A COMPARATIVE ANALYSIS OF CBERS AND LANDSAT DATA

Shrinidhi Ambinakudige, Assistant Professor Jinmu Choi, Assistant Professor Sami Khanal, Graduate Student Department of Geosciences, Mississippi State University Mississippi State, MS 39762 [email protected], [email protected], [email protected]

ABSTRACT

Failure of and aging LANDSAT 5 have created a data gap for the studies that depend on medium resolution satellite data. This raised the need to explore the potential of using satellites other than LANDSAT series to bridge the data gap. In this study, we examined the potential of using Chinese Brazilian Earth Resource Satellite (CBERS) images in the absence of LANDSAT images. CBERS images were selected for comparison because of their close resemblance in electromagnetic spectrum range with LANDSAT images. Images of LANDSAT TM and CBERS of the same area in Brazil were used to evaluate the comparative performances based on the spectral vegetation indices. The results showed that in most cases the LANDSAT images and the CBERS images are comparable. This study also indicated that CBERS images could fill the data gaps in Landsat images for land-cover studies.

INTRODUCTION

Landsat, the longest running enterprise for acquisition of imagery of the earth from space, has acquired millions of imagery which form a unique resource for applications in agricultural, geology, forestry, regional planning, and global change studies for more than 36 years. It is being used in a number of regional, national and international monitoring projects in detection of temporal and spatial land use land cover changes. Multiple versions of Landsat satellites were launched in different time period with enhanced features and operated with in certain time period. Landsat 5 and Landsat 7 are currently operated. However, both Landsat 5 and Landsat 7 have experienced technical problems during the past three years. Landsat 5, launched in 1984, operated successfully till October, 2005 when it faced a problem in solar array. After several times of suspension, its operation was resumed on January 2006 (USGS, 2008). Similarly, Landsat 7, launched in 1999, went flawlessly until May 2003 when a Scan Line Corrector (SLC) failure left wedge shaped spaces of missing data on either side of landsat 7’s scene. Without operating SLC, Landsat 7 now traces zig-zag pattern which has led to approximately 22 percent loss of data. To fulfill the expectation of full coverage single scene, data from multiple acquisitions are being merged to resolve SLC off data gap (USGS, 2008). Due to the aging of Landsat 5 and degraded quality of Landsat 7, future of Landsat imagery is uncertain. Both Landsat 5 and Landsat 7 are estimated to run out of fuel around 2010 (Wu et al., 2006). At the same time data continuity is critical to continue various monitoring programs. Most remote sensing research studies (including environmental studies, vegetation studies, land-use change, precision agriculture, landscape ecology, and urban planning) in the world within the last 30 years has dependent on Landsat technology. There is an immediate problem and must be addressed. Therefore, there is a need for studies to find a viable alternative to Landsat which is fairly inexpensive to acquire and has similar image quality/dependability. This study examines the potentials of using the Chinese-Brazilian earth resource satellite (CBERS) images as a viable data source for replacement of Landsat imagery especially in vegetation studies. For this study Landsat + ETM and CBERS 2 images are used. In the next section data and methodology for comparison of CBERS and Landsat images are discussed. Results section shows the results of comparison of CBERS and Landsat images. Key ideas and findings are summarized in the conclusion.

ASPRS 2009 Annual Conference Baltimore, Maryland Š March 9-13, 2009 DATA AND METHODOLOGY There are three satellite sensors (SPOT, IRS –P6 and CBERS) available that cover similar wavelengths as Landsat. Table 1 shows the sensor comparison between Landsat TM with IRS, Spot and CBERS 2 sensors. The CBERS 2 is also similar to Landsat in visible bands and infrared bands. Although IRS-P6 and Spot satellites also covers wavelengths very similar to Landsat, only CBERS 2 images are used in this study. CBERS 2 CCD images are selected for the comparison because of their close resemblance in the range of electromagnetic spectrum covered to that of Landsat satellites.

Table 1. Sensor comparison between Landsat +ETM and similar satellites.

LANDSAT +ETM CBERS 2 (CCD) IRS-P6 SPOT-4 Spectral Reso- Spectral Reso- Spectral Reso- Spectral Reso- Bands (μm) lution Bands (μm) lution Bands (μm) lution Bands (μm) lution 0.45-0.52 30m 0.45-0.52 20m 0.52-0.60 30m 0.52-0.59 20m 0.52-0.59 24m 0.50 – 0.59 20m 0.63-0.69 30m 0.63-0.69 20m 0.62-0.68 24m 0.61– 0.68 20m 0.76-0.90 30m 0.77-0.89 20m 0.77-0.86 24m 0.79 – 0.89 20m 1.55-1.75 30m 0.51-0.73 20m 1.55-1.70 24m 1.58 – 1.75 20m 10.4-12.4 120m 2.08-2.35 30m 0.52-090 15m

