation. Through sampling data from the year 2023, the coverage area was also inverted using EVI, and the weight data were measured using an electronic balance (Figure 15). It was used as a verification at a lower spatial scale.
Using Equation (14), the conversion coefficient was calculated as 1.1215 for an area of 0.0063 m2 and a side length of 0.079 m (as shown in Figure 16a). After the conversion step, the result we obtained was calculated and presented in Figure 16b.
In addition, in the sample survey, Wieger, Lin et al. believed that the larger the sample, the more stable the mean biomass tends to be [51,52]. Yang et al. believe that the small sample size is greatly affected by boundary effects and can result in excessively high biomass estimates [53]. This is also the case in intertidal seaweed surveys (Figure 14a). The larger the sample, the smaller the difference between the results. Conversely, the smaller the sample, the greater the difference in the extreme values of the results.

5. Conclusions

In this study, we utilized UAV and satellite retrieval technologies to quantitatively retrieve the biomass values of Ulva pertusa and Sargassum thunbergii species present in the intertidal zone of Gouqi Island, Zhejiang Province, China. We used the biomass retrieved by UAV as a proxy for the corresponding pixel values in Sentinel-2 and PlanetScope images. We employed five non-collinear VIs, including NDVI, PVI, OSAVI, MSAVI, and VARI, calculated from Sentinel-2B’s 8a and edge2 bands, as the feature set. We also used the biomass values of Sentinel-2 pixels retrieved by UAV as the label set. The VIs used for the PlanetScope imagery were RVI, PVI, MSAVI, CIred edge, and VARI. The GBDT model established in this study caused significant improvements to the biomass assessment of Ulva pertusa and Sargassum thunbergii species using Sentinel-2 and PlanetScope imagery. For Sentinel-2, the R2 were 0.74 and 0.88 for Ulva pertusa and Sargassum thunbergii, respectively, with corresponding RMSE values of 134.26 and 226.83 g/m2, MAE values of 105.32 and 196.85 g/m2, and adjusted R2 values of 0.73 and 0.87. Similarly, for PlanetScope, the R2 values were 0.80 and 0.69 for Ulva pertusa and Sargassum thunbergii, respectively, with corresponding RMSE values of 97.74 and 907.72 g/m2, MAE values of 72.78 and 725.67 g/m2, and adjusted R2 values of 0.79 and 0.68. The average biomass of Sargassum thunbergii and Ulva pertusa in the intertidal zone of Gouqi Island is 2606.60 g/m2 and 456.84 g/m2, respectively, the total resource is 1,411,796,718 g and 348,610,248 g, respectively, and the carbon sequestration weight is 241,417,238.78 g and 19,208,424.66 g, respectively. Furthermore, the integration of the multi-source data, such as UAV and in situ images, revealed evidence of a positive correlation between seaweed biomass and its coverage, supporting the validity of evaluating Sargassum thunbergii biomass levels in the image. The aggregation distribution of seaweed in the intertidal zone of Gouqi Island makes it hard to use the average of intra pixel plots as satellite pixel biomass (p < 0.01). By using the UAV retrieval results as auxiliary data, the method proposed in this paper was shown to effectively improve the retrieval accuracy and reduce the impact of mixed pixels during this process. The results of this study provide a new approach for the large-scale assessment of intertidal seaweed biomass via satellite remote sensing methods and for studying the patch distribution of intertidal seaweed in selected areas.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/rs15184428/s1, File S1: Derivation of the relationship between spectral index and biomass; File S2: Spatial distribution patterns of intertidal seaweed.

Author Contributions

All authors conceived the initial design of the research. Conceptualization, S.Z. and K.W. J.C. (Jianqu Chen) acquired the UVA data and the in situ sampling data. J.C. (Jianqu Chen) processed/analyzed the data and together with X.L., X.Z. and X.C. interpreted the results. J.C. (Jie Chen) and X.L. prepared the manuscript. All authors contributed to the review of the manuscript. Project administration, K.W. and J.L. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by Key Laboratory of Marine Environmental Survey Technology and Application, Ministry of Natural Resources, P. R. China (MESTA-2022-B010), Guangdong Provincial Key Laboratory of Marine Biotechnology (No. GPKLMB202201), Fujian Key Laboratory of Island Monitoring and Ecological Development (Island Research Center, MNR) (NO: 2022ZD03), Key Laboratory of Marine Ranching, Ministry of Agriculture and Rural Affairs, P.R. China (KLMR-2022-04), Key Laboratory of Marine Ecological Conservation and Restoration, Ministry of Natural Resources/Fujian Provincial Key Laboratory of Marine Ecological Conservation and Restoration (EPR2023002) and Fund of the Key Laboratory of Tropical Marine Ecosystem and Bioresource, MNR, (2022QN04).

