A One . Step Algorithm for Gorrection and Calibration of AVHRR Level b Data

NoAA-AvHRn level 1b data have been widely used for environmental rcsearch at rcgional and global scales' There are, however, problems in preprocessing level 1b tape data in small sysiems, including a general lack of avnna-spectfc softwaie possessing suitable levels of sophistication, efficiency, and geographic covercge. This paper desctibes a onestep preprocessing algorithm which combines simple tape reading with geomebic conection, tadiomettic conection, and cilibration using the auxiliary patametets appended in the level lb tape as primary input' Out algorithm is fast, memory-efficient, and rc compatible. lntroduction In the past several years, one of the most importalt developments in remote sensing has been the widespread use of meteorological satellite data for non-meteorological applicaiions. The National Oceanic and Atmosptreric Adminisiiation (NoAA) series of polar-orbiting meteorological satellites have received considerable attention because of the onboard sensor; the Advanced Very High Resolution Radiometer (AVHRR). AVHRR data have been used extensively to study regional, continental, and global phenomena. The most popular application of these dita has been to monitor and evaluate vegetation over the land surface. avnnn data have three different formats: GAC, HRPT, and LAc. cAc, or Global Area Coverage, is intended to allow worldwide coverage with a manageable amount of data' The soatial resolution for GAC is 16 km, Local Area Coverage (l,ac) ana High Resolution Picture Transmission (rnrr) ac-iually are thJ same kind of data but differ in the manner of tranimission from satellite to ground station. Hnpt data are sent to Earth continuously in real time, while LAC data are selectively recorded on-board for subsequent playback. -Both HRPT andLAC have a spatial resolution of r.t km at nadir and a spectral resolution of ro bits (NOAA, 1990)' Because of the higher spatial and spectral resolutions than that of cac, bottr HRPT and LaC data are widely used in regional or continental environmental research. However, rat dat" are unique in that researchers can study geographic areas where no ground station is available. tac data are usually stored on computer compatible tapes (ccrs) as level 1b daia for distribution (NOAA, 1990). The level 1b data contain raw AVHRR spectral data as well as calibration coefficients, solar-zenifh angles, Earth location, and other auxiliary Liping Di is with The Cooperative Institute for-Research in Envir6nmental Sciences (CIRXS), University of Colorado at Boulder, Campus Box 449, Boulder CO 80309-0449 D.C. Rundquiit is with The Center for Advanced Land Management Inlormation Technologies (CALMIT), University of Nebraska-Lincoln, Lincoln, NE 68588-0517 PE&RS data. However, the spectral data contain both geometric and radiometric errors which must be removed in order to quantitatively analyze AvHRR data and overlay other data sets' Therefoie, preprocessing is a prerequisite to using AvHItR level 1b data in research' The Problem There are many commonly used software packages to allow users to easily read avnnn level 1b data from a computer compatible tape (ccr) into a disk file for further analyses; for example, mavunn in ERDAS (ERDAS, 1990)' RFU'C in ELAS INASA, 1990), and LAcIN in LAS (USGS' 1990). However' ihese packages do not have any geometric and radiometric co.reciiott finctions designed specifically for AVHRR data, so all auxiliary information, such as solar zenith angles, geometric locaiions, and calibration coefficients, are lost (i'e" these packages do not store the auxiliary data). A?ter the tape data have been read into a disk file by. means of the software noted above, conventional geometriccorrection algorithms, such as rubber sheeting andpollmomial fitting, miy be applied to raw AVHRR data to achieve 8eoreferenclng. dne pr66lem is that thosealgorithms typically. involve thE selection of ground control points. Because of the coarse spatial resolution of I-Ac, 1100 by 1100 metres at the nadir, t6e process of locating ground-control points (ccpsJ in an AvHRR image is difficult [peters, 19Bg). 4 second problem is that there aie no commonly available software packageswhich nrovide suitable radiometric corrections, such as solar zenith angle, atmospheric attenuation, and data calibration for conveiting the riw digital numbers (nNs) to ground reflectance, albedo, surface temperature, and so on' Despite the general lack of software for AVHRR-preprocessing, ccrs containing AVHRR laC datado provide the necessary parameters for both geometric and radiometric adjustments (NOAA, 1990). Instead of performing c-orrections after the tape has been read as in conventional methods, it is nossible to use those parameters to do geometric and radiometric corrections while reading an AVHRR tape. The development of a one-step algorithm combi-ning eometric correction, radiometric iorrection, and calibration with tape reading seems to be a worthwhile objective. This article introduces such an algorithm. Geometric Correction ofAVHRR Data Remotely sensed data usually contain both systematic and nonsystematic geometric errors (Jensen, 19BO). The purposes Photogrammetric Engineering & Remote Sensing, Vol. 60, No. 2, February 1994, pp. 165-171. oosg -7 772 II 4/600 2-1 6 5$ 03.00/0 @1994 American Society for Photogrammetry and Remote Sensing of geometric correction are to remove these errors and to relate the digital remote-sensing data to a map projection. The traditional geometric correction of remotely sensed data almost always i-nvolves relating the pixel coordinates (row and column) of ground