5: REMOVING SUN GLINT FROM COMPACT AIRBORNE SPECTROGRAPHIC IMAGER (CASI) IMAGERY. Aim of Lesson. Objectives. Background Information

Transcription

1 Lesson 5: Removng sun glnt from CASI magery 5: REMOVING SUN GLINT FROM COMPACT AIRBORNE SPECTROGRAPHIC IMAGER (CASI) IMAGERY Am of Lesson To learn how to remove sun glnt from hgh resoluton arborne and satellte magery to reveal bottom features and mprove classfcaton. Objectves 1. To understand the ratonale underlyng the method of sun glnt removal developed by Hochberg et al. (2003) as modfed by Hedley et al. (2005). 2. To choose a reference near nfra-red (NIR) band and fnd out the ambent NIR sgnal n the absence of sun glnt. 3. To select areas of deep water wth varyng sun glnt ntenstes and fnd out the lnear relatonshp (regresson) between each vsble band and the chosen NIR band pxels. 4. To use the nformaton from 2 and 3 above to remove the component of vsble band values that s due to sun glnt and thus deglnt each pxel of the vsble bands. Background Informaton Ths lesson relates to recent research publshed by Hochberg et al. (2003) and developed by Hedley et al. (2005). You are recommended to consult these papers for further detals of the technques nvolved. Ths lesson uses the modfed approach of Hedley et al. (2005) to show how to remove sun glnt from Compact Arborne Spectrographc Imager (CASI) magery flown n the vcnty of the sland of St John n the US Vrgn Islands. The Blko 3 mage processng software Famlarty wth Blko 3 s requred to carry out ths lesson. In partcular, you wll need experence of stackng mages and creatng colour compostes and of usng Formula documents to carry out mathematcal manpulatons of mages. Tutorals 2 and 10 n the Introducton to usng the Blko 3 mage processng software are partcularly relevant and provde a good groundng for ths lesson. A famlarty wth the Mcrosoft Excel spreadsheet package or the Mntab statstcal package s also hghly desrable. You wll need access to one or other of these two packages to complete ths lesson. Image data A Compact Arborne Spectrographc Imager (CASI) nstrument was mounted on a lght arcraft and flown at 4100 ft (1250 m) alttude over shallow coral reef areas off the coast of St John n the USVI at a speed of 130 nautcal mles per hour (240 km.h -1 ) n an East West drecton. An ncdent lght sensor (ILS) was fxed to the fuselage so that smultaneous measurements of rradance could be made and radance measurements thus converted to reflectances. A Dfferental Global Postonng System (DGPS) was mounted to provde a record of the arcraft s flght path. Data were collected at a spatal resoluton of 2 x 2 m (4 m 2 ) n 19 wavebands (Table 5.1). The mages were geometrcally corrected usng a UTM projecton and WGS 84 spherod/datum. The wdth of the groundtrack recorded by the CASI was approxmately 1 km. The fnal geometrcally and radometrcally corrected mages were stored n a 16-bt unsgned nteger format. For the purposes of ths lesson the coordnate nformaton has been removed from the mages. However, f you wsh to replace ths, the coordnates of the upper left pxel are Eastng m 1

2 Applcatons of satellte and arborne mage data to coastal management Northng m (UTM zone 20 N). The pxel DX: = 2.0 m and DY: = 2.0 m. (If you correctly edt the coordnates you should fnd that the lower rght pxel has an Eastng of m and Northng of m.) Each mage conssts of 581 rows and 1545 columns of pxels. Table 5.1. Band settngs used on the CASI. Bands 8-16 have not been ncluded wth the lesson data. Band Part of electromagnetc spectrum Md-wavelength (nm) Wdth of waveband (nm) Regons of peak senstvty n eye 01 Blue Blue cones 02 Blue Blue cones 03 Blue Green Green Green Green Green cones 08 Green Green cones 09 Green Green Green Yellow Red cones 13 Orange Red cones 14 Orange Red Red Near nfra-red Near nfra-red Near nfra-red Geometrcally and radometrcally corrected Compact Arborne Spectrographc Imager (CASI) data for 10 bands are provded for the purposes of ths lesson as a set (St_John_USVI_CASI.set) of 10 Blko.dat fles (St_John_USVI_CASI#01.dat St_John_USVI_CASI#19.dat). These fles are unsgned 16-bt nteger fles (each pxel recordng a reflectance value converted to an nteger between 0 and 65535). Ths means that there are two bytes needed per