Data Availability Statement

Additional supporting information may be found online in the Supporting Information section at the end of the article.

Acknowledgments

We appreciate the support of our funding agency. We also thank the editor and the anonymous reviewers, whose comments significantly improved the manuscript. And we thank the management and staff of the ESA and PLANET, for the free provision of the satellite data used in this research.

Conflicts of Interest

The authors declare no conflict of interest.

Appendix A

Table A1. Calculation Formula of VI.
Table A1. Calculation Formula of VI.
SequenceSpectral IndexFormulaReferences
1NDVI N I R R N I R + R [54]
2RVI N I R R [55]
3DVINIR-R[56]
4PVI N I R 10.489 R 6.604 ( 1 + 10.489 2 ) [57]
5EVI 2.5 × N I R R ( N I R + 6 R 7.5 B + 1 ) [58]
6GNDVI N I R G N I R + G [21]
7RDVI N I R R ( N I R + R ) [59]
8SAVI ( N I R R ) ( 1 + L ) N I R + R + L [57]
9OSAVI N I R R N I R + R + L [60]
10NLI N I R 2 R N I R 2 + R [61]
11GRVI N I R G [62]
12GBNDVI N I R ( G + B ) N I R + ( G + B ) [63]
13GRNDVI N I R ( G + R ) N I R + ( G + R ) [64]
14BNDVI N I R B N I R + B [65]
15MSAVI 2 N I R + 1 ( 2 N I R + 1 ) 2 ( 8 N I R R ) 2 [59]
16RBNDVI N I R ( R + B ) N I R + ( R + B ) [66]
17Pan NDVI N I R ( G + R + B ) N I R + ( G + R + B ) [67]
18GCI N I R G 1 [68]
19CI red edge N I R R e d   e d g e 1 [69]
20VARI G R G + R B [70]