control points (ccps) with their corresponding map coordinates (e.g., the latitude/longitude positionJ. A ccp is a point on the surface of the Earth that can be identified on both an image (in rows and columns) and a map (in degrees of latitude and longitude, feet, or metres). The geometric relationship between the input pixel Iocation [row and column) and the associated map cooidinate (x,y) can be determined by a group of ccps. The typical proje-ction equations relating the map coordinates and image coordinates are polynomials: that is, x ' : eo* o.x* ary * a"x2 t aqry + asy2 + . . , /4r y' : bo + bry + b"y * b"xz * bqU * b"y2 * ... trJ where x and y are the positions in the rectified image or map and.rC and y' represent the corresponding positions in the original input image. The coefficients in Equation 1 are determined by regression analysis of ccps. The order of the polynomial in Equation 1 is decided by both the magnitude bf the distortion in the raw image and the number of available ccPs. A typical rectification of a satellite image involves third-order polynomials and 20 to 30 ccps. With Equation 1, a pixel position in the rectified image can be projected into the distorted image coordinate and a resampling algorithm is used to retrieve the spectral value from the distorted image for the Dosition. Geometric correction usuallv is done row-bv-row or block-by-block of the output (reitified) image, Because of the geometric distortion, a row in the output image may cross several rows in the distorted image. Therefore, any part of the distorted image must be directly accessible (e.g., the distorted image must be resident in a disk file or computer system memory). However, computer tapes are accessed sequentially. Therefore, it is impossible to use traditional methods to do geometric correction while reading an At{HItR tape. AVHRR level 1b data contain Earth location information, with a fixed number of ground reference points appended to each scan Iine (NOAA, 1990). There arc ZO4B pixels in an I.a,c/HRpr scan line. The Earth location data (laiitude and longitude) are sampled every 40 points starting at pixel zS (25, 65, 105, .,., 1945, 1985, and 2025), so there are 51 possible ground reference values for each scan line. Latitude and longitude values are each stored in two-byte fields in 128th of a degree (East positive), which is less than the size of one pixel (about 0.0133 degree under nadir). Based on those Earth locations, we developed a method to geometrically correct AVHRR data while reading the tape. Spatial Interpolation Suppose that we want to obtain a rectangular area (window) bounded by minimum longitude X-r, maximum longitude X-*, minimum latitude Y-,., and maximum latitude Y-* . The image size for the window, in terms of rows and columns, can be expressed as nrow : (X-* X-J/4. + 0.5 ,6t ncolumn : (y-o y^h)lpy + O.S l.t where P, is the pixel size in the X direction and P" is the pixel size in the Y direction. Suppose a scan line is read from the tape, allowing us to 166 obtain the 51 Earth locations sampled at every 40 pixels, For the ith pixel of the scan line, whiih is located between Earth location i and i+t, a linear interpolation equation can be used to determine the geographiC location of the pixel: that is , Xt : (X,*, X)0 column(i))l4o + X, Yt -(Y,*, n(i column(i))leO + 4 lrl column(i) : (I ' 7) * 40 + 25 i : 7,2,...,5O i : column(i),column(i) + 1,,..,( i + 1) where Xl the interpolated Earth location (longitude) for the ith pixel of the current scan line, )j = the interpolated Earth location (latitude) for the lth pixel of the current scan line, X = the ith appended Earth location (longitude), and Yr = the ith appended Earth location (latitude). Equation3 is applicable from pixel 25 ( column(1) ) to pixel 2025 ( column(sl) I of the current scan line. For pixels 1 to 24, another interpolation equation is used: Xt = (X" X,JU 25)l4O + Xl Yt (Y, Yr)U 25)l4O + Y, j : 7,2, ..,, 24 (4) Pixels 2026 to 2048 are interpolated by equation Xi : {Xu X,")U 19Bb)/40 + Xso Y : V", )i")U 19Bs)/40 + %o j : 2 0 2 a , 2 0 2 7 , ' . , 2 0 4 8 ( 5 ) Both Equations 4 and 5 are extensions of Equation 3. The geographic location of the ith pixel of the current scan line is then converted to the row and column numbers of the rectified output image based on window coordinate and pixel size: that is, L, : (Y^* Y1)lP" + o.s rar Ct : 6i X-i")/{ + 0.5 IU, where ! r-s the row number for pixel I in the rectified output image and C, is the column number. Note that a row increment in the output image is the inverse of the y (latitude) increment. The first row in the output image has the maximum latitude possible in the image. The row and column numbers are then used to compare with the specified image area (window) to decide whetlier or not pixel 7' is in the se-lected output area: [ l e * i . r d o * 1 < l r < n r o w 1 l < C r < n c o l u m n ( 7 ) [i G window otherwise Once pixel i of the current scan line is in the output window area, the (Li, C;) location of the rectified output image will be assigned the intensity value of the input pixel i. In order to reduce the calculation time, it is wise to check the 51 points of Earth locations first. If none are in the output window, discard the cunent scan line and read next scan line. The procedure for spatial interpolation of avtnn data can been described as follows: (1) Read in one scan line from an AvHRR tape. If there are no scan lines left in the tape, exit the loop. (2) Check whether or not the scan line falls within the selected window. We might only check the 51 Earth locations of the scan line. If none falls within the window area, go to step (1) .