Multitemporal Cloud Masking in the Google Earth Engine
<p>Geographic location of the 96 images from the Landsat-8 Biome dataset [<a href="#B31-remotesensing-10-01079" class="html-bibr">31</a>] ingested on the Google Earth Engine.</p> "> Figure 2
<p>Multitemporal cloud detection scheme implemented on the Google Earth Engine platform.</p> "> Figure 3
<p>Illustration of the Cloud detection scheme. Comparison between the ground truth and the proposed cloud mask algorithm and the FMask. Discrepancies are shown in blue when the proposed method detects ‘cloudy’, and in orange when pixels are classified as ‘cloud-free’.</p> "> Figure 4
<p>This figure shows ROC curves for the four proposed background estimation methods. Crosses show the TPR and FPR values for the proposed threshold (0.04) and for FMask [<a href="#B5-remotesensing-10-01079" class="html-bibr">5</a>] and ACCA (BQA) [<a href="#B27-remotesensing-10-01079" class="html-bibr">27</a>] on the same dataset.</p> "> Figure 5
<p>Landsat 8 image (LC81820302014180LGN00 006_013) acquired on 10 June 2013. Rural area in Bulgaria presenting crops misclassified as clouds when the threshold in reflectance is not applied.</p> "> Figure 6
<p>(<b>Left</b>) Scatter plot of the clusters. Norm of TOA reflectance of the visible bands on the Y-axis (<math display="inline"><semantics> <mi>γ</mi> </semantics></math>) and norm of difference in reflectance on the X-axis (<math display="inline"><semantics> <mi>α</mi> </semantics></math>). Each point corresponds to one of the 10 clusters from each of the 2661 image patches. Vertical and horizontal lines show the proposed thresholds. (<b>Right</b>) ROC curves with and without the extra threshold on reflectance <math display="inline"><semantics> <mi>γ</mi> </semantics></math>. The median is used for background estimation in both cases.</p> "> Figure 7
<p>Average accuracy over the image patches for each of the 23 different Landsat-8 acquisitions selected from the Biome dataset.</p> "> Figure 8
<p>Patches of 500 × 500 pixels from Biome dataset. From left to right: RGB scene, RGB scene with ground truth cloud mask in yellow, differences between ground truth and the proposed cloud mask (using the median as background estimation), difference between ground truth and FMask, and estimated cloud-free image.</p> "> Figure 9
<p>Root mean square error between the estimated and actual background pixels in the cloud-free areas. Distribution of the errors is shown separately for all Landsat bands for the four background estimation approaches.</p> ">
Abstract
:1. Introduction
- Data access and image retrieval. Most of the proposed methods assume that a sufficiently long time series of collocated images is available. It is worth pointing out that retrieving the images to build the time series in an easy and operational manner is technically difficult. We need access to the full catalog and powerful enough GIS software to select and co-register the images. We overcome this limitation using the Google Earth Engine (GEE) platform.
- Computational cost. Most of the proposed methods require a sufficiently long time series of images to operate: at least 15 in the case of TMask [16] and at least 10 in the case of MTCD [14]. This is a critical problem if the algorithm cannot be implemented using parallel computing techniques. We again solve this issue ensuring that our algorithm can be implemented on the GEE cloud computing platform.
- Validation of results. A consistent drawback in most cloud detection studies is the lack of quantitative validation over a large collection of independent images. On the one hand, as we have mentioned in the two previous points, if the multitemporal algorithm is computationally demanding and required images are hard to retrieve, it will be difficult to test the method over a large dataset. On the other hand, simultaneous collocated observations informing about the presence of clouds or independent datasets of manually annotated clouds are often not available. Therefore, without a comprehensive ground truth, validation of cloud detection results is usually limited to a visual inspection of the generated cloud masks. In this work we take advantage of the recently released Landsat-8 Biome Cloud Masks dataset [10], which contains manually generated cloud masks from 96 images from different Biomes around the world.
2. Satellite Data and Ground Truth
2.1. Landsat-8 Data
2.2. Google Earth Engine Platform
2.3. Cloud Detection Ground Truth
3. Methodology
3.1. Background Estimation
- Nearest date: It consists of taking the nearest cloud-free image in time as the background. This is the approach used in Ref. [15], however they rely on human intervention to assess that the image does not present any cloud.
- Median filter: It takes the pixel-wise median over time using the three previous cloud-free images. This is the approach suggested in TMask [16] for pixels where the time series is not long enough.
- Kernel regression: The nonlinear version of the former method. It is based on a specific kernel ridge regression (KRR) formulation for change detection presented in Ref. [18].