References

  1. Diazpulido, G.J.; Mccook, L. Macroalgae (Seaweeds). The State of the Great Barrier Reef On-Line; Australian Government: Great Barrier Reef Marine Park Authority: Townsville, Australia, 2008. [Google Scholar]
  2. Gellenbeck, K.W.; Chapman, D.J. Seaweed uses: The outlook for mariculture. Endeavour 1983, 7, 31–37. [Google Scholar] [CrossRef]
  3. Gao, K.; Mckinley, K.R. Use of macroalgae for marine biomass production and CO2 remediation: A review. J. Appl. Phycol. 1994, 6, 45–60. [Google Scholar] [CrossRef]
  4. Macreadie, P.I.; Nielsen, D.A.; Kelleway, J.J.; Atwood, T.B.; Ralph, P.J. Can we manage coastal ecosystems to sequester more blue carbon? Front. Ecol. Environ. 2017, 15, 206–213. [Google Scholar] [CrossRef]
  5. Kwan, V.; Fong, J.; Chin, N.; Huang, D. Temporal and spatial dynamics of tropical macroalgal contributions to blue carbon. Sci. Total Environ. 2022, 828, 154369. [Google Scholar] [CrossRef] [PubMed]
  6. Albert, E.; Assis, J.; Filbee-dexter, K.; Burrows, M.; Gattuso, J.P.; Duarte, C.M.; Krause-Jense, D.; Moore, P.J.; Smale, D.A.; Wernberg, T. Global seaweed productivity. Sci. Adv. 2022, 8, eabn2465. [Google Scholar]
  7. Starko, S.; Neufeld, C.J.; Gendall, L.; Timmer, B.; Campbell, L.; Yakimishyn, J.; Druehl, L.; Baum, J.K. Microclimate predicts kelp forest extinction in the face of direct and indirect marine heatwave effects. Ecol. Appl. 2022, 32, e2673. [Google Scholar] [CrossRef]
  8. Menge, B.; Gravem, S.; Johnson, A.; Robinson, J.; Poirson, B. Increasing instability of a rocky intertidal meta-ecosystem. Proc. Natl. Acad. Sci. USA 2022, 119, e2114257119. [Google Scholar] [CrossRef]
  9. Brodie, J.; Kunzig, S.; Agate, J.; Yesson, C.; Robinson, L. The Big Seaweed Search: Evaluating a citizen science project for a difficult to identify group of organisms. Aquat. Conserv. Mar. Freshw. Ecosyst. 2022, 33, 44–45. [Google Scholar] [CrossRef]
  10. Cappelatti, L.; Griffin, J.; Mauffrey, A. Functional diversity of habitat formers declines scale-dependently across an environmental stress gradient. Oecologia 2020, 194, 135–149. [Google Scholar] [CrossRef]
  11. Johnson, M. Estimating intertidal seaweed biomass at larger scales from quadrat surveys. Mar. Environ. Res. 2020, 156, 104906. [Google Scholar] [CrossRef]
  12. Terada, R.; Abe, M.; Abe, T.; Aoki, M.; Dazai, A.; Endo, H.; Kamiya, M.; Kawai, H.; Kurashima, A.; Motomura, T.; et al. Japan’s nationwidelong-term monitoring survey of seaweed communities known as the “Monitoring Sites 1000”: Ten-year overview and future perspectives. Phycol. Res. 2021, 69, 12–30. [Google Scholar] [CrossRef]
  13. Wraase, L.; Reuber, V.; Kurth, P.; Fekadu, M.; Demissew, S.; Miehe, G.; Opgenoorth, L.; Selig, U.; Woldu, Z.; Zeuss, D.; et al. Remote sensing-supported mapping of the activity of a subterranean landscape engineer across an afro-alpine ecosystem. Remote Sens. Ecol. Conserv. 2022, 9, 195–209. [Google Scholar] [CrossRef]
  14. Zoffoli, L.; Gernez, P.; Oiry, S.; Godet, L.; Dalloyau, S.; Davies, B.; Barillé, L. Remote sensing in seagrass ecology: Coupled dynamics between migratory herbivorous birds and intertidal meadows observed by satellite during four decades. Remote Sens. Ecol. Conserv. 2022, 9, 420–433. [Google Scholar] [CrossRef]
  15. Bell, T.; Siegel, D. Nutrient availability and senescence spatially structure the dynamics of a foundation species. Proc. Natl. Acad. Sci. USA 2022, 119, e2105135118. [Google Scholar] [CrossRef]
  16. Morgan, B.E.; Chipman, J.W.; Bolger, D.T.; Dietrich, J.T. Spatiotemporal Analysis of Vegetation Cover Change in a Large Ephemeral River: Multi-Sensor Fusion of Unmanned Aerial Vehicle (UAV) and Landsat Imagery. Remote Sens. 2021, 13, 51. [Google Scholar] [CrossRef]
  17. Haro, S.; Jesus, B.; Oiry, S.; Papaspyrou, S.; Lara, M.; Gonzalez, C.; Corzo, A. Microphytobenthos spatio-temporal dynamics across an intertidal gradient using Random Forest classification and Sentinel-2 imagery. Sci. Total Environ. 2021, 804, 149983. [Google Scholar] [CrossRef]
