Day 4

Detailed paper information

Back to list

Paper title Monitoring chlorophyll-a dynamics from Sentinel-2 imagery in high-mountain lakes
  1. Joana Maria Llodrà-Llabrés University of Granada Speaker
  2. Domingo Alcaraz-Segura University of Granada
  3. Carmen Pérez-Martínez University of Granada (ESQ1818002F)
Form of presentation Poster
  • A7. Hydrology and Water Cycle
    • A7.06 EO for monitoring water quality and ecological status in inland waters
Abstract text High-mountain lakes are among the most vulnerable ecosystems to climate change, particularly in Mediterranean climates. The Mediterranean region is suffering from an exacerbated rate of climate change compared to the global trends, being considered as a climate change hotspot in the world. The characteristic summer droughts of the Mediterranean climate are being intensified due to the increase in the mean annual air temperature, specially during the summer months, and the decrease in annual rainfall. As a whole, it may produce a decline in the snow accumulation in those regions and an earlier melting of the snow that could eventually affect the hydrology of the ecosystems, among other affected processes. Moreover, high-mountain ecosystems are experiencing elevation-dependent warming, whereby the warming rate is amplified with altitude.
Sierra Nevada National Park (Granada, Spain) is the southernmost mountain range in Europe and constitutes a biodiversity hotspot. In Sierra Nevada there are around 50 small glacially-formed lakes at an elevation of 2800-3100 m above sea level. They are small (surface ranges between 0.01-2.1 ha) and shallow (depth ranges between 0.3-8 m) oligo- to meso-oligotrophic lakes. Because of the high sensitivity of remote lakes to environmental changes, they are considered as excellent sentinels of climate change.
These lakes are hardly accessible. Hence, continuous and regular monitoring of these water bodies is difficult, but essential for their correct management. In this study we will mainly focus on chlorophyll-a (chl-a) since it is a key indicator of phytoplankton biomass and water quality. Remote sensing techniques represent an alternative to the current field samplings that are not always possible or not as frequently as desirable. One of the satellites with the greatest potential is the Multispectral Instrument (MSI) Sentinel-2, due to its high spatio-temporal resolution. Its spatial resolution of up to 10 m may allow the analysis of small waterbodies, and its revisit time of five days might allow the characterisation of temporal dynamics. However, a lack of data for such small lakes as ours has been detected, despite being very frequent ecosystems.
Hence, the aims of this work are (a) to explore the potential of remote sensing techniques using the Sentinel-2 imagery to estimate water quality parameters, mainly chlorophyll, in shallow and small high-mountain lakes in a regional scale and (b) to develop a chl-a estimate model as a tool for eutrophication monitoring since it may increase as a consequence of climate change. This may, in turn, allow the characterization of the lakes in terms of susceptibility to eutrophication since each lake is affected differently by variables such as livestock pressure, tourism, Sahara dust deposition and lake morphometry and watershed features. Finally, (c) this model is intended to be used by the management of the Sierra Nevada National Park and take necessary measures in order to maintain a good ecological status of the lakes.
Achieving our objectives would represent a major breakthrough since until now Sentinel-2 imagery has only been used for this purpose on lakes much larger and deeper than the small high-mountain lakes of Sierra Nevada. This work might represent a baseline for further remote studies of similar ecosystems.
The Sentinel-2 images were obtained and processed through Google Earth Engine (GEE). A first approach was made to select lakes with water-pure pixels. Sseven lakes met this requirement: Río Seco Lake, Yeguas Lake, Caldera Lake, Larga Lake, Mosca Lake, Vacares Lake and Caballo Lake.
Field sampling campaigns were conducted during the ice-free period of 2020 and 2021 obtaining 8 and 40 samples, respectively. An optimal time gap of ±3 days and a maximal of ±5 days were established between the in-situ measurements and the satellite overpass. In each lake an integrated water sample of 1.2 m was collected at a point in which the adjacency and bottom effects were minimized. The samples were stored in dark conditions until we arrived at the laboratory where the chl-a concentration, colored dissolved organic matter (CDOM) and total suspended solids (TSS) were analyzed. Chl-a and CDOM were determined through the filtration of the water samples using pre-combusted Whatman GF/F. Chl-a concentration was assessed by pigment extraction from the filter using ethanol and it was spectrophotometrically analysed. CDOM was determined spectrophotometrically from the filtered water. Finally, TSS were determined from the pre- and post-filtering Whatman GF/F filter weight.
Around 1500 papers relating chl-a and Sentinel-2 published until October 2021 were reviewed. It is worth noting that none of them was conducted specifically in high-mountain lakes. We selected and tested on Sierra Nevada the potential models that had already shown a good performance in oligo- and mesotrophic waters like ours. Traditional empirical, semi-analytical and novel machine learning models were tested. The use of machine learning, a type of artificial intelligence, is increasing in several scientific branches and it is in constant evolution. Hence, it represents a novel approach to chl-a retrieval and an increasing number of papers are showing its high performance on this purpose. According to the literature, the chl-a models that have performed best in waters with similar characteristics to ours use the red band (665 nm), red edge band 1 (705 nm) and red edge band 2 (740 nm). Some of these models are 2BDA (Moses, 2009), 3BDA (Gitelson, 2009), MCI (Gower, 2005) and Toming (2016), among others. However, the model FLH (Fluorescence Light Height) (Buma and Lee, 2020) has shown its high performance in similar waters and it uses the blue band (490 nm), green band (560 nm) and red band (665 nm). Finally, different atmospheric correction algorithms previously published in the literature were tested in combination with the developed chl-a models. Regarding the atmospheric algorithms cited in the bibliography, the algorithms that perform best in clear waters like ours are Polymer and iCor. The last one is the only one that includes correction for the adjacency effect, which is almost unavoidable in our study area. By contrast, the algorithm C2RCC has been shown to fail in presence of adjacency effect.
Taking into account the limited chl-a data taken in-situ during 2020, the model that performed best in our study area was FLH with a determination coefficient R2=0.59. By introducing the data collected during the 2021 field-sampling campaign and increasing the number of experimental replications as well as the number of sampled lakes it is expected to obtain a more accurate model for our study area. Moreover, more pre-existing models developed during 2021 will be tested and different atmospheric corrections will be introduced.