pxel. The problem of sun glnt The mappng of benthc features can be serously mpeded by sun glntng on the water surface. When skes are clear and the water surface s rppled, specular reflecton of the ncdent radaton occludes the benthc sgnal wth brght sun glnt. Unfortunately the problem of sun glnt s partcularly acute under condtons when remote sensng mght otherwse be most effectve: clear skes, shallow waters (whch, when wnd-blown, form waves), and when mages are collected at a hgh spatal resoluton. Typcally, sun glnt forms bands of whte along wave edges on the wndward sde of near shore envronments (Fgure 5.1). These whte bands confound the vsual dentfcaton of bottom features, and wll confuse both supervsed and unsupervsed classfcatons (wth the presence or absence of sun glnt domnatng the categorsaton n affected areas). 2

3 Lesson 5: Removng sun glnt from CASI magery Fgure 5.1. A colour composte (f prnted, ths may be n shades of grey), showng the CASI mage before sun glnt removal. Specular reflecton along wave edges s obscurng much of the mage and sun glnt s generally bad along the southern sde of the ground-track. Three areas of deepsh water showng varyng sun glnt are outlned by boxes. Although n areas wth sun glnt the recorded sgnal appears almost entrely composed of the water surface specular reflectance sgnal, the component caused by water leavng radance may be recoverable provdng that the sensor remans spectrally unsaturated. A recent paper by Hedley et al. (2005) presents a conceptually smple method whereby sun glnt can be removed from vsble wavelength spectral bands of remotely sensed mages usng a spectral band n the near nfra-red (NIR) part of the electromagnetc spectrum (around nm). The method s partcularly applcable to hgh-resoluton magery such as that obtaned usng arborne sensors such as CASI or satellte sensors such as IKONOS. Image pxels are adjusted to remove the sun glnt component of the recorded sgnal, thereby leavng only the component derved from benthc reflectance and processes wthn the water column. Ths leads to a substantal vsual mprovement n the deglnted mages and can also lead to an ncrease n user accuraces on maxmum lkelhood classfcatons of such mages. Outlne of the deglntng method The deglntng method reles on two smple assumptons: 1) That the sgnal n the NIR s composed only of sun glnt and a spatally constant ambent sgnal. In partcular, there s no spatally varant benthc contrbuton to the NIR. 2) That the amount of sun glnt n the vsble bands s lnearly related to the sgnal n the NIR band. The frst assumpton s justfed by the fact that water s relatvely opaque to NIR wavelengths ( nm) (Mobley, 1994), so that even shallow waters (e.g. those only m deep) have a low water-leavng radance n the NIR regardless of bottom type. Although a mnmum NIR sgnal over deep water mght be expected to be zero, n practce the mnmum NIR (Mn NIR ) sgnal s usually greater than zero. In partcular, f mages are not atmosphercally corrected ths resdual or ambent NIR sgnal corresponds to NIR backscatter n the atmosphere. The method assumes a constant ambent NIR (Mn NIR ) sgnal level, whch s removed from all pxels durng the analyss. The assumpton of a lnear relatonshp between the NIR sgnal and the amount sun glnt n the vsble bands holds because the real ndex of refracton (whch governs reflecton) s nearly equal for NIR and vsble wavelengths (Mobley, 1994). Therefore the amount of lght reflected from the water surface n the NIR s good ndcator of the amount of lght that wll be reflected n vsble wavelengths, and a lnear relatonshp exsts between the two. The deglntng method proceeds by establshng the lnear relatonshp between NIR sgnal and the amount of sun glnt n each vsble band. Ths nformaton, combned wth the NIR sgnal n each mage pxel, s used to work out how much to reduce the sgnal n each band to remove the sun glnt n each pxel. 3