  18. Li, X.; Kai, W.; Shouyu, Z.; Meiping, F. Distribution and Flora of Seaweed Beds in the Coastal Waters of China. Sustainability 2021, 13, 3009. [Google Scholar] [CrossRef]
  19. Diruit, W.; Bris, A.; Bajjouk, T.; Richier, S.; Helias, M.; Burel, T.; Lennon, M.; Guyot, A.; Gall, E. Seaweed Habitats on the Shore: Characterization through Hyperspectral UAV Imagery and Field Sampling. Remote Sens. 2022, 14, 3124. [Google Scholar] [CrossRef]
  20. Chen, J.; Xunmeng, L.; Kai, W.; Shouyu, Z.; Jun, L.; Jian, Z.; Weicheng, G. Variable Optimization of Seaweed Spectral Response Characteristics and Species Identification in Gouqi Island. Sensors 2022, 22, 4656. [Google Scholar] [CrossRef]
  21. Candiago, S.; Remondino, F.; De Giglio, M.; Dubbini, M.; Gattelli, M. Evaluating Multispectral Images and Vegetation Indices for Precision Farming Applications from UAV Images. Remote Sens. 2015, 7, 4026–4047. [Google Scholar] [CrossRef]
  22. Lucas, P.O.; Mauro, S.A.; José, M.J.; Buceli, D.N.; Paula, M.R.; Akemi, S.M.; Nobuhiro, I.N.; Roberto, P.D.; Eduardo, I.N.; Roberto, P.D.; et al. A convolutional neural network approach for counting and geolocating citrus-trees in UAV multispectral imagery. ISPRS J. Photogramm. Remote Sens. 2020, 160, 97–106. [Google Scholar]
  23. Guo, Y.; Chen, S.; Li, X.; Maro, C.; Senthilnath, J.; Davide, C.; Yongshou, F. Machine Learning-Based Approaches for Predicting SPAD Values of Maize Using Multi-Spectral Images. Remote Sens. 2022, 14, 1337. [Google Scholar] [CrossRef]
  24. Qi, G.; Chang, C.; Yang, W.; Zhao, G. Soil salinity inversion incoastal cotton growing areas: A integration method of satellite-ground spectral fusion and satellite-UAV collaboration. Land Degrad. Dev. 2022, 13, 2289–2302. [Google Scholar] [CrossRef]
  25. Román, A.; Navarro, G.; Caballero, I.; Tovar-Sanchez, A. High-spatial resolution UAV multispectral data complementing satellite imagery to characterize a chinstrap penguin colony ecosystem on deception island (Antarctica). GIScience Remote Sens. 2022, 59, 1159–1176. [Google Scholar] [CrossRef]
  26. Ma, Y.; Zhu, W.; Zhang, Z.; Chen, H.; Zhao, G.; Liu, P. Fusion level of satellite and UAV image data for soil salinity inversion in the coastal area of the Yellow River Delta. Int. J. Remote Sens. 2022, 43, 7039–7063. [Google Scholar] [CrossRef]
  27. Ling, C.; Sun, H.; Zhang, H.; Lin, H.; Ju, H.; Liu, H. Study on Above-Ground Biomass Estimation of East Dong Ting Lake wetland Based on Worldview-2 Data. In Proceedings of the 3rd International Workshop on Earth Observation and Remote Sensing Applications, EORSA 2014—Proceedings, Changsha, China, 11–14 June 2014. [Google Scholar] [CrossRef]
  28. Han, L.; Liu, D.; Cheng, G.; Zhang, G.; Wang, L. Spatial distribution and genesis of salt on the saline playa at Qehan Lake, Inner Mongolia, China. CATENA 2019, 177, 22–30. [Google Scholar] [CrossRef]
  29. Granadeiro, J.P.; Belo, J.; Henriques, M.; Catalao, J.; Catry, T. Using Sentinel-2 Images to Estimate Topography, Tidal-Stage Lags and Exposure Periods over Large Intertidal Areas. Remote Sens. 2021, 13, 320. [Google Scholar] [CrossRef]
  30. Chen, J.; Xunmeng, L.; Kai, W.; Shouyu, Z.; Jun, L. Estimation of Seaweed Biomass Based on Multispectral UAV in the Intertidal Zone of Gouqi Island. Remote Sens. 2022, 14, 2143. [Google Scholar] [CrossRef]
  31. Ge, H.; Zheng, J.; Xu, H. Advances in Machine Learning for High Value-Added Applications of Lignocellulosic Biomass. Bioresour. Technol. 2022, 369, 128481. [Google Scholar] [CrossRef]
  32. Akira, O.; James, G.; Mitchel, J.G. Patchy Distribution and Diffusion. Diffus. Ecol. Probl. Mod. Perspect. 2001, 14, 268–297. [Google Scholar]
  33. Fales, R.J.; Smith, J.R. Long-term change in a high-intertidal rockweed (Pelvetiopsis californica) and community-level consequences. Mar. Biol. 2022, 169, 04022. [Google Scholar] [CrossRef]
