In order to investigate bio-physical parameters associated with water quality, a model based on calibrated and atmospherically corrected Remotely Sensed data has been implemented. Secchi Disk depth and chlorophyll concentration parameters were estimated in a system of very small lakes at Monticchio (Italy) using Landsat TM data. The model was developed through the use of stepwise multiple regression and gave high coefficients of determination (R2 = 0.82 for Secchi Disk and R2 = 0.72 for chlorophyll). Values for water transparency were strongly correlated with chlorophyll a concentration: a linear relationship between the two parameters showed a high coefficient of determination (R2 = 0.93). The case study shows that the application of this approach on lakes with a small surface area, such as the Monticchio lakes in Southern Italy, is effective and the developed model well describes the water quality parameters.
Assessing the quality of surface water in lakes, rivers and reservoirs is a key issue for environmental monitoring and management. In Europe, lakes are classified according to eco-regions (
Remotely sensed data from satellites have been used for the monitoring of water quality (
Remote sensing water quality assessment, is usually investigated using the optical bands in the region from blue to near infrared. These data are then used to explore the relation between the reflectance of water bodies and biophysical parameters such as: transparency, chlorophyll concentration (phytoplankton), and the organic and mineral suspended sediments.
The first approach in the lake water quality assessment has been the analysis of chromatic coordinates in the visible part of the electromagnetic spectrum. This approach is based on the idea that an increase of radiance or reflectance at long wavelengths is an indicator of a decrease of spectral response, whereas a reduction of the spectral response towards the shorter wavelengths (blue) is related to an increase in chlorophyll or suspended solid sediments. Although this methodology has been applied by several studies (
Several studies investigated the correlation between spectral response of internal waters with water transparency using single Landsat bands in the visible region (
Common statistical techniques have been used to study the correlation between spectral data and limnological parameters such as chlorophyll concentration (
The atmospheric path between the satellite and the water surface affects the remotely sensed reflectance: therefore data need to be adequately corrected before statistical analysis can be carried out (
The aim of this study is to assess the reliability of a model based on calibrated and atmospherically corrected remotely sensed data on two adjacent and small volcanic lakes (Lago Grande and Lago Piccolo) in Basilicata Region (Southern Italy) through the analysis of the relationships between Landsat TM data and chlorophyll concentration and water transparency, gathered in a previous study (
In the present paper the main innovation is represented by the implementation of the proposed Landsat TM based methodology on very small lakes, while so far it has been applied only for very large lakes. In fact, the use of Landsat TM dataset available since the 70s, even if characterised by a coarse resolution, allows a multitemporal analysis of the internal water quality.
The proposed approach leads also to the implementation of a different functional model (
The Monticchio lakes system consists of two volcanic basins located in the Vulture mountain range in Basilicata, south Italy (40° 55’ 48.29” N; 15° 36’ 16.61” E). The catchment area is about 400 ha and ranges between 1262 m and 653 m a.s.l., that represents the inlet of the Lago Grande effluent. The Monticchio lakes system within the whole area of Vulture mountain range belong to the Natura 2000 network of protected areas (European Birds Directive 79/409/CEE and European Habitats Directive 92/43/ CEE).
The two lake basins, with different and relatively considerable depths (
The limnological data on water transparency and chlorophyll
Referring to
Among these sampling points, the closest measurements to the coastline were excluded to avoid comparison with Landsat pixels with an ambiguous spectral response due to contamination by any vegetation around the lakes occurring in the pixel footprint. Moreover, given the close proximities of some stations, some sampling points fell into the same pixel. In such cases an arithmetic mean of the measurements was calculated. Finally, as regards the chlorophyll concentration measurements, samples collected near the bottom of the lake were excluded because of the influence of the depth on the spectral response. The statistical properties of the two limnological parameters are shown in
It is clear from the data that the conditions of the two lakes are significantly different: Lago Grande has a higher chlorophyll
The remotely sensed data were acquired by the TM sensor onboard Landsat 5. The scene (path 188, row 32) dates back to the 13/05/2001 (field data collection) and covers the whole Basilicata region. Weather conditions were stable between field data collection and satellite data, with no rain from 6 to 23 May, as reported from the Venosa meteorological station. The sensor’s spectral characteristics are such that TM data is amongst the most used to monitor water quality.