4 Applcatons of satellte and arborne mage data to coastal management Frstly a lnear relatonshp s establshed between a NIR band and each vsble band usng lnear regresson. To do ths one or more regons of the mage are selected whch provde a range of sun glnt, but where the underlyng sgnal would be expected to be consstent (areas of deep water are deal for ths, Fgure 5.1). For each vsble band all the selected pxels are ncluded n a lnear regresson of the vsble band sgnal (y-axs) aganst the NIR sgnal (x-axs) (Fgure 5.2) pxel to be deglnted 5000 R NIR - Mn NIR R Vsble band pxel value R' deglnted pxel slope b Band = NIR R - R' regresson lne sample of pxels wth sunglnt 0 Mn NIR Near IR band pxel value R NIR Fgure 5.2. Graphcal llustraton of how the deglnt method works. A regresson s carred out between a sample of vsble band (band ) pxels from areas of varyng sun glnt ntensty and correspondng pxels n the chosen NIR band. The assumpton s made that all the NIR pxels would have the ambent NIR (Mn NIR ) value n the absence of sunglnt. Knowng the slope (b ) of the regresson and the value of Mn NIR, you can then work out the proportonate reducton n the vsble band sgnal (R ) requred to remove the component of the sgnal that s due to sun glnt; the magntude of the glnt beng obtanable from the NIR band (R NIR Mn NIR ). Ths allows you to calculate a deglnted value (R' ) for each vsble band pxel. If the slope of the regresson lne for band s b, then all the pxels n the mage can be deglnted n band by the applcaton of followng equaton: R ' = R - b (R NIR - Mn NIR ) [Equaton 5.1] whch smply means: reduce pxel sgnal n band (R ) by the product of regresson slope (b ) and the dfference between the pxel s NIR sgnal (R NIR ) and the ambent NIR level (Mn NIR ). Mn NIR essentally represents the NIR sgnal n a pxel wth no sun glnt and can be estmated by the mnmum NIR found n the regresson sample or alternatvely as the mnmum NIR found across the whole mage. In general, the mnmum NIR pxel s less prone to problematc outlers than the maxmum NIR pxel. 4

5 Lesson 5: Removng sun glnt from CASI magery If you fnd problems wth understandng the dervaton of equaton 5.1, we can look at the dagrammatc nterpretaton of the method n Fgure 5.2 and derve the equaton step by step. By defnton, the slope (b ) of the regresson lne s just the amount of change n the vsble band value dvded by the amount of change n the NIR band value, between any two ponts on the regresson lne. Thus f, between two ponts on the regresson lne, the vsble band changed from 1000 to 4000 (change = = 3000) whlst the NIR band changed from 1500 to 3000 (change = = 1500) then the slope of the lne would be ( )/( ) = 3000/1500 = 2. For each pxel n the mage we know ts NIR value (R NIR ) and ts value n the vsble band (R ). We also know the ambent NIR value (Mn NIR ) and assume that all reflectance n the NIR above ths value s due to specular reflecton at the sea surface. After carryng out regresson analyss between selected pxels n our vsble band and the correspondng pxels n the NIR band, we know the slope of the regresson lne b. Ths allows us to work out how much of the sgnal of each pxel n band s due to specular reflecton. From fgure 5.2 we can see that: ( R R ) = b ( R Mn ) NIR NIR whch translates as: Amount of sgnal due to specular reflecton n vsble band Amount of sgnal due to specular reflecton n NIR band = Slope The only value we do not know n the equaton s the value (R' ) of the deglnted pxel. We thus need to rearrange the equaton so that t can be solved to gve us R'. Ths can be done n 3 steps. Frstly, we must multply each sde by (R NIR Mn NIR ). Ths gves: R R = b R Mn ) ( ) ( NIR NIR Secondly, to solate R' on the left-hand sde of the equaton, we next need to subtract R from each. Ths gves us: R = b ( RNIR MnNIR ) R Fnally, to make R' postve, we need to multply each sde of the equaton by 1. Ths gves us equaton 5.1: R = R b ( R Mn ) NIR NIR Snce the method reles on a user-based selecton of a sample set of pxels t s not necessary to mask out non-submerged or cloud pxels pror to deglntng. It s prudent to ensure that the sample pxels do not contan any non-submerged objects, but the regresson wll nevertheless mtgate the mpact of solated nvald pxels. However, non-submerged areas wll not contan vald data after deglntng snce the algorthm s vald only for submerged pxels. Note also, that as the method operates purely on the relatve magntudes of values, the absolute unts of the pxel values are unmportant. Therefore, there s no need to transform pxel values nto radance and deglntng can be appled to the orgnal mage dgtal numbers. It s however advsable to ensure floatng-pont arthmetc s used n order to correctly handle fractonal values and negatve numbers. 5