  34. Zongling, W.; Chao, Y.; Xuelei, Z.; Yongjuan, L.; Mingzhu, F.; Jie, X. Interannual variations of Sargassum blooms in the Yellow Sea and East China Sea during 2017–2021. Harmful Algae 2023, 126, 102451. [Google Scholar]
  35. Melville, B.; Fisher, A.; Lucieer, A. Ultra-high spatial resolution fractional vegetation cover from unmanned aerial multispectral imagery. Int. J. Appl. Earth Obs. Geoinf. 2019, 78, 14–24. [Google Scholar] [CrossRef]
  36. Roca, M.; Mar, D.; Martha, R.; Caballero, I.; Zoffoli, M.; Gernez, P.; Navarro, G. Monitoring the marine invasive alien species Rugulopteryx okamurae using unmanned aerial vehicles and satellites. Front. Mar. Sci. 2022, 9, 1004012. [Google Scholar] [CrossRef]
  37. Lewis, P.H.; Roberts, B.P.; Moore, P.J.; Pike, S.; Scarth, A.; Medcalf, K.; Cameron, I. Combining unmanned aerial vehicles and satellite imagery to quantify areal extent of intertidal brown canopy-forming macroalgae. Remote Sens. Ecol. Conserv. 2023, 9, 540–552. [Google Scholar] [CrossRef]
  38. Wu, Z.; Zhang, S. Effect of Typhoon on the Distribution of Macroalgae in the Seaweed Beds of Gouqi Island, Zhejiang Province. J. Agric. Sci. Technol. 2019, 21, 159–168. [Google Scholar]
  39. Tang, G.; Li, Y.; Wills, P.; Hanisak, D.; Ouyang, B. Development of a macroalgal biomass sensor for an integrated multi-trophic aquaculture (IMTA) system. Conf. Big Data III Learn. Anal. Appl. 2021, 6, 1173007. [Google Scholar]
  40. Allen, D.M. The Relationship Between Variable Selection and Data Agumentation and a Method for Prediction. Technometrics 1999, 16, 125–127. [Google Scholar] [CrossRef]
  41. Stone, M. Cross-Validatory Choice and Assessment of Statistical Predictions. J. R. Stat. Soc. Ser. B 1974, 36, 111–133. [Google Scholar] [CrossRef]
  42. Kogan, F.; Kussul, N.; Adamenko, T.; Skakun, S.; Kravchenko, O.; Kryvobok, O.; Shelestov, A.; Kolotii, A.; Kussul, O.; Lavrenyuk, A. Winter wheat yield forecasting in Ukraine based on Earth observation, meteorologicaldata and biophysical models. Int. J. Appl. Earth Obs. Geoinf. 2013, 23, 192–203. [Google Scholar]
  43. Li, Y.; Guan, K.; Yu, A.; Peng, B.; Zhao, L.; Li, B.; Peng, J. Toward building a transparent statistical model for improving crop yield prediction: Modeling rainfed corn in the U.S. Field Crops Res. 2019, 234, 55–65. [Google Scholar] [CrossRef]
  44. Wen, Y.; Lin, J.; Yang, G. Study on hydrodynamic effect of suspended mussel aquaculture facilities and detritus transportation of epiphytic seaweed. J. Shanghai Ocean Univ. 2022, 31, 1549–1561. [Google Scholar]
  45. Martina, M.; Jacob, S.; Gil, R. Biomass calibration of nine dominant native and non-native Levantine seaweeds. Aquat. Bot. 2022, 178, 103496. [Google Scholar]
  46. Sara, B.; Wickham, C.; Darimont, C.; Reynolds, J.; Starzomske, B. Species-specific wet-dry mass calibrations for dominant Northeastern Pacific Ocean macroalgae and seagrass. Aquat. Bot. 2019, 152, 27–31. [Google Scholar]
  47. Li, X.; Zhao, X.; Yuan, H.; Guo, Y.; Li, J.; Zhang, S.; Chen, J.; Wang, Z.; Wang, K. Diversity and Carbon Sequestration of Seaweed in the Ma’an Archipelago, China. Diversity 2023, 15, 12. [Google Scholar] [CrossRef]
  48. Walter, J.; Hallett, L.; Sheppard, L.; Anderson, T.; Zhao, L.; Hobbs, R.; Suding, K.; Reuman, D. Micro-scale geography of synchrony in a serpentine plant community. J. Ecol. 2020, 2, 750–762. [Google Scholar] [CrossRef]
  49. Yan, G.; Li, L.; Coy, A.; Mu, X.; Chen, S.; Xie, D.; Zhang, W.; Shen, Q.; Zhou, H. Improving the estimation of fractional vegetation cover from UAV RGB imagery by colour unmixing. ISPRS J. Photogramm. Remote Sens. 2019, 158, 23–34. [Google Scholar] [CrossRef]
  50. Peter, K.D.; D’Oleire-Oltmanns, S.; Ries, J.B.; Marzolff, I.; Hssaine, A.A. Soil erosion in gully catchments affected by land-levelling measures in the Souss Basin, Morocco, analysed by rainfall simulation and UAV remote sensing data. Catena 2014, 113, 24–40. [Google Scholar] [CrossRef]
