Calibration of Landsat thermal data and application to water resource studies

Transcription

1 Remote Sensing of Environment 78 (2001) Calibration of Landsat thermal data and application to water resource studies John R. Schott*, Julia A. Barsi, Bryce L. Nordgren, Nina Gibson Raqueño, Dilkushi de Alwis Center for Imaging Science, Rochester Institute of Technology, 54 Lomb Memorial Drive, Rochester, NY , USA Received 13 February 2000; received in revised form 2 January 2001; accepted 30 April 2001 Abstract The newest in the Landsat series of satellites was launched April 15, The imagery collected by Landsat is used for a myriad of applications, from coral reef studies to land management. In order to take advantage of Landsat 7 data, the Enhanced Thematic Mapper+ (ETM+) instrument must be calibrated. This study focuses on the immediate postlaunch calibration verification of the Landsat 7 thermal band (Band 6), specifically so that it can be useful in water resource studies. Two year s worth of thermal calibration results using a combination of underflight data and ground truth show the ETM+ to be extremely stable, though the prelaunch calibration produces an offset of W/m 2 sr mm. This paper focuses on the details of the calibration process, including problems faced with ground truth instrumentation. While the technical emphasis in this paper is the calibration of Landsat thermal data, it is presented in the context of the water resource studies for which calibrated thermal data are required. At certain times in the year, water quality in large lakes, particularly the spatial structure of water quality, is driven by temperature of lake waters. During the spring warming, a phenomena called the thermal bar drives the current and sedimentation of large water bodies. A long-term goal of this study is to use thermally driven hydrodynamic models of lake processes to better understand and monitor water quality in large lakes. This paper presents the hydrodynamic model and the relationship between temperature and water quality in the Great Lakes as one example of why high-resolution, well-calibrated data are critical to earth observing. D 2001 Elsevier Science Inc. All rights reserved. 1. Background 1.1. The Great Lakes as a natural resource The Laurentian Great Lakes hold 18% of the world s fresh water. The US coast line of the Great Lakes exceeds that of the Atlantic coast. Roughly 10% of the US population and 32% of the Canadian population live in the Great Lakes drainage basin, approximately 35 million people. Pollution loads, both from within the region and introduced with precipitation, represent an enormous burden with both local manifestations, as well as an ongoing regional threat. Yet, despite all this, the Great Lakes are surprisingly understudied as a system and poorly understood. The enormous surface area, volume, and dynamic nature of the lakes make it very difficult to study just one, much less all five of the Great Lakes. The first, and last, serious attempt to study Lake Ontario as a whole was in 1972 during the * Corresponding author. Tel.: ; fax: address: schott@cis.rit.edu (J.R. Schott). International Field Year on the Great Lakes before much of the current monitoring technology existed (Aubert & Richards, 1981). Even localized studies of shoreline processes and river discharges are relatively scarce because of problems associated with sampling the receiving waters and the length of the coastline. The Great Lakes are often referred to as inland oceans; however, unlike oceans, they are small enough that given a thorough study, it may be possible to understand the primary drivers and processes that impact lake conditions. Our interest in studying the Great Lakes is twofold. The first is the obvious need to understand, monitor, and forecast water quality conditions impacting this precious resource. The second is our belief that the size of the lakes makes them a good metric for global process monitoring [i.e., they are large enough to damp out localized phenomena yet small enough (compared to oceans) to generate observable responses] The Great Lakes as a thermal target The thermal structure in the Great Lakes has a tremendous influence on the hydrodynamics and water quality in the /01/$ see front matter D 2001 Elsevier Science Inc. All rights reserved. PII: S (01)00253-X

