Multiplatform optical monitoring of eutrophication in temporally and spatially variable lakes
Introduction
Lakes are valuable watersystems. They are, for instance, intensively used for production of drinking water, for fisheries and recreation. The ecological value of many lakes, however, has deteriorated, mainly as a consequence of eutrophication (Scheffer, 1998). The occurrence of blooms of (potentially) toxic cyanobacteria has increased globally (Chorus et al., 2000). Restoration programmes aiming at reducing the impact of eutrophication have been successful in some, but not in all cases. All of the above necessitates proper monitoring of lake water quality. Monitoring programmes have been initiated in many countries. The aim of the main national water-monitoring programme in the Netherlands (MWTL) is to: (i) establish long-term trends in the ecological quality of water systems; and (ii) to evaluate national water management policies through periodical assessment of the (ecological) quality of water systems (Ibelings et al., 1998). MWTL is based upon in situ sampling of the water systems. Algae and cyanobacteria are the principal biological receptors of eutrophication, such that changes in structure and activities of the phytoplankton determine secondary effects of eutrophication observed at higher levels. Proper monitoring of algae (for which chlorophyll is widely used as an overall estimate, despite variable cellular levels, e.g. Cracknell et al., 2001) is essential, and remote sensing of chlorophyll patterns will be emphasised in this paper.
In situ samples of spatially heterogeneous parameters like total suspended matter, chlorophyll-a, vertical diffuse attenuation of irradiance over PAR and Secchi depth are often not representative for the spatial mean of these parameters (Dekker et al., 2001, Pulliainen et al., 2001). Hence, the spatial patterns cannot be represented using traditional in situ sampling techniques. Uncertainties in lakewide estimates of the mean chlorophyll concentration results in inaccuracies in nutrient loading—chlorophyll prediction models that are extensively used in lake restoration schemes, (e.g. Galat and Verdin, 1989). For large lakes these problems are even more significant than for small lakes, since spatial heterogeneity is often more prominent in large lakes, with the lake regularly even being subdivided in several basins. Remote sensing; either by sensors on a satellite or an aeroplane; gives a (more) synoptic view of these spatially heterogeneous parameters. However, lakes are optically complex waters (Kallio et al., 2001), and extracting useful information from the images is not always straightforward. Our study focuses on the two largest lakes in the Netherlands, Lakes IJssel and Marken (1136 and 702 km2, respectively). Spatial variability in chlorophyll in these lakes is mainly attributed to a patchy distribution of the zebra mussel, Dreissena polymorpha, an efficient filter feeder.
Fig. 1 in Schofield et al. (1999) shows the ecologically important temporal and spatial scales that affect phytoplankton blooms. The authors put forward that extracting information at all of the relevant scales requires the use of a multiplatform sampling network. In this paper we study the added value—to the standard in situ monitoring efforts—of such a multiplatform approach in the monitoring of chlorophyll and total suspended matter (TSM) in Lakes IJssel and Marken. The synoptic maps in our study are based upon SeaWiFS images, a sensor on board of the SeaStar satellite. In addition, we use hyperspectral optical data from the airborne EPS-a sensor and a shipboard PR-650 spectroradiometer (Gons, 1999, Gons et al., 1998, Rijkeboer, 1999). Hyperspectral field radiospectrometers have proven useful to validate remote sensing algorithms for retrieval of total suspended matter and chlorophyll, and the PR-650 data will be used in such a way in this study (Hakvoort et al., 2002). Satellite and airborne sensors can be complementary to each other. Aeroplanes with hyper spectral sensors are valuable for the accurate observation of small-scale patchiness in chlorophyll in lakes (Dekker, 1993, Hoogenboom et al., 1998, Gege, 1998). However, the aeroplane cannot cover large lakes with a few flight tracks only, and the costs and personnel effort involved strongly limit the frequency of flights. Therefore, operational multi-spectral sensors on board of satellites that daily revisit Lakes IJssel and Marken are important tools for monitoring water quality. The disadvantage of SPOT and Landsat TM satellite sensors is that they miss frequency bands for chlorophyll-a detection in case II waters (see Dekker et al., 2001). The SeaWiFS sensor is also considered not to be optimal for chlorophyll-a retrieval from case II waters (Ruddick et al., 2000, Hu et al., 2000, Vos and Rijkeboer, 2000). Vos and Rijkeboer (2000) developed an algorithm that is suited for case II waters with high chlorophyll content, and this algorithm is used in this paper. Despite the frequent cloud cover in the Netherlands the frequency of good images is approximately 10 per year, which is sufficient to show the main seasonal variations. Further improvement may be achieved by combining remote sensing with computational models that may meet the demand for a sufficiently high sampling frequency by deterministic interpolation of the data in time. In our work on Lakes IJssel and Marken we compared remote sensing maps with the output of an algal growth model, DBS (van der Molen et al., 1994), and calculated a Goodness of Fit for each SeaWiFS pixel. Although the technical aspects of the model integration will be discussed in a separate paper, we will discuss the main findings of our study.