6 Applcatons of satellte and arborne mage data to coastal management Lesson Outlne Before proceedng wth the deglntng you should examne the CASI mages to judge the magntude of the sun glnt problem. Open the set St_John_USVI_CASI.set (select Fle, Open and then select SETS (*.set) from the Fles of type: drop-down menu). Make sure the Extract checkbox s unchecked and n the Redsplay Image dalog box check the Null Value(s): == 0 checkbox and select an and Auto lnear stretch as the stretch to use, before clckng on the button to apply the stretch to all of the mages n the set. You should have a stack of 10 mages dsplayed. Holdng down the <Ctrl> key, double-clck on the top mage to zoom out so that you can see the whole mage. Then use the <Tab> key to look at each mage n turn, notng how detals of the seabed become progressvely less dstnct n bands #5 #7 and abruptly dsappear when you move from band #7 to band #17 n the NIR. Note also that sun glnt s concentrated on the southern half of the ground-track. For bands #17, #18 and #19 n the NIR part of the electro-magnetc spectrum, there should n theory be no return over deep (> c. 1 m depth) water and brght pxels should be due entrely to sun glnt. In bands #17 and #18 note the seres of parallel wave-fronts orentated n a south-west to north-east drecton. To see the extent of the sun glnt more clearly and provde a reference mage aganst whch you can judge the success of the deglntng, you need to create a false colour composte mage of bands #1, #3 and #5. [Ths combnaton was found to gve a reasonable mage and the fnal comparson wll be wth a composte made from deglnted bands #1, #3 and #5.] Clck on Image, Connect and select St_John_USVI_CASI#01.dat, St_John_USVI_CASI#03.dat and St_John_USVI_CASI#05.dat as the three mages to be connected. Do not check the Stacked checkbox and use the Selector toolbar to desgnate band #05 as mage 1 (@1: red gun), band #03 as mage 2 (@2: green gun) and band #01 as mage 3 (@3: blue gun). [Ths s acheved by clckng each button n turn from rght to left.] When you have done ths, select Image, Composte to generate the colour composte. Zoom out so that you can see the whole mage. Note the serous sun glnt, whch s obscurng the coral reefs along the southern part of the CASI groundtrack. Save ths mage for reference as a Wndows.bmp mage. [Note: a Wndows btmap (.bmp) mage can be readly nserted nto a Word document usng Insert, Pcture, From Fle menu n Word.] Choosng a NIR band Three CASI bands were recorded n the near nfra-red (NIR): bands #17, #18 and #19. Of these, band #17 s on the border between the far red and NIR part of the spectrum but we would not expect any bottom reflectance for water more than c. 1.0 m deep for any of these wavelengths (see Table 5.1). You need to decde whch of the three NIR bands wll be best to use for deglntng. Normally you would need to check whch NIR band gves best results n the deglntng process. There appear to be at least two crtera for decdng ths: () how good the deglnted vsble bands look, and () the goodness of ft of the regresson lnes between the sample(s) of pxels n the vsble bands and the NIR bands. To fnd () out, you would need to carry out the whole deglntng process usng all three NIR bands; to fnd () out you can regress a couple of sample bands on each NIR band. Ths nvolves carryng out later parts of the deglntng process; thus to save you tme ths has been done for you for bands #2 and #7. The results are shown n Table

7 Lesson 5: Removng sun glnt from CASI magery Table 5.2. Comparsons of lnear regressons of vsble bands #2 and #7 on each of NIR bands #17, #18 and #19. Slopes of regressons lnes and coeffcents of determnaton (R 2 ), whch ndcate the proporton of the varance accounted for by the regressons, are shown. Test bands for Band 17 (698.4 nm) Band 18 (756.6 nm) Band 19 (809.3 nm) regresson Slope R 2 Slope R 2 Slope R 2 Band 2 (blue) Band 7 (green) Table 5.2 shows that coeffcents of determnaton are hghest for band #17 and worst for band #19. If you study the band #19 mage you wll notce that t appears less focused (less sharp) than bands #17 and #18. Use