Documente Academic
Documente Profesional
Documente Cultură
1Article
4Jianhu Zhao 1,2, Xinglei Zhao 1,2,*, Hongmei Zhang 3 and Fengnian Zhou 4
5 1 School of Geodesy and Geomatics, Wuhan University, 129 Luoyu Road, Wuhan 430079, China;
6 jhzhao@sgg.whu.edu.cn
7 2 Institute of Marine Science and Technology, Wuhan University, Wuhan 430079, China
8 3 Automation Department, School of Power and Mechanical Engineering, Wuhan University,
9 Wuhan 430072, China; hmzhang@whu.edu.cn
10 4 The Survey Bureau of Hydrology and Water Resources of Yangtze Estuary, Shanghai 200136, China;
11 cjkfnzhou@126.com
12 * Correspondence: xingleizhao@whu.edu.cn; Tel.: +86-130-0616-9852
13 Academic Editor: name
14 Received: date; Accepted: date; Published: date
15 Abstract: Airborne lidar bathymetry (ALB) is efficient and cost-effective in obtaining shallow water
16 topography, but often suffers the low accuracy sounding solution due to the impacts of measurement and
17 ocean hydrological parameters. In the bathymetry estimates, peak shifting of green bottom return induced by
18 pulse stretching effect induce a depth bias that is the largest error source in ALB depth measurements. The
19 traditional depth bias model is often applied to weaken the depth bias, but is insufficient for the various ALB
20 system parameters and ocean environments. Therefore, it is necessary to establish an accurate model by
21 considering all of the necessary influencing factors. In this study, a comprehensive depth bias model is
22 proposed with taking water depth, scanning angle of laser beam, sensor height, and suspended sediment
23 concentration into account using stepwise regression. The two models are used for an experiment finally.
24 The experimental results show that systematic deviation of depth bias corrected by the traditional and
25 comprehensive model has been weakened significantly. Standard deviations of 8.3 and 6.3 cm are obtained
26 by the traditional and comprehensive models, respectively. The accuracy of the ALB-derived depth corrected
27 by the comprehensive model is improved relative to that by the traditional model.
28 Keywords: airborne lidar bathymetry; depth bias correction; the comprehensive depth bias model; the
29 measurement and hydrological parameters
30
311. Introduction
32 Airborne lidar bathymetry (ALB) is an accurate, cost-effective, and rapid technique for shallow
33water measurements [13]. In addition to traditional nautical charting, ALB is widely used for
34monitoring engineering structures, sand movement, environmental changes, and resource
35management and exploitation [4]. Furthermore, ALB can also be used to produce other
36environmental products such as seafloor reflectance images, seafloor classification maps, and water
37column characterization maps [5].
38 Figure 1 depicts the principle of ALB measurement. In the measurement, bathymetric accuracy
39is one of the essential qualities required for a successful ALB system, which is primarily affected by
40ALB measurement and ocean hydrological parameters. The ALB bathymetric errors can be resolved
41into two components: depth bias and residuals [6]. For an integrated IR and green ALB system
42which an additional IR laser is used to detect water surface accurately, depth bias is mainly induced
43by the pulse stretching of the green bottom return [2,7,8]. Geometric dispersion and multiple
44scattering lead to temporal stretching of the received green bottom return, which is known as the
45pulse stretching effect [13,710]. The pulse waveform is distorted (e.g., peak shifting) by the pulse
46stretching effect. Peak shifting induces a bias in the bathymetry estimates that is based on a peak
47detection of up to 92% of the true water depth [8], and this depth bias thus becomes the largest
48source of error in ALB depth measurements. Depth bias can be corrected by the methods, such as
50theoretical analysis [2,8], empirical modeling [6,11,12] and error statistics [13,14].
