Kari Kallio

Lake water quality classification with airborne hyperspectral spectrometer and simulated MERIS data

We study the use of airborne and simulated satellite remote sensing data for classification of three water quality variables: Secchi depth, turbidity, and chlorophyll a. An extensive airborne spectrometer and ground truth data set obtained in four lake water quality measurement campaigns in southern Finland during 1996–1998 was used in the analysis. The class limits for the water quality variables were obtained from two operational classification standards. When remote sensing data is used, a combination of them proved to be the most suitable. The feasibility of the system for operational use was tested by training and testing the retrieval algorithms with separate data sets. In this case, the classification accuracy is 90% for three Secchi depth classes, 79% for five turbidity classes, and 78% for five chlorophyll a classes. When Airborne Imaging Spectrometer for Applications (AISA) data was spectrally averaged corresponding to Envisat Medium Resolution Imaging Spectrometer (MERIS) channels, the classification accuracy was about the same as in the case of the original AISA channels.

Retrieval of water quality from airborne imaging spectrometry of various lake types in different seasons

The suitability of the AISA airborne imaging spectrometer for monitoring lake water quality was tested in four surveys carried out in southern Finland in 1996-1998. Altogether, 11 lakes were surveyed and the total number of stations with concurrent remote sensing and limnological measurements was 127. The ranges of the water quality variables were: the sum of chlorophyll a and phaeophytin a 1-100 microg l(-1), turbidity 0.4-26 FNU, total suspended solids 0.7-32 mg l(-1), absorption coefficient of aquatic humus at 400 nm 1.2-14 m(-1) and secchi disc transparency 0.4-7 m. For the retrieval analyses, 24 AISA channels in the 450-786 nm range with a channel width of 6-14 nm were used. The agreement between estimated and observed water quality variables was generally good and R2 for the best algorithms was in the range of 0.72-0.90 over the whole dataset. The channels used for May were, in most cases, the same as those for August, but the empirical parameters of the algorithms were different. After seasonal grouping, R2 varied from 0.84 to 0.95. The use of apparent reflectance instead of radiance improved the estimation of water quality in the case of total suspended solids and turbidity. In the most humic lake, the empirical algorithms tested were suitable only for the interpretation of total suspended solids and turbidity.

Detection of water quality using simulated satellite data and semi-empirical algorithms in Finland.

The aim of the study was to test the feasibility of the band combination of the TERRA MODIS and ENVISAT MERIS instruments for operational monitoring of lakes and coastal waters in Finland. Also simulated LANDSAT TM data were tested. Satellite bands were simulated using airborne measurements with AISA imaging spectrometer. Semi-empirical algorithms with simulated satellite data were tested against field observations using regression analysis. Interpretation of chlorophyll a, suspended matter, turbidity and secchi-disk depth was included in the analyses. The data for this study were gathered in campaigns carried out in May and August 1997 and August 1998 both for lakes in southern Finland and coastal waters of the Baltic Sea. The data set included 85 in situ observations for lakes and 107 for coastal waters. Our results show that the band combination to be included in the ENVISAT MERIS instrument enables the interpretation of water quality, including chlorophyll a concentration using semi-empirical algorithms both for lakes and coastal waters. MERIS band 9 centred at 705 nm is proven to be of vital importance for the detection of chlorophyll a in local surface waters.

A semi-operative approach to lake water quality retrieval from remote sensing data

A semi-operative approach to retrieve chlorophyll-a concentration from airborne/spaceborne spectrometer observations has been developed and tested using the airborne imaging spectrometer (AISA) data from 11 lakes located in southern Finland. The retrieval approach is empirical and requires nearly simultaneous in situ training data on water quality for the determination of regression coefficients. However, the training data does not have to be collected from every lake under investigation. Instead, the results obtained indicate that reliable estimates on the level of chlorophyll-a (chl-a) for an individual lake can be achieved without employing in situ data representing this specific lake. This enables the estimation of water quality from remotely sensed data for numerous lakes with the aid of reference data only for a few selected lakes representing the region under investigation. In addition, it is shown that the remotely sensed spectrum shape characteristics are highly affected by the trophic and humic state of the lake water.

A hyperspectral model for interpretation of passive optical remote sensing data from turbid lakes

A hyperspectral model was developed for the interpretation of remote sensing data collected above inland waters. Specific absorption and scattering coefficients proposed by other authors were not suitable for modelling of the irradiance reflectance in 12 studied lakes. Therefore, special studies were carried out to estimate absorption and scattering coefficients as well as backscattering probability of suspended matter in turbid waters. AC-9 and Li-1800UW results were used for these purposes. The algorithms obtained were used to improve the model, which was then tested in forward and inverse modes.