the <Tab> key to move between bands #17, #18 and #19 n the stack of CASI bands and <Shft>+<Tab> key to move back agan. Note that the sun glnt on band #19 appears less clearly delneated than that on bands #17 or #18. Gven both the relatvely low coeffcents of determnaton and slght blurrness of band #19, you should not consder t further but concentrate on band #17 as your frst choce NIR band. Gven the relatvely short IR wavelength of band #17, you should, however, carry out deglntng usng band #18 for comparson. Determnng the mnmum NIR value over deep water To determne the mnmum NIR (Mn NIR ) value of an mage you need to select an area of water that s (a) relatvely dark and (b) reasonably deep ( 2 m depth) so that there s no chance of bottom reflectance from coral or sand close to the surface. You want a representatve sample of pxels wth whch to calculate the Mn NIR but can ft a maxmum of 256 columns n an Excel spreadsheet. Thus the suggested sample sze s 256 x 100 pxels from an area wth top left coordnates at X: 1200 and Y: 5 whch covers a substantal area of reasonably deep water on the north sde of the ground-track n the darkest part of the mage. You wll now determne the Mn NIR for bands #17 and #18 usng Excel. Queston: Inspect bands #17 and #18 and note that the north-east of the mages appears darkest. Start Excel and open the Excel workbook called Lesson05_deglntng_St_John.xls, makng sure to clck the Enable Macros button (f ths appears). Select the MnNIR worksheet by clckng on ts tab. Rows 2 to 101 are empty, ready to receve the sample of band #17 pxels needed to determne Mn NIR for band #17. Returnng to Blko, select band #17 (St_John_USVI_CASI#17.dat) as the actve mage n the stack and then select Edt, Go To from the menu. In the Go To dalog box, make sure Selecton Type: s set to Box Selecton and then set X: to 1200, Y: to 5, DX: to 256 and DY: to 100 and clck OK. Copy the block of pxels usng the Copy button or by pressng <Ctrl>+C. Swtch to Excel, clck on cell A2 of the MnNIR worksheet and paste the cells (usng Paste button, <Shft>+<Insert> or <Ctrl>+V). Scroll down to row 103 and you wll see that a formula n cell B103 has calculated the mnmum NIR value for band #17. [Make a note of ths as t wll be needed later; t should be 208.] Note that rows 108 to 207 are empty and ready to receve the sample of band #18 pxels needed to determne Mn NIR for band #18. Returnng to Blko, press the <Tab> key to make band #18 the actve mage n the stack. The box selecton should already be n place so you just need to copy the sample of band #18 pxels and paste t to cell A What s the Mn NIR for band #18? [Consult the spreadsheet below where you pasted the band #18 sample.] 7

8 Applcatons of satellte and arborne mage data to coastal management Selectng areas of varyng sunglnt to calculate regresson of each vsble band on the chosen NIR band For the purposes of ths lesson you wll select a 50 x 50 pxel area from each of three areas on the mage; one from a low sun-glnt area, one from an area wth hgh sun glnt and one from an area wth ntermedate sun-glnt problems (Fgure 5.1). The top left coordnates of each of these areas are gven n Table 5.3. Table 5.3. Column and row coordnates for three areas of varyng sun glnt. (Settngs for Go To dalog box.). Level of sun glnt Top left X: coordnate Top left Y: coordnate DX: DY: Low Intermedate Hgh Queston: 5.2. What s the area n both hectares (ha) and m 2 of each of the areas of sun glnt? How many pxels are n each area sampled? In the Excel workbook called Lesson05_deglntng_St_John.xls clck on the tab labelled Band samples and you wll see that for all bands except band 5, the 7500 pxels from the three areas of varyng sun glnt have already been entered. Your task s to enter those for band #5. The top 2500 rows below the headngs contan the pxels from the low sun glnt area (Table 5.3 and Fgure 5.1), the next 2500 rows contan those from the ntermedate sun glnt area, and the fnal 2500 rows contan those from the hgh sun glnt area. In each case the rectangular sample of pxels has been coped and pasted nto a Mntab worksheet and then stacked nto one column. If you have access to the Mntab statstcal package you can do the same. If you only have access to the Excel spreadsheet