49
Sea surface
t2
t1
Sea bottom
(a) (b)
t0
d
t3
51 Figure 1. Principle of ALB measurement. Depth bias d is induced by the peak shifting of green
52 bottom return. The red and green colors represent the infrared (IR) and green lasers, respectively: (a)
53 denotes the waveform detections of the IR laser and green laser; (b) shows the propagation ways of
54 the two lasers and the bias induced by using only the green laser; t 0t3 denote the initial emission
4
5Remote Sens. 2017, 9, x FOR PEER REVIEW 3 of 14
55 time of the laser pulse, round trip time of the IR surface return, round trip time of the actual green
56 bottom return distorted by pulse stretching effect, and round trip time of the ideal green bottom
57 return, respectively.
58 In theoretical analysis methods, the Monte Carlo numerical method and analytical approach
59are the two classic methods [9]. A Monte Carlo simulation is used to estimate depth biases by
60impulse response function (IRF) which is a function of beam scanning angle, sea depth, phase
61function, optical depth and single-scattering albedo [2]. The single-scattering albedo is a main
62parameter in IRF, and can be obtained by estimating the scattering coefficient. However, it is difficult
63to estimate scattering coefficients accurately and efficiently. Depth bias induced by peak shifting was
64also analyzed by the Wa-LID simulator [8]. Wa-LID is developed to simulate the reflection of lidar
65waveforms from water across visible wavelengths [15]. The relationship between the time shifts of
66the waveform peaks and the bottom slope, the water depth and the footprint size is modeled by the
67Wa-LID simulator [8]. However, the model is based on the assumptions of a vertical incident beam
68and homogeneous water clarity, which are inconsistent with actual circumstances.
69 Integrated the bathymetric data of ALB and sonar, an empirical model for depicting the ALB
70depth bias is formed by performing a regression analysis, which is only a function of water depth
71[6,11]. The ALB depth corrected by this model can only meet the requirements of some low-
72accuracy applications because the two parameters are not sufficient in the expression of the ALB
73depth bias. Wright et al. [12] improved the depth bias model by adding a constant term. This model
74is still only a function of water depth and is simple to be established, but its adaptability to complex
75and variable measurement and ocean hydrological parameters is weak. Therefore, it is necessary to
76establish an accurate model by considering all of the necessary influencing factors.
77 The error statistics method is often used for correcting the ALB depth biases produced in the
78river measurements [13,14]. Magnitude and spatial variation of the depth bias can be evaluated by
79referring the ground surveying result, and can be obtained by subtracting the ground surveying
80elevations from the ALB-derived elevations. The error statistics shows that the depth bias has a low
81relevance with local topographic variance or flow depth, and can be corrected by subtracting the
82mean bias from the raw ALB-derived depths. The method is efficient and simple in correcting the
83depth biases in water areas with small depth variations, which was used successfully in rivers by
84Hilldale et al. [13] and Skinner et al. [14], but may suffer challenge in the water areas with
85complicated depth variations.
86 The methods mentioned as above improve the ALB bathymetric accuracy to a certain extent.
87The error statistics method has the advantages in its simplicity and high efficiency, but its
88adaptability to the complex ocean hydrological environment is weak. The theoretical analysis
89method provides an understanding of the physical processes involved, but it is limited by the
90simplified assumptions that may be inconsistent with the actual ocean hydrological environment,
91therefore needs to be further improved. The empirical modeling method can be applied to establish
92a relationship model between the ALB depth bias and the influencing factors in a specific
93environment, which is simple and easy to be implemented. However, in the empirical model,
94parameters influencing the depth bias need to be found completely and their significances need to
95be fully analyzed. Otherwise, the empirical model established may result in low-accuracy
96correction.
97 Based on the analyses of the ALB bathymetric mechanisms, and the factors influencing ALB
98depth accuracy and the performances of these methods used for the ALB depth bias correction, a
99comprehensive model for ALB depth bias correction is proposed in this paper with taking
100parameters such as water depth, turbidity, beam scanning angle, and sensor height into account.
101The structure of this paper is as follows. Section 1 introduces the applications of ALB systems and
102analyzes the existing correction methods for ALB depth bias. Section 2 provides the detailed
103method of building the depth bias model. Section 3 presents the validation and analysis of the
104proposed method through experiments. Section 4 provides the corresponding discussion. Lastly,
6
7Remote Sens. 2017, 9, x FOR PEER REVIEW 4 of 14
105Section 5 presents several beneficial conclusions and recommendations that are drawn from the
106experiments and discussions.