The comparison of in situ data, airborne data and satellite data—where applicable combined with model output—revealed problems related to the comparison of data with different spatial and temporal scales. Extrapolating individual point measurements to spatially representative values for lakes proves to be problematic. Comparison of data with different scales can lead to large differences that are not a consequence of erroneous measurements, but are a primary property of the lake system. This aspect of different scales is fundamental in problems with validation of remote sensing images. Several papers (e.g. Cracknell et al., 2001, Kallio et al., 2001) emphasise the need for concurrent in situ sampling. Sufficient in situ data, however, rarely are available. Even if they would be the question may be asked: ‘what is the value of the so-called ground truth?’ How to validate a SeaWiFS image of Lake IJssel, consisting of many pixels of 1.1×1.1 km on basis of a single or limited number of point measurements in a patchy environment? Our paper explores some ways for validation of remote sensing when ground truth data are not or insufficiently available.
Section snippets
Retrieval of water quality parameters for hyperspectral sensors
The remote sensing parameter used in this paper to derive water quality parameters is the subsurface irradiance reflectance, R(0−). Remote sensing sensors measure the water leaving radiance above water. Corrections for the effects from an air–water interface and atmospheric radiance are required to derive R(0−). Once the observed R(0−) is obtained the next step is to relate it to the optically active water constituents. This is done with an analytical optical model by fitting the observed
Study sites
Lake IJssel and Lake Marken are shown in Fig. 1 together with the EPS-a flight tracks. These lakes are of prime socio-economic importance to the country through their role in the production of drinking water, recreation, shipping and fisheries. Moreover, both lakes are listed as protected wetlands (Ramsar sites) by the IUCN (International Union for Conservation of Nature and Natural Resources). Lake IJssel is a large fresh water lake of 1136 km2 in the centre of The Netherlands with an average
Validation of the retrieval algorithms by field spectrometry
The shipborne R(0−) spectra collected with a hyperspectral PR-650 spectroradiometer, were used to mimic satellite and airborne sensors in order to test the accuracy of the SeaWiFS and EPS-a algorithms for retrieval of water quality parameters. The data set (including in situ samples) was taken from Rijkeboer (1999). Retrieval algorithms were tested for various selected frequencies, namely:
- a
A WLSQ algorithm using all frequency bands of the PR-650 sensor (at 4-nm intervals) from 400 to 750 nm;
- b
A
Discussion
The main national monitoring program for aquatic ecosystems in the Netherlands (MWTL) is based upon a wide range of variables, ranging from concentrations of heavy metals to abundance of macrophytes or birds. In our study, we focus solely on the monitoring of phytoplankton. Phytoplankton abundance has increased in many Dutch lakes as a result of eutrophication; nuisance cyanobacteria have become the dominant group in many lakes in (late) summer. It follows that the key incentives for monitoring
Acknowledgements
We thank BCRS, RWS-WONS and Meetstrategie 2000+ for financial support of the study. We acknowledge the help of Harry Hosper (RIZA – Lelystad) in completing the final report of the study, which formed the basis for this publication. We thank Directorate IJsselmeergebied of Rijkswaterstaat (Lelystad) for use of their in situ data for Lake IJssel and Lake Marken. Pascal Boderie of Delft Hydraulics set up the algal growth model that was used in a Goodness of Fit test with SeaWiFS. We also thank
References (49)
- et al.
Comparison of remote sensing data, model results and in situ data for total suspended matter (TSM) in the southern Frisian lakes
Sci Total Environ
(2001) - et al.
Detection of water-quality using simulated satellite data and semi-empirical algorithms in Finland
Sci Total Environ
(2001) - et al.
Coupling remote sensing with computational fluid dynamics modelling to estimate lake chlorophyll-a concentration
Remote Sens Environ
(2002) - et al.
Atmospheric correction of SeaWiFS imagery over turbid coastal waters: a practical method
Remote Sensing Environ
(2000) - et al.
Estimation of phytoplankton production from space: current status and future potential of satellite remote sensing
J Exp Mar Biol Ecol
(2000) - et al.
Retrieval of water quality from airborne imaging spectrometry of various lake types in different seasons
Sci Total Environ
(2001) Comparison of two inversion techniques of a semi-analytical model for the determination of lake water constituents using imaging spectrometry data
Sci Total Environ
(2001)Perspectives on combining ecological process models and remotely sensed data
Ecol Model
(2000)- et al.
A semi-operative approach to lake water quality retrieval from remote sensing data
Sci Total Environ
(2001) - et al.
Photosynthetic rates derived from satellite-based chlorophyll concentration
Limnol Ocenaogr
(1997)