NSCAT Views the Earth

|A spaceborne scatterometer is a radar instrument used to measure the radar backscatter of the earth's surface. Data retrieved by the NASA Scatterometer (NSCAT) has been used to create enhanced resolution images of the land and polar regions of the earth using the SIRF algorithm developed by Brigham Young University's Microwave Earth Remote Sensing Laboratory. We have developed a standard product suite of NSCAT-derived SIRF images. This report describes these products and their generation. INTRODUCTION NASA Scatterometer (NSCAT) is a spaceborne scatterometer mounted aboard the National Space Development Agency of Japan's ADvanced Earth Observing Satellite (ADEOS). A spaceborne scatterometer is a radar instrument designed to measure and record the radar backscatter of the earth's surface. Satellite mounted scatterometers such as Seasat, ERS1, and ERS2, have been use to collect information that has primarily been used in oceanic and wind studies. NSCAT is the latest such scatterometer and o ers the highest resolution and measurement accuracy. Backscatter measurement data taken by NSCAT, although of higher resolution than data collected by previous spaceborne scatterometers, is still of insu cient resolution to be used for many land applications. The Scatterometer Image Reconstruction with Filter (SIRF) algorithm developed by the Brigham Young University Microwave Earth Remote Sensing (MERS) laboratory provides a means whereby the o values by several revolutions of the scatterometer can be used to produce images of much higher resolution than the intrinsic measurement resolution. These images can be used to study land and polar regions. This paper explores the general process by which these images are made, and describes the standard NSCAT SIRF image product suite. The le naming convention and image storage formate are also described. THE SIRF ALGORITHM AND THE NSCAT SIRF PRODUCT The Scatterometer Image Reconstruction with Filtering (SIRF) resolution enhancement algorithm is a multivariate nonlinear resolution enhancement algorithm based on modi ed algebraic reconstruction and maximum entropy techniques. This algorithm was originally developed for application to Seasat scatterometer (SASS) measurements [1] and has recently been optimized for application to NSCAT measurements [2]. NSCAT measures the o of the Earth's surface at several azimuth angles and over a variety of incidence angles at nominal resolution of 25 km [?]. Over a limited incidence angle ( ) range of [20 ,55 ], o (in dB) over land and ice is a approximately a linear function of , o( ) = A+ B( 40 ) where A and B are functions of surface characteristics, azimuth angle, and polarization. A is the o value at 40 incidence and is termed the incidence angle normalized o. B describes the dependence of o on . The SIRF algorithm generates image of A and B on a high resolution grid from the NSCAT o measurements using multiple satellite passes [1]. The A and B images are the primary SIRF product. Computation of the A and B images are described in detail in [1] and [2]. Several useful ancillary images are also produced as part of the NSCAT SIRF standard product, including a set of non-enhanced images. The ancillary images can be used for diagnositc evaluation of the A and B images and can be useful in scienti c studies. Table 1 lists the image product types and their units. Ancillary Images In applying the SIRF algorithm we assume that the surface doesn't change over the imaging interval. When it does, SIRF nonlinearly weights the measurements when combining them to estimate an e ective A and B over the imaging interval. When combined with variations in the temporal sampling, the question of what e ective time should be Type Character Code Code Number Units Description A image a 1 dB o 40 incidence angle B image b 2 dB/deg slope of o vs. incidence angle counts C 8 (none) number of measurements hitting each pixel err image E 21 dB mean o reconstruction err inc std I 7 deg incidence angle standar deviation mean inc J 9 deg mean incidence angle time image p 11 min time estimate from start of interval STD image V 22 dB o reconstruction err standard deviation longitude image x 30 deg longitude of lower-left (SW) corners of pixels latitude image y 31 deg latitude of lower-left (SW) corners of pixels Table 1: Image type codes and units used for standard regions for SIRF products assigned to the image arises. We have developed a simple time inference algorithm. The algorithm computes the time corresponding to a linear (in dB) average of the measurements assuming uniform motion. This time index is can vary over the image and is thus computed for each pixel. The `time' image gives this time estimate as the time in minutes from the start of the imaging interval (given by iyear, isday and ismin in the image header { see Appendix 2). The di erence (in dB) between the o measurement and its back projection from A and B images is always non-zero due to noise in the measurements. Azimuth variation in o, temporal variation in the surface characteristics, and limitations in the resolution