«Remote estimation of chlorophyll concentration in productive waters: Principals, algorithm development and validation A.A. Gitelson, Y.Z. Yacobi, D. ...»
Gitelson et al., 1994b; Yacobi et al., 1995). At this point, chlorophyll absorbance is offset by scattering of the cell walls and is a point of minimum sensitivity of reflectance to algal density and Chl concentration.
In waters with low Chl concentrations, a peak of solar-induced Chl fluorescence was observed in reflectance spectra near 685 nm (Figs. 1a and b). With increase in Chl concentration, emitted fluorescence increased as well as the magnitude of the peak. The peak was used as an indicator of Chl concentration (e.g., Neville and Gower, 1977; Gower, 1980; Doerffer, 1981; Fischer & Kronfeld, 1990; GKSS, 1986).
With increase in Chl concentration above 15-20 mg/m3, re-absorption of fluorescence signal occurs and magnitude of fluorescence peak decreases (Kishino et al. 1986; Gitelson, 1992; 1993), and the nature of the peak is very different from that for low Chl concentration (Gitelson et al., 1986; Vos et al., 1986; Gitelson 1992; 1993;
Gitelson et al., 1993a,b). This peak is an outcome of an interaction between strong absorption by chlorophyll and water and scattering by algal cells and other sestonic matter (Gitelson et al., 1986; Vos et al., 1986; Gitelson, 1992; 1993).
Fig. 3: Demonstration of accessory pigment effects on reflectance spectra. Two representative spectra from Carter Lake for June 4 and July 29, 1996. Chl concentrations were almost the same, but on July 29, phycocyanin (PC) concentration was much higher. As a result of higher PC concentration, the depth of a trough near 625 nm was at least twice more than that on June 4. The spectrum taken in July 29 shows a shifted position of the “green” peak towards shorter wavelengths.
The magnitude of the peak, as well as its position, depends strongly on Chl concentration (Gitelson 1992;
1993; Gitelson et al. 1994a; Yacobi et al. 1995, Schalles et al. 1998). The peak magnitude depends on scattering by all suspended matter and, thus, increases with increase in phytoplankton biomass. It correlated with Chl concentration via the link between Chl and algal biomass. As the algal biomass increases, or actually the active cell surface, scattering and the reflectance increase. In this range of the spectrum, combined absorption by Chl and water is minimal; therefore, the scattering surplus to the basic scattering by non-organic suspended matter may be attributed to phytoplankton cells surface (Gitelson et al. 1994a; Yacobi et al. 1995).
Chl has low but significant absorption between 690 and 715 nm. With increasing Chl concentration, the pigment absorption offsets cell scattering at progressively higher wavelengths, and the position of the peak shifts toward longer wavelengths (Figs. 2a,b, see also Gitelson et al., 1986; 1993a,b; Vos et al., 1986; Gitelson 1992; Rundquist et al., 1995; Yacobi et al., 1995, Schalles et al. 1998).
It is worth to note, that the distinctive spectral features of reflectance (Fig. 2a) characterized Peridinium bloom in Lake Kinneret. These reflectance spectra were conspicuously different from that observed in other seasons (reflectance spectra of the lake in non-Peridinium period resemble spectra obtained in Okoboji lakes, Fig. 1a, and in Haifa Bay, Fig. 1a, with Chl 10 mg/m3; see also Gitelson et al. 1994b). Thus, even for single lake, different approaches have to be used to estimate remotely Chl concentrations in different seasons.
Algorithm development and validation Low Chl concentration. In productive, turbid waters with Chl concentration below 20 mg/m3, the peak of Chl fluorescence near 685 nm was found to be the best spectral reflectance feature for estimating of Chl concentration. The height of the peak above a base line between 650 and 730 nm was used successfully for remote detection of Chl in Case II waters (e.g., Neville and Gower, 1977; Gower, 1980; Doerffer, 1981; Fischer & Kronfeld, 1990; GKSS, 1986). We slightly modified this algorithm placing the base line between 670 and 730 nm. It increased in certain degree the sensitivity of Chl detection (Gitelson et al. 1994b, Mayo et al., 1995).
