A More Accurate Field-to-Field Method towards the Wind Retrieval of HY-2B Scatterometer
<p>Locations of TAO (blue) and NDBC (red) buoys used in this paper.</p> "> Figure 2
<p>Schematic representation of two NN-based approaches: P2P (<b>a</b>) and F2F (<b>b</b>). What is happening in grey picture? The P2P retrieval method inputs the parameters of a single vector cell (NRCS, etc.) into the P2P TF to obtain the corresponding single wind field (wind speed and direction). The results obtained by the P2P method lack continuity and there are ambiguous solutions. The F2F method inputs the parameters of multiple wind vector cells (m × n in (<b>b</b>)) within a certain range into the F2F TF, and retrieve m × n wind fields at the same time. The F2F TF extracts the spatial continuity characteristics of the wind field and applies it to the retrieval process. At the same time, the entire continuous wind field composed of multiple wind vector cells is obtained, which can fundamentally eliminate the ambiguous solutions, and the wind vector cells of the obtained wind field are smoother and more continuous. The F2F TF is a retrieval model, which contains mathematical formulas for the process (retrieval) from the measured parameter data of the scatterometer to the wind field data. The retrieval process is completed by a neural network, and the continuity characteristics of the wind field will be applied to this process. TF represents the transfer function. m × n is the base size, i.e., the number of cells that serve as NN inputs and/or outputs. The case for m = n = 3 is shown.</p> "> Figure 3
<p>Wind direction statistics (<b>a</b>) and wind speed statistics (<b>b</b>). In (<b>a</b>), the interval of the abscissa of the wind direction is 10°, starting from 0, every 10° is divided into one category for statistics. In (<b>b</b>), starting from 0, every 1 m/s is counted as a category.</p> "> Figure 4
<p>The schematic diagram of our CNN. The structure and parameters of the wind speed neural network are different from the wind direction neural network. The structure and parameter settings of the two neural networks are shown in the figure (The data in the brackets are the structural parameters of the wind direction neural network).</p> "> Figure 5
<p>Variations of the wind direction RMSE (<b>a</b>) and wind speed RMSE (<b>b</b>).</p> "> Figure 6
<p>Comparison of label data and F2F-CNN method results (<b>a</b>), label data and HY-2B L2B results (<b>b</b>). It can be clearly seen that compared to the HY-2B L2B data, the F2F-CNN wind speed is more confirmed to the ECMWF ERA5 wind speed.</p> "> Figure 7
<p>(<b>a</b>,<b>d</b>) show the wind speeds of ECMWF below 10 m/s and above 10 m/s, respectively. (<b>b</b>,<b>e</b>) show the difference map between F2F-CNN wind speed and ECMWF wind speed. (<b>c</b>,<b>f</b>) show the difference map between HY-2B wind speed and ECMWF wind.</p> "> Figure 8
<p>Comparison of F2F-CNN method results and ECMWF wind direction (<b>a</b>), HY-2B L2B results and ECMWF wind direction (<b>b</b>). It can be clearly seen that results of the F2F-CNN method are better fitted to the ECMWF wind direction and poses to smaller deviations.</p> "> Figure 9
<p>(<b>a</b>,<b>c</b>,<b>e</b>) show the comparison between the F2F-CNN wind direction and the ECMWF wind direction, and (<b>b</b>,<b>d</b>,<b>f</b>) show the comparison between the HY-2B wind direction and the ECMWF wind direction.</p> "> Figure 10
<p>(<b>a</b>) shows the absolute difference between the F2F-CNN wind speed and the ECMWF data, and (<b>b</b>) shows the absolute difference between the HY-2B L2B wind speed and the ECMWF data. It is obvious seen from the figure that the wind speed of the HY-2B L2B data at the center of the typhoon is significantly higher than the ECMWF value, and the overall wind speed error of the F2F-CNN is more uniform and smaller.</p> "> Figure 11
<p>The wind direction map derived by F2F-CNN vs. Target (<b>a</b>) and HY-2B L2B vs. Target (<b>b</b>).</p> "> Figure 12
<p>The wind speed and wind direction from F2F-CNN (<b>a</b>), HY-2B L2B (<b>b</b>) and ECMWF (<b>c</b>).</p> ">
Abstract
:1. Introduction
2. Datasets
2.1. HY-2B Scatterometer (HY-2B SCAT) Wind Field Data
2.2. Advanced Scatterometer (ASCAT) Wind Field Data
