The availability of cloud free Landsat TM and ETM+ land observations and implications for global Landsat data production

Transcription

1 14475 The availability of cloud free Landsat TM and ETM+ land observations and implications for global Landsat data production *V. Kovalskyy, D. Roy (South Dakota State University) SUMMARY The NASA funded Web enabled Landsat Data (WELD) project generates 30m composited Landsat 7 Enhanced Thematic Mapper Plus (ETM+) mosaics of the conterminous United States and Alaska using terrain corrected (Level 1T) data. Recent WELD prototyping has demonstrated the feasibility of generating global Landsat monthly products. In this paper the global availability of cloud free Landsat ETM+ and Thematic Mapper (TM) data and the monthly probability of acquiring a cloud free land surface observation for the two instruments independently and fused together are reported for climate year The analysis uses more than 90,000 Landsat Level 1T scene metadata records obtained from the US Landsat archive. The results demonstrate the utility of fusing both Landsat data streams which increases the mean monthly global probability of a cloud free land surface observation by up to 37% relative to using Landsat ETM+ data only. Recommendations for global Landsat processing and implications for global Landsat data fusion with similar polar orbiting high spatial resolution sensors are described. XI th International Conference on Geoinformatics Theoretical and Applied Aspects May 2012, Kiev, Ukraine

2 Introduction A number of projects are developing continental and global coverage 30m Landsat products that take advantage of the recent free US Geological Survey (USGS) Landsat data policy. The NASA funded Web-enabled Landsat Data (WELD) project is generating 30m composited Landsat 7 Enhanced Thematic Mapper Plus (ETM+) mosaics of the conterminous United States and Alaska for a 10 year period ( using terrain corrected (Level 1T) data available in the USGS Landsat archive (Roy et al. 2010a). Recent WELD prototyping has demonstrated the feasibility of global Landsat processing to generate global monthly 30m Landsat ETM+ products. However, as with any optical wavelength satellite sensor, cloud contamination greatly compromises Landsat image usability for land surface studies. Spatio-temporal image availability is further reduced by selective Landsat scene acquisition due to payload, ground station and mission cost constraints and by the 2003 failure of the ETM+ scan line corrector. In this paper the global availability of cloud-free Landsat ETM+ and Thematic Mapper (TM) data and the monthly probability of acquiring a cloud-free clear land surface observation for the two instruments independently and fused together are reported for The analysis is performed at 0.05º resolution using geographic, temporal and cloud fraction scene metadata for the more than 96,000 Landsat Level 1T scenes available in the USGS Landsat archive. The results demonstrate the utility of fusing both Landsat data streams in order to take advantage of their different acquisition patterns and to mitigate the deleterious impact of the Landsat ETM+ scan line failure. Fusion of Landsat ETM+ and TM data increases the mean monthly global probability of a cloud-free land surface observation by up to 37% relative to using Landsat ETM+ data only. Recommendations for global Landsat processing and implications for global data fusion with other Landsat resolution systems including the planned ESA Sentinel-2 system are described. Data Metadata describing the geographic location (latitude and longitude of the four scene corners, scene path and row), date of acquisition, and sensor type (Landsat 5 TM or Landsat 7 ETM+) for 2010 for all daytime Level 1T acquisitions archived by the U.S. Landsat project on 5 th January 2012 from WWW1 were used in this study. The date of metadata access is noted because the U.S. Landsat project has started repatriating Landsat data from other agencies and the volume of available data for these years may increase in the future (Wulder et al. 2012). Only L1T data were used in the analysis, since systematically generated large area Landsat products require the input data to be precisely coregistered (Bindschadler et al. 2008), especially if temporally composited Landsat products are generated (Roy et al. 2010a, Roy, 2000). In this paper the standard climate year definition adopted by the climate modeling community is used, where the year 2010 is defined by December 1 st 2009 to November 30th 2010, so that winter is defined by the months: December, January and February. Globally, a total of 96,301 L1T metadata records in ,440 and 64,861 were obtained for Landsat 5 TM and Landsat 7 ETM+ respectively. The much greater proportion of available Landsat ETM+ data in the U.S. Landsat archive has been documented before and reflects that in the Landsat TM era there was a smaller US data recording and ground system capacity and no systematic acquisition (Roy et al. 2010b, Wulder et al. 2012). Methods A global grid of regularly spaced points was defined in the sinusoidal projection as it provides an uninterrupted equal area projection (Snyder 1987). The sinusoidal grid spacing was set as equivalent to 5.56 km (equivalent to 0.05 at the Equator) and is less than half the minimum across-track overlap observed between adjacent Landsat paths and so ensures that across-track scene overlap is captured in the analysis. Every grid point falling within the 9233 non-antarctic Landsat scenes defined in the U.S. Landsat project Land Definition data base (WWW2) was considered, a total of 6,120,644 land grid points.

3 The probability of there being at least one cloud-free observation of a land grid point within a year and a month were derived from the probability of all of the overpasses over the period being cloudy as: P = 1- n p one+ i [1] i=1 where P one+ is the probability of there being at least one cloud-free observation of the land grid point, n is the number of satellite overpasses in the period of interest over the land grid point, and p i is defined as: p = f 0.05 [2] i i + where p i is the probability of the i-th Landsat acquisition being cloudy, and f i is the Landsat cloud fraction metadata (0,1 9), to give possible cloud probabilities of 0.05, 0.15, , 0.95 (Ju and Roy 2008). To model the impact of the scan line corrector failure removal of 22% of each Landsat ETM scene (Markham et al. 2004), the cloud probability is defined as: p = ( f 0.05) [3] i i + where p i is the probability of the i-th Landsat 7 ETM acquisition being cloudy and f i is the cloud fraction metadata (0,1 9). If there were no Landsat observations of the land grid point within the given period then P one+ is set as zero. Results Figure 1 shows the probability of there being at least one cloud-free observation of each land grid point for climate year 2010 (left column) and for July 2010 (right column). The rows of Figure 1 show the probabilities for Landsat 5 TM (top), Landsat 7 ETM+ (middle), and both sensors combined (bottom). Where there were no acquisitions the land grid locations are colored white. The annual probabilities are high (typically >0.95) and clearly show that the majority of the land is sensed by at least one Landsat ETM+ acquisition. The magnitude of the annual probabilities are broadly similar between Landsat 5 TM and Landsat 7 ETM+ at path/rows acquired by both sensors, with lower probabilities in cloudier regions. This is expected as there is no reason why the Landsat sensors, which have the same overpass time but sense the same location on different days, would capture different cloud conditions when considered over large areas and time periods. In July 2010 the magnitude of the Landsat 7 ETM+ probabilities are generally lower than Landsat 5 TM at path/rows acquired by both sensors because of the impact of the Landsat ETM+ scan line corrector failure (Equation [3]). Combination of Landsat 5 TM and Landsat 7 ETM+ appears to provide higher probabilities than considering one sensor alone. Figure 2(a) shows the monthly mean global probability of there being at least one cloud-free observation of each land grid point for December 2009 to November Figure 2(b) shows the relative difference in the monthly mean global probabilities with respect to Landsat 7 ETM+ as it had more than twice as many as Landsat 7 ETM+ acquisitions in 2010 compared to Landsat 5 TM. The seasonal variation in the probabilities is correlated with the monthly number of Landsat acquistions and monthly mean cloud cover. For all months, the maximum monthly Landsat 7 ETM+ probabilities are greater than for Landsat 5 TM. This is despite the 22% of modelled missing Landsat 7 ETM+ pixels and is because of the much greater Landsat 7 ETM+ acquisition density. The maximum probability is 0.52 and occurs in the northern hemisphere summer. When the two sensors are combined the monthly mean global probability of there being at least one cloud-free observation is always greater than any one sensor alone. Fusion of Landsat ETM+ and TM data increases the mean monthly global probability of a cloud-free land surface observation by up to 37% relative to using Landsat ETM+ data only.

4 Figure1. Probability of there being at least one cloud-free observation of each land grid point (Equation 1) for the entire year of L1T acquisitions sensed in 2010 (left column) and for July 2010 (right column) by Landsat 5 TM (top row), Landsat 7 ETM+ (middle row), and by both Landsat 5 TM and Landsat 7 ETM+ (bottom row). Probabilities colored as 0 = White, 0< Dark Blue <0.25, 0.25 Light Blue <0.5, 0.5 Green<0.75, 0.75 Yellow <0.85, 0.85 Orange <0.95, 0.95=< Red=<1. Grid points not defined as land by the global land mask are shown as grey. Figure 2 Summary monthly global statistics of a) the probability of there being at least one cloud-free observation of each land grid point and b) probability of there being at least one cloud-free observation expressed relatively to ETM+ values in % for Landsat 5 TM (blue circles), Landsat 7 ETM+ (Yellow squares), both Landsat 5 TM and Landsat 7 ETM+ (Green Triangles) at the left hand side.

5 Conclusions Global Landsat data sets have been developed through NASA and USGS data buys but only a fraction of the archive has been exploited (Tucker et al. 2004, Gutman et al., 2008). With the advent of the free Landsat data policy (Wulder et al. 2012) it is now feasible to consider using all the U.S. Landsat archive to generate global gridded 30m Landsat data sets. The metadata analysis described in this paper shows that the use of two Landsat sensors is advantageous for production of monthly gridded remote sensing products. Global 30m observations have been provided by the Thematic Mapper (TM) on Landsat satellites 4 and 5 from 1982 to present, and by the Enhanced Thematic Mapper Plus (ETM+) onboard Landsat 7 since 1999, and possibilities for fusion to provide improved temporal and spatial coverage will likely continue with the launch of the Landsat Data Continuity Mission and Sentinel-2. The remaining challenges are primarily concerned with the provision of sufficient computational satellite data processing and distribution resources. Acknowledgements We thank the WELD team, Landsat 7 Mission Operations Center, USGS EROS Data Center Staff members, and NASA for making this research effort possible. References Bindschadler, R., P. Vornberger, A. Fleming, A. Fox, J. Mullins, D. Binnie, S. J. Paulsen, B. Granneman, and D. Gorodetzky (2008), The Landsat Image Mosaic of Antarctica, Remote Sensing of Environment, 112(12), Gutman, G., Byrnes, R., Masek, J., Covington, S., Justice, C., Franks, S., and Headley, R Towards monitoring Land-cover and land-use changes at a global scale: the global land survey Photogrammetric Engineering and Remote Sensing. 74(1):6-10. Ju, J., and D. P. Roy (2008), The availability of cloud-free Landsat ETM+ data over the conterminous United States and globally, Remote Sensing of Environment, 112(3), Lee, D. S., J. C. Storey, M. J. Choate, and R. W. Hayes (2004), Four years of Landsat-7 on-orbit geometric calibration and performance, Geoscience and Remote Sensing, IEEE, 42(12), Markham, B., Storey, J., Williams, D. and Irons, J., 2004, Landsat sensor performance: history and current status. IEEE Transactions on Geoscience and Remote Sensing, 42, pp Roy, D. P. (2000), The impact of misregistration upon composited wide field of view satellite data and implications for change detection, Geoscience and Remote Sensing, IEEE, 38(4), Roy, D. P., J. Ju, K. Kline, P. L. Scaramuzza, V. Kovalskyy, M. Hansen, T. R. Loveland, E. Vermote, and C. Zhang (2010a), Web-enabled Landsat Data (WELD): Landsat ETM+ composited mosaics of the conterminous United States, Remote Sensing of Environment, 114(1), Roy, D. P., J. Ju, C. Mbow, P. Frost, and T. Loveland (2010b), Accessing free Landsat data via the Internet: Africa's challenge, Remote Sensing Letters, 1(2), Snyder, J. P. (1987), Map projections--a working manual, 383 pp., United States Government Printing Office, Washington, DC. Tucker, C.J., Grant, D.M., Dykstra, J.D NASA s Global Orthorectified Landsat Dataset. Photogrammetric Engineering & Remote Sensing. 70(3), Wulder, M. A., J. G. Masek, W. B. Cohen, T. R. Loveland, and C. E. Woodcock (2012), Opening the archive: How free data has enabled the science and monitoring promise of Landsat, Remote Sensing of Environment, In Press. WWW1 (2012), Landsat Metadata, hosted by USGS at WWW2 (2012), Land Definition Table, hosted by USGS at