CBERS Satellites CBERS, a joint program under development by Brazil and China, is an earth observation satellite similar to Landsat (Wu et al., 2006). On-board sensors include CCD Camera, IRMSS (infrared MSS) and WFI (Wide Field Imager). IRMSS has three spectral bands with 78m resolution and one at 156m resolution. WFI sensors have two bands with a resolution of 258m. The first in CBERS series, CBERS 1 was launched on October, 1999 and CBERS 2 was launched on October 2003. CBERS 2B was launched on September 2007 and successfully generating images. Currently CCD images received are concentrated mainly over China and Brazil. They are not generally available to other countries. CBERS has a sun-synchronous orbit with an altitude of 778 km, with a temporal resolution of about 24 days and swath width of 113 km for CCD sensors. WFI cameras has 5 days revisit period. CBERS-2 CCD images are available for free from the National Institute for Space Research (INPE) for entire Latin America (http://www.cbers.inpe.br/en/index_en.htm). Landsat +ETM scene was purchased from the USGS EROS web site. It was difficult to find the images from two sensors that are recorded at the same time. The images from CBERS and Landsat used in the study are of about 4 months apart (Table 2). Unfortunately Landsat +ETM image downloaded was a SLC off image and there are missing data in the image. The images covered the area close to Sao Paulo in Brazil (Fig 1).

Table 2. Characteristics of the images used in the study

CBERS Landsat +ETM Path 156 220 Row 124 75 Date April 11, 2003 Aug. 11, 2003

ASPRS 2009 Annual Conference Baltimore, Maryland Š March 9-13, 2009

Figure 1. Study area.

CBERS scene was geo-rectified and geo-referenced to Landsat image in order to rectify the possible locational shift in the satellite images. To allow comparison between two scenes with different spatial resolution from different sensor, CBERS (20m) and Landsat (30m) pixels were re-sampled to coarser spatial resolution (60m) (Goward et al, 2003). It also helps to minimize the effect of any unknown radiometric artifacts in CBERS and Landsat images.

Figure 3. CBERS scene of Path -156 and Rows 124 Figure 2. Landsat scene of Path -220 and Rows 75 acquired on Aug. 11, 2003. acquired on April 11, 2003.

Because of lower spatial resolution of Landsat compared to CBERS, Landsat had higher area of coverage than CBERS (Fig 2 and Fig 3). So, the Landsat image was clipped to the size of CBERS. Clouds in both images were masked out and excluded in the analysis. Areas of missing data in the Landsat image were masked out in both Landsat and CBERS satellite images and excluded from the study. In this study three types of comparisons were carried out. To begin with, band to band correlation between Landsat and CBERS were conducted. Inter band correlations were also calculated. In the second step, there

ASPRS 2009 Annual Conference Baltimore, Maryland Š March 9-13, 2009 vegetation indices were calculated in both the satellite images and compared with each other. The third method used was the land cover classification of the images. In order to compare the images, an unsupervised classification was conducted (Ambinakudige 2006, Alves & Skole, 1996) and classified both images into urban, vegetation, open field and water. Comparison of Landsat and CBERS sensor observations on the per-spectral-band basis provides a first order assessment of their relative measurement performance. Further insight into the relative measurement performance is revealed through the inspection of multi-band combinations such as spectral vegetation indices (SVI). There are a multitude of transformations available for visible-near infrared spectral measurements for monitoring vegetation. The purpose of these indices is to compensate for variable background (e.g. soil and litter) reflectance and some forms of atmospheric attenuation, while emphasizing vegetation spectral features (Trishchenko et al., 2002).In this study, three SVIs are used to compare Landsat with CBERS images. 1. The simple ratio (IR/R): ρ SR = nir ρvis 2. Normalized Difference Vegetation Index (NDVI) ρ − ρ NDVI = nir vis ρnir + ρvis 3. Transformed Normalized Difference Vegetation Index (TNDVI). This index has a more complex ratio form for calculating the vegetation but still only uses visible reflectance and infrared reflectance. The equation is: ρ − ρ TNDVI = nir vis + .5 ρnir + ρvis ρ ρ In all the above equations nir = near infrared reflectance; vis = visible reflectance. Square root of infrared reflectance was also calculated.

RESULTS: COMPARISON OF CBERS AND LANDSAT

The results of inter and intra band correlations between two images showed the similarities in bands of Landsat and CBERS (Table 3). The intra band correlation in Landsat and CBERS are very similar.

Table 3. Intra Band Correlation in CBERS and Landsat images.