2 J.R. Schott et al. / Remote Sensing of Environment 78 (2001) lakes. This is best exemplified by the spring thermal bar. The thermal bar occurs in dimictic temperate zone lakes, resulting from the fact that the maximum density of water is 4C. To understand the thermal bar, consider lakes in the winter. At this time, lakes are in winter stratification with cold water on top and often with some degree of ice coverage. As insolation increases in the spring, both direct heat input and indirect input from runoff begin to thaw the ice. Once the ice has melted, warm water runoff and insolation effects dominate the shallow nearshore processes and a localized summer stratification sets up with warm water on top and cooler water underneath. A ring of water in summer stratification forms around the lake. The central core, still in winter stratification, slowly warms from the surface. The surface water sinks, setting up a very slow mixing process in the core of the lake. Where the two bodies (cold core and warm shore waters) meet, they equilibrate towards 4C and the water sinks. The result is the formation of the thermal bar, a stable boundary between the warmer nearshore water and the cold core, persisting for many weeks into early summer (cf. Fig. 1). During this period, the thermal bar moves out from shore and slowly constricts. Eventually, the thermal bar breaks down and the whole lake goes into summer stratification with relatively well-mixed water on the surface, a fairly sharp thermocline and cooler stratified water underneath. Because there is very little mixing across the thermal bar, it is a controlling factor in water quality conditions. Nutrient- and pollution-rich runoff is trapped in the nearshore region. While the thermal bar persists, localized studies have shown the nearshore water to be more turbid and nutrient-, pollution-, and bacteria-rich (Menon, Dutka, & Jurkovic, 1971; Rogers, 1968, 1971; Stoermer, 1968). While this entire process is of interest, we are particularly interested in the early nearshore formation of the thermal bar. During this early period, both stream discharges and nonpoint runoff carry the spring flush of pollutants into the lake to be trapped in a very localized area where concentrations dramatically increase. Traditional sensing and monitoring methods have not been able to study this very localized (nearshore) yet simultaneously lake-wide (i.e., the entire shore) process except in isolated cases. Yet it is clear from Fig. 1 that remote sensing offers a potential to observe, at a minimum, the surface turbidity effects at both localized and whole lake scales The use of remote sensing for water quality studies To improve monitoring and mapping of regional water quality parameters and processes in the Great Lakes, there are a number of requirements: the need to be able to extend from a spatially localized (usually nearshore) study to the entire lake and from a temporally localized (days to weeks) study to long-term predictions of temporal processes and conditions; the need to relate what is learned from one localized study to a similar, but different site, tens or hundreds of miles down shore; and the need to look at the impact of aggregate sites Fig. 1. A selection of Landsat 5 images of Lake Ontario assembled from nearly cloud-free scenes during the spring warming season. The thermal bar progression and collapse is readily apparent in the thermal band and water quality gradients coincide with the location of the warm water in the visible color composite. and nonpoint source contributions on the entire regional resource. Remote sensing offers the synoptic perspective to study an entire lake and potentially all the lakes as a system. By combining different remote sensing resources, both small-scale, site-specific phenomena and lake-wide phenomena can be studied. The daily coverage of many sensors allows us to generate cloud-free composites at temporal intervals to monitor dynamic processes. With the current generation of sensors [AVHRR, SeaWiFS, and Landsat TM/ Enhanced Thematic Mapper+ (ETM+)], sediments (suspended solids), algae (chlorophyll), and gelbstoffe (dissolved organic compounds) have successfully been monitored in the open ocean, and to some extent, in large lakes (Bukata, Jerome, Kondratyev, & Pozdnyakov, 1991; Hudson, Moore, Bale, Dyer, & Aiken, 1994; Piech, Schott, & Stewart, 1978; Strong, 1978). With the next generation of sensors (AVIRIS, MODIS, NEMO, WARFIGHTER, etc.), the potential to detect and measure critical water quality parameters should improve (Murphy, Moore, Wilson, Youngs, & Morris, 1996; Richardson & Ambrosia, 1996).