package, a macro s provded to allow you to do the stackng n the worksheet labelled Stackng_sheet. Instructons wll be gven to assst you to prepare the band #5 sun glnt samples n both Excel and Mntab. Choose whch package you wll use and follow the approprate set of nstructons below. Instructons for Excel are gven frst, followed by those for Mntab. Stackng the band 5 sun glnt samples for regresson analyss usng Excel In the Excel workbook Lesson05_deglntng_St_John.xls clck on the tab labelled Stackng_sheet. Ths worksheet should be empty. Swtch to Blko and select the mage stack. Use the Selector toolbar to St_John_USVI_CASI#05.dat as the actve mage. Now use Edt, Go To to select the low sun glnt area, referrng to Table 5.3 for the coordnates. When the box selecton s n place, Copy the selecton. Swtch to Excel and paste the values to cell A1 (.e. make sure cell A1 s hghlghted when you clck on the Paste button). Checkpont: You should see the value 1534 n cell A1 of the Stackng_sheet worksheet. If you press <Ctrl>+<End> you should fnd the cell AX50 hghlghted and ths cell should contan the value If all s well, congratulatons! If not, undo the paste and check that () you have the rght mage n the stack and () the box selecton s correctly postoned. Then try agan. Wth Stackng_sheet as the actve worksheet n Excel select Tools, Macro, Macros. Select the macro called Stack_columns and clck the button. The macro 8

9 Lesson 5: Removng sun glnt from CASI magery takes each column of 50 pxels n turn and stacks them n column A of the worksheet. The next step s to transfer the 2500 stacked cells to the Band#05 column n the Band samples worksheet. To do ths select cell A1 n the Stackng_sheet worksheet and press <Ctrl>+<Shft>+<Down arrow>. Ths wll select the whole column of cells. (You should see that cell A2500 now has the value 1305.) Clck on the Cut button (or Edt, Cut) and then clck on the Band samples tab to make ths the actve worksheet. Now clck on cell E3 at the top of the blank part of the Band#05 column and clck on the Paste button to paste the stacked cells. Before dong anythng else press <Ctrl>+<Down arrow> whch should take you to cell E2502 whch should have the value Note the thck horzontal lne through neghbourng columns, whch marks the end of the low sun glnt pxel sample. If you now clck on the band#5_17 tab you wll see that these pxels now feature on the graph. In order to complete the graph and obtan the regresson equaton and hence slope for band #5, you need to repeat ths procedure for the ntermedate and hgh sunglnt samples, addng the stacked pxel values to the Band#05 column untl all 7500 pxels are present. Make Stackng_sheet the actve worksheet agan. Then swtch to Blko and select the ntermedate sun glnt sample n band #5 usng Table 5.3 for gudance. Copy and paste the cells to the Stackng_sheet (makng sure that the frst value s pasted n cell A1) and then run the macro agan to stack the values nto one column. Select the column wth <Ctrl>+<Shft>+<Down arrow> and cut and paste t to cell E2503 n the Band samples worksheet. If you now clck on the band#5_17 tab you wll see that these ntermedate sun glnt pxels now feature on the graph. Checkpont: The value of E2503 should be 1572 and the value of E5002 (press <Ctrl>+<Down arrow> to move to bottom of column) should be Repeat the exercse for the hgh sun glnt sample area (Table 5.3). Ths tme the stacked column of pxel values wll be pasted to cell E5002 n the Band samples worksheet. If you now clck on the band#5_17 tab you wll see that these hgh sun glnt pxels now feature on the graph. Checkpont: The value of E5003 should be 2034 and the value of E7502 should be If all s well, congratulatons! You can now progress to the secton headed Deglntng the vsble bands. Stackng the band 5 sun glnt samples for regresson analyss usng Mntab Leave the Excel workbook open at the worksheet labelled Band samples. Start Mntab and make sure you have a blank worksheet showng as the actve wndow. Swtch to Blko and select the mage stack. Use the Selector toolbar to St_John_USVI_CASI#05.dat as the actve mage. Now use Edt, Go To to select the low sun glnt area, referrng to Table 5.3 for the coordnates. When the box selecton s n place, Copy the selecton. Swtch to Mntab and paste the values to row 1 of column C1 (.e. make sure ths cell s hghlghted when you clck on the Paste button). 9

