Water Temperature Determination of Inland Surface Waters Using Landsat 7 ETM+ TIR Images
by Tom de Boer
This research evaluates the potential of thermal infrared (TIR) remote sensing to determine the horizontal water temperature distribution of inland surface waters. Usually, monitoring temporal and spatial variability of water temperature takes place by measurement networks of in-situ gauges, but these networks are limited by sparse sampling in time and space. For this reason the use of remote sensing in water management has increased. By remote sensing in the TIR spectrum the TIR radiation leaving from the top surface water layer (< 0.1 mm) is measured, and used to determine the radiant water temperature distribution in the horizontal plane. It is necessary to evaluate the accuracy and precision of the remotely sensed water temperatures. Therefore, the goal of this study is to determine how and with what accuracy and precision TIR radiant water temperature measurements (Tremote) can be used as an approximation for the horizontal distribution of the kinetic water temperature, based on comparisons of Tremote with in-situ kinetic water temperature measurements (Tin-situ). The bias (Tremote – Tin-situ) statistics are indicative for the obtained accuracy and precision.
Landsat 7 ETM+ images measured in the TIR spectrum (λ=8-14 μm) are used for determination of Tremote (with Planck’s Law). The effects of emissivity; atmospheric absorption, emission and scattering; and surface effects and thermal stratification are evaluated and, if possible, corrected for. Uncertainty contained by the ETM+ instrument is taken into account by applying a 95% confidence interval over the obtained surface water body. Reduction of thermal pollution by land-originating TIR radiation of radiant water temperatures is well established by such a 95% confidence interval water body.
The correction for atmospheric circumstances took place by means of the Atmospheric Correction Parameter Calculator (see atmcorr.gsfc.nasa.gov/). Results have been verified with the newly derived atmospheric correction algorithm for ETM+ TIR images, developed with use of the MODIS In-Scene Split Window Method. For clear-sky images, on which this study focuses, the uncertainty contained by the atmospheric correction is up to ±0.8 °C inland, which can increase up to ±1.5 °C near the coast. The surface effects and thermal stratification are influenced by many factors and processes which are difficult to address. The combined result of the surface effects and thermal stratification lead to an uncertainty in winter and summer of ±1.6 and ±3.2 °C, respectively.
The best procedure to approximate the horizontal kinetic water temperature distribution of inland surface waters with ETM+ TIR images makes use of a 95% Confidence Interval Water Mask, and an Emissivity and Atmospheric Correction. The accuracy and precision of the horizontal water temperature distribution display an average bias of +1.5 °C with σ = 1.5 °C and SE = 0.1 °C. No physical relation could be derived between Tremote and Tin-situ. The numerous, complex processes that affect the measurement of Tremote make it difficult to derive a (physical) relation or formula that connects Tremote to Tin-situ. The seasonal influence, expressed by a difference between winter and summer, could be captured by means of the statistical analysis. During winter, Tremote over-predicts Tin-situ on average by 0.8 ° with σ = 0.8 °C and SE = 0.2 °C. During summer, Tremote over-predicts Tin-situ on average by 1.8 ° with σ = 1.6 °C and SE = 0.2 °C. The bias statistics and the statistical seasonal relation between Tremote and Tin-situ can be used as an approximation for the horizontal kinetic water temperature distribution. The proposed systematic correction becomes: in winter Tremote - 0.8 and in summer Tremote – 1.8. The bias spread statistics form a first and reasonable quantification for the uncertainty contained by the obtained approximation.
It is recommended to reduce the uncertainty contained by the approximations. Research towards a better Atmospheric Correction, which accounts for the spatial variations in atmospheric circumstances, could mean a major improvement for the remotely sensed water temperature approximations. Other research to improve the approximations is to assess the thermal stratification and the surface effects. Understanding the thermal water processes, and improving our insight in the relationship between kinetic water temperature in the water column, kinetic water temperature at the surface and the TIR radiant surface water temperature measured locally would help to improve the approximations.
Student: Tom de Boer
Contact: tomdeboer89@gmail.com
Graduation Committee: Prof. dr. ir. N.C. van de Giesen, Prof. dr. M. Menenti, Dr. ir. S.C. Steele-Dunne, Dr. ir. E.L.A. Wolters, Drs. P.C.M. Jeurissen