2.3. Reanalysis of Wind Field Data
2.4. Buoys and Other Data
3. Methodology
3.1. F2F and CNN
3.2. Data Matching
3.3. Establishing and Training the Neural Network
4. Results and Analysis
4.1. Wind Speed Validation Using ECMWF Data for the First Set of Test Data
4.2. Wind Speed Validation Using Buoy Data for the First Set of Test Data
4.3. Wind Direction Validation Using ECMWF Data for the First Set of Test Data
4.4. Wind Direction Validation Using Buoy Data for the First Set of Test Data
4.5. Test the F2F-CNN in a Cyclone Wind Field
4.6. Summary of Results Analysis
5. Discussion
6. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Acknowledgments
Conflicts of Interest
References
- Carvajal, G.K.; Eriksson, L.; Ulander, L. Retrieval and Quality Assessment of Wind Velocity Vectors on the Ocean With C-Band SAR. IEEE Trans. Geosci. Remote Sens. 2014, 52, 2519–2537. [Google Scholar] [CrossRef]
- Chelton, D.B.; Freilich, M.H.; Sienkiewicz, J.M.; Ahn, J. On the Use of QuikSCAT Scatterometer Measurements of Surface Winds for Marine Weather Prediction. Mon. Weather Rev. 2006, 134, 2055–2071. [Google Scholar] [CrossRef]
- Kite-Powell, H. The Value of Ocean Surface Wind Information for Maritime Commerce. Mar. Technol. Soc. J. 2011, 45, 75–84. [Google Scholar] [CrossRef]
- Choisnard, J.; Power, D.; Davidson, F.; Stone, B.; Randell, C. Comparison of Cband SAR algorithms to derive surface wind vectors and initial findings in their use in marine search and rescue. Can. J. Remote Sens. 2007, 33, 1–11. [Google Scholar] [CrossRef]
- Christiansen, M.B.; Koch, W.; Horstmann, J.; Hasager, C.B.; Nielsen, M. Wind resource assessment from C-band SAR. Remote Sens. Environ. 2006, 105, 68–81. [Google Scholar] [CrossRef] [Green Version]
- Cornford, D.; Nabney, I.T.; Bishop, C.M. Neural Network-Based Wind Vector Retrieval from Satellite Scatterometer Data. Neural Comput. Appl. 1999, 8, 206–217. [Google Scholar] [CrossRef]
- Schroeder, L.C.; Boggs, D.H.; Dome, G.; Halberstam, I.M.; Jones, W.L.; Pierson, W.J.; Wentz, F.J. The relationship between wind vector and normalized radar cross section used to derive SEASAT-A satellite scatterometer winds. J. Geophys. Res. 1982, 87, 3318–3336. [Google Scholar] [CrossRef]
- Hersbach, H. Comparison of C-Band Scatterometer CMOD5.N Equivalent Neutral Winds with ECMWF. J. Atmos. Ocean. Technol. 2010, 27, 721–736. [Google Scholar] [CrossRef]
- Arnús, M.P. Wind Field Retrieval from Satellite Radar Systems. TDX 2004, 453, 70–95. [Google Scholar]
- Schultz, H. A circular median filter approach for resolving directional ambiguities in wind fields retrieved from spaceborne scatterometer data. J. Geophys. Res. Ocean. 1990, 95, 5291–5303. [Google Scholar] [CrossRef]
- Long, D.G.; Mendel, J.M. Model-Based Estimation of Wind Fields Over the Ocean from Wind Scatterometer Measurements, Part I: Development of the Wind Field Model. IEEE Trans. Geosci. Remote Sens. 1990, 28, 349–360. [Google Scholar] [CrossRef] [Green Version]
- Richaume, P.; Badran, F.; Crépon, M.; MejìA, C.; Roquet, H.; Thiria, S. Neural network wind retrieval from ERS-1 scatterometer data. Neurocomputing 2000, 30, 37–46. [Google Scholar] [CrossRef]
- Thiria, S.; Mejìa, C.; Badran, F.; Crépon, M. A neural network approach for modeling nonlinear transfer functions: Application for wind retrieval from spaceborne scatterometer data. J. Geophys. Res. 1993. [Google Scholar] [CrossRef]
- Chen, K.S.; Tzeng, Y.C.; Chen, P.C. Retrieval of ocean winds from satellite scatterometer by a neural network. IEEE Trans. Geosci. Remote Sens. 2002, 37, 247–256. [Google Scholar] [CrossRef]
- Kasilingam, D.; Lin, I.I.; Khoo, V.; Hock, L. A neural network-based model for estimating the wind vector using ERS scatterometer data. In Proceedings of the IGARSS’97: 1997 IEEE International Geoscience and Remote Sensing Symposium Proceedings. Remote Sensing—A Scientific Vision for Sustainable Development, Singapore, 3–8 August 1997. [Google Scholar]
- Lin, M.; Song, X.; Jiang, X. Neural network wind retrieval from ERS-1/2 scatterometer data. Acta Oceanol. Sin. Engl. Ed. 2006, 25, 35–39. [Google Scholar]
- Bishop, C.M. Mixture Density Networks. 1994. Available online: https://publications.aston.ac.uk/id/eprint/373/1/NCRG_94_004.pdf (accessed on 10 March 2021).