3.2. Change Detection and Clouds Identification
3.3. Remarks
4. Experimental Results
4.1. Cloud Detection Example
4.2. Parameters and Errors Analysis
4.3. Cloud Detection Results
4.4. Cloud Removal
5. Algorithm Implementation in the Google Earth Engine
6. Discussion
7. Conclusions
Supplementary Materials
Author Contributions
Funding
Conflicts of Interest
References
- Gómez-Chova, L.; Camps-Valls, G.; Calpe, J.; Guanter, L.; Moreno, J. Cloud-Screening Algorithm for ENVISAT/MERIS Multispectral Images. IEEE Trans. Geosci. Remote Sens. 2007, 45, 4105–4118. [Google Scholar] [CrossRef] [Green Version]
- Malenovsky, Z.; Rott, H.; Cihlar, J.; Schaepman, M.E.; García-Santos, G.; Fernandes, R.; Berger, M. Sentinels for science: Potential of Sentinel-1, -2, and -3 missions for scientific observations of ocean, cryosphere and land. Remote Sens. Environ. 2012, 120, 91–101. [Google Scholar] [CrossRef]
- Irish, R.R.; Barker, J.L.; Goward, S.N.; Arvidson, T. Characterization of the Landsat-7 ETM+ Automated Cloud-Cover Assessment (ACCA) Algorithm. Photogramm. Eng. Remote Sens. 2006, 72, 1179–1188. [Google Scholar] [CrossRef]
- Zhu, Z.; Woodcock, C.E. Object-based cloud and cloud shadow detection in Landsat imagery. Remote Sens. Environ. 2012, 118, 83–94. [Google Scholar] [CrossRef]
- Zhu, Z.; Wang, S.; Woodcock, C.E. Improvement and expansion of the Fmask algorithm: Cloud, cloud shadow, and snow detection for Landsats 4–7, 8, and Sentinel 2 images. Remote Sens. Environ. 2015, 159, 269–277. [Google Scholar] [CrossRef]
- Mei, L.; Vountas, M.; Gómez-Chova, L.; Rozanov, V.; Jäger, M.; Lotz, W.; Burrows, J.P.; Hollmann, R. A Cloud masking algorithm for the XBAER aerosol retrieval using MERIS data. Remote Sens. Environ. 2017, 197, 141–160. [Google Scholar] [CrossRef]
- Hughes, M.J.; Hayes, D.J. Automated Detection of Cloud and Cloud Shadow in Single-Date Landsat Imagery Using Neural Networks and Spatial Post-Processing. Remote Sens. 2014, 6, 4907–4926. [Google Scholar] [CrossRef] [Green Version]
- Hollstein, A.; Segl, K.; Guanter, L.; Brell, M.; Enesco, M. Ready-to-Use Methods for the Detection of Clouds, Cirrus, Snow, Shadow, Water and Clear Sky Pixels in Sentinel-2 MSI Images. Remote Sens. 2016, 8, 666. [Google Scholar] [CrossRef]
- Iannone, R.Q.; Niro, F.; Goryl, P.; Dransfeld, S.; Hoersch, B.; Stelzer, K.; Kirches, G.; Paperin, M.; Brockmann, C.; Gómez-Chova, L.; et al. Proba-V cloud detection Round Robin: Validation results and recommendations. In Proceedings of the 2017 9th International Workshop on the Analysis of Multitemporal Remote Sensing Images (MultiTemp), Brugge, Belgium, 27–29 June 2017; pp. 1–8. [Google Scholar]
- Foga, S.; Scaramuzza, P.L.; Guo, S.; Zhu, Z.; Dilley, R.D.; Beckmann, T.; Schmidt, G.L.; Dwyer, J.L.; Joseph Hughes, M.; Laue, B. Cloud detection algorithm comparison and validation for operational Landsat data products. Remote Sens. Environ. 2017, 194, 379–390. [Google Scholar] [CrossRef] [Green Version]
- Loveland, T.R.; Dwyer, J.L. Landsat: Building a strong future. Remote Sens. Environ. 2012, 122, 22–29. [Google Scholar] [CrossRef]
- Drusch, M.; Bello, U.D.; Carlier, S.; Colin, O.; Fernandez, V.; Gascon, F.; Hoersch, B.; Isola, C.; Laberinti, P.; Martimort, P.; et al. Sentinel-2: ESA’s Optical High-Resolution Mission for GMES Operational Services. Remote Sens. Environ. 2012, 120, 25–36. [Google Scholar] [CrossRef]
- Wang, B.; Ono, A.; Muramatsu, K.; Fujiwara, N. Automated Detection and Removal of Clouds and Their Shadows from Landsat TM Images. IEICE Trans. Inf. Syst. 1999, 82, 453–460. [Google Scholar]