120where b is the constant item and, is the depth bias coefficient, Turb is the water turbidity, is the
121beam scanning angle, and F is the lidar footprint size.
122 The correlations between the turbidity and suspended sediment concentration (SSC) have been
123investigated through extensive experiments [1622]. Although turbidity depends on SSC, as well as
124particle composition and size distribution, many experiments have shown that a good linear
125relationship exists between turbidity and SSC [16,18,2022]. Therefore, Turb is expressed as follows:
Turb kC m (4)
126where C is SSC, and k and m are the coefficients that vary with different regions and time. However,
127these coefficients can be regarded as constants in the same region and in a short
128period [16,18,2022].
129 F can be calculated using the sensor height H and lidar divergence angle [8] using the
130following equation:
F H tan (5)
131 The ALB system determines the transmitted pulse characteristics (i.e., its initial radius and its
132divergence) [23]. Thus, should no longer be included in the depth bias model as an independent
133variable. We obtain the following equation when Equations (4) and (5) are substituted into Equation
134(3):
d d b
(6)
= 1 2 3 2 4 H 5 H 2 6C 7C 2
135where is the model coefficient, which is a function of the beam scanning angle , sensor elevation
136H, and suspended sediment concentration C. Equation (6) is the initial depth bias model proposed
137in this study which should be further optimized.
140 (1) Seabed elevations derived from the ALB and sonar at the same locations.
141 (2) Ocean hydrological parameters of the ALB survey water (i.e., suspended sediment
142concentration C), and ALB measurement parameters (i.e., beam scanning angle , sensor elevation
143H).
144 ALB depth bias d can be calculated by using the water depths derived from the ALB and
145sonar at the same location as follows:
d DALB Dsonar (9)
146where DALB is the ALB-derived water depth after chart datum correction and Dsonar is the sonar-
147derived water depth after chart datum correction.
148 For the seabed topography produced from the ALB and sonar, we have
d ( H ALB
B
H datum ) H Sonar
B
H datum (10)
149where Hdatum is the ellipsoid height of the chart datum, HBALB and HBSonar are the seabed ellipsoid
150heights derived by the ALB and sonar, respectively. Therefore, d can be obtained by comparing
151HBALB and HBSonar at the same location as follows:
d H ALB H Sonar (11)
B B
152 For the ALB measurement on the spot, Equation Error: Reference source not found can be
153transformed into the following:
d ( H ALB +d ALB ) (TSonar +d Sonar ) (12)
S
154where HSALB, dALB are the sea surface ellipsoid heights derived by the ALB IR channel and the ALB-
155derived depth at time of ALB measurement respectively. Tsonar and dsonar are the tidal level defined on
156the ellipsoid surface and the sounding result at the time of the sonar sounding respectively. dALB can
157be calculated by subtracting HSALB from HBALB as follows:
d ALB H ALB H ALB (13)
B S
158 In the preceding data, HBALB, , and H can be extracted from the ALB records; HBSonar can be
159extracted from the sonar records; and C can be obtained through the on-site sampling and
160laboratory analysis. After obtaining these data, the comprehensive depth bias model presented in
161Equation (6) can be established. The matrix form can be expressed as follows:
Vm1 =Bm7 X 71 l m1 (14)
162where m is the number of point pairs of the ALB and sonar surveying points with same locations.
d1 1d1 12 d1 H1d 1 H12 d 1 C1d1 C12d1 1
d 2 2 d 2 d
2 2 2
2 2 H 2 d2 H d 2 2 C2 d 2 C d 2 2 1
d3 3 d3 d 2
H 3 d3 H d 2
C3 d3 C d 2
1
3 3 3 3 3 3
L L L L L L L L
d
m d
m m 2
mdm H m dm H m2 d m Cm d m Cm2 d m 1 m8
l d1 d m m1
T
d 2 d 3 L
X 1 8
T
2 3 4 5 6 7
X = BT B B T l
1
(15)
168of the data. Determining this subset is called the variable selection problem. The goal of variable
169selection becomes one of parsimony: achieving a balance between simplicity (i.e., as few regressors
170as possible) and fit (i.e., as many regressors as needed) [24]. The variable selection of the depth bias
171model can be determined using a stepwise regression procedure.