- Zou, J.; Lin, M.; Zhang, Y.; Mu, B.; Cui, S. Operational regrouping algorithm for HSCAT. J. Remote Sens. 2017, 21, 825–834. [Google Scholar]
- Wang, H.; Zhu, J.; Lin, M.; Zhang, Y.; Chang, Y. Evaluating Chinese HY-2B HSCAT Ocean Wind Products Using Buoys and Other Scatterometers. IEEE Geosci. Remote Sens. Lett. 2020, 17, 923–927. [Google Scholar] [CrossRef]
- Wu, Q.; Chen, G. Validation and intercomparison of HY-2A/MetOp-A/Oceansat-2 scatterometer wind products. Chin. J. Oceanol. 2015, 33, 1181–1190. [Google Scholar] [CrossRef]
- Krasnopolsky, V.M. The Application of Neural Networks in the Earth System Sciences; Springer: Berlin/Heidelberg, Germany, 2013. [Google Scholar]
- Kussul, N.; Lavreniuk, M.; Skakun, S.; Shelestov, A. Deep Learning Classification of Land Cover and Crop Types Using Remote Sensing Data. IEEE Geosci. Remote Sens. Lett. 2017, 14, 778–782. [Google Scholar] [CrossRef]
- Gómez-Ríos, A.; Tabik, S.; Luengo, J.; Shihavuddin, A.; Krawczyk, B.; Herrera, F. Towards Highly Accurate Coral Texture Images Classification Using Deep Convolutional Neural Networks and Data Augmentation. Expert Syst. Appl. 2019, 118, 315–328. [Google Scholar] [CrossRef] [Green Version]
- Yu, Y.; Zhang, W.; Wu, Z.; Yang, X.; Cao, X.; Zhu, M. Assimilation of HY-2A scatterometer sea surface wind data in a 3DVAR data assimilation system—A case study of Typhoon Bolaven. Front. Earth Sci. 2015, 9, 192–201. [Google Scholar] [CrossRef]
- Chen, K.; Dong, X.; Xingou, X.; Lang, S. The Study on Oceanic Vector Wind Field Retrieve Technique based on Neural Networks of Microwave Scatterometer. Remote Sens. Technol. Appl. 2017, 32, 683–690. [Google Scholar]
- Ribal, A.; Young, I.R. Calibration and Cross Validation of Global Ocean Wind Speed Based on Scatterometer Observations. J. Atmos. Ocean. Technol. 2020, 37, 279–297. [Google Scholar] [CrossRef]
Technical Parameters | Values |
---|---|
Working frequency/GHz | 13.256 |
Observation swath/km | Outer beam: 1700 |
Inner beam: 1350 | |
Ground resolution/km | 25 |
Backscattering coefficient accuracy/dB | 0.5 |
Backscattering coefficient range/dB | −40∼20 |
Wind speed range/(m/s) | 2∼24 |
Wind speed retrieval accuracy/(m/s) | ±2 (10%) |
Wind direction retrieval accuracy/(°) | ±20 |
Algorithm | F2F-CNN | HY-2B | MetOp-A | MetOp-B |
---|---|---|---|---|
RMSE (m/s) | 0.4269 | 1.2042 | 0.7806 | 0.7428 |
Bias (m/s) | 0.0528 | −0.0465 | −0.0546 | 0.0843 |
SI | 0.1415 | 0.1530 | 0.1363 | 0.1537 |
r | 0.9964 | 0.9500 | 0.9314 | 0.9570 |
Algorithm | ECMWF | F2F-CNN | HY-2B | MetOp-A | MetOp-B |
---|---|---|---|---|---|
RMSE (m/s) | 0.7223 | 0.7137 | 1.5044 | 0.9929 | 0.9585 |