  51. Wiegert, R.G. The Selection of an Optimum Quadrat Size for Sampling the Standing Crop of Grasses and Forbs. Ecology 1962, 43, 125–129. [Google Scholar] [CrossRef]
  52. Lin, D.; Lai, J.; Mi, X.; Ren, H.; Ma, K. Spatial variation in community structure of a subtropical evergreen broad-leaved forest: Implications for sampling design. Chin. Sci. Bull. 2013, 58, 1181–1186. [Google Scholar] [CrossRef]
  53. Yang, K.; Guan, D. Selection of gaining quadrat for harvesting the undergrowth vegetation and its biomass estimation modeling in forest. Acta Ecol. Sin. 2007, 27, 705–714. [Google Scholar]
  54. Su, H.; Karna, D.; Fraim, E.; Fitzgerald, M.; Dominguez, R.; Myers, J.S.; Coffland, B.; Handley, L.R.; Mace, T. Evaluation of Eelgrass Beds Mapping Using a High-Resolution Airborne Multispectral Scanner. Photogramm. Eng. Remote Sens. 2006, 9, 789–797. [Google Scholar] [CrossRef]
  55. Xiao, Y.; Zhang, J.; Cui, T.; Gong, J.; Liu, R.; Chen, X.; Liang, X. Remote sensing estimation of the biomass of floating Ulva prolifera and analysis of the main factors driving the interannual variability of the biomass in the Yellow Sea. Mar. Pollut. Bull. 2019, 140, 330–340. [Google Scholar] [CrossRef]
  56. Che, S.; Du, G.; Wang, N.; He, K.; Mo, Z.; Sun, B.; Chen, Y.; Cao, Y.; Wang, J.; Mao, Y. Biomass estimation of cultivated red algae Pyropia using unmanned aerial platform based multispectral imaging. Plant Methods 2021, 17, 12. [Google Scholar] [CrossRef]
  57. Elvidge, C.D.; Chen, Z. Comparison of broad-band and narrow-band red and near-infrared vegetation indices. Remote Sens. Environ. 1995, 54, 38–48. [Google Scholar] [CrossRef]
  58. Sakamoto, T.; Yokozawa, M.; Toritani, H.; Shibayama, M.; Ishitsuka, N.; Ohno, H. A crop phenology detection method using time-series MODIS data. Remote Sens. Environ. 2005, 96, 366–374. [Google Scholar] [CrossRef]
  59. Driss, H.; John, R.M.; Elizabeth, P.; Zarco-Tejada, P.; Strachan, I. Hyperspectral vegetation indices and novel algorithms for predicting green LAI of crop canopies: Modeling and validation in the context of precision agriculture. Remote Sens. Environ. 2004, 90, 337–352. [Google Scholar]
  60. Daughtry, C.S.T.; Walthall, C.L.; Kim, M.S.; de Colstoun, E.B.; McMurtrey, J.E. Estimating Corn Leaf Chlorophyll Concentration from Leaf and Canopy Reflectance. Remote Sens. Environ. 2000, 74, 229–239. [Google Scholar] [CrossRef]
  61. Feng, W.; Wu, Y.; He, L.; Ren, X.; Wang, Y.; Hou, G.; Wang, Y.; Liu, W.; Guo, T. An optimized non-linear vegetation index for estimating leaf area index in winter wheat. Precis. Agric. 2019, 20, 1157–1176. [Google Scholar] [CrossRef]
  62. Motohka, T.; Nasahara, K.N.; Oguma, H.; Tsuchida, S. Applicability of Green-Red Vegetation Index for Remote Sensing of Vegetation Phenology. Remote Sens. 2010, 2, 2369–2387. [Google Scholar] [CrossRef]
  63. Rumora, L.; Majić, I.; Miler, M.; Medak, D. Spatial video remote sensing for urban vegetation mapping using vegetation indices. Urban Ecosyst 2021, 24, 21–33. [Google Scholar] [CrossRef]
  64. Wang, W.; Liu, R.; Gan, F.; Zhou, P.; Zhang, X.; Ding, L. Monitoring and Evaluating Restoration Vegetation Status in Mine Region Using Remote Sensing Data: Case Study in Inner Mongolia, China. Remote Sens. 2021, 13, 1350. [Google Scholar] [CrossRef]
  65. Merwe, D.; Price, K.P. Harmful Algal Bloom Characterization at Ultra-High Spatial and Temporal Resolution Using Small Unmanned Aircraft Systems. Toxins 2015, 7, 1065–1078. [Google Scholar] [CrossRef] [PubMed]
  66. Shinya, T.; Seijiro, G.; Masayasu, M.; Akiyama, T.; Muramoto, Y.; Yoshida, K. Estimation and validation of leaf chlorophyll concentration in winter wheat at heading to anthesis stage using ground-based and aerial hyperspectral data. J. Jpn. Soc. Photogramm. Remote Sens. 2008, 47, 39–49. [Google Scholar]