Feasibility of airborne imaging spectrometry for lake monitoring—a case study of spatial chlorophyll a distribution in two meso-eutrophic lakes

The spatial distribution of the sum of chlorophyll a and phaeophytin a concentrations (chl-a) under light wind (0–2 m s−1) conditions was studied in two lakes with an AISA airborne imaging spectrometer. Chl-a was interpreted from AISA radiance data using an algorithm based on the near-infrared (700–710 nm) to red (660–665 nm) ratio. The results of Lake Lohjanjärvi demonstrate that the use of one monitoring station can result in over- or underestimation by 29–34% of the overall chl-a compared with an AISA-based estimation. In Lake Hiidenvesi, the AISA-based estimation for the mean chl-a with 95% confidence limits was 25.19±2.18 µg l−1. The use of AISA data together with chl-a measured at 15 in situ sampling stations decreased the relative standard error of the mean chl-a estimation from 20.2% to 4.0% compared with the use of 15 discrete samples only. The relative standard error of the mean chl-a using concentrations at the three routine monitoring stations was 15.9 µg l−1 (63.1%). The minimum and maximum chl-a in Lake Hiidenvesi were 2 and 101 µg l−1, 6 and 70 µg l−1 and 11 and 66 µmg l−1, estimated using AISA data, data from 15 in situ stations and data from three routine in situ stations, respectively.

Analysis on the feasibility of multi-source remote sensing observations for chl-a monitoring in Finnish lakes.

Chlorophyll-a (chl-a) concentration of lake water can be measured with airborne (or spaceborne) optical remote sensing instruments. The rmse obtained here with empirical algorithms and 122 measurement points was 8.9 microg/l (all points used for training and testing). Airborne Imaging Spectrometer for Applications (AISA) was used in four lake water quality measurement campaigns (8 measurement days) in southern Finland during 1996-1998 with other airborne instruments and extensive in situ data collection. As empirical algorithms are employed for chl-a retrieval from remote sensing data, temporally varying factors such as surface reflection and atmospheric effects degrade the estimation accuracy. This paper analyzes the quantitative accuracy of empirical chl-a retrieval algorithms available as methods to correct temporal disturbances are either included or excluded. The aim is to evaluate the usability of empirical chl-a retrieval algorithms in cases when no concurrent reference in situ data are available. Four methods to reduce the effects of temporal variations are investigated. The methods are: (1) atmospheric correction; (2) synchronous radiometer data; (3) wind speed data; and (4) bidirectional scattering model based on wind speed and sun angle data. The effects of different correction methods are analyzed by using single-date test data and multi-date training data sets. The results show that the use of a bidirectional scattering model and atmospheric correction reduces the bias component of the measurement error. Radiometer data also appear to improve the accuracy. However, if concurrent in situ reference data are not available, the retrieval algorithms and correction methods should be improved for reducing the bias error.

Water quality classification of lakes using 250-m MODIS data

The traditional method used in the water quality classification of Finnish lakes includes the collection of water samples from lakes and their analysis in laboratory conditions. The classification is based on statistical analysis of water quality parameter values and on expert opinion. It is possible to acquire similar information by using radiance values measured with the Earth Observing System Terra/Aqua Moderate Resolution Imaging Spectroradiometer (MODIS). In this letter, the classification accuracy with MODIS data is about 80%. Only about 0.2% of the 20 391 pixels were misclassified by two or more classes, as a four-class classification system is used.

A filtering approach for estimating lake water quality from remote sensing data

Abstract In this paper we consider the estimation of lake water quality constituent distributions from hyperspectral remote sensing data. We present a computational approach that can be used to assimilate information from mathematical evolution models into data processing. The method is based on a reduced order iterated extended Kalman filter, and a convection–diffusion model is used to describe the movement of the water quality constituents. The performance of the technique is evaluated in a simulation study. The results show that the filter approach with an appropriate evolution model yields estimates that have better spatial and temporal resolutions than those obtained with conventional methods. Furthermore, the use of a feasible evolution model may make it possible to obtain information also on the concentrations in the lower parts of the lake.

Simulation of dissolved phosphorus from cropped and grassed clayey soils in southern Finland

Abstract Agricultural phosphorus (P) loading is a major contributor to eutrophication of surface waters in Finland. Of the various forms of P in runoff from cultivated fields, dissolved reactive phosphorus (DRP) is immediately available for algal growth and can directly accelerate eutrophication. The applicability of an empirical model developed in southeastern USA was evaluated by simulating DRP in surface runoff from two cropped and grassed clayey soils (Vertic Cambisols) in southwestern Finland. The model relates DRP in a runoff event, e.g. to desorbable soil P (PD), runoff volume (V) and the concentration of total suspended solids (TSS) in runoff. The model overestimated the mean concentration of DRP by a factor of 110–1645. In addition, the observed and simulated mean concentrations of DRP in plots with different winter covers, crops and P status did not correlate with each other. The predictions improved when PD was estimated by water extractions instead of Bray extractions, and the model was simplified by excluding the dependence of DRP concentration on V. The correct level of results could, however, only be achieved by calibration. In order to improve the model fit, the dependence of DRP concentration on V, runoff duration and TSS should be assessed under Finnish conditions and the model should be modified accordingly.