Historic, Archive Document Do not assume content reflects current scientific knowledge, policies, or practices. ► 7 Ax i t- I’ «' Reserve aG70 .5 .T76M37 2007 United States Creating Cloud-Free Landsat Department of Agriculture ETM+ Data Sets in Tropical Forest Service Landscapes: Cloud and International Institute of Tropical Forestry Cloud-Shadow Removal General Technical Report IITF-GTR-32 February 2007 Sebastian Martinuzzi, William A. Gould, and Olga M. Ramos Gonzalez 67°15'0''W 67°0'0"W 66M5'0"W 66'’30'0"W 66°15'0"W 66°0'0”W 65”45'0"W 65°30'0"W 65°15'0"W N 0'' 6' 3 8° 1 PUER O RICO N 0'' 7' 2 8° CULEB RA 1 N 0” 8' 1 8° 1 VIEQUES N 0" 9' 8° 1 N 18”0'0" N Kilomelers O" 5r 7“ 1 67”15'0"W 67°0'0"W 66°45'0"W 66°30'0"W 66°15'0"W 66”0'0"W 65°45'0"W 65"30'0"W 65°15'0"W The Forest Service of the U S. Department of Agriculture is dedicated to the principle of multiple use management of the Nation’s forest resources for sustained yields of wood, water, forage, wildlife, and recreation. Through forestry research, cooperation with the States and private forest owners, and management of the National Forests and National Grasslands, it strives—as directed by Congress—to provide increasingly greater service to a growing Nation. The U.S. Department of Agriculture (USDA) prohibits discrimination in all its programs and activities on the basis of race, color, national origin, age, disability, and where applicable, sex, marital status, familial status, parental status, religion, sexual orientation, genetic information, political beliefs, reprisal, or because all or part of an individual’s income is derived from any public assistance program. (Not all prohibited bases apply to all programs.) Persons with disabilities who require alternative means for communication of program information (Braille, large print, audiotape, etc.) should contact USDA’s TARGET Center at (202) 720-2600 (voice and TDD). To file a complaint of discrimination, write USDA, Director, Office of Civil Rights, 1400 Independence Avenue, SW, Washington, DC 20250-9410 or call (800) 795-3272 (voice) or (202) 720-6382 (TDD). USDA is an equal opportunity provider and employer. Authors Sebastian Martinuzzi is an ecologist, Laboratorio de Investigacion de Sistenias Ecologicos y Ambientales, Universidad Nacional de La Plata, Diagonal 113 s/n Escuela Superior de Bosques, CP 1900, Argentina, and visiting scholar at the International Institute of Tropical Forestry. William A. Gould is a research ecologist, and Olga M. Ramos Gonzalez is a geographic information system analyst, U.S. Department of Agriculture, Forest Serv ice, International Institute of Tropical Forestry, Jardin Botanieo Sur, 1201 Calle Ceiba, Rio Piedras, PR 00926-1119. SEP 2 4 SEC'B Abstract Martinuzzi, Sebastian; Gould, William A.; Ramos Gonzalez, Olga M. 20(16. Creating cloud-free Landsat ETM+ data sets in tropica! landscapes: cloud and cloud-shadow remtwal. Gen. Tech. Rep. llTF-32. Rio Piedras, PR: U.S. Depart¬ ment of Agriculture, Forest Service, International Institute of Tropical Forestry. 12 p. Clouds and cloud shadows are common features of visible and infrared remotely- sensed images collected from many parts of the world, particularly in humid and tropical regions. We have developed a simple and semiautomated method to mask clouds and shadows in Fandsat FTM-i- imagery, and have developed a recent cloud-free composite of multitemporal images for Puerto Rico and its adjacent islands that can be used for a variety of landscape analyses. Our assumption is that if clouds and shadows can be identified in a reference image, they can be replaced with data from other dates. We created cloud masks by using Fandsat FTM+ band 1 (blue) and thermal band 6.1. Additionally, Fandsat FTM-i- band 4 (near infrared) and parameters of sun angle, topography, and cloud-shadow projection were used for directing and masking shadows. This methodology was applied to a set of 18 images from 1999 to 2003 to develop an island-wide image that is 96.5 percent cloud free. We considered the seasonality of the imagery when selecting reference images and building the mosaic in order to minimize variation in reflectance related to dry or wet season canopy characteristics. We developed a higher resolution data set by merging the 15-m resolution panchromatic band with the 30-m resolution Fandsat FTM+ data. The methodology developed is simple and straightforward to use wherever obtaining cloud-free image data sets is desirable and can be integrated into other efforts that demand an accurate method for the identification of clouds and shadows. Keywords: Remote sensing, clouds, shadows, mask, Fandsat FTM-i-, Puerto Rico. Creating Cloud-Free Landsat ETM* Data Sets in Tropical Landscapes: Cloud and Cloud-Shadow Removal Introduction Clouds are