The quantitative accuracy of the technique is limited by the varying fluorescence efficiency of different phytoplankton populations and by changes in water constituent absorption and scattering that reduces the light available for excitation of fluorescence. This technique was found to be useful for Chl detection, however, it is difficult to generalize and make comparisons based on previous studies, especially for inland waters with highly variable biooptical properties. Nevertheless, the technique proved expedient in Lake Kinneret during non-Peridinium period when Chl concentration were below 10 mg/m3 (Gitelson et al. 1994b) and in Iowa lakes (Fig. 7a). In lake Kinneret, an estimation error of Chl, ranged from 3.8 to 16, was less than 1 mg/m3; in Iowa lakes in the range of Chl concentration from 2 to 55 mg/m3, an estimation error of 7.8 mg/m3 was achieved.
Fig. 4: Algorithms for estimation of Chl concentration. (a) Ratio of the peak magnitude to reflectance at 670 nm; (b) Reflectance height above the base line between 670 and 750 nm; (c) Area above the base line between 670 and 750 nm.
Moderate to high Chl concentration. The semi-analytical algorithms developed and tested in our work are all based on the use of reflectance in the red and NIR range of the spectrum, since other portions of the spectrum are irrelevant in productive, coastal and freshwater ecosystems (Gitelson et al. 1986; Vos et al. 1986; Gitelson, 1992; Mittenzwey and Gitelson, 1988; Mittenzwey et al. 1992; Millie et al. 1992; Quibell, 1992; Dekker 1993; Gitelson et al. 1993a,b;
1994a; Goodin et al. 1993; Boxall & Mathews, 1994; Han et al. 1994; Richardson et al. 1995; Rundquist et al. 1995, 1996; Yacobi et al. 1995; Schalles et al. 1997, 1998). The basic concept of those algorithms is the inclusion of the spectrum range which shows the maximal sensitivity to changes in Chl concentration and the range with the minimum sensitivity to variation of Chl concentration (Gitelson et al. 1986; 1993a,b; 1994a; Yacobi et al. 1995). The latter accounts for non-pigmented suspended matter that causes variation in the reflectance. Magnitude and position of the reflectance peak near 700 nm were found to be the most sensitive variable for algorithms, and the reflectance at 670 nm was the least sensitive to changes in algal density, especially for Chl 15-20 mg/m3.
A simple reflectance ratio R700/R670 was first used as a predictor of Chl concentration (Gitelson et al. 1986;
Mittenzwey & Gitelson, 1988; Mittenzwey et al. 1992; Gitelson et al. 1993a,b). The R700/R670 ratio (Fig. 4a) was applied to our data obtained in highly diverse aquatic ecosystems dominated by different algal assemblages: Anabaena sp. (Fig. 5a), Micricystis aeruginosa (Fig. 5b), and Peridinium gatunense (Fig. 6). The relationship between the R700/R670 ratio and Chl concentration was linear up to concentrations of approximately 180-200 mg/m3, but turned exponential at higher concentrations (fig. 5b). To estimate accuracy of Chl prediction in lake Kinneret during Peridinium bloom, the combined dataset (reflectance spectra and Chl concentrations) was separated into modeldevelopment and model-testing subsets. For the model-development subset, data collected in March 1993 were used (Fig. 6). Validation was done by data, collected in April 1993. Predicted Chl concentrations were calculated using reflectance from model-testing subset with regression coefficients for the model-development dataset equation (solid line in Fig. 6). The accuracy of Chl prediction was made against Chl concentrations actually measured. Root-mean square deviation of predicted Chl values from actually measured did not exceed 18 mg/m3 (insert in Fig. 6).
Fig. 5: Validation of the reflectance ratio algorithm, R700/R670, for estimation of Chl concentration.
(a) Carter Lake, March 1995-March 1996. The solid line represents best-fit function; the dotted lines represent standard error of chlorophyll estimation; (b) Fishponds in the Jordan Valley, Israel, February 1998. For Chl 180 mg/m3, the relationship ratio vs. Chl was linear, and then leveled off.