enhancement can increase the di erence. For each pixel, the mean di erence between the o measurement values for the o measurements which touch the pixel and the measurement's backprojections is termed the `reconstruction error'. The standard deviation (` o reconstruction standard deviation') of the di erence is a particularly use parameter for discriminating ice and ocean. Large pixel values in the standard deviation image indicate areas of temporal and/or azimuth variation. As a diagnostic tool for the SIRF images, \images" of the number of o measurements which hit each pixel (the `counts' or `number of measurements' image), the mean incidence angle of the measurements (`mean incidence image'), and the standard deviation of the incidence angles (`incidence standard devation image') are produced. The output of the rst iteration of the SIRF, known as AVE [1], is a simple resolution enhancement algorithm. AVE A and B are provided as a comparison product. Low resolution comparison images are provided in the standard NSCAT SIRF product suite. This images are created by `binning' or `gridding' the o measurements into a low resolution ( 25 km) grid according to the center point of the cell. A and B are computed using linear regression (in dB). In addition to this product and to aid in comparison, a `non-enhanced' image product is producted by expanding the gridded image to the same size as the SIRF images. For user convenience, \images" of the longitude and latitude of each pixel is computed. Thus, the user can avoid directly computing the lat/lon of each pixel if desired, since it precomputed and stored in these images. DATA EXTRACTION, SORTING AND STORAGE Backscatter data collected by NSCAT is distributed to MERS by Jet Propulsion Laboratories (JPL) NSCAT project on 8mm data tapes, having been formatted and processed to several di erent levels. Level 1.5 (L1.5) data is used to create SIRF land and polar images. The L1.5 data tapes contain a week's worth of data divided into multiple les. Each le contains the data collected during one complete orbit revolution and are typically around 20Mbytes in size. Each le contains all the viable o values collected during the revolution of the satellite. Since only land o measurements are generally used for creating images, the data is sorted before processing. All measurements in the polar regions are retained to enable imaging of both land and sea-ice. Ocean data over nonpolar regions is discarded in the sorting. After reading from tape, the L1.5 measurements are separated into multiple categories: the polar regions, arctic and antarctic, and land-only. Once the data has been sorted into one of the four categories, it is copied onto 660 Mbyte CD's according to Julian day for later processing. Region Name Region Number Region Abbreviation LL-Lat LL-Long UR-Lat UR-Lon Antarctic 100 Ant -90.0 -180.0 -52.0 180.0 Arctic 110 Arc 60.0 -180.0 90.0 180.0 Greenland 202 Grn 59.0 -74.0 84.5 -11.0 Alaska 203 Ala 50.0 -180.0 3.0 -130.0 Cntrl-Amer 204 CAm 5.0 -115.0 30.0 -57.0 North-Amer 205 NAm 25.0 -135.0 65.0 -50.0 South-Amer 206 SAm -58.0 -83.0 15.0 -32.0 North-Afri 207 NAf 2.0 -20.0 40.0 65.0 South-Afri 208 SAf -38.0 5.0 10.0 53.0 Siberia 209 Sib 50.0 60.0 75.0 180.0 Europe 210 Eur 35.0 -12.0 72.0 65.0 South-Asia 211 SAs 5.0 60.0 30.0 130.0 Chin-Japan 212 CnJ 25.0 60.0 55.0 150.0 Indonesia 213 Ind -15.0 93.0 10.0 165.0 Australia 214 Aus -48.0 110.0 -10.0 180.0 Table 2: Names, ID numbers used, and spatial coordinates (lower-left [LL] and upper-right [UR] corners in longitude and latitude) of the 15 standard regions for SIRF land/polar image processing REGIONS Because it is impractical to create global images at full resolution, the earth has been divided into overlapping regions for which full resolution images are practical. To create enhanced land and polar images, rst, a setup le for each region of the earth is created. An additional setup le is created from which non-enhanced images are generated. Each region of the 15 standard regions is de ned by its name, speci ed region number, latitudinal and longitudinal coordinates, and are rectangular with the exception of the two polar regions (Arctic and Antarctic) which are circular. Table 2 lists the names, region numbers, and spatial coordinates of the 15 di erent regions. Images from these 15 regions constitute the standard product suite. For reference, Figures 2 through 16 give examples of A images for all the regions that comprise the standard set. Separate images are made for V-pol and H-pol. Images are created for di erent time periods based on region. Imaging Periods The SIRF resolution enhancement algorithm is based the spatial overlap of the o measurements made by multiple passes of the scatterometer. Since the surface may vary during the \imaging interval", there is a tradeo between resolution enhancement and the imaging interval. Because of the coverage versus latitude characteristics of NSCAT, the optimum imaging interval can vary depending on the imaging location. For example, the polar regions are very frequently covered and since sea-ice can move or evolve