Influence of atmospheric correction and number of sampling points on the accuracy of water clarity assessment using remote sensing application

Summary

The main objectives of the research described in this paper are to assess the influence of atmospheric correction and the number of sampling points on the accuracy of lake water clarity using remote sensing. For this purpose field experiments were carried out at Bung Boraphet – the largest fresh water lake in Central Thailand. Two clarity parameters, secchi disk transparency (SDT), at 80 locations for three events on April 15th 2008 and March 24th and 31st 2009, and suspended sediment concentration (SSC) for the last two events were collected. These field data were collected nearly contemporaneous or contemporaneous to three Landsat 5 TM images, which were acquired on April 13th 2008 and March 24th and 31st 2009, respectively. Digital numbers of the Thematic Mapper (TM) image bands were transformed to at-sensor radiance by a radiometric correction process and then atmospheric correction was made using the Second Simulation of Satellite Signal in the Solar Spectrum (6S) code to acquire the surface reflectance. Linear regression equations between the log-transformed of clarity parameters and surface reflectance of each band and their band ratios at the same pixels were then determined. The band combinations TM1/TM3 and TM1 to estimate ln(SDT), and TM3/TM1 and TM1 to determine ln(SSC) for all satellite images were selected. These equations were later used to estimate the SDT and SSC values distributed across the whole lake. The results showed that lake wide average values of SDT for the three images are approximately 37, 97 and 81 cm, respectively, while the values of SSC for the second and third images are around 11 and 13 mg/l, respectively. The application of atmospheric correction to the TM data has proved to have the effect on the average values of SDT and SSC, and especially on their maximum and minimum values. Finally, it is possible to reduce the ground observation data of SDT and SSC substantially and still obtain reliable empirical relationships for the estimation of lake wide SDT and SSC values from TM data within the lake.

Introduction

Lakes are valuable freshwater resources that can be used for various purposes such as for drinking water, agriculture, fishing, recreation and tourism (Oyama et al., 2009). Unfortunately, water quality in a number of lakes around the world is becoming degraded by a variety of anthropogenic causes.

Monitoring water quality is a vital aspect of lake management to determine its suitability for human uses and to inform on the need to take corrective actions in the event that water is not fit for use. Conventional field methods for lake water sampling are time consuming and expensive, however satellite imagery is another source of information with great potential to be used for assessment of lake water quality. Processing and interpretation of satellite imagery are now becoming relatively inexpensive and easy to perform with today’s powerful desktop computers and sophisticated software.

However, empirical relationships between satellite data and contemporaneous ground observations are always necessary for evaluating the spatial variation of water quality variables within the lake. Among several satellite systems that have been used for water quality assessment, the Landsat system is particularly useful for assessment of inland lakes (Kloiber et al., 2002). Several investigations have developed reliable empirical relationships between Thematic Mapper (TM) data and ground observations of water quality characteristics (Brown et al., 1977, Giardino et al., 2001, Jie et al., 2006, Kloiber et al., 2002, Lathrop and Lillesand, 1986, Olmanson et al., 2008, Ostlund et al., 2001). Regardless of the method they are constructed, the empirical and semi-empirical relationships are generally site specific (Liu et al., 2003). Up to the present efforts to produce standard prediction equations for water quality parameters applicable to images collected on different dates at the same location have not been successful. Use of a consistent equation form to relate ground observations and satellite data is however preferable because it allows for easier comparison of the results from different images (Kloiber et al., 2002). However the relationships will vary from one event to another because the relationships are empirical by nature.

This study was conducted at Bung Boraphet to derive empirical relationships for estimation of secchi disk transparency (SDT) and suspended sediment concentration (SSC) from three Landsat 5 TM images and in situ sampled data. Three TM images covering Bung Boraphet were acquired on April 13th 2008 and March 24th and 31st 2009. SDT ground observation data were collected on April 15th 2008 and March 24th and 31st 2009, while SSC data were collected only on the last two days. Ground observation data for the first event is nearly contemporaneous to the satellite image, while the last two events are contemporaneous to the images. SDT was chosen because it is used frequently to identify trends in lake conditions due to its simplicity and relatively low cost (Heiskary et al., 1994). SSC was selected because it reflects the physical and chemical property of water and relates to the total primary productivity of the lake (Jie et al., 2006).

Pattiaratchi et al. (1994) found that log-transformed SDT data produced strong correlations to TM data and follow a normal distribution. Klemas et al. (1974) found that a log-transformed SSC relationship to the Landsat Multispectral Scanner (MSS) data is better than a relationship in the original SSC domain. Therefore, log-transformed SDT and SSC versus untransformed TM data were therefore used in deriving the regression equations.

Prior to deriving these relationships, geometric, radiometric and atmospheric corrections were applied to the Landsat 5 TM images. The atmospheric correction was performed using 6S (Second Simulation of the Satellite Signal in the Solar Spectrum), which is a physically-based atmospheric correction model. Several researchers have found that 6S is an effective model for correcting satellite images affected by atmospheric factors. For example, Stroeve et al. (1997) and Zhao et al. (2001) found significant improvements in albedo after atmospheric corrections were applied on their AVHRR and Landsat TM data. Zhao et al. (2001) claimed that there was around 6% improvement in surface albedo after the correction. Also Sharma et al. (2009) concluded that corrected reflectance data better separated ground features such as water bodies and crop fields compared to uncorrected data.

This study has two main objectives. Firstly, to test whether atmospheric correction is needed for improving the reliability of the estimated values of SDT and SSC distributed within the lake. The estimated values of SDT and SSC for all pixels of each image, with and without atmospheric correction, were compared for their differences. Secondly, to investigate the minimum number of ground sampling points of SDT and SSC necessary for constructing reliable empirical relationships. To do this, subsets of points were chosen randomly from the overall set of sampling points and empirical relationships between SDT and SSC values versus TM image data were prepared. Then SDT and SSC values calculated using the subsets of sampling points were compared to values determined using the whole set to see whether it is possible to reduce the numbers of ground observation points but still provide reliable relationships.