3 110 J.R. Schott et al. / Remote Sensing of Environment 78 (2001) Remote sensing, however, also has limitations. Optical remote sensing records only the surface of the world. In the case of water there is some penetration of the surface, but deeper water is obscured, limiting studies of vertical stratification of the thermal bar and transport of materials. A critical issue identified with the thermal bar is restricted mixing; the bar prevents nearshore waters from mixing evenly with core lake water as it normally would, trapping nutrient- and pollutant-rich runoff nearshore. Adding a hydrodynamic model to predict what cannot be seen would increase the ability to follow subsurface progressions of thermal bars and sediments and would add the capability to further understand the physics of the processes, including inputs, outputs, and driving factors Hydrodynamic modeling To augment imaging studies, Rochester Institute of Technology (RIT) is utilizing a hydrodynamic model initially developed by the Department of Energy to predict the movement and dissipation of thermal plumes discharged into cooling lakes, rivers, and estuaries (Garrett & Hayes, 1997; Garrett et al., in press). The four-dimensional (x,y,z,t) finite difference hydrodynamic model, termed ALGE, is capable of predicting temperatures, flow vectors, and material transport. Cooling lake simulations include recirculation, buoyancy-driven flow, and sediment deposition. ALGE also simulates wind-driven circulation and can combine wind stress effects with tidal and buoyancy forces. Atmospheric energy exchange is modeled through turbulent sensible and latent heat transfer, including the effects of clouds. ALGE was designed to produce high-resolution simulations for node-to-node matching with aircraft and satellite imagery. The ALGE model has recently been adapted for use with Lake Ontario (Fig. 2). Modifications have enabled modeling vertical stratification, formation, and development of the thermal bar, inclusion of the Niagara River (80% of the total influx to Lake Ontario), the St. Lawrence River outflow, and general lake circulation (including coriolis effects). The model incorporates lake bathymetry, inflow and outf low (temperature, volume and extent), hourly air temperature, wind speed, wind direction, insolation (including cloud cover effects), relative humidity, and radiosonde inputs for upper air effects. These meteorological data are interpolated to a lake-wide mean using up to seven meteorological sites around the lake. The input parameters to ALGE are the initial lake temperature and the date of the spring turnover when the lake is assumed to be well mixed. These parameters can be determined using remotely sensed images of surface temperature. The starting date and initial start temperature have a dramatic effect on the ALGE s prediction of a thermal bar s initial formation and progression. Remotely sensed data can both improve inputs to the hydrodynamic model and empirically calibrate the model. With the spatial resolution of Landsat, the images can be used both as feedback and as verification to the model using both lake-wide and localized (e.g., stream discharge) phenomena. Presently, AVHRR-derived temperature maps are used as input for a start date and initiation temperature for ALGE. Images are observed until the surface of the lake appears to be ice-free and isothermal. The surface temper- Fig. 2. Example outputs of the ALGE 3-D hydrodynamic model with two validation images. The surface temperature map images show the formation and the two-phase propagation of the thermal bar (water temperature of C) in Lake Ontario. Images (a) (d) were taken from a time sequence of modeled temperatures for spring warming in 1998 before the Niagara inflow was included. Images (e) (h) are for the same conditions after the Niagara inflow and St. Lawrence outflow were added. Images (k) and (l) are east west cross-sections of the lake corresponding to the surface images (g) and (h). Images (i) and (j) are AVHRR-derived temperature maps using a different color code and illustrate the need for incorporating the Niagara inflow.

4 J.R. Schott et al. / Remote Sensing of Environment 78 (2001) ature on this date is used as the initial water temperature for Lake Ontario in the model. Landsat s calibration and resolution should provide more accurate temperature maps, providing the model more accurate inputs. Once modeling has predicted the location of the thermal bar during its propagation, improved Landsat temperature maps can be used to verify that ALGE is modeling the surface hydrodynamics correctly. In particular, interactions of local discharges with the thermal bar can be observed with Landsat. Since each discharge has a separate set of inputs, they can be independently adjusted until the ALGE predictions match the detailed thermal patterns observed in that region by Landsat. Additionally, Landsat s spatial resolution allows a closer look at local effects which are not presently in ALGE. At this point, the model of Lake Ontario only includes one inflow, the Niagara River (Fig. 2). This is the primary source of water for the lake, but smaller discharges may provide the bulk of sediments and pollutants. With the higher spectral and spatial resolution, effects of smaller discharges can be observed and significant effluents that should be incorporated into ALGE can be identified. In summary, a hydrodynamic model has been developed and implemented that models, on a whole-lake basis, the formation and development of the thermal bar. It also includes the major inflow and outflow and can be expanded to include local discharges and nonpoint pollutants. This model is also capable of mass transport calculations throughout the lake. However, the model is driven by and must be calibrated and validated by accurate local and lakewide temperature measurements. The circulation in large lakes is driven largely by buoyancy effects caused by thermal variations in the water. One primary way to evaluate and control the model s performance is to compare measured and observed temperatures at the surface, the one place lake-wide measurements are possible. To accomplish this, we need a well-calibrated, high-spatial resolution imaging sensor capable of mapping the detail associated with nearshore phenomena (early thermal bar formation and storm discharges) as well as whole-lake processes. This is just one example of why we need well-calibrated, high-resolution thermal sensors on-board instruments like Landsat. The rest of this paper focuses on the thermal calibration of the Landsat ETM+ instrument which is critical to the future success of studies