CHOOSING OF OPTIMAL REFERENCE SAMPLES FOR BOREAL LAKE CHLOROPHYLL A CONCENTRATION MODELING USING AERIAL HYPERSPECTRAL DATA
- 1Faculty of Information Technology, University of Jyväskylä, P.O. Box 35, 40014 Jyväskylä, Finland
- 2Luode Consulting Oy, Sinimäentie 10 B, 02630, Finland
- 3Finnish Geospatial Research Institute FGI, Geodeetinrinne 2, 02430 Masala, Finland
Keywords: Aerial remote sensing, lake water color, water quality monitoring, hyperspectral imaging, chlorophyll a
Abstract. Optical remote sensing has potential to overcome the limitations of point estimations of lake water quality by providing spatial and temporal information. In open ocean waters the optical properties are dominated by phytoplankton density, while the relationship between color and the constituents is more complicated in inland waters varying regionally and seasonally. Concerning the difficulties relating to comprehensive modeling of complex inland and coastal waters, the alternative approach is considered in this paper: the raw digital numbers (DN) recorded using aerial remote hyperspectral sensing are used without corrections and derived by means of regression modeling to predict Chlorophyll a (Chl-a) concentrations using in situ reference measurements. The target of this study is to estimate which number of local reference measurements is adequate for producing reliable statistical model to predict Chl-a concentration in complex lake water ecosystem. Based on the data collected from boreal lake Lohjanjärvi, the effect of standard deviation of Chl-a concentration of reference samples and their local clustering on predictability of model increases when number of reference samples or bands used as model variables decreases. However, the 2 or 3 band models are beneficial and more cost efficient when compared to 5 or 7 band models when the standard deviation of Chl-a concentration of reference samples is over certain level. The simple empirical approach combining remote sensing and traditional sampling may be feasible for regional and seasonal retrieval of Chl-a concentration distributions in complex ecosystems, where the comprehensive models are difficult or even impossible to derive.