Bias (m/s) | 0.0601 | 0.0517 | −0.1222 | −0.1102 | 0.1016 |
SI | 0.1538 | 0.1501 | 0.1504 | 0.1418 | 0.1447 |
r | 0.9846 | 0.9834 | 0.9395 | 0.9481 | 0.9582 |
Algorithm | ECMWF | F2F-CNN | HY-2B | MetOp-A | MetOp-B |
---|---|---|---|---|---|
RMSE (m/s) | 0.6779 | 0.7065 | 0.9211 | 0.9061 | 0.7324 |
Bias (m/s) | −0.3209 | −0.2949 | −0.2163 | 0.2350 | 0.2556 |
SI | 0.0803 | 0.0860 | 0.1454 | 0.1115 | 0.1272 |
r | 0.9591 | 0.9558 | 0.8747 | 0.8999 | 0.9566 |
Algorithm | F2F CNN | HY-2B | MetOp-A | MetOp-B |
---|---|---|---|---|
RMSE (rad) | 0.1331 | 0.2822 | 0.2629 | 0.2310 |
Bias (rad) | −0.0561 | −0.0489 | 0.0443 | 0.0288 |
SI | 0.1303 | 0.1510 | 0.1636 | 0.1489 |
r | 0.9982 | 0.9855 | 0.9876 | 0.9813 |
Algorithm | ECMWF | F2F-CNN | HY-2B | MetOp-A | MetOp-B |
---|---|---|---|---|---|
RMSE (rad) | 0.1819 | 0.1775 | 0.3318 | 0.2476 | 0.2286 |
Bias (rad) | 0.0140 | 0.0182 | −0.0192 | −0.1375 | −0.0945 |
SI | 0.0588 | 0.0572 | 0.0838 | 0.0840 | 0.0709 |
r | 0.9961 | 0.9959 | 0.9898 | 0.8972 | 0.8987 |
Algorithm | ECMWF | F2F-CNN | HY-2B | MetOp-A | MetOp-B |
---|---|---|---|---|---|
RMSE (rad) | 0.1698 | 0.1696 | 0.2506 | 0.2188 | 0.1946 |
Bias (rad) | 0.0172 | 0.0306 | 0.2137 | −0.1218 | −0.0634 |
SI | 0.0361 | 0.0356 | 0.0268 | 0.0367 | 0.0215 |
r | 0.9964 | 0.9963 | 0.9915 | 0.9626 | 0.9891 |
Algorithm | ECMWF | F2F-CNN | HY-2B | MetOp-A | MetOp-B |
---|---|---|---|---|---|
RMSE (m/s) | 0.7118 | 0.6961 | 1.5333 | 0.9829 | 0.9585 |
Bias (m/s) | 0.3841 | 0.3371 | −0.2731 | −0.2422 | 0.1649 |
SI | 0.1053 | 0.1079 | 0.2140 | 0.1418 | 0.1047 |
r | 0.9422 | 0.9721 | 0.8903 | 0.8998 | 0.9582 |
Algorithm | ECMWF | F2F-CNN | HY-2B | MetOp-A | MetOp-B |
---|---|---|---|---|---|
RMSE (rad) | 0.1841 | 0.1764 | 0.3116 | 0.2868 | 0.2278 |
Bias (rad) | 0.0710 | 0.1329 | −0.2266 | −0.2510 | −0.1387 |
SI | 0.0788 | 0.0494 | 0.0967 | 0.0631 | 0.0509 |
r | 0.9456 | 0.9571 | 0.9102 | 0.9566 | 0.9590 |
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |
© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Shi, X.; Duan, B.; Ren, K. A More Accurate Field-to-Field Method towards the Wind Retrieval of HY-2B Scatterometer. Remote Sens. 2021, 13, 2419. https://doi.org/10.3390/rs13122419
Shi X, Duan B, Ren K. A More Accurate Field-to-Field Method towards the Wind Retrieval of HY-2B Scatterometer. Remote Sensing. 2021; 13(12):2419. https://doi.org/10.3390/rs13122419
Chicago/Turabian StyleShi, Xinjie, Boheng Duan, and Kaijun Ren. 2021. "A More Accurate Field-to-Field Method towards the Wind Retrieval of HY-2B Scatterometer" Remote Sensing 13, no. 12: 2419. https://doi.org/10.3390/rs13122419