172 The depth bias model shown in Equation (6) is only an initial model that considers the main
173influencing factors and can be optimized by stepwise regression. Moreover, t-test is adopted to
174conduct significance tests on the regression coefficients of the depth bias model. Detailed
175descriptions of stepwise regression and t-test theories are appended in Appendix A.
Parameter Specification
Operating altitude 400 m (nominal)
Aircraft speed 140 kts (nominal)
Pulse repetition frequency 10 kHz
Maximum depth single pulse KdDmax=3.75-4.0 daytime (bottom reflectivity > 15%)
Minimum depth <0.15 m
Bathymetric accuracy [0.32 + (0.013 depth)] m, 2
Horizontal accuracy (3.5 + 0.05 depth) m, 2
Depth density 2 m 2 m nominal
Scan angle 20 (fixed off-nadir, circular pattern)
Swath width 294 m nominal
12
13Remote Sens. 2017, 9, x FOR PEER REVIEW 7 of 14
193
34520N
0 1
34520N
Kilometers
34510N 119
119170E 119
119170E
14
15Remote Sens. 2017, 9, x FOR PEER REVIEW 8 of 14
194 Figure 2. Locations and scopes of the different measurements. The yellow, green, and blue colors
195 denote the land, the bottom points of ALB and the bottom points of sonar respectively. The black
196 triangles denote the locations of three SSC sampling stations.
211 If the ALB sounding point has approximately the same location as the sonar sounding point,
212the two sounding points from CZMIL and HY1600 are defined as a point pair. There are 344 point
213pairs found in the surveying water area. 287 point pairs are selected randomly to construct the
214depth bias model, and the remaining 57 point pairs are used to assess the model established. The
215measurement and ocean hydrological parameters used in the model construction are obtained and
216shown in Table 3 and Figure 3. Evidently, d ranges from -0.16 m to 1.22 m and has a mean of 0.16 m
217and standard deviation of 0.29 m. This change implies that raw ALB-derived depth bias is
218significant. Thus, a depth bias model should be built to improve the accuracy of the ALB-derived
219depth.
220 Table 3. Statistical parameters of the five-type data used in the modeling. d: depth bias; d: water
221 depth; : beam scanning angle; H: sensor height; C: SSC
1 1
0.8 0.8
Depth bias (m)
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
15 16 17 18 19 20 21 390 400 410 420 430 440 450
Beam scanning angle (Degree) Senosr height (m)
(a) (b)
16
17Remote Sens. 2017, 9, x FOR PEER REVIEW 9 of 14
1.2 1.2
1 1
0.8 0.8
Depth bias (m)
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
160 165 170 175 180 185 190 195 -4.6 -4.4 -4.2 -4 -3.8 -3.6 -3.4 -3.2 -3
SSC (mg/L) Water depth (m)
(c) (d)
223
224 Figure 3. Depth bias d obtained by comparing HBALB and HBSonar at the same location (see Equation
225 11) changes with different parameters. (a), (b), (c), and (d) denote depth biases changing with beam
226 scanning angle, sensor height, SSC, and water depth respectively.
227 Using above data, the initial depth bias model depicted in Equation (6) can be constructed. To
228evaluate the significance of the model parameters, the t-test is adopted. The test results are shown in
229Table 4. To guarantee that these coefficients are significant in the model, t-test is adopted to conduct
230the hypothesis tests on these regression coefficients. The standard error (SE), t-statistics (t), and p-
231value (p) of each coefficient are calculated and shown in Table 4. Accordingly, comparing the p-
232values of these coefficients shows that all coefficients, except 5, 6, and 7, are higher than the
233standard = 0.05 cutoff, thereby indicating multicollinearity in the initial model shown in Equation
234(6) and that the model cannot substantially represent depth bias. To solve this problem, stepwise
235regression is adopted to optimize the initial depth bias model. Equation (16) shows the
236comprehensive depth bias model optimized by stepwise regression. Relative to Equation (6), the
237comprehensive model coefficient drops H, and C2, thereby indicating that the impact of the SSC on
238depth bias can be described through a simple linear function. Upon optimization, the p-values of
239the remaining parameters in the comprehensive model are below . Therefore, the parameters in
240Equation (16) are statistically significant and should be included in the model. This result also
241shows the necessity of taking ALB measurement and ocean hydrological parameters into account
242when constructing the depth bias model.