common features of visible and infrared remotely-sensed images col¬ lected from many tropical, humid, mountainous, and coastal regions of the world. Puerto Rico is a mountainous tropical island with pressing needs for readily appli¬ cable methods to develop Landsat ETM+ cloud-free scenes for analyses of current land cover and land-cover change. Landsat data has been used in Puerto Rico for high temporal resolution analyses of land-cover change related to urban expansion and regeneration of forests over abandoned agricultural land (Grau et al. 2003; Hel- mer 2004; Helmer and Ruefenacht 2005; Lopez et al. 2001; Thomlinson et al. 1996; Ramos-Gonzalez 2001), estimation of forested area, identification of forests types (Helmer et al. 2002), and current land-cover classification, a key component for mapping habitats and biodiversity as part of the Puerto Rico Gap Analysis Program (Gould 2005). Cloud-free data acquisition for a single date is difficult or impossible for the entire island, particularly for higher elevations where a high percentage of the protected forest resources are located. In the past, the continued presence of clouds in imagery has reduced the amount of island-wide data available, and as a result has restricted analyses to those years for which low cloud cover images exist (1985, 1991, and 1992). Cloud-free image data sets are useful both for the creation of composites and for comparison of temporal variation in spectral signals. In the creation of composite imagery, data can be radiometrically corrected, mosaicked, and analyzed as a single image for classification of general land cover types, or scenes can be classified individually and mosaicked postclassification. The Puerto Rican landscape represents a challenge for clouds and shadow identification in Landsat ETM+ images owing to the complexity and heterogene¬ ity of land cover (including different forest types and a variety of high-reflectance surfaces such as urban, barren, quarries, rocks, salt mining, sand, pastures and agricultural lands), and topography. In the visible and near-infrared spectral regions of Landsat ETM+ images, clouds are generally characterized by reflectance higher than the underlying (land or ocean) surface. However, under certain surface con¬ ditions, such as urban/built-up areas (as for iee and snow in temperate and cold regions), the reflectance difference between cloud and underlying surface can be insufficiently low or even reversed (Li et al. 1999, Schroder et al. 2002). Optical remote sensing images also have the problem of cloud shadows. The spectral irradi- ance reaching the ground in cloud-shadow areas is both reduced in intensity and changed in its spectral properties (Choi and Milton 1999), similar to that of topo¬ graphic shadows. The presence of two additional thermal bands in Landsat imagery 1 GENERAL TECHNICAL REPORT IITF-GTR-32 (bands 6.1 and 6.2) allows for the analysis of the surface in terms of ground or canopy surface temperatures (Melesse and Jordan 2002). In these bands, clouds, tree foliage, and damp ground appear in dark tones (cooler), whereas grassland and other low-lying vegetation, dry ground, concrete, and asphalt appear in brighter tones (warmer) (Jensen 2000). The development of cloud-free composites involves both the identification of cloud features as well as distinguishing cloud shadows from topographic shadows. Various techniques have been developed to minimize the effect of clouds and cloud-shadows (Ackerman et al. 1998, Choi and Milton 1999, Helmer and Ruef- enacht 2005, Li et al. 1999, Liew et al. 1998, Milton and Choi 1999, Schroder et al. 2002, Song and Civco 2002). We used optical and thermal spectral characteristics as well as geographic position to identify cloud and cloud shadows. Analysis of spectral reflectance in different bands is essential to identifying features in the images. Song and Civco (2002) used the brightness value in Landsat TM bands 1 (blue) and 4 (near infrared) to detect clouds and shadows in Madagascar. We added thermal information from Landsat ETM band 6.1 to spectrally separate clouds from similarly responding features, such as urban/built-up lands. Although cloud shad¬ ows and topographic shadows are not spectrally distinguishable in optical images, they can be identified by geographic position. The direction of the shadow occur¬ rence relative to the cloud is the same for the entire scene as it depends on the sun illumination angle, a constant for any given scene. The location, form, and dimen¬ sions of any cloud shadow depend on the cloud shape and elevation, topography, and the sun illumination angle. The development of masks for identifying clouds and cloud shadows is our solution to develop a final cloud- and shadow-free Landsat ETM+ composite