like the one discussed here Thermal calibration Validation of the thermal calibration of Landsat ETM+ is a fundamental goal of this study. To minimize error in this vicarious method of calibration, the appropriate targets must be selected. The qualities of a good thermal calibration target are: (1) a well-known emissivity, (2) large (enough to cover at least a few pixels), (3) homogeneous, and (4) thermally stable. In the right proportions, water satisfies all of these requirements. The emissivity of water is nearly flat through Fig. 3. Variation in the emissivity of water over the 8 14-mm spectral range (from Melchor, 1941). the long-wave infrared region at (Fig. 3) (Melchor, 1941), even under the varying conditions typically experienced on a large body of water (variable wind speed, Singh, 1994; and differing material concentrations, Masuda, Takishima, & Takayama, 1988). Large lakes incorporate more than a few pixels. Wind across large lakes produces a nearly constant mixing effect, tending to bring the lake to a uniform temperature and making the temperature extremely stable. The need for data over a large lake to perform vicarious calibration drives the water quality research. Already, its apparent that any one of the Great Lakes would make an ideal target for both calibration and water quality study. However, the fact that the Great Lakes also develop thermal bars makes them an even more desirable target. The thermal bar provides dramatic thermal contrast for the calibration study, as well as driving the deterioration of nearshore water quality in the spring. 2. Technical approach and results Over the two vicarious calibration collection seasons, summer 1999 and summer 2000, Landsat 7 Band 6 has proven to be incredibly stable. The offset has remained constant since launch and the internal gain, if it is changing, is changing at a rate of 0.04% per year (Markham, 2000). The focus of this study was to verify the absolute radiometry of the internal gain and offset. A combination of airborne instrument measurements and ground samples was collected and the corresponding Landsat scenes were acquired to compare the image predicted sensor-reaching radiance to the ground truth predicted sensor-reaching radiance On-board thermal calibration The ETM+ on-board thermal calibration uses essentially the same two-step approach that was used to calibrate

5 112 J.R. Schott et al. / Remote Sensing of Environment 78 (2001) Landsat 4 and 5 TM thermal bands (Barker, 1985). A pair of on-board thermal reference levels is introduced by the calibration wand during the scan mirror reversal at the end of each line (Fig. 4). This provides two known radiance levels (one from the monitored wand temperature and one from a monitored blackbody whose radiance is reflected onto the detectors by a mirror on the calibration wand). Thus, the digital count (DC)-to-radiance calibration at the wand has a full two-point update for every mirror oscillation. This would complete the calibration process if the calibration wand were ahead of all of the ETM+ optics. However, as shown in Fig. 4, the scan mirror, telescope, and scan line corrector optics are all ahead of the calibration wand. These optical elements and their support structure will attenuate the radiance reaching the scan mirror and emit additional radiance into the optical path within the instrument. Since viewing the calibration wand does not include these effects, the contribution of these elements was estimated using a fore-optics radiometry model developed prior to launch. The radiometry model is empirically calibrated based on a number of thermal sensors that monitor the temperature of various elements in the optical assembly ahead of the calibration wand. This calibration generates a correction to the per scan calibration that should account for the fore-optics contribution. Thus, by monitoring the temperature of the calibration wand, on-board blackbody, and optical components, the image data can theoretically be calibrated. This study provides an independent check of the calibration by comparing targets of known radiance to imagederived Landsat radiances. Using simultaneous airborne image and ground truth acquisitions, the surface-leaving radiance was predicted and extrapolated to space-reaching radiance using the MODTRAN radiation propagation model (Berk, Bernstein, & Robertson, 1989). Schott, Gallagher, and Barsi (1997) describe the details behind this approach. For large, uniform, thermally stable targets, the predicted and observed airborne image radiances should differ only by a small correction for the effect of the atmosphere above the aircraft. This can be estimated using MODTRAN according to (Eq. (1)): L s ðsþ ¼tðs hþl a ðhþþl u ðs hþ ð1þ where L s (s) and L a (h) are the spectral radiance values predicted to reach the spacecraft and observed by the aircraft or measured on the ground, respectively, s and h indicate the sensor s location in space (s) or at some elevation (h), t(s h) is the transmission from altitude h to space and L u (s h) is the path radiance due to the air column between the aircraft and the spacecraft. For airborne acquisitions, since most of the atmospheric effects are in the lower atmosphere, the correction to space is small and any errors due to lack of knowledge of the intervening atmosphere should also be quite small (Table 1). For some dates where calibrated aircraft data were not available, surface temperatures were propagated to space using MODTRAN Calibration campaign 1999 The summer of 1999 was the first flight season for RIT s Modular Imaging Spectrometer Instrument (MISI), flown in a Piper Aztec aircraft. The MISI instrument has four thermal channels, as well as a 64-channel visible near-infrared imaging spectrometer. Our emphasis here will be on the thermal channels, two of which are close matches to the Landsat ETM+ Band 6 [one is shifted slightly ( 0.5 mm lower) and the other uses a spare Landsat spectral filter for a very close match]. The thermal sensors have a 2 mrad field of view, a system noise level better than 0.1 K and MISI was flown at altitudes up to 5000 ft yielding a 10 = ft (3 m) spatial resolution. The details of the MISI instrument can be found in Feng, Schott, and Gallagher (1994) and Schott, Gallagher, Nordgren, Sanders, and Barsi (1999). MISI s on-board calibration system is similar to that of Landsat, a two-point calibration system updating every scan line. The essential difference is that MISI s two full aperture Fig. 4. The Landsat 7 optical path. The calibration wand moves into the optical path once every scan line, providing both visible and thermal calibration targets.