243 Table 4. Coefficients and their t-test results of the initial and comprehensive depth bias models. SE:
244 standard error; t: t-statistics
d d b
(16)
= 1 2 3 2 5 H 2 6C
248varying with d, , H, and C. Figure 4 shows these relationships. In a relatively small range of , d
249increases with , whereas d decreases gradually with when increases to a certain range (see
250Figure 4a). In terms of the CZMIL instruments, the turning point of depth bias changing with the
251beam scanning angle appears at as 18.8 when water depth is 4 m, sensor height is 420m, and SSC
252is 180 mg/L. This phenomenon can be explained by Guenthers theory [2]. Two competing effects
253exist: the path lengthening due to multiple scattering, and a path shortening duo to energy
254returning early from the undercutting region [2]. The relative magnitudes of these effects depend
255strongly of the beam scanning angle [2]. That is , the geometric dispersion effect which cause a path
256shortening is dominant when beam scanning angle is smaller than 18.8 and multiple scattering
257effect which causes a path lengthening is dominant when beam scanning angle is larger than 18.8.
258The impact of the suspended sediment concentration and the sensor elevation on the depth bias is a
259positive correlation (see Figure 4b and c). Namely, the depth biases increase with the two
260parameters respectively. The larger the SSC is, the more expand of the laser beam is and the lager
261the depth bias is. The higher the sensor height is, the larger the laser spot size is and the larger of
262the depth bias is. In Figure 4d, the depth bias decreases linearly with water depth, which consistent
263with the actual measurement data shown in Figure 3d.
264
20
21Remote Sens. 2017, 9, x FOR PEER REVIEW 11 of 14
0.74 0.85
0.72 0.8
0.7 0.75
0.68 0.7
0.66 0.65
d=4m d=4m
0.64 H = 420 m 0.6 = 20
C = 180 mg/L C = 180 mg/L
0.62 0.55
16 17 18 19 20 21 400 405 410 415 420 425 430 435 440
Beam scanning angle (Degree) Sensor height (m)
(a) (b)
1.6
0.8
1.4 = 20
0.75 1.2
H = 420 m
C = 180 mg/L
0.7 1
0.8
0.65
0.6
0.6
0.4
0.55 d=4m
0.2
= 20
0.5 0
H = 420 m
0.45 -0.2
160 165 170 175 180 185 190 -5 -4.5 -4 -3.5 -3
SSC (mg/L) Water depth (m)
(c) (d)
265 Figure 4. The regression curves of depth biases d varying with ALB measurement and ocean
266 hydrological parameters. (a)~(d) show that d varying with scanning angle of laser beam, sensor
267 height, SSC and water depth follows parabola, monotonically increasing, monotonically increasing
268 and monotonically decreasing, respectively.
278 Table 5. Coefficients and their t-test results of the traditional depth bias model.
279 Table 6. Statistical parameters of the depth biases corrected by different models
Depth Bias (cm) Max. Min. Mean Std. The worst Meets IHO
Raw ALB-derived 122.1 -16.7 16.2 28.9 74
Corrected by the traditional model 17.4 -21.2 0.3 8.3 16.9
Corrected by the comprehensive model 14.1 -14.5 0.2 6.3 12.8
22
23Remote Sens. 2017, 9, x FOR PEER REVIEW 12 of 14
0.20 0.20
0.15 0.15
Residual depth bias (m)
0.05 0.05
0.00 0.00
-0.05 -0.05
-0.10 -0.10
-0.15 -0.15
-0.20 -0.20
-0.25 -0.25
160 165 170 175 180 185 190 195 160 165 170 175 180 185 190 195
SSC (mg/L) SSC (mg/L)
(a) (b)
280 Figure 5. (a) and (b) show the distributions of depth bias corrected by the traditional and
281 comprehensive models varying with SSC respectively.