- Hagolle, O.; Huc, M.; Pascual, D.V.; Dedieu, G. A multi-temporal method for cloud detection, applied to FORMOSAT-2, VENμS, LANDSAT and SENTINEL-2 images. Remote Sens. Environ. 2010, 114, 1747–1755. [Google Scholar] [CrossRef] [Green Version]
- Jin, S.; Homer, C.; Yang, L.; Xian, G.; Fry, J.; Danielson, P.; Townsend, P.A. Automated cloud and shadow detection and filling using two-date Landsat imagery in the USA. Int. J. Remote Sens. 2013, 34, 1540–1560. [Google Scholar] [CrossRef]
- Zhu, Z.; Woodcock, C.E. Automated cloud, cloud shadow, and snow detection in multitemporal Landsat data: An algorithm designed specifically for monitoring land cover change. Remote Sens. Environ. 2014, 152, 217–234. [Google Scholar] [CrossRef]
- Frantz, D.; Röder, A.; Udelhoven, T.; Schmidt, M. Enhancing the Detectability of Clouds and Their Shadows in Multitemporal Dryland Landsat Imagery: Extending Fmask. IEEE Geosci. Remote Sens. Lett. 2015, 12, 1242–1246. [Google Scholar] [CrossRef]
- Gómez-Chova, L.; Amorós-López, J.; Mateo-García, G.; Muñoz-Marí, J.; Camps-Valls, G. Cloud masking and removal in remote sensing image time series. J. Appl. Remote Sens. 2017, 11, 015005. [Google Scholar] [CrossRef]
- Candra, D.S.; Phinn, S.; Scarth, P. Cloud and cloud shadow removal of landsat 8 images using Multitemporal Cloud Removal method. In Proceedings of the 2017 6th International Conference on Agro-Geoinformatics, Fairfax, VA, USA, 7–10 August 2017; pp. 1–5. [Google Scholar]
- Bovolo, F.; Bruzzone, L. The Time Variable in Data Fusion: A Change Detection Perspective. IEEE Geosci. Remote Sens. Mag. 2015, 3, 8–26. [Google Scholar] [CrossRef]
- Melgani, F. Contextual reconstruction of cloud-contaminated multitemporal multispectral images. IEEE Trans. Geosci. Remote Sens. 2006, 44, 442–455. [Google Scholar] [CrossRef]
- Lin, C.H.; Tsai, P.H.; Lai, K.H.; Chen, J.Y. Cloud Removal From Multitemporal Satellite Images Using Information Cloning. IEEE Trans. Geosci. Remote Sens. 2013, 51, 232–241. [Google Scholar] [CrossRef]
- Hu, G.; Li, X.; Liang, D. Thin cloud removal from remote sensing images using multidirectional dual tree complex wavelet transform and transfer least square support vector regression. J. Appl. Remote Sens. 2015, 9, 095053. [Google Scholar] [CrossRef]
- Chen, B.; Huang, B.; Chen, L.; Xu, B. Spatially and Temporally Weighted Regression: A Novel Method to Produce Continuous Cloud-Free Landsat Imagery. IEEE Trans. Geosci. Remote Sens. 2017, 55, 27–37. [Google Scholar] [CrossRef]
- Gorelick, N.; Hancher, M.; Dixon, M.; Ilyushchenko, S.; Thau, D.; Moore, R. Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sens. Environ. 2017, 202, 18–27. [Google Scholar] [CrossRef]
- Chander, G.; Markham, B.L.; Helder, D.L. Summary of current radiometric calibration coefficients for Landsat MSS, TM, ETM+, and EO-1 ALI sensors. Remote Sens. Environ. 2009, 113, 893–903. [Google Scholar] [CrossRef] [Green Version]
- Scaramuzza, P.L.; Bouchard, M.A.; Dwyer, J.L. Development of the Landsat Data Continuity Mission Cloud-Cover Assessment Algorithms. IEEE Trans. Geosci. Remote Sens. 2012, 50, 1140–1154. [Google Scholar] [CrossRef]
- Dean, J.; Ghemawat, S. MapReduce: Simplified Data Processing on Large Clusters. In Proceedings of the Sixth Symposium on Operating System Design and Implementation (OSDI’04), San Francisco, CA, USA, 6–8 December 2004; pp. 137–150. [Google Scholar]
- U.S. Geological Survey. L7 Irish Cloud Validation Masks; Data Release; U.S. Geological Survey: Reston, VA, USA, 2016.