Pre-processing of the Landsat data consisted of radiometric calibration and atmospheric correction: necessary for quantitative studies (
Radiometric calibration consists of a series of equations used to convert the stored quantized energy signal (digital number: DN) of the TM data into radiance values at the sensor. Landsat images were converted into radiance values at the satellite by using the following equation (
where Lsens is the radiance at satellite level of a specific band (W m-2 sr-1 μ-1); DN is the value of the digital number; G is the gain and B is the bias. The line intercept, described by the bias, takes into account the fact that even with a null input signal (Lsens = 0) the acquisition system can still give an electric output signal fundamental to the acquisition system. The gain and offset values for the single bands were obtained from the CCT header file (
Atmospheric correction is needed because electromagnetic radiation travels through the atmosphere along its two paths from the sun to the earth surface and from there to the sensor, undergoing alterations to the radiometric signal. For water, in particular, the sensor recorded radiometric signal is very weak compared to the atmosphere contribution (
where ρsup (λ) is the surface reflectivity; Latm↑(λ) is the atmosphere spectral radiance (path radiance) diffused upward as W m-2 sr-1 μm-1; Lsens(λ) is the sensor-measured spectral radiance; E0(λ) is the solar spectral irradiance as W m-2 μm-1; Eatm↓(λ) is the downward spectral irradiance incident on the surface due to the diffusion of solar radiation through the atmosphere as W m-2 μm-1; τ↑(λ) is the atmospheric transmittance along the sun-surface route; θz is the zenith angle of the sun.
The at-satellite radiance (Lsens) results from the radiometric calibration; E0(λ) is the extra-atmospheric solar constant [Esol = E0(λ) cosθz], where the zenith angle of the sun (θz) was calculated from the date, position and time of satellite overpass while θv (view angle) was assigned a value of 0°, which is considered an acceptable approximation for narrow FOV systems such as Landsat. The solution of the unknown terms is carried out by using several methods, which can be grouped in two categories: image based methods (used in this study) and radiative transfer models. The latter solve the equation on the base of the theory of the radiation transfer of electromagnetic energy interacting with the atmosphere but need
In the method we use, spectral irradiance due to solar radiation in the atmosphere is considered null [Eatm↓(λ) = 0], while the path radiance [Latm↑(λ)] is obtained from the radiance measured by satellite of a pixel having no or very little reflectivity (dark object). In this study the dark object used was a pixel related to deep water off the Tyrrhenian coast (
Downward atmospheric transmittance [τ↓(λ)] is approximated from the zenith angle of the sun (
an approximation valid for wavelengths up to 1.1 μm, while for greater wavelengths (
The upward transmittance is calculated from the following equation (
assuming there is only Rayleigh and not aerosol scattering. The Rayleigh optical thickness [δR(λ)] was calculated according to the equation (
where λ is the wavelength in μm.
The corrected Landsat bands were subset to the area of interest and georeferred to the UTM-WGS84 system using 20 GCP before extracting the radiometric values of the limnological data at the sampling points. Average radiometric values were calculated for a 3x3 window centred on the sampling point (
Common statistical techniques were used in order to determine the relationship between electromagnetic energy and water quality parameters (
First, an analysis using Pearson’s correlation was carried out in order to detect the existence of significant relationships between Landsat bands and both transparency and chlorophyll concentration. Previous studies (
As for the independent variables, all the visible bands and their ratios were included in the multiple regression. The procedure to select significant variables is the stepwise one, with p>0.10 as a limit for factor removal.