282
Raw ALB-derived
6
Corrected by
5 the traditional model
Corrected by
4 the comprehensive model
Density
0
-0.2 0 0.2 0.4 0.6 0.8 1 1.2
Depth bias (m)
283 Figure 6 PDF curves of the depth biases of raw ALB-derived, corrected by the traditional and
284 comprehensive depth bias models respectively.
285 The Order-1 specification of International Hydrographic Organization (IHO) defines the
286maximum allowable Total Vertical Uncertainty (TVU) at the 95% confidence level, which can be
287computed as follows:
24
25Remote Sens. 2017, 9, x FOR PEER REVIEW 13 of 14
299improved relative to that by the traditional model. Moreover, it can be found from Figure 5 that the
300depth biases corrected by the traditional model vary linearly with SSS, while those corrected by the
301comprehensive model change slightly with SSS. The result is due to ignoring SSS variation in the
302traditional model and considering that in the comprehensive model. The depth biases corrected the
303two models are relatively large when SSS is in 175~177. Because only 3 SSS stations are around the
304measurement area, the linear interpolation adopted for getting the SSS among these SSS stations
305may result in small SSS bias with the actual. The influence leads to large depth bias in the
306traditional correction method, but relative small depth bias in the proposed method.
3074. Discussion
308 The proposed method provides a good way to weaken depth bias and obtain accurate water
309depth by ALB. The following factors will influence the applications and accuracies of the proposed
310method.
311 (a) Different ALB systems
312 ALB systems are categorized into two according the used lasers: integrated IR and green ALB
313systems and green ALB systems [2931]. For an integrated IR and green ALB system which an
314additional IR laser is used to detect water surface accurately, the comprehensive depth bias model
315can be used directly to correct the ALB-derived depth. However, for a green ALB system, the
316primary IR laser is no longer used and the green surface return cant accurately represent water
317surface. The height models of green ALB systems proposed by Jianhu Zhao et al. [31] which
318considers near water surface penetration (NWSP) of green laser should be used firstly to correct
319green water surface and water bottom heights. Then, the comprehensive depth bias model can be
320used to correct the ALB-derived depth bias.
321 (b) Effects of surface wave and bottom slope
322 Surface wave and bottom slope will bring effects on ALB depth. Surface wave slope can be
323estimated by referring wave height and wavelength caused by the wind speed, and its effect can be
324compensated by adding the surface wave slope into beam scanning angle [31]. Bottom slope can
325change the bottom incident angle of laser beam and have an influence on the pulse stretching of
326green bottom return, as shown in Figure 7a [8]. Literature [8] proposed a model to estimate the
327effect of bottom slope on ALB depth, and the effect varying with bottom slope is shown in Figure
3287b. It can be seen that the effect becomes large with the increase of bottom slope. If the limitation is
329defined as 5 cm, the effect of bottom slope less than 7 can be ignored when the beam footprint
330radius is less than 2 m and the water depth is less than 10 m, otherwise, the effect needs to be
331considered. Using the model proposed in literature [8], the effect can be estimated by integration of
332beam footprint, water depth and bottom slope for the compensation of ALB depth. Residual of the
333compensation will be remained in the depth bias and should be further compensated by the
335comprehensive model proposed in this paper.
334
0.1
2m
0.08 5m
8m
Depth bias (m)
0.06
0.04
0
0 2 4 6 8 10
Bottom slope (Degree)
(a) (b)
26
27Remote Sens. 2017, 9, x FOR PEER REVIEW 14 of 14
336 Figure 7. (a) denotes the effect of bottom slope on peak shifting of green bottom return and (b)
337 denotes the corresponding depth bias varying with bottom slope.
367 2017 by the authors. Submitted for possible open access publication under the
368 terms and conditions of the Creative Commons Attribution (CC BY) license
369 (http://creativecommons.org/licenses/by/4.0/).
28