rapidly a short imaging interval is used. In the equitorial regions the imaging interval must be longer since 2-3 days can bass between revisits. NSCAT makes also three times as many V-pol measurements as H-pol. Thus, a longer imaging interval is typically required for H-pol measurements as for V-pol. For the standard NSCAT SIRF product suite, polar region images are produced using six day imaging periods for both polarizations while other regions use seven days for V-pol and 14 days for H-pol. Polar Images are time overlapped at 1/2 the imaging period, i.e., 6 day images are producd every 3 days in the polar regions. Projection types The images are a two-dimensional projection of the surface of the globe. While in general any projection can be used, images are distributed only in ve standard projections. All of these are supported in the standard le format described later. The projection options include: Rectangular array (image data only, no projection information) A rectangular lat/lon array Two di erent types of Lambert equal-area projections which can be used in both non-polar and polar projections Polar stereographic projections (northern or southern hemisphere) EASE grid polar projections with various resolutions EASE global projection with various resolutions While the enhanced images generally have a nomial pixel resolution of approximately 4.5 km, the actual resolution is dependent on the choice of projection. As part of the standard NSCAT SIRF product suite, SIRF images are producted using the Lambert xed radius projection for the non-polar regions and the polar stereographic projections (reference latitude 70 deg) for polar (Arctic and Antarctic) images. Utilities are available to transform standard product images to other supported projections. Global images made at lower resolution from the individual region images use the rectagular Lat/Lon array. Figure 1 is an example of a low resolution global image. Images are stored in the BYU SIR image le format described below. The le format includes a header containing all of the projection information. IMAGE FILE NAMING SCHEME A standardized UNIX naming convention has been developed which allows key information about a particular image to be determined by looking at the name. Note that the le header includes the same information as well as additional information in the le. Each image le produced is named using the same format: SENS-T-REGYR-DY1-DY2.RCNEXT The string `SENS' is a four character abbreviation that denotes the remote sensor from which the image is derived. In the case of NSCAT data, there are two possibilities for this catagory `nscv' and `nsch', which refers to vertically polarized data and horizontally polarized data. The character `T' refers to the type of image the le contains. The types generated for NSCAT data include `a', `b', `C', `E', `I', `J', `p', and `V': (see Table 1) a: A image. o in dB at a 40 incidence angle. b: B image. T the slope of o in dB/degree. C: count of the number of measurements hitting each pixel. E: mean o reconstruction error in dB. I: incidence angle standard deviation for the measurements hitting each pixel in deg. J: mean incidence angle for the measurements hitting each pixel in deg. p: image pixel time estimate in minutes from the start of the imaging interval. V: o reconstruction error standard deviation in dB. x: longitude for each pixel. y: latitude for each pixel. Examples of each of these images and their non-enhanced counterparts for a given region are shown in Figures 17 to 23. Note that the longitude and latitude images do not vary with time so the day range isn't really needed. The string `REG' in the le name format gives the three letter region abbreviation (refer to Table 2). The strings `YR', `DY1' and `DY2' give the year as two digits, the three digit Julian day of the year at the start of the imaging interval, and the three digit Julian day of the year at end of the imaging interval. The string `RCN' denotes the image construction technique. The four choices for RCN are `sir', `ave', `grd' and `non': sir: SIRF-produced enhanced resolution image. These are the highest resolution products (see [2]). ave: SIRF-produced AVE image. An AVE image corresponds to the rst iteration of SIRF and has lower e ective resolution than SIRF but higher than gridding or non-enhanced images (see [1]). grd: non-enhanced image made by gridding or binning the measurements onto a low resolution ( 25 km) grid. non: image at same pixel size as sir or ave image but with resolution of grd image. Made from grd image by pixel replication. Note that the `sir' lat/lon images are identical for `ave' and `non' so separate versions are not required. The nal string `EXT' is optional extension and denotes additional post SIRF processing. Standard extensions include `.Lmsk' `.Imsk' `.dif' and `.ed': .Lmsk: land masked image. Ocean areas blacked out. For non-polar images this is the default unless overridden by ice masking and thus may not be explicitly stated. .Imsk: ice masked image. Non-land and ice-free ocean areas blacked out. .dif: a di erence image between to images