As for transparency, the variables selected by the statistical analysis were TM1 and the TM3/TM2, TM1/TM2 e TM2/TM1 ratios. Therefore, the functional model is (
where SD is the Secchi Disk measure in m and TMi are the Landsat bands, with radiometric and atmospheric correction, as W m-2 sr-1 μm-1. In
Data analysis confirms the ability of radiometric values, as derived from the Landsat bands, to predict lake water transparency. Very high R2 values were found and were in agreement with similar studies (
In order to assess model performance we used two indices: the Index of Agreement (IA -
On the other hand, the ratio TM3/TM2 was very significant accounting for 66% of SD variation; adding further terms to the functional model improves the prediction significantly reducing RMSE. The model, however, tends to slightly overestimate SD when values are low. The residuals values, the difference between predicted and observed values, show higher variability when the limnological values are low,
The variability of chlorophyll
where Chl-a is the chlorophyll
The determination coefficient R2 has quite a high value, in line with similar studies (
Having identified the regression functions for transparency and chlorophyll
The correlation between transparency and chlorophyll
Finally, the maps created of chlorophyll
Most of Lago Grande has transparency values lower than 1.5 m, which is typical of a hypertrophic lake and 83.2% of the chlorophyll concentration values have characteristics of mesotrophic lakes; only 17% (mainly concentrated in lake western part), has a concentration typical of eutrophic lakes.
Most of Lago Piccolo has transparency values exceeding 3 m, which is typical of mesotrophic lakes, although an important part (around 35%) has eutrophic lake characteristics. Chlorophyll
From the results, it can be stated that satellite remotely sensed data, especially data from Landsat, can be used effectively to assess the limnological parameters of lake waters. Radiometric data from Landsat can be used to map the areal distribution of some water biophysical parameters even for small basins such as the Monticchio lakes.
Data from the new sensors, such as Modis, can have a high temporal frequency (
A strong, significant, relationship was found between the spectral data in the TM bands with both chlorophyll concentration (R2 = 0.72) and with transparency (R2 = 0.82) measured as Secchi Disk depth. Multiple regression analysis on remote sensing data allowed to efficiently define the models and identify statistically significant variables. The ratios among visible bands and, most significantly, between TM3 and TM2 were good predictors both of transparency and chlorophyll concentration. The TM3 and TM2 bands have opposing trends in reflectance as a function of chlorophyll concentration: an increase of chlorophyll
The two parameters can be mapped for the whole lake surface by identifying the best set of band combinations to describe the relationship linking spectral response to limnological data. The same method can be applied to the trophic conditions. In our case the synoptic view showed a different states of two small lakes providing a basis for a targeted intervention strategy.
This work was supported by EC - Interreg III B WETMUST project (code A.1.042).
Study area (40° 55’ 48.29” N; 15° 36’ 16.61” E) with sampling points of Secchi Disk (triangles) and chlorophyll
Observed and predicted Secchi Disk values with the 95% confidence and prediction limits (dashed and dash dotted lines, respectively) of the regression model.
Observed and predicted values of chlorophyll
Transparency (left) and chlorophyll
Correlation between transparency (SD) and chlorophyll
Morphometric characteristics of the Monticchio lakes.
Parameters | Lago Grande | Lago Piccolo |
---|---|---|
Catchment basin size ( |
240 | 107 |
Volume ( |
3 270 | 2 460 |
Height ( |
653.7 | 656 |
Surface ( |
41.3 | 13.9 |
Maximum depth ( |
36 | 38 |
Mean depth ( |
8.9 | 17.9 |
Interlake channel flow ( |
10 |
Descriptive statistics of chlorophyll and transparency in the two lakes.
