Agricultural University of Athens
Water quality assessment in Greek lakes by using remote sensing and statistical modelling
Abstract
dc:descriptionThe aim of the present PhD thesis is the achievement of continuous monitoring and assessment of water quality (WQ) and trophic state of Greek lakes by exploiting the implementation of the Water Framework Directive (WFD) in Greece in synergy with satellite RS, providing parallel support to sustainable water resources management at a national scale. Continuous WQ monitoring is the most crucial aspect for lake management. Therefore, the methodological framework developed herein has as an ultimate goal the generation of lake WQ quantitative models while the practical use of this approach was developed and evaluated in a total of 50 lake water bodies (natural and artificial) from 2013–2018, constituting the National Lake Network Monitoring of Greece in the context of the WFD. Concerning the utilized Earth Observation (EO) data, images from Landsat 7 Enhanced Thematic Mapper Plus (ETM+) and Landsat 8 Operational Land Imager (OLI) sensors have been combined with co-orbital WQ in-situ measurements (Chl-a, Secchi Disk depths and Total phosphorus-TP- concentrations) with the main objective of delivering robust WQ assessment models. From the statistical point of view, principal component analysis (PCA) was performed to explore Greek lakes’ interrelationships among their Chl-a values and certain criteria, e.g. their characteristics (artificial/natural), WFD typology, climatic type (according to the Köppen-Geiger climate classification), season of water samplings and the date difference between sampling and satellite overpass. PCA highlighted the lake characteristics (natural/artificial) and WFD typology as the variables that mostly contribute to the variance of Chl-a concentration; thus, numerous stepwise multiple regression analyses (MLRs) among different groups of cases, formed by the PCA criteria, were implemented with basic aim the generation of different remote sensing-derived Chl-a algorithms for different types of lakes. Moreover, correlation analysis among in-situ co-orbital WQ data was conducted to explore and detect their inter-relationships. Subsequently, based on correlation analysis’s results, further stepwise MLRs employing available in-situ TP and Secchi depth datasets were further implemented to establish optimal quantitative models. Eventually, trophic status classification was conducted herein, calculating Carlson’s Trophic State Index (TSI) of each lake, initially throughout all lakes and then oriented toward natural-only and artificial-only lakes. The proposed scheme resulted in the development of models separately for natural (R = 0.78) and artificial (R = 0.76) lakes, while the model developed without criteria proved weaker (R = 0.65) in comparison to the other ones examined. MLRs among Landsat data and Secchi depths resulted in 3 optimal models concerning the assessment of Secchi depth of all lakes (Secchi general; R = 0.78; RMSE = 2 m), natural (Secchi natural; R = 0.95; RMSE = 1.87 m) and artificial (Secchi artificial; R = 0.62; RMSE = 1.36 m), with reliable accuracy. Study findings showed that TP-related MLR analyses failed to deliver a statistically acceptable model for the reservoirs; nevertheless, they delivered a robust TP general model for all lakes (R = 0.71; RMSE = 0.008 mg/L) and a TP natural model for natural lakes (R = 0.93; RMSE = 0.018 mg/L). Subsequently, regarding the TSI results, the higher deviation of satellite-derived TSI values in relation to in-situ ones was detected in reservoirs and shallower lakes (mean depth < 5 m), indicating noticeable divergences among natural and artificial waterbodies. Next significant key questions that were answered are initially whether Landsat-based empirical WQ algorithms can be efficiently applied to Sentinel 2 images and then whether the combined use of multi-sensor data improves those algorithms’ prediction accuracy. Hence, Sentinel 2 images of 2018 with concurrent dates with those of field measurements were utilized to facilitate a WQ models’ efficiency evaluation and comparison with the respective Landsat’s validation results. Concerning the results, in particular for general models of all WQ elements (Chl-a, Secchi depth and TP), all models were more efficient and accurate when were accompanied by Landsat images while no improvement was observed by using multi sensor images. Chl-a and TP models (for natural lakes) presented lower values of error metrics when employing Sentinel 2 images (RMSE Chl-a=16.4 μg/l vs 21.5 μg/l; RMSE TP=0.03 mg/l vs 0.031 mg/l) and only Secchi natural model performed better with Landsat data (2.8 m vs 2.9 m). Concerning artificial lakes, performance of Chl-a model was better by exploiting Landsat data (RMSE= 3.7 μg/l vs 7.7 μg/l of Sentinel 2) while Secchi model achieved slightly better efficiency with Sentinel 2 images (RMSE= 1.5 m vs 1.6 m of Landsat). The largely worse performance of Chl-a models compared to rest of WQ elements (median MAPE values ranged from 42 % to 58%, Secchi depth from 24% to 44% and TP from 22% to 38%), emphasized once again the complexity that mapping of Chl-a in Case 2 waters (coastal and/or inland waters) hides. Today, open source Cloud Computing platforms have emerged as a valuable tool for geospatial analysis of image data