6 J.R. Schott et al. / Remote Sensing of Environment 78 (2001) Table 1 Error propagation of the effects of atmospheric uncertainty on sensor-reaching radiance (apparent temperature) using MODTRAN T (K) t t error L u (W/m 2 sr mm) L u error (W/m 2 sr mm) Error in sensor-reaching radiance (app. temp.) MODTRAN (1.5 km to space) MODTRAN (3 km to space) MODTRAN (7 km to space) Targets at two different temperatures are propagated from various altitudes to space and the error in radiance-reaching space is predicted based on uncertainties in estimating the atmospheric profile. blackbodies are ahead of all the optics (Fig. 5). Whereas Landsat required a prelaunch radiometric model to offset the propagation of radiance through the optical system, MISI s calibration includes the effect of the optical elements. The MISI instrument successfully acquired data four times in 1999 under Landsat 7. These data were acquired along the south shore of Lake Ontario and the north shore of Lake Erie. The data collection program involved flying the aircraft at multiple altitudes over the same targets from several hundred feet to about 5000 ft. The timing of the Landsat launch, coupled with a very warm winter and early development of the thermal bar, resulted in only one collection of a weak thermal bar (low thermal contrast) in Lake Ontario. In addition to being one of our prime scientific interests, the thermal bar is also an ideal calibration target since it provides multiple large regions of uniform temperature water (i.e., many Landsat pixels). In the absence of the thermal bar, the underflights focused on flight lines that included large bays, open lake, major river discharges into the lake, and power plant thermal discharges. These provide a gradient of temperatures for calibration, but are less than ideal because a uniform large area is harder to obtain. Extensive ground truth campaigns were conducted concurrent with the underflights. The ground truth included deployment of reflectance panels, field and water spectral reflectance measurements, water sampling and laboratory analysis of coloring agents, and most importantly, for our purposes, surface water temperature measurements. Measurements were made from piers and several small boats deployed in the embayments and coastal waters. Fig. 6 shows sample MISI thermal images and ground truth points. Measurements included Global Positioning System (GPS) location and surface temperature readings from thermistors floated just at the surface on the bottom of small styrofoam floats. Because surface waters in these lakes are well mixed by the nearly constant wind and long fetch, a surface gradient is not expected in these waters and the thermistor measurements of temperature are used as surface temperature values. Regrettably, the MISI laboratory blackbody calibration describing the temperature-to-dc relationship was discovered to be invalid due to a difference in the in-flight and laboratory readout electronics. By the time this was discovered at the end of the season, changes had been made to the readout circuitry so a proper recalibration using the flight circuitry was impossible. To overcome this limitation, an empirical calibration of the MISI was performed using ground truth data from three collections spanning the 1999 collection season. For each collection, ground truth data were converted to in-band sensor reaching spectral radiance estimates using an equation of the form (Eq. (2)): L a ðhþ ¼ R ðtðh; lþ½eðlþltl þð1 eðlþþl dl ŠþL ul ðh; lþþbðlþd l R bðlþdl ð2þ Fig. 5. The MISI line scanner design. The scan mirror rotates in the crosstrack direction while the aircraft moves forward. The calibration sources are located in the backscan of the mirror and are imaged every scan line. A pyramid mirror lies at the focal point of the Cassegrain telescope and currently redirects the incoming energy to two spectrometers, a broadband visible focal plane and the cooled thermal focal plane. where L a (h) is the in-band spectral radiance [w/cm 2 sr mm] reaching the aircraft, t(h,l), L dl, and L ul (h) are MODTRAN-derived spectral estimates of transmission, downwelled spectral radiance and upwelled spectral radiance, respectively, e(l) is the spectral emissivity of water, L Tl is the Planckian spectral radiance from a target at temperature T, and b(l) is the spectral response function