  67. Wang, F.-M.; Huang, J.-F.; Tang, Y.-L.; Wang, X.-Z. New Vegetation Index and Its Application in Estimating Leaf Area Index of Rice. Chin. J. Rice Sci. 2007, 21, 159–166. [Google Scholar] [CrossRef]
  68. Hassan, M.A.; Yang, M.; Rasheed, A.; Jin, X.; Xia, X.; Xiao, Y.; He, Z. Time-Series Multispectral Indices from Unmanned Aerial Vehicle Imagery Reveal Senescence Rate in Bread Wheat. Remote Sens. 2018, 10, 809. [Google Scholar] [CrossRef]
  69. Fernández, C.I.; Leblon, B.; Haddadi, A.; Wang, K.; Wang, J. Potato Late Blight Detection at the Leaf and Canopy Levels Based in the Red and Red-Edge Spectral Regions. Remote Sens. 2020, 12, 1292. [Google Scholar] [CrossRef]
  70. Zhou, X.; Zheng, H.B.; Xu, X.Q.; He, J.Y.; Ge, X.K.; Yao, X.; Cheng, T.; Zhu, Y.; Cao, W.X.; Tian, Y.C. Predicting grain yield in rice using multi-temporal vegetation indices from UAV-based multispectral and digital imagery. ISPRS J. Photogramm. Remote Sens. 2017, 130, 246–255. [Google Scholar] [CrossRef]
Figure 1. Methodological scheme. The figures in () are the corresponding chapters in this paper. § represents the content in each section.
Figure 1. Methodological scheme. The figures in () are the corresponding chapters in this paper. § represents the content in each section.
Remotesensing 15 04428 g001
Figure 2. Study area. (a) Approximate location of the study area; (b) Gouqi Island, sampling sites, UAV image and PlanetScope-RGB stack; (c) tidal position during satellite imaging; (d) spectral reflectance fitting curve of the sampling site in the PlanScope; (e) multispectral UAV orthophotos.
Figure 2. Study area. (a) Approximate location of the study area; (b) Gouqi Island, sampling sites, UAV image and PlanetScope-RGB stack; (c) tidal position during satellite imaging; (d) spectral reflectance fitting curve of the sampling site in the PlanScope; (e) multispectral UAV orthophotos.
Remotesensing 15 04428 g002
Figure 3. Spectral differences between UAV and satellite sensor information (seaweed spectral data obtained from Chen et al., 2022 [20]).
Figure 3. Spectral differences between UAV and satellite sensor information (seaweed spectral data obtained from Chen et al., 2022 [20]).
Remotesensing 15 04428 g003
Figure 4. Quantitative inversion of UAV biomass (g/m2).
Figure 4. Quantitative inversion of UAV biomass (g/m2).
Remotesensing 15 04428 g004
Figure 5. Schematic diagram of LOOCV algorithm.
Figure 5. Schematic diagram of LOOCV algorithm.
Remotesensing 15 04428 g005
Figure 6. Relationship between biomass retrieved by UAV and average biomass of quadrat. Sentinel-2: (a) Ulva pertusa and (b) Sargassum thunbergii; PlanetScope: (c) Ulva pertusa and (d) Sargassum thunbergia.
Figure 6. Relationship between biomass retrieved by UAV and average biomass of quadrat. Sentinel-2: (a) Ulva pertusa and (b) Sargassum thunbergii; PlanetScope: (c) Ulva pertusa and (d) Sargassum thunbergia.
Remotesensing 15 04428 g006
Figure 7. Change in correlation coefficient between spectral index and biomass. (a,b) Correlation coefficients of the average quadrat biomass values of Ulva pertusa, Sargassum thunbergii, and 20 VIs. (c,d) Correlation coefficients of the biomass of Ulva pertusa, Sargassum thunbergii, and 20 VIs retrieved by UAV. (eh) is the correlation value corresponding to the PlanetScope. * represents α < 0.05; ** represents α < 0.01.
Figure 7. Change in correlation coefficient between spectral index and biomass. (a,b) Correlation coefficients of the average quadrat biomass values of Ulva pertusa, Sargassum thunbergii, and 20 VIs. (c,d) Correlation coefficients of the biomass of Ulva pertusa, Sargassum thunbergii, and 20 VIs retrieved by UAV. (eh) is the correlation value corresponding to the PlanetScope. * represents α < 0.05; ** represents α < 0.01.
Remotesensing 15 04428 g007
Figure 8. The 8 VIFs between 20 VIs (the coordinate axis serial number is the corresponding VI value in Appendix A, and the figure number is represent a VIF value and has been converted to an integer value, with 0 representing Inf). (a) Sentinel-2; (b) PlanetScope.