from various satellites while the Google Earth Engine (GEE) platform is the most widespread in the scientific field of satellite RS. Newest launches of various satellites in combination with the GEE platform, facilitate in a great extent national-scale lake monitoring. Next step was to test the transferability and performance of hereby-developed WQ algorithms when employing Landsat (7 +ETM/8 OLI) and Sentinel 2 surface reflectance (SR) values embedded in GEE and subjected to different atmospheric correction (AC) methods from those used as they were developed. More particularly, GEE-Landsat and -Sentinel 2 SR of year 2018 was retrieved from the WFD lake sampling stations while were atmospherically corrected by the methods LaSRC (Landsat 8 OLI), LEDAPS (Landsat 7 ETM+) and Sen2Cor (Sentinel 2). Those SR values (GEE) were matched with WQ in-situ data of 2018 within ±7 days (from sampling date) of satellite overpasses, while the same pairs were created with SR derived from manually downloaded and pre-processed, with AC DOS1 method, respective images. Empirical WQ models of Chl-a, Secchi depth and TP (for all and separately for natural and artificial lakes), were applied twice employing both types of SR (DOS1- and rest in GEE- corrected). Furthermore, double application of WQ models was conducted separately for Landsat (7 ETM+/OLI) and Sentinel 2 data. Double application of WQ models resulted in double quantifications of each studied WQ element in each sampling station while those double WQ values were inserted in a linear regression analysis. Yielded linear equations (corrected WQ models), for each sensor, were accompanied by strong associations (R2 ranging from 0.68 to 0.98). Initial and corrected sensor-specific WQ models were validated based on available in-situ WQ datasets of 2019 and 2020. Sensor-specific correction of WQ models was proven essential for some of them while RMSE values ranged for Chl-a from 11.68 μg/l (Landsat) to 14.88 μg/l (Sentinel 2), for Secchi depth from 2.02 m (Landsat) to 2.57 m (Sentinel 2) and TP from 0.14 mg/l (Landsat) to 0.09 mg/l (Sentinel 2), values that confirmed the stability and transferability of hereby WQ models even when employ differently-from-DOS1 method corrected SR. One more ambiguous question that has been examined is whether WQ universal models are efficient for WQ monitoring of oligotrophic Case-2 waters. The classification of waters in Case 1 (oceanic) and Case 2 (coastal regions, rivers, and lakes), is characterized by great importance; Case 1 waters are determined by phytoplankton and co-varying substances, while Case 2 waters are more complex concerning their composition and optical properties. Oligotrophic lakes are classified as Case-2 rather than Case-1 waters since they typically receive significant levels of terrigenous input and their water clarity is primarily controlled by the concentration of Dissolved Organic Carbon (DOC). In purview of the above, lake WQ models were applied to Landsat 8 OLI images, with available in-situ WQ data, illustrating two (2) Greek oligotrophic waterbodies, Trichonis and Amvrakia lakes. Conclusively, their application was ineffective: Chl-a general model yielded values of RMSE=1.9 μg/l, NRMSE=1.6 and median MAPE=256.8 %, Chl-a natural model yielded values of RMSE=1.8 μg/l, NRMSE=1.5 and median MAPE=176.6 % while results of all models of Secchi Disk and Total Phosphorus were statistically insignificant. Based on the previous approach, an effort has been made to develop special designed WQ algorithms in Trichonis lake based on a dense sampling network (22 stations). Subsequently, the most statistically promising quantitative models – accrued from statistical elaboration of 2014 data- were applied to another satellite image of 2013 (with available in-situ WQ data). Results from the validation process showed a relatively variable statistical relationship between the in-situ data and reflectances (R logchl-a: 0.4, R NH4 +: 0.7, R Chl-a: 0.5, R CDOM at 420 nm: 0.3). Hereby findings were concurrent with other studies in international literature, indicating that estimations for oligotrophic are less accurate than eutrophic and mesotrophic lakes, owing to the lack of suspended particles that are detectable by satellite sensors. Conclusively, background information required, suitable spectral bands and essential circumstances are described for the most optimal designation of the WQ monitoring in oligotrophic waterbodies.
Degree
thesis:*- Grantor dc:publisher
- Agricultural University of Athens
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Markogianni, Vassiliki
- Μαρκογιάννη, Βασιλική
Subjects
dc:subject × 26- Ποιότητα υδάτων λιμνών
- Ανάλυση κύριων συνιστωσών
- Πολλαπλή γραμμική παλινδρόμηση
- Τροφική κατάσταση
- Περίπτωση 2
- Οδηγία πλαίσιο για τα ύδατα 2000/60/ΕΕ
- Δορυφορικά δεδομένα Landsat
- Δορυφορικά δεδομένα Sentinel 2
- Google Earth Engine
- Ολιγοτροφικό
- lake water quality
- Principal component analysis (PCA)
- Multiple linear regression analysis
- Trophic status of lakes
- Case 2
- Water framework directive 2000/60/EU
- LANDSAT
- Sentinel 2
- Google earth engine (GEE)
- Oligotrophic
- Γεωεπιστήμες και Επιστήμες Περιβάλλοντος
- Φυσικές Επιστήμες
- Υδάτινοι πόροι
- Earth and Related Environmental Sciences
- Natural Sciences
- Water Resources
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/53644