Indices | Transparency (Secchi Disk) ( |
Chlorophyll |
||
---|---|---|---|---|
Lago Grande | Lago Piccolo | Lago Grande | Lago Piccolo | |
Minimum | 0.25 | 3.00 | 4.63 | 1.11 |
Maximum | 1.00 | 3.75 | 11.35 | 4.57 |
Mean | 0.83 | 3.25 | 6.93 | 2.45 |
Range | 0.75 | 0.77 | 6.72 | 3.46 |
St. Dev. | 0.186 | 0.199 | 1.656 | 1.051 |
Gain and offset for Landsat TM5 bands.
Bands | 1 | 2 | 3 | 4 | 5 | 7 |
---|---|---|---|---|---|---|
Offset | -1.5 | -2.8 | -1.2 | -1.5 | -0.37 | -0.15 |
Gain | 0.6024 | 1.1749 | 0.8059 | 0.8145 | 0.1081 | 0.0570 |
Pearson’s
Bands | Transparency (SD) |
Chlorophyll |
---|---|---|
TM1 | 0.56** | -0.59** |
TM2 | 0.15 | -0.32** |
TM3 | -0.28 | 0.09 |
TM4 | 0.11 | -0.21 |
TM5 | 0.21 | -0.28 |
TM7 | 0.21 | -0.29 |
TM1/TM2 | 0.20 | -0.06 |
TM2/TM1 | -0.14 | 0.04 |
TM1/TM3 | 0.71** | -0.54** |
TM3/TM1 | -0.67** | 0.47** |
TM3/TM2 | -0.81** | 0.68** |
Regression analysis between limnological parameters and Landsat bands: (A) Secchi Disk depth; (B) chlorophyll
(A) DependentVariable | SD(SD [m]; TMi [W m-2 sr-1 μm-1]) | |
---|---|---|
Previsional Model | SD = a + bTM3/TM2 + cTM1/TM2 + dTM1 + eTM2/TM1 | |
Coefficients | a | 10.752 |
b | -5.467 | |
c | -3.783 | |
d | 401.21 | |
e | -5.581 | |
R2 | 0.82 | |
R2 change | TM3/TM2 | 0.66 |
TM1/TM2 | 0.72 | |
TM1 | 0.80 | |
TM2/TM1 | 0.82 | |
RMSE | 0.540 | |
NSE | 0.943 | |
IA | 0.985 | |
(B) DependentVariable | Chl-a(Chl-a [mg m-3]; TMi [W m-2 sr-1 μm-1]) | |
Previsional Model | Chl-a = a + bTM3/TM2 + cTM1/TM2 + dTM2 + eTM2/TM1 | |
Coefficients | a | -47.515 |
b | 9.516 | |
c | 20.952 | |
d | - 873.0 | |
e | 34.889 | |
R2 | 0.72 | |
R2 change | TM3/TM2 | 0.46 |
TM1/TM2 | 0.49 | |
TM2 | 0.63 | |
TM2/TM1 | 0.72 | |
RMSE | 1.300 | |
NSE | 0.754 | |
IA | 0.925 |
The area based chlorophyll concentration and transparency according to the OECD classification ranges.
RangeOECD | Transparency (SD) m | RangeOECD | Chlorophyll |
||||||
---|---|---|---|---|---|---|---|---|---|
Lago Grande | Lago Piccolo | Lago Grande | Lago Piccolo | ||||||
Area(ha) | % | Area(ha) | % | Area(ha) | % | Area(ha) | % | ||
<1.5 | 39.7 | 96.2 | 0.25 | 1.8 | ≤ 1 | - | - | 0.67 | 4.8 |
1.5 - 3 | 1.6 | 3.8 | 4.91 | 35.3 | ≤ 2.5 | - | - | 7.66 | 55.1 |
3 - 6 | - | - | 8.74 | 62.9 | 2.5 - 8 | 34.6 | 83.2 | 5.57 | 40.1 |
6 - 12 | - | - | - | - | 8 - 25 | 6.9 | 16.8 | - | - |
>12 | - | - | - | - | > 25 | - | - | - | - |