7 114 J.R. Schott et al. / Remote Sensing of Environment 78 (2001) Fig. 6. The Ginna Nuclear Power Plant discharge plume as imaged by MISI from 4000 ft. White dots represent areas where ground truth was collected. This image has not been roll corrected. of the MISI spectral channel. The MODTRAN runs used radiosondes from the nearest airport (usually Buffalo, NY corrected for local surface temperature and water vapor values; Schott 1997). The counts-to-radiance relationship for each flight line could then be solved for using: DCðiÞ ¼mL a ðh; iþþb ð3þ where DC(i) is the count for the ith target, L a (h,i) is the predicted sensor reaching spectral radiance for the ith target and m and b are sensor gain and bias (which, for this study, must be assumed constant over a flight line). Since DC values for each on-board blackbody and the voltage readout from their temperature monitoring thermistors were recorded, the thermistor readout was calibrated using the following procedure. Blackbody DCs were converted to spectral radiance and corresponding blackbody temperature using Eq. (3) and a Planckian lookup table built for the particular spectral channel. Temperatures, which were thus estimated for the blackbody, could be regressed against the voltage recorded for that blackbody to generate a voltage-totemperature calibration. Fig. 7 shows a plot of temperature versus voltage derived in this manner using flight lines where ground truth data were available. The plot indicated that the temperature readout circuit was very stable (as expected) over the flight season and provided the basis for analysis of all subsequent data. The error in blackbody equivalent temperature values used to generate this curve is approximately 1.2 K as opposed to a value of a few tenths Kelvin that could have been expected using a proper laboratory calibration. This empirical calibration procedure for the MISI blackbodies was evaluated by inverting MISI aircraft radiance values to ground temperature values for ground truth sites. This inversion used the multialtitude or profile calibration technique described by Schott (1979). This is a wholly in- Fig. 7. Empirically derived blackbody temperature versus thermistor voltage for MISI calibration in Linearity indicates the system is stable over the collection season.

8 J.R. Schott et al. / Remote Sensing of Environment 78 (2001) Fig. 8. The results of the 1999 calibration validation campaign using empirically calibrated MISI data from three dates, 11 May 1999, 07 June 1999, and 03 September scene technique based on the availability of radiometrically calibrated data. The root mean square error between the ground truth and profile-derived temperatures obtained for 18 samples from 3 days and four flight lines was 1.3 K with most of the error due to lack of knowledge of the onboard calibration. With MISI finally calibrated using ground truth, the aircraft-derived radiances were extrapolated to space using a MODTRAN atmosphere. The preliminary results of the comparison to Landsat from 1999 are shown in Fig. 8. While still exhibiting the linear behavior expected of a thermal instrument, there is definitely an error in the offset and appears to be a significant error in gain. The errors in the MISI data are such that, while a bias of 2 3 K is strongly suspected, the error in gain error could have potentially been due to MISI (though later it will be associated with Landsat). If the error does, in fact, include both a bias and a gain effect, errors at about 300 K are only 2 K but grow to 4 K for colder (280 K) targets, far too high to be acceptable for any quantitative studies with the data. These error estimates are consistent with initial results from Palluconi (1999) for a single target at about 292 K, who estimated that the Landsat observed radiances are high by 2.3 K. The estimate using the regression results in Fig. 8 would suggest a difference of 2.2 K for an apparent at satellite temperature of 292 K. After one year, the conclusion was to agree that there was an offset error, and rely on more data to verify the slope was not an artifact Calibration campaign 2000 The analysis to date for the summer of 2000 season is based primarily on surface temperature measurements. Ground truth surface temperatures were converted to surface-leaving radiances and extrapolated to space using MODTRAN and Equation 5. Essentially, this is the same procedure as used with the underflights, except for the introduction of a greater error in predicting the effects of the lower atmosphere. Fig. 9 illustrates the results from three collection dates in While the offset error still appears, the slope of the line is essentially unity, meaning the gain error is gone. This is in much closer agreement with the Palluconi data from 1999 and upon inquiries, with the Palluconi data from 2000 (Palluconi, 2000). The offset is still on the order of 2 K, but the error is no longer temperature dependent Compiled calibration results Compiling both year s data still generates an error in gain. Since this gain error did not appear in the Palluconi Fig. 9. The results of the 2000 calibration validation campaign using ground measurements extrapolated to space from three dates, 05 July 2000, 26 July 2000, and 07 September 2000.