10 Applcatons of satellte and arborne mage data to coastal management Checkpont: You should see the value 1534 n the frst cell of the worksheet. If you press <Ctrl>+<End> you should fnd the row 50 cell n column C50 hghlghted and ths cell should contan the value If all s well, congratulatons! If not, undo the paste and check that () you have the rght mage n the stack and () the box selecton s correctly postoned. Then try agan. Select Manp, Stack/Unstack, Stack Columns from the Mntab menu. In the Stack Column dalog box enter C1-C50 n the Stack the followng columns: text box and C1 n the Store the stacked data n: text box and clck on OK. All the data s now stacked n column C1 so you can delete the data n columns C2-C50. To do ths, select Manp, Erase Varables and enter C2-C50 n the Columns, constants, and matrces to erase: text box. You are now left wth just a sngle column of the data you want. The next step s to transfer the 2500 stacked cells to the Band#05 column n the Band samples worksheet n Excel. To do ths clck on the frst cell n column C1 and press <Ctrl>+<Shft>+<End> to select the whole column of 2500 cells. Clck on the Cut button (or Edt, Cut Cells) and swtch to the Band samples worksheet n the Excel workbook. Clck on cell E3 at the top of the blank part of the Band#05 column and clck on the Paste button to paste the stacked cells. Before dong anythng else press <Ctrl>+<Down arrow> whch should take you to cell E2502 whch should have the value Note the thck horzontal lne through neghbourng columns, whch marks the end of the low sun glnt pxel sample. If you now clck on the band#5_17 tab you wll see that these pxels now feature on the graph. In order to complete the graph and obtan the regresson equaton and hence slope for band #5, you need to repeat ths procedure for the ntermedate and hgh sunglnt samples, addng the stacked pxel values to the Band#05 column untl all 7500 pxels are present. Swtch to Blko and select the ntermedate sun glnt sample n band #5 usng Table 5.3 for gudance. Copy and paste the cells to the Mntab worksheet (makng sure that the frst value s pasted n row 1 of column C1). Stack the contents of columns C1-C50 n C1 as before, and then erase columns C2-C50. [Note that your prevous settngs are preserved n the two dalog boxes, whch makes lfe easer the second tme round!] Wth the frst cell of C1 hghlghted, select the stacked column wth <Ctrl>+<Shft>+<End> and cut and paste t to cell E2503 n the Band samples Excel worksheet. If you now clck on the band#5_17 tab you wll see that these ntermedate sun glnt pxels now feature on the graph. Checkpont: The value of E2503 should be 1572 and the value of E5002 (press <Ctrl>+<Down arrow> to move to bottom of column) should be Repeat the exercse for the hgh sun glnt sample area (Table 5.3). Ths tme the stacked column of pxel values wll be pasted to cell E5002 n the Band samples worksheet. If you now clck on the band#5_17 tab you wll see that these hgh sun glnt pxels now feature on the graph. Well done! that was the hardest part of the practcal. You can now close Mntab. Checkpont: The value of E5003 should be 2034 and the value of E7502 should be

11 Lesson 5: Removng sun glnt from CASI magery Deglntng the vsble bands You now have the full sample of 7500 band #5 pxels n column E of the Band samples worksheet. Ths s plotted aganst the correspondng pxels n the NIR band #17 (n column I) n the band#5_17 chart. A lnear regresson (trendlne) has been ftted to the data and the slope and ntercept of the regresson are automatcally dsplayed on the chart. Queston: 5.3. What s the slope and ntercept for the regresson of band #5 pxel values on the correspondng NIR band 17 pxel values? You now have the necessary nformaton to carry out the deglntng of the 7 vsble bands usng band #17 to estmate the amount of specular reflecton n each pxel. Ths s done usng Equaton 5.1: R ' = R - b (R NIR - Mn NIR ) expressed n a Formula document. To save tme most of the formula document has been wrtten for you. You only need to enter the Mn NIR value and the slope values for bands #5 #7 (b 5, b 6 or b 7 n the equaton). Open the formula document Deglntng_band17.frm. Study the formula document and note how the b and Mn NIR values are set up as a seres of constants n the CONST = statements. You need to fll n the Mn NIR value for band #17 and also the regresson slope values for bands #5, #6 and #7, whch can be found on the relevant charts n the Lesson05_deglntng_St_John.xls spreadsheet. Do ths, rememberng that each statement must end wth a ;. Note that there s a separate formula for each band. Let us brefly examne the frst of the formulas. IF (@1 == 0) 0 ELSE (@1 - Band01_Slope * (NearInfraRed - MnNIR)) ; The background n each mage (areas outsde the CASI groundtrack) have been set to zero and thus should not be processed. Thus the frst part of each formula says that f the vsble band pxel value s equal to zero then the output mage pxel wll be 0. Else (otherwse) for all other pxels Equaton 5.1 s appled. Thus each pxel (R 1 ) of (= band#01) mage has the Band01_Slope (b 1 ) multpled by the dfference between the correspondng band#17 (@8 mage) pxel (R NIR ) and the Mn NIR value for band #17 (= 208). Ths effectvely removes the component of the sgnal due to specular reflecton. The formula just mmcs Equaton 5.1. When you have completed the formula document, save t. Then select Optons! from the menu, set the Output Image Type: to Floatng pont 32-bt, and uncheck the Use specal handlng for Nulls checkbox (f checked) because the formula deals wth these already. Copy the formula and paste t to the stack of connected mages. You should get 7 deglnted mages produced. Connect the frst (deglnted band#01), thrd (deglnted band#03) and ffth (deglnted band#05) of the output mages (usng Image, Connect) but do not stack them. Use the Selector toolbar to desgnate the deglnted band#05 as mage 1 (@1: red gun), the deglnted band#03 as mage 2 (@2: green gun) and the deglnted band#01 as mage 3 (@3: blue gun). When you have done ths, select Image, Composte to generate the false colour composte. Save the deglnted composte as ether a Blko.set or Wndows.bmp fle. When you have done ths, you should close the ndvdual deglnted mages, leavng only the composte open, n order to reduce clutter. The deglnted composte should look smlar to Fgure 5.3. Zoom out to 50% and compare the deglnted composte wth the orgnal raw composte you saved as a Wndows.bmp mage at the start of the lesson. 11

12 Applcatons of satellte and arborne mage data to coastal management Queston: 5.4. In what specfc ways has the deglnted composte mage mproved compared to the raw composte? Fgure 5.3. A colour composte (f prnted, ths may be n shades of grey), showng the CASI mage after deglntng. At ths pont you have seen the effectveness of the deglntng method and hopefully understood the concepts underlyng t. If you are feelng ambtous or curous or both, you may wsh to dscover whether band #18 s a better NIR band to choose for deglntng. As we noted earler, band #17 s only just nto the near nfra-red part of the electro-magnetc spectrum and one mght acheve better results wth band #18, despte the coeffcents of determnaton (R 2 ) n the test regressons beng less than for band #17 (Table 5.2). To allow you to test ths farly easly, the formula Deglntng_band18.frm s ncluded. All you need to add s the Mn NIR value for band #18, whch you calculated earler, nto the CONST statement at the start. Ether fnsh by closng all documents (Wndow, Close All) or deglnt the stack of mages usng the Deglntng_band18.frm formula wth the CONST MnNIR = ; statement completed. Create a colour composte of deglnted bands #1, #3 and #5 and compare wth the one you created usng band #17 as the NIR band. When you are fnshed, close all documents. You wll fnd that there s lttle to choose between deglntng wth band #17 or band #18. References Hedley, J.D., Harborne, A.R. and Mumby, P.J Smple and robust removal of sun glnt for mappng shallow-water benthos. Internatonal Journal of Remote Sensng (n press). Hochberg, E.J., Andréfouët, S. and Tyler, M.R Sea surface correcton of hgh spatal resoluton Ikonos mages to mprove bottom mappng n near-shore envronments. IEEE Transactons on Geoscence and Remote Sensng 41 (7): Mobley, C. D Lght and Water. Academc Press. 12

13 Applcatons of satellte and arborne mage data to coastal management 14