Several algorithms were developed for estimation of Chl concentration. They were based on the following variables: (a) the magnitude of the peak near 700 nm; (b) the height of the peak near 700 nm above the baseline drawn from 670 to 750 nm (Fig. 4b); (c) the area delimited by the reflectance curve and the mentioned baseline (Fig. 4c); (d) the position of the peak in NIR range. These algorithms were based on following concept. The trough at 670 nm is the wavelength of maximum absorbance by Chl in the red range of the spectrum. At this wavelength, Chl absorbance and scattering by cell walls are almost in equilibrium, and R670 shows minimum sensitivity to Chl concentration. For a Chl 20 mg/m3, R670 primarily depends on concentration of non-organic suspended matter (Gitelson et al. 1993a; 1994b;
Dekker, 1993; Yacobi et al. 1995). Reflectance beyond 750 nm depends on both organic and non-organic suspended matter concentrations and is insensitive to algal pigments (Han et al. 1994); the variation of R750 is comparatively small because of strong water absorption in the NIR range. Thus, the slope of the baseline between 670 and 750 nm depends primarily on scattering by water constituents, but phytoplankton: with variation of non-organic and non-pigmented organic suspended matter concentration, the slope of the baseline changes but it has minimal influence on the height and area of the 700 nm peak above the base line. Therefore, the height of the peak and the area above the baseline between 670 and 750 nm depends mainly on phytoplankton density and was used as its quantitative measure.
The regression of reflectance height at 700 nm above the baseline from 670 to 750 nm (Fig. 4b), as well as area above this baseline (Fig. 4c), and the position of the NIR peak (Gitelson 1992; 1994), against Chl a concentrations yielded high correlation coefficients, always r2 0.90 (Gitelson et al. 1994a; Yacobi et al. 1995). The developed algorithms were validated by recurrent experiments in Lake Kinneret, as well as in other environments, including Iowa lakes (Fig. 7a) dominated by diatom algae and Haifa Bay with dinoflagellates and diatoms (Fig. 7b), both with small to moderate Chl concentrations (Table 1). The algorithms were also used to estimate Chl concentrations in wastewater ponds, where chlorophytes algae dominated with extremely high (70-520 mg/m3) Chl concentration (Figs. 7c). In all aquatic systems studied, the algorithms proved expedient as tool for Chl concentration estimation.
Fig. 6: Validation of the reflectance ratio algorithm, R700/R670, for estimation of Chl concentration in lake Kinneret. Chl concentrations were predicted by ratio R700/R670, measured in April 1993. Relationship R700/R670 vs. Chl (solid line), obtained in March 1993, was used to calculate Chl predicted. Then, Chl predicted was compared with actually measured Chl concentrations. Error of Chl prediction, determined as root-mean square deviation of predicted Chl values from actually measured, did not exceed 18 mg/m3. The slope of the relationship R700/R670 vs. Chl changed in various experiments (see Figs 5 and 6). In the cases presented herein, slope of the relationship for Lake Kinneret was at least 50% higher than that of the fishponds.
Basically, in all instances the same algorithms were applied, however coefficients of the relationships between Chl concentration and remotely measured variables should be adjusted for each case separately (Table 2). The suitability of those algorithms for completely different water bodies studied, from oligotrophic and mesotrophic Iowa lakes to hypereutrophic wastewater ponds, underlines the physical rationale behind the choice of the spectral reflectance features used.
For all inland waters studied, excluding wastewater ponds, the coefficients b in the relationships Chl vs.
reflectance line height were in agreement (they ranged from 31.8 to 47.2 mg/m3/% - Table 2). Minimal b value (31.8 mg/m3/%) was found for Carter Lake, Nebraska in a full annual cycle. Taking into account extremely diverse trophic status of water bodies, the existence of robust relationship between Chl concentration and suggested variables of reflectance must to be taken as proved. Worth to note, that the suitability of the algorithms to estimate the density of phytoplankton in water bodies dominated by blue-greens is of great importance, considering the significance of that algal group as potential environmental hazard.
Fig. 7: Validation of the algorithm, height of the peak near 700 nm above a baseline drawn from 670 nm to 750 nm, for estimation of Chl concentration. (a) Lakes in northwestern Iowa, USA, September 1996; (b) Haifa Bay, Southeastern Mediterranean Sea, June 1995; (c) Waste water ponds, Israel, May 1998. The difference in Chl concentrations between the extremes (Fig 7a. and 7c) was at least ten-fold. Nevertheless, in all the systems examined in our studies, we found a linear relationship between the height above the base line and Chl concentration. Note, however, that the coefficients of the relationships changed between experiments (Table 2).
Table 2: Comparison of regression statistics for several studies in which total chlorophyll was regressed against the
predictor - reference peak height near 700 nm above baseline from 670 nm to 750 nm (RLH670-750):
Chl = a + b*RLH670-750; r2 is determination coefficient.