- U.S. Geological Survey. L8 SPARCS Cloud Validation Masks; Data Release; U.S. Geological Survey: Reston, VA, USA, 2016.
- U.S. Geological Survey. L8 Biome Cloud Validation Masks; Data Release; U.S. Geological Survey: Reston, VA, USA, 2016.
- Fawcett, T. An introduction to ROC analysis. Pattern Recognit. Lett. 2006, 27, 861–874. [Google Scholar] [CrossRef]
- Meng, F.; Yang, X.; Zhou, C.; Li, Z. A Sparse Dictionary Learning-Based Adaptive Patch Inpainting Method for Thick Clouds Removal from High-Spatial Resolution Remote Sensing Imagery. Sensors 2017, 17, 2130. [Google Scholar] [CrossRef] [PubMed]
- Oreopoulos, L.; Wilson, M.J.; Várnai, T. Implementation on Landsat Data of a Simple Cloud-Mask Algorithm Developed for MODIS Land Bands. IEEE Geosci. Remote Sens. Lett. 2011, 8, 597–601. [Google Scholar] [CrossRef] [Green Version]
- Mahajan, D.; Girshick, R.; Ramanathan, V.; He, K.; Paluri, M.; Li, Y.; Bharambe, A.; van der Maaten, L. Exploring the Limits of Weakly Supervised Pretraining. arXiv, 2018; arXiv:1805.00932. [Google Scholar]
True Positive Rate | TPR | |
False Positive Rate (Commission Error) | FPR | |
Omission Error | 1- TPR | |
Overall Accuracy |
Method | Overall Accuracy | Kappa Statistic | Commission Error | Omission Error |
---|---|---|---|---|
FMask [5] | 88.18% | 0.7550 | 16.64% | 2.62% |
ACCA (BQA) [27] | 90.45% | 0.7933 | 9.90% | 8.86% |
Nearest date | 94.18% | 0.8733 | 6.31% | 4.87% |
Median filter | 94.13% | 0.8720 | 6.36% | 4.94% |
Linear regression | 93.66% | 0.8593 | 4.78% | 9.32% |
Kernel regression | 93.56% | 0.8572 | 4.82% | 9.53% |
Thresholds | Overall Accuracy | Kappa Statistic | Commission Error | Omission Error | |
---|---|---|---|---|---|
Difference () | Reflectance () | ||||
0.02 | 0.000 | 85.54% | 0.7076 | 21.53% | 0.95% |
0.02 | 0.150 | 89.52% | 0.7820 | 15.16% | 1.54% |
0.02 | 0.175 | 92.59% | 0.8415 | 9.61% | 3.21% |
0.03 | 0.000 | 89.56% | 0.7823 | 14.90% | 1.93% |
0.03 | 0.150 | 91.43% | 0.8187 | 11.83% | 2.35% |
0.03 | 0.175 | 93.63% | 0.8624 | 7.76% | 3.74% |
0.04 | 0.000 | 91.63% | 0.8217 | 10.86% | 3.63% |
0.04 | 0.150 | 92.45% | 0.8382 | 9.43% | 3.95% |
0.04 | 0.175 | 94.13% | 0.8720 | 6.36% | 4.94% |
0.05 | 0.000 | 92.50% | 0.8378 | 8.54% | 5.53% |
0.05 | 0.150 | 92.97% | 0.8474 | 7.70% | 5.76% |
0.05 | 0.175 | 94.26% | 0.8738 | 5.32% | 6.54% |
FMask | 88.18% | 0.7550 | 16.64% | 2.62% | |
ACCA (BQA) | 90.45% | 0.7933 | 9.90% | 8.86% |
© 2018 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 (http://creativecommons.org/licenses/by/4.0/).
Share and Cite
Mateo-García, G.; Gómez-Chova, L.; Amorós-López, J.; Muñoz-Marí, J.; Camps-Valls, G. Multitemporal Cloud Masking in the Google Earth Engine. Remote Sens. 2018, 10, 1079. https://doi.org/10.3390/rs10071079
Mateo-García G, Gómez-Chova L, Amorós-López J, Muñoz-Marí J, Camps-Valls G. Multitemporal Cloud Masking in the Google Earth Engine. Remote Sensing. 2018; 10(7):1079. https://doi.org/10.3390/rs10071079
Chicago/Turabian StyleMateo-García, Gonzalo, Luis Gómez-Chova, Julia Amorós-López, Jordi Muñoz-Marí, and Gustau Camps-Valls. 2018. "Multitemporal Cloud Masking in the Google Earth Engine" Remote Sensing 10, no. 7: 1079. https://doi.org/10.3390/rs10071079