Figure 8. The 8 VIFs between 20 VIs (the coordinate axis serial number is the corresponding VI value in Appendix A, and the figure number is represent a VIF value and has been converted to an integer value, with 0 representing Inf). (a) Sentinel-2; (b) PlanetScope.
Remotesensing 15 04428 g008
Figure 9. Sentinel-2 biomass retrieval results using UAV data. Sentinel-2: (a) Ulva pertusa, and (b) Sargassum thunbergii; PlanetScope: (c) Ulva pertusa, and (d) Sargassum thunbergii.
Figure 9. Sentinel-2 biomass retrieval results using UAV data. Sentinel-2: (a) Ulva pertusa, and (b) Sargassum thunbergii; PlanetScope: (c) Ulva pertusa, and (d) Sargassum thunbergii.
Remotesensing 15 04428 g009
Figure 10. Spectral data measured from (a) reef, (b) beach, and (c) seawater above seaweed, as well as multispectral data obtained from satellites.
Figure 10. Spectral data measured from (a) reef, (b) beach, and (c) seawater above seaweed, as well as multispectral data obtained from satellites.
Remotesensing 15 04428 g010
Figure 11. Mixed spectral reflectances between seaweed and various terrestrial materials; spectral reflectances extracted from satellite imagery.
Figure 11. Mixed spectral reflectances between seaweed and various terrestrial materials; spectral reflectances extracted from satellite imagery.
Remotesensing 15 04428 g011
Figure 12. Distribution of seaweed biomass in the intertidal zone of Gouqi Island (g/m2). (a) Biomass of Sargassum thunbergii; (b) biomass of Ulva pertusa.
Figure 12. Distribution of seaweed biomass in the intertidal zone of Gouqi Island (g/m2). (a) Biomass of Sargassum thunbergii; (b) biomass of Ulva pertusa.
Remotesensing 15 04428 g012
Figure 13. Extraction of Ulva pertusa coverage areas at 0.25, 0.5, 3, 5.8 and 10 m and biomass at 3, 508, 10 m (g/m2).
Figure 13. Extraction of Ulva pertusa coverage areas at 0.25, 0.5, 3, 5.8 and 10 m and biomass at 3, 508, 10 m (g/m2).
Remotesensing 15 04428 g013
Figure 14. Relationship between coverage and biomass. (a) Coverage-biomass fitting curves with different spatial resolutions; (b) fitting curve of conversion coefficient; (c) coverage-biomass fitting curves with converted spatial resolutions.
Figure 14. Relationship between coverage and biomass. (a) Coverage-biomass fitting curves with different spatial resolutions; (b) fitting curve of conversion coefficient; (c) coverage-biomass fitting curves with converted spatial resolutions.
Remotesensing 15 04428 g014
Figure 15. Extraction of coverage for 0.06 × 0.105 m area.
Figure 15. Extraction of coverage for 0.06 × 0.105 m area.
Remotesensing 15 04428 g015
Figure 16. Relationship between coverage and biomass: (a) coverage–biomass fitting curves at 0.0625 and 0.0063 m2; (b) coverage–biomass fitting curves with converted spatial resolutions.
Figure 16. Relationship between coverage and biomass: (a) coverage–biomass fitting curves at 0.0625 and 0.0063 m2; (b) coverage–biomass fitting curves with converted spatial resolutions.
Remotesensing 15 04428 g016
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Chen, J.; Wang, K.; Zhao, X.; Cheng, X.; Zhang, S.; Chen, J.; Li, J.; Li, X. Satellite Imagery-Estimated Intertidal Seaweed Biomass Using UAV as an Intermediary. Remote Sens. 2023, 15, 4428. https://doi.org/10.3390/rs15184428

AMA Style

Chen J, Wang K, Zhao X, Cheng X, Zhang S, Chen J, Li J, Li X. Satellite Imagery-Estimated Intertidal Seaweed Biomass Using UAV as an Intermediary. Remote Sensing. 2023; 15(18):4428. https://doi.org/10.3390/rs15184428

Chicago/Turabian Style

Chen, Jianqu, Kai Wang, Xu Zhao, Xiaopeng Cheng, Shouyu Zhang, Jie Chen, Jun Li, and Xunmeng Li. 2023. "Satellite Imagery-Estimated Intertidal Seaweed Biomass Using UAV as an Intermediary" Remote Sensing 15, no. 18: 4428. https://doi.org/10.3390/rs15184428

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop