Ocean surface temperature- and colour studies from satellites JAN-PE‘ITER PEDERSEN Pedersen, J.-P. 1990: Ocean surface temperature- and colour studies from satellites. Polar Research 8, 5 9. This paper introduces satellitc remote sensing of sca-wrface temperatures and ocean colour studies. The basic radiative transfer cquations and available satellite sensor systems are presented. The final part of the paper focuses on tcmpcraturc applications. Sea surface temperature data derived from availahle satellite data are presented and discussed. Jan-Perter Pedersen. Department of Information Technology, Foundation of A p p l i e d Research at the Unioersity o f Tromsm ( F O R U T ) , P . 0. Box 2806 Elurrh@y. N-9001 Trontsm, N o r w a y ; February 1989 (revised June 1989) The role of satellites in the monitoring of con- ditions of the Earth’s atmosphere and surface is becoming increasingly important. Satellite measurements are particularly useful in disci- plines such as oceanography and meteorology/ climatology. Due to the frequent and the sim- ultaneous large area coverage, vast areas of ocean and atmosphere can be studied simultaneously. Oceanographic applications of satellite remote sensing mainly include studies of surface para- meters such as: temperatures, currents, winds and waves. Also of interest are studies of ocean colour which is an index of (near) sub-surface parameters such as ocean productivity and/or concentrations of suspended material. The application areas of a remote sensor are mainly determined from the operational charac- teristics of the sensor. The remote sensing of ocean colour requires that the sensor can detect the radiation coming from below the ocean surface. This radiation carries information about sub-surface chemical and/or physical conditions. Such radiation is present in narrow spectral bands in the blue and the blue-green part of the elec- tromagnetic spectrum (Swain & Davis 1978). It is therefore a requirement that an ocean colour observing sensor must operate narrow spectral bands in the blue/blue-green part of the spectrum. The surface temperature determines the ther- mal infrared radiation emitted from the sea. The radiation is emitted according to Planck’s radi- ation law (Swain & Davis 1978). The surface temperature is observable by a remote sensor operating in the thermal infrared, o r in the micro- wave part of the electromagnetic spectrum. Application of traditional oceanographic methods in combination with satellite remote sensing has proved a high degree of correlation between the observed sea surface temperature and the near sub-surface productivity. The areas of highest productivity normally correspond to the areas of lowest surface temperatures (NASA 1984b). Regions of strong tidal mixing and upwell- ing of deeper water appear as local cool areas exposing different temperature gradient patterns. The temperature gradients (fronts) reflect the physical processes (currents and mixing) which affect the local biological productivity. Ocean colour information derived from spaceborne data has been used by fishermen off the coast of Cali- fornia as a tool for increasing the tuna catches (NASA 1 9 8 4 ~ ) . In addition to fish stocks, areas of enhanced productivity have been observed as feeding areas for marine birds and other predators (Schneider this volume). Radiative transfer equations The radiance detected by a remote sensor is com- prised of different components, originating from different parts of the total sensor-atmosphere- ocean system: - the radiance originating from the atmosphere - t h e radiance reflected at the ocean surface - the radiance originating from and below the The individual component contribution to the (atmospheric) (reflected) surface (emitted) 4 Jan-Petter Pedersen n 025 I j AMOS- m c 0.9 3 E M m Fig, 1. Components of the total signal detected by a remote sensor in the visible and in the thermal infrared wavelength regions. The numbers indicate the relative fractional con- tribution limits for the different components (Maul 1981). total signal is strongly wavelength dependent. Fig. 1 illustrates schematically the fractional com- ponent contribution to the total signal at the visible and the thermal infrared wavelength regions. The numbers indicate the relative frac- tional contribution limits for the different com- ponents. The figure shows that in the visible region the atmospheric component of the total detected signal will comprise 6@90%. When deal- ing with quantitative remote sensing of ocean near-surface conditions, the interesting com- ponent is that representing the near-surface emission. This means that the components repre- senting the atmospheric contribution and the sur- face reflection have to be quantified and compensated for. The equation of radiative transfer The electromagnetic radiative transfer in a sensor- atmosphere-ocean system can be described math- ematically by the equation of radiative transfer. This equation describes all effects inherent upon the electromagnetic radiation. Of special import- ance are the effects of the atmosphere (attenu- ation) between the sensor and the measured surface. The atmospheric attenuation depends on the wavelength of the electromagnetic radiation and the composition of the atmosphere along the sensor's line of sight. The complete form of the equation of radiative transfer is complicated (Liou 1980), and only a simplified component version of the equation is presented here (Pedersen 1987): + RE@, h, $91 + R,(A, h, p, Cp) (1) where R(A, h, p, Cp) = the radiance detected by a sen- sor at altitude h, at a given wave- length A, in the direction (p = cos-'O, 4). 8, $J represent the zenith and azimuth angles, respectively. RR(A, h, p, Cp) = the radiance component due to reflection of solar radiation at the surface. RE(& h, p, Cp) = the radiance emitted from the surface. Rp(h, h, p, Cp) = the atmospheric radiance (emitted and reflected). $A, h, p, Cp) = the atmospheric attenuation of the surface radiance. Solving equation (1) implies applications of proper boundary conditions at the top of the atmosphere, at the atmosphere-ocean interface and at the sea bottom (if visible) (Guzzi et al. 1987). In practice this means that the equation is almost impossible t o solve without any sim- plifications. The simplifications of the complete equation are strongly dependent upon the actual wave- length bands applied. For example, Fig. 1 shows that a quantitative application of the emitted radi- ance component in the visible region requires that the dominating atmospheric component must be determined and corrected for (Pedersen 1987). In this region the surface reflection will also con- tribute significantly. A similar application in the thermal infrared region means that the reflected component can be ignored without introducing any significant errors (cf. Fig. 1). The atmospheric component is also less dominating in this region. Compared to the visible region, the atmospheric contribution is more easily determined in the thermal infrared region (Pedersen 1982, 1987). The quantitative expressions are outside the scope of this paper. Available satellite sensors The NOAA-series of polar orbiting, sunsynch- ronous satellites from which data are read out at Ocean surface temperature- and colour studies from satellites 5 TromsG Satellite Station (TSS), offer the oppor- tunity t o study surface phenomena in arctic regions with a high frequency of repetition. The primary sensor of the NOAA satellites is the Advanced Very High Resolution Radiometer (AVHRR) observing in the visible and in the thermal infrared region. The thermal infrared data from the NOAA satellites are frequently used for studying currentsand the sea surface temperatures (SST). The spatial resolution of the NOAA-data of 1 km limits their application to open ocean areas. The new generation of satellites represented by the Landsat/Thematic Mapper (TM) offers, however, the opportunity t o study surface phenomena at an increased spatial reso- lution. Compared to the NOAA satellites, the 120 meter resolution of the TM thermal channel is more adaptable for coastal-zone applications. The important operating characteristics for the satellites are presented in Table 1. The Landsat satellites also operate another visible/near-infra- red radiometer, the Multispectral Scanner (MSS). Due t o the spectral location and bandwidth of the MSS bands, this radiometer is not suitable for quantitative ocean colour applications. Until now, there has been only one operational satellite dedicated to ocean colour studies. This satellite, NIMBUS-7, was launched in 1978, and was operational for about five years. The ocean colour sensor on board NIMBUS-7 was a passive radiometer, CZCS (Coastal Zone Colour Scanner), operating at narrow visible and near- infrared spectral bands (see Table 1). The spatial resolution of the sensor was 800 meters, which again limits the applications mainly to open oceans. Sensors analogue to the CZCS are plan- ned to be put on future Earth observation plat- forms. These sensors will, however, not become operational until the mid-1990s. Geophysical satellite data applications Ocean colour studies During its operational lifetime, data from NIMBUS/CZCS were applied for a number of ocean colour studies (Gordon & Clark 1980; Sturm 1982; Gordon et al. 1983). CZCS data have, however, not been applied for quantitative ocean colour studies in Norway. A study presented by Gordon et al. (1983) discussed the application of CZCS data for the determination of phytoplankton pigment con- centrations in the Middle Atlantic Bight off the East Coast of the USA and in the Sargasso Sea. The pigment' concentrations determined from sat- ellite images were compared to measurements performed by ships. The results obtained sug- gested that over the 0.08-1.5 mg/m3 range, the error in the retrieved pigment (mainly Chloro- phyll a) concentration is of the order 3 0 4 0 % for a variety of atmospheric conditions. Fig. 2 shows a comparison of ship and CZCS measured pigment concentrations (Gordon et al. 1983). The main problem in ocean colour studies is t o obtain a correct atmospheric correction. The atmosphere is a strongly varying medium both in time and space, and these variations are hardly described by any given law. The variations must therefore rely on observations. There exists no Table I . Actual ocean colour/temperature observing satellites operational characteristics. AVHRR TM CZCS Spectral bands (p) Spatial resolution Swath width Operational status 0.58-0.68 0.7CL1.1 3.55-3.93 10.3-11.3 11.5-12.5 l x l k m c. 2,500 km Active 0.45-0.52 0.5>0.M) 0.6W.69 0.764.90 1.55-1.75 2.08-2.35 10.412.5 30 x 30m 120 x 120m thermal IR 185 km Active 0.4N.45 0.51-0.53 0.54-0.56 0.660.68 0.704.80 10.5&12.50 0.8 x 0.8 km c. 2,500 km Passive 6 Jan-Petter Pedersen -2 -I I 2 Log (Ship Pigmento) Fig. 2. Comparisons of ship and CZCS measured pigment concentrations for four different NIMBUS orbits (3157, 3171, 3240, 3351) (Gordon et al. 1983). developed observation network in ocean areas, and present satellite sensors do not give the radio- metric accuracy necessary for operational atmospheric corrections of high quality. Sea-surface temperature studies The thermal channels of the AVHRR and the TM are t o be considered for this application. The AVHRR 10.3-11.3 pm, the 11.5-12.5 pm chan- nels, and the TM 10.4-12.5 pm channel are all in the same part of the electromagnetic spectrum. Notice, however, that the spectral width of the TM channel is twice that of the AVHRR channels. Due to the spectral coincidence of the TM and AVHRR channels, they are applicable for comparable studies of sea surface tempera- tures. According to the pre-launch specifications, the temperature sensitivities are <0.12 K and 0.5 K at a surface temperature of 300 K for the AVHRR and TM, respectively (NASA 1984a; Schwalb 1979). The most significant difference between the channels is their spatial resolution, 120 meters for TM compared to 1 km for AVHRR. The spatial resolution of the AVHRR limits its applications primarily to open oceans. For studies in the coastal zone and within the fjords, the TM ther- mal channel is the most suitable. Different algorithms have been proposed for retrieval of sea surface temperatures (SST) from thermal infrared satellite data. These algorithms range from physical solutions of the equation of radiative transfer to algorithms based completely upon statistical regression analysis. The absolute accuracies, as compared to in situ data, vary from approximately 1 deg. Kelvin for the single band direct solution algorithm to a few tenths of a Kelvin for the split-window algorithms (Pedersen 1982). Weinreb & Hill (1980) describe an algorithm applying single band thermal infrared data. From the general equation of radiative transfer in an attenuating medium (cf. equation l ) , the absolute surface temperature is derived after having cor- rected for the atmospheric influence. The atmospheric correction is calculated from an input of atmospheric temperature- and humidity pro- files. This algorithm has been implemented for operational use at T r o m s ~ Satellite Station (Pedersen 1982). For testing the implemented SST algorithm, a NOAA data set from the summer 1981 covering the island Jan Mayen was applied (see Fig. 3) (Pedersen 1982). Radiosonde data from the meteorological station at the island were applied as input for atmospheric corrections of the sat- ellite data. In situ SST measurements from the area around Jan Mayen made by ships were also available from the Norwegian Institute for Marine Research. The resulting atmospheric corrected SST image is presented in Fig. 4. The scale at the top of the image gives the relationship between the grey levels and the surface temperatures. The scale number 1.25 means that the surface temperature is within the range 1-1.5deg. centigrade. The island Jan Mayen is located almost in the centre of the image. Comparisons of the satellite and the in situ temperatures show an average absolute difference of 1 deg Kelvin. For all comparisons, the satellite measurements were systematically below the in situ measurements. The observed difference agrees well with the commonly accepted error achievable from a physical model, single thermal infrared band algorithm (Pedersen 1982). The increased spatial resolution of the TM ther- mal infrared channel, as compared to the AVHRR, offers the possibility of applying the TM for studies of the surface temperatures in the coastal zone and within the fjords. The image presented in Fig. 5 is a I,andsat-S/TM thermal infrared sub-scene covering the T r o m s ~ area. TromsB is located in the upper right corner of the Ocean surface temperature- and colour studies f r o m satellites 7 Fig. 3. Approximate geographical location of data sets presented in this paper. A = Jan Mayen, B = Tromse. image. The scale in the bottom left corner gives the correspondence between the grey levels and the surface temperatures. The colour-tempera- ture scale was calibrated using surface tem- peratures measured at known image locations. By combining the thermal infrared band with a near- infrared band (0.764.90 pmj, areas of land have been blanked out. When the data set was acquired (3 June 1984), the sea current around the island Tromsp, was moving north (up in the image). Relatively warm surface water (c. 10deg. centigrade) is trans- ported upwards from Balsfjord (centre right in the image). Two bridges extend from the island of T r o m s ~ , one to the east and one to the west. As the water flows under the bridges, the supports cause a turbulent mixing of the warm surface water with the colder sub-surface water which is clearly observed in the image. In the centre of the image, there is an area 8 Jan- Petter Pedersen Fig. 4. SST image from Jan Mayen (centre of image). Image based on NOAA/AVHRR data read out at Troms0 Satellite Station 20 August 1981. Thc scale at the top of the image gives the relationship between the grey levels and the temperatures. A scale value of e.g. 1.25 means that the surface temperature is within 1 . and 1.5 deg. centigrade. where the surface water temperature is approxi- mately 2-3 deg. below that of the surrounding areas. This is caused by strong turbulence in the narrow sound between the two large islands. Notice also that there is a small island located in Fig. 5. Landsat-S/Thematic Mapper SST image covering the Tromse area. Data read out at Kiruna read out station 3 June 1984. For discussion of the image, see the text. the sound which strongly influences the water transport through the sound. In the bottom right of the image, a fjord which is an outlet for cold water can be seen. The cold water results from the river MBlselva transporting cold snow-melt water from the nearby mountains into the fjord. Conclusions Ocean colour and surface temperature studies are best performed using optical remote sensing data. The major limitation regarding the optical data availability is the dependence upon the weather conditions. Darkness and/or cloud cover strongly limit the data acquisition. Experience from optical remote sensing in arctic regions show that, on average, there are only a few days during the year which are completely cloud free over large areas. Despite these limitations, a number of important and interesting optical remote sensing appli- cations have been demonstrated. The results obtained from NIMBUS/CZCS data for ocean colour studies indicate that this category of remote sensing data represents an important potential for the mapping of the pro- ductivity of remote, but important, ocean areas. The atmosphere represents the major problem in order to obtain acceptable geophysical data accuracy for operational applications. The at- mosphere is a temporary and spatially varying medium, and the variations can not be given from a known law. In each case, the atmosphere needs to be described and the variations detected before reliable atmospheric corrections of the sub-sur- face data can be obtained. For SST studies, the medium-resolution NOAA/AVHRR have been very useful in the research towards a better understanding of oce- anic processes. Although the coastal current off the western coast of Norway has been known for long, NOAA/AVHHR surface temperature data offered the possibility t o perform simultaneous, multitemporary large area current studies. Sat- ellite data combined with model studies have increased the understanding of the coastal current generation mechanisms which is a combination of hydrographical and meteorological conditions. This research has resulted in an operational sys- tem for coastal current forecasting. A number of algorithms are available for the SST generation. The single band, physical model Ocean surface temperature- and colour studies f r o m satellites 9 achieves an accuracy of approximately 1 deg. Kel- vin, while the multiple band algorithms claim an accuracy of a few tenths of a Kelvin (Pedersen 1982). The results presented in this paper show that the satellite derived surface temperatures are consistently below the ship measured tempera- tures. This is a very important result, and the absolute accuracy may for example be enhanced by the addition of a bias term into the algorithm. The reason €or not applying a bias term in this algorithm was primarily t o understand and dem- onstrate the single band physical model for radi- ative transfer at thermal infrared wavelengths in a coupled sea-atmosphere-sensor system (Pedersen 1982). The medium resolution of the AVHRR data limits the applications mainly to open ocean areas. The improved resolution of the Landsat/TM offers the possibility to perform SST studies in the coastal zone and within the fjords. These applications have become more and more impor- tant during recent years, because of the increased economical exploitation of the coastal areas. References Gordon, H. R. & Clark, D. K. 1980: Atmospheric effects in the remote sensing of phytoplankton pigments. Bound. Layer Meteorology 18, 295L313. Gordon, H. R . , Clark, D. K . , Brown, J . W., Brown, 0. B., Evans, R. H. & Broenkow, W. W. 1983: Phytoplankton pigment concentrations in the Middle Atlantic Bight: com- parison of ship determinations and CZCS estimates. Applied Optics 22(1), 2&36. Guzzi, R., Rizzi, R. & Zibordi, G. 1987: Atmospheric cor- rection of data measured by a flying platform over the sea: elements of a model and its experimental validation. Applied Optics 26(15), 3043-3051. Liou, K. 1980: A n Introduction to Atmospheric Radiation. Aca- demic Press Inc., New York. Maul, G. A. 1981: Application of GOES Visible-Infrared Data to Quantifying Mesoscale Ocean Surface Temperatures. J. Geophys. Res. 86(C9), 8007-8021. NASA 1984a: A Prospectus for Thematic Mapper Kesearch in the Earth Sciences. NASA Technical Memorandum 86149. NASA, GSFC, Greenbelt, Md. NASA 1984b: Phytoplankton & Surface Patterns. NASA, JPL 400-221 2/84. NASA 1984c: Tuna Catch & Ocean Color. NASA, JPL 4W 225 2/84. Pedersen, J.-P. 1982: Atmospheric effects on infrared sea-surface temperature measurements from satellites. Cand. real. thesis, Univ. of Tromscl (in Nonuegian). Pedersen, J.-P. 1987: A model for estimation and removal of the atmospheric effects on Landsat/Thematic Mapper subsurface water radiances. Internal report. IMR, Univ. of Tromsb. Schwalb, A. 1979: The TIROS-N/NOAA Satellite Series. N O A A Technical Memorandum NESS 95. Swain, P. & Davis, S . 1978: Remote sensing: The quantitative approach. Addison-Wesley Publ. Company, Reading, Mass. Sturm, B. 1982: The atmospheric correction of remotely sensed data and the quantitative determination of suspended matter in marine water surface layers. Paper presented at Remote Sensing Summer School at Dundee, Scotland. Weinreb, M. P. & Hill, M. L. 1980: Calculation of Atmospheric Radiances and Brightness Temperatures in Infrared Window Channels of Satellite Radiometers. N O A A Technical Report N E S S 80.