9 116 J.R. Schott et al. / Remote Sensing of Environment 78 (2001) Fig. 10. The combined results from the two years of validation of Landsat 7. Because 11 May 1999 and 07 June 1999 were during the initial check-out period, they were removed from the analysis. This validation reveals that the Landsat calibration is predicting image radiances that are too high by W/m 2 sr mm. data or the 2000 ground truth data, the possible source of error focused on the use of MISI. More recently, however, the Landsat Project Science Office has found that because the Landsat instrument was functioning in a very different operating mode during the first three months after launch, the initial check-out data may be significantly different from data collected in normal operations. Because the imager was only on for limited amounts of time during that period, the temperatures of the instrument components were much cooler than those once the instrument was in normal operation mode. Since the only two scenes from this check-out period are in the 1999 data set (none of the Palluconi data were), the check-out period calibration may be different than normal operation calibration. Given that the validation needs to apply to the normal operational period, the 5/11/99 and 6/7/99 data were removed from the analysis, in order to compile a final offset value. Fig. 10 shows the compiled 1999 and 2000 validation data. The final offset value determined by this study indicates that Landsat image radiances are too high by W/m 2 sr mm, about 2 K. The difference between this and the Palluconi study is W/m 2 sr mm, about 0.4 K at 300 K. Compiling all the data, the final offset value is W/m 2 sr mm. The Landsat Project Science Office has accepted these results and begun the process of making the changes in the appropriate software and calibration files. For scenes purchased from the Eros Data Gateway (EDG) before December 2000, the user will have to manually subtract this offset error from the radiance image product. Scenes purchased from EDG after December 2000 will include this correction, so it will include the latest in calibration efforts. 3. Conclusions and recommendations Most uses of Landsat data require calibrated data, including the validation and calibration of the ALGE hydrodynamic model presented here. The thermal band itself has proven to be extremely stable though both calibration validation investigations have shown an error in the offset. The W/m 2 sr mm bias observed here agrees with an independent validation effort to within W/m 2 sr mm or 0.4 K. The average of the two studies, W/m 2 sr mm, is the correction that the Landsat Project Science Office will implement in its processing system. The newly calibrated images will be available through the EDG mid-december 2000, but images already processed and purchased can easily be corrected for the calibration error by subtracting the offset from the image radiance. A model has been created to map lake-wide thermallydriven hydrodynamic processes in Lake Ontario and the potential exists for using Landsat as both input and validation to the model. Due to two warm winters, the thermal bar was weaker than normal so a detailed mapping of the early formation of a thermal bar using Landsat 7 has not been observed yet. Long-term quantitative thermal data will be required to refine and calibrate the models that will allow us to understand and forecast hydrodynamic processes and material transport within large lakes. In order to continue refining the hydrodynamic model and for use in other studies, Landsat must continue to be a resource for quantitative studies. While the instrument appears to have remained stable since launch, there is as yet no well-accepted source for the bias error documented here. Thus, continued efforts to monitor and update the thermal calibration of the ETM+ sensor are recommended. References Aubert, E. J., & Richards, T. L. (1981). IFYGL The International Field Year for the Great Lakes (p. 410). Ann Arbor, MI: National Oceanic and Atmospheric Administration. Barker, J. L. (Ed.) (1985). Proceedings of Landsat-4 Science Characterization Early Results Symposium, 2/83, NASA Goddard, NASA Conference Publication 2355, NASA/GSFC. Berk, A., Bernstein, L. Sl., & Robertson, D. C. (1989). MODTRAN: a moderate resolution model for LOWTRAN 7. GL-TR , Spectral Sciences. Bukata, R. P., Jerome, J. H., Kondratyev, K. Y., & Pozdnyakov, D. V. (1991). Estimation of organic and inorganic matter in inland waters: optical cross sections of Lakes Ontario and Ladoga. Journal of Great Lakes Research, 17, Feng, X., Schott, J. R., & Gallagher, T. W. (1994). Modeling the performance of a high speed scan mirror for an airborne line scanner. Optical Engineering, 33 (4), Garrett, A. J., & Hayes, D. W. (1997). Cooling lake simulations compared to high resolution thermal imagery and dye tracer data. Journal of Hydraulic Enginering, 123, Garrett, A. J., Irvine, J., Evers, M., Smyre, T. K., King, J., Ford, A. D., & Levine, C. (1998). An imagery-based hydrodynamic simulation of an effluent stream entering the Clinch River. Photogrammetric Engineering and Remote Sensing. Hudson, S. J., Moore, G. F., Bale, A. J., Dyer, K. R., & Aiken, J. (1994). An operational approach to determining suspended sediment distributions in the Humber estuary by airborne multi-spectral imagery. Proceedings of the First International Airborne Remote Sensing Conference, 3,

10 J.R. Schott et al. / Remote Sensing of Environment 78 (2001) Markham, B. (2000). Personal communication. NASA/GSFC, Goddard, MD. Masuda, K., Takishima, T., & Takayama, Y. (1988). Emissivity of pure and sea waters for the model of sea surface in the infrared window regions. Remote Sensing of Environment, 24, Melchor, C. V. (1941). The refractive index of liquid water in the near infrared spectrum. Journal of Optical Society of America, 31, Menon, A. S., Dutka, B. J., & Jurkovic, A. A. (1971). Preliminary bacteriological investigation of the Lake Ontario Thermal Bar. Proceedings 14th Conference on Great Lakes Research, Murphy, R., Moore, G., Wilson, A., Youngs, K., & Morris, K. (1996). Remote sensing of coastal estuaries. In: SeaWiFS Technical Report, Vol. 33, NASA Tech Memo , Vol. 33. Palluconi, F. (1999). Personal communication. NASA/JPL, Pasadena, CA. Palluconi, F. (2000). Personal communication. NASA/JPL, Pasadena, CA. Piech, K. R., Schott, J. R., & Stewart, K. M. (1978). The blue-to-green reflectance ratio and lake water quality. Photogrammetric Engineering and Remote Sensing, 44, Richardson, L., & Ambrosia, V. G. (1996). Algal accessory pigment detection using AVIRIS image-derived spectral radiance data. In: Proceedings of the 6th Annual Airborne Earth Science Workshop. Jet Propulsion Laboratory, NASA. Rogers, G. K. (1968). Heat advection within Lake Ontario in spring and surface water transparency associated with the thermal bar. Proceedings 11th Conference on Great Lakes Research, Rogers, G. K. (1971). Field investigation of the thermal bar in Lake Ontario: precision temperature measurements. Proceedings 14th Conference on Great Lakes Research, Schott, J. R. (1979). Temperature measurement of cooling water discharged from power plants. Photogrammetric Engineering and Remote Sensing, 45 (6), Schott, J. R. (1997). Remote sensing: the image chain approach. Oxford University Press. Schott, J. R., Gallagher, T. W., & Barsi, J. (1997). Calibration procedures for evaluation of in-flight radiometry performance of thermal infrared satellite sensors. Presented at EOS and SPIE Europto Conference, Aerospace Remote Sensing, London. Schott, J. R., Gallagher, T. W., Nordgren, B., Sanders, L., & Barsi, J. (1999). Radiometric calibration procedures and performance for the Modular Imaging Spectrometer Instrument (MISI). Proceedings of the Earth International Airborne Remote Sensing Conference, ERIM. Singh, S. M. (1994). Effect of surface wind speed and sensor view zenith angle dependence on emissivity for SST retrieval from thermal infrared data: ATSR. International Journal on Remote Sensing, 15, Stoermer, E. F. (1968). Nearshore phytoplankton populations in the Grand Haven, Michigan vicinity during thermal bar conditions. Proceedings 11th Conference on Great Lakes Research, Strong, A. E. (1978). Chemical whitings and chlorophyll distributions in the Great Lakes as viewed by Landsat. Remote Sensing of Environment, 7,