CBERS Band 1 Band 2 Band 3 Band 4 Landsat Band 1 Band 2 Band 3 Band 4 Band 1 1.00 0.99 0.97 0.97 Band 1 1.00 0.99 0.95 0.93 Band 2 0.99 1.00 0.99 0.96 Band 2 0.99 1.00 0.97 0.93 Band 3 0.97 0.99 1.00 0.94 Band 3 0.95 0.97 1.00 0.82 Band 4 0.97 0.96 0.94 1.00 Band 4 0.93 0.93 0.82 1.00

The correlation coefficient between Bands 2 and 3 in CBERS is high. The same pattern was also observed in Band 2 and Band 3 of Landsat image. Band 1 is also highly correlated with all other three bands in both sensors.

ASPRS 2009 Annual Conference Baltimore, Maryland Š March 9-13, 2009 Table 4. Inter band correlation coefficient between Landsat and CBERS images.

Landsat

Band 1 Band 2 Band 3 Band 4 Band 1 0.90 0.89 0.83 0.86 Band 2 0.88 0.88 0.84 0.84 Band 3 0.88 0.87 0.84 0.82 CBERS Band 4 0.93 0.87 0.81 0.86

The inter band correlation coefficients also showed that the bands of Landsat images are highly correlated to the corresponding bands in CBERS.

Comparison of Land-cover Classes Four land cover classes were identified in the study area using an unsupervised classification method. These classes include urban, vegetation, open field and water. It was found that CBERS image was classified more pixels as vegetation than the Landsat; Landsat image was classified more pixels as urban (Table 5). However, some of these differences could be attributed to the differences in spatial resolution and also the quality of Landsat image.

Table 5. Land-Cover classification (ha)

CBERS Landsat % Difference Urban 5099 6026 -18 Vegetation 4993 4214 16 Open 1385 1243 10

Comparison of Vegetation Indices NDVI, SR, TNDVI and SQRTIR were also calculated from CBERS and Landsat images. Correlation coefficients were calculated on vegetation index layers created from Landsat and CBERS images. The results of these experiments (Table 6) indicated that vegetation indices created from Landsat and CBERS are correlated each other. Low correlation between NDVI values may be attributed to the poor quality of Landsat 7 image used in this study. It may be also because of the time difference between the acquisitions of images.

Table 6. Correlation between the vegetation indices calculated from CBERS and Landsat images.

Vegetation Index Correlation between CBERS and Landsat NDVI 0.56 TNDVI 0.88 SQRTIR 0.91 SR 0.77

CONCLUSIONS

Landsat and CBERS are very similar in various aspects including image quality, spectral bandwidth though there are some basic differences like spatial resolution. Similar band ranges and high resolution images could make CBERS a better substitute for LANDSAT images. In terms of spectral similarity, intra and inter band correlations of Landsat and CBERS images show higher than 90%. The correlations of multiple vegetation indices are about and

ASPRS 2009 Annual Conference Baltimore, Maryland Š March 9-13, 2009 above 80% except NDVI. Also, the results of land use land cover classification show the differences of average 15% in the final land use land cover classes. These experiments show how CBERS image is similar spectrally to Landsat image. The correlation values of vegetation indices, especially NDVI, presented in this paper might be affected by the difference in acquisition date of the two images. Four-month difference in image acquisition can be a really a major factor and thus limitation of this study. In order to exclude some of the discrepancies introduced in this study because of the differences in acquisition dates and also poor quality of Landsat 7 data, a study of comparing the Landsat 5 and CBERS 2 images that are acquired within 5 days interval is underway.

REFERENCES

Alves, D.S. and D.L. Skole, 1996. Characterizing land cover dynamics using multi-temporal imagery, International Journal of Remote Sensing, 17: 835-839. Ambinakudige, S., 2006. Differential Impacts of Commodification of Agriculture in the Western Ghats of India: An Extended Environmental Entitlement Analysis, Tallahassee, Florida State University, U.S. Goward, S.N., et al., 2003. Empirical comparison of Landsat 7 and Ikonos multispectral measurements for selected Earth Observation System (Eos) validation sites, Remote Sensing and Environment, 88: 80-99. Trishchenko, A.P., et al., 2002. Effects of spectral response function on surface reflectance and NDVI measured with moderate resolution satellite sensors, Remote Sensing of Environment, 81(1): 1-18. USGS, 2008. Landsat: A Global Land-Observing Program, U.S. Geological Survey http://egsc.usgs.gov/isb/pubs/factsheets/fs02303.html accessed on Nov-29, 2008. Wu, X., J. Guob, J. Wallacea, S.L. Furbya, and P. Caccettaa, 2006. Evaluation of CBERS image data: Geometric and radiometric aspects, The 13th Australasian Remote Sensing and Photogrammetry Conference.

ASPRS 2009 Annual Conference Baltimore, Maryland Š March 9-13, 2009