image. Our principle is to identify clouds and shadows in a reference image and replace them with cloud- and shadow-free parts from other dates. Masks facilitate opera¬ tions such as subtraction and replacement of specific areas, as well as buffering, intersection, and union. We have developed simple semiautomated procedures to generate cloud and shadow masks in tropical regions for use in the creation of cloud-free composite Landsat ETM+ imagery, by using multiple recent images acquired at different times over the same region. Our goals in developing these methods were to make use of spectral and geographic information to facilitate the identification of cloud cover, separate cloud and topographic shadows, and minimize the needed masked areas. 2 Creating Cloud-Free Landsat ETM+ Data Sets in Tropical Landscapes: Cloud and Cloud-Shadow Removal In addition to the masking method, we have incorporated additional processes for the creation of the final imagery data set. We sought to minimize differences in reflectance owing to atmospheric conditions by correcting for Rayleigh scatter and differences in seasonality of vegetation by using, whenever possible, imagery acquired during the same season. Aside from Rayleigh scatter correction, we have not incorporated any methods to normalize systematic scene-to-scene radiometric variation such as dark object subtraction (Chavez 1996). This would be advisable when classifying a composite image but unnecessary when classifying individual scenes before mosaicking. We also improved our ability to separate subpi.xel features (for urban or mangrove areas) by enhancing the spatial resolution incorpo¬ rating the higher (15-m) resolution of the panchromatic band into the multispectral images. We have developed a useful image data set for the islands of Puerto Rico, Vieques, Culebra, and additional small islands within the Puerto Rican Archi¬ pelago. The methods to develop this data set will work equally well throughout the Caribbean basin and similar regions, and with Landsat ETM+ imagery acquired over a range of dates. Data and Methods Four Landsat ETM+ path-rows (4-47, 4-48, 5-47, and 5-48) are necessary to cover Puerto Rico and its adjacent islands of Vieques and Culebra. Our data set is com¬ posed of 18 images acquired between September 1999 and February 2003. During 2002, no images were incorporated owing to persistent dense cloud cover. The complete process leading to the resulting cloud- and cloud-shadow-free composite image can be accomplished in four major steps outlined in our methods, and subsequently applied to our 18 images. These include atmospheric correction, cloud and cloud-shadow masking, resampling and addition of the panchromatic band, and image selection and composition. Atmospheric Correction Atmospheric correction is necessary to put multitemporal data on the same radiometric scale, in order to create an image data set with compatible scenes that is useful for classification and monitoring terrestrial surfaces over time (Song et al. 2001). We used a model developed by the Institute for Astronomy and Space Physics lAFE-CONlTET to convert radiance to reflectance corrected by Rayleigh scatter (Paolini et al. 2004). The input parameters required for running the model are included in the image header and include gain, bias, sun elevation angle, solar azimuth angle, and date. The model is described by the following formula: 3 GENERAL TECHNICAL REPORT IITF-GTR-32 0 Surface reflectance = 6 x (L^ - L Rayleigh) x r“/Ey (cos/6), Where = at satellite radiance = gain^; x dN x bias^, LRayieioh “ radiance owing to Rayleigh scatter, = extraterrestrial irradiance, r = Earth-Sun distance, and 6 = Sun elevation angle. Each scene was proeessed using this formula before proceeding with the cloud and shadow masking. Cloud and Cloud-Shadow Masking This involves two main proeedures: (1) identifying clouds and developing a mask by using Landsat ETM+ brightness values (DN) in optical band 1 (blue) and in thermal band 6.1 and (2) developing a final eloud and cloud-shadow mask by using the cloud mask, the brightness values of shadowed areas in band 4 (near infrared), and the geometrie relationship between shadow position and the clouds (i.e., parameters of cloud projection) (fig. 1). Cloud mask > Cloud and cloud-shadow mask t step(T) Parameters of cloud projection Figure l-Steps for masking clouds and cloud shadows. In step 1, the range of reflectance values for clouds are determined by visually inspecting the range of cloud types and digital numbers (DNs) in band 1 (blue) and in thermal band 6.1. and then selecting DNs in this range. In step 2, band 4 (near infrared), the geometric relationships of shadows to clotids, and the clouds mask, were used in the identification of cloud shadows. 4