Monitoring of Calcite Precipitation in Hardwater Lakes with Multi-Spectral Remote Sensing Archives

water

Article Monitoring of Calcite Precipitation in Hardwater Lakes with Multi-Spectral Remote Sensing Archives

Iris Heine 1,*, Achim Brauer 2, Birgit Heim 3, Sibylle Itzerott 1, Peter Kasprzak 4, Ulrike Kienel 2,5 and Birgit Kleinschmit 6

1 Helmholtz Centre Potsdam, GFZ German Research Centre for Geosciences, Section 1.4 Remote Sensing, Telegrafenberg, 14473 Potsdam, Germany; [email protected] 2 Helmholtz Centre Potsdam, GFZ German Research Centre for Geosciences, Section 5.2 Climate Dynamics and Landscape Evolution, Telegrafenberg, 14473 Potsdam, Germany; [email protected] (A.B); [email protected] (U.K.) 3 Alfred Wegener Helmholtz Center for Polar and Marine Research, Telegrafenberg, 14473 Potsdam, Germany; [email protected] 4 Department of Experimental Limnology, Leibniz-Institute of Freshwater Ecology & Inland Fisheries, Alte Fischerhütte 2, OT Neuglobsow, 16775 Stechlin, Germany; daphnia@igb-berlin.de 5 Institute for Geography and Geology, University Greifswald, Friedrich-Ludwig-Jahn-Street 16, 17487 Greifswald, Germany 6 Geoinformation in Environmental Planning Lab, Technische Universität Berlin, Straße des 17. Juni 145, 10623 Berlin, Germany; [email protected] * Correspondence: [email protected]; Tel.: +49-331-288-1763; Fax: +49-331-288-1192

Academic Editor: Y. Jun Xu Received: 4 October 2016; Accepted: 13 December 2016; Published: 3 January 2017

Abstract: Calcite precipitation is a common phenomenon in calcium-rich hardwater lakes during spring and summer, but the number and spatial distribution of lakes with calcite precipitation is unknown. This paper presents a remote sensing based method to observe calcite precipitation over large areas, which are an important prerequisite for a systematic monitoring and evaluation of restoration measurements. We use globally archived satellite remote sensing data for a retrospective systematic assessment of past multi-temporal calcite precipitation events. The database of this study consists of 205 data sets that comprise freely available Landsat and Sentinel 2 data acquired between 1998 and 2015 covering the Northeast German Plain. Calcite precipitation is automatically identified using the green spectra and the metric BGR area, the triangular area between the blue, green and red reflectance value. The validation is based on field measurements of CaCO3 concentrations at three selected lakes, Feldberger Haussee, Breiter Luzin and Schmaler Luzin. The classification accuracy (0.88) is highest for calcite concentrations ≥0.7 mg/L. False negative results are caused by the choice of a conservative classification threshold. False positive results can be explained by already increased calcite concentrations. We successfully transferred the developed method to 21 other hardwater lakes in Northeast Germany. The average duration of lakes with regular calcite precipitation is 37 days. The frequency of calcite precipitation reaches from single time detections up to detections nearly every year. False negative classification results and gaps in Landsat time series reduce the accuracy of frequency and duration monitoring, but in future the image density will increase by acquisitions of Sentinel-2a (and 2b). Our study tested successfully the transfer of the classification approach to Sentinel-2 images. Our study shows that 15 of the 24 lakes have at least one phase of calcite precipitation and all events occur between May and September. At the lakes Schmaler Luzin and Feldberger Haussee, we illustrated the influence of ecological restoration measures aiming at nutrient reduction in the lake water on calcite precipitation. Our study emphasizes the high variance of calcite precipitation in hardwater lakes: each lake has to be monitored individually, which is feasible using Landsat and Sentinel-2 time series.

Water 2017, 9, 15; doi:10.3390/w9010015 www.mdpi.com/journal/water Water 2017, 9, 15 2 of 31

Keywords: calcium-rich hardwater lakes; Landsat Time series analysis; Sentinel 2; Northeast German Plain; evaluation of ecological restoration measures

1. Introduction Calcite (or calcium carbonate) precipitation events in lakes are a common phenomenon in calcium-rich hardwater lakes. They are also described as “whiting”, “milky water phenomenon” or “seasonal clouding” [1–3]. The complex process of calcite precipitation has been intensively studied [1,3–19]. Calcite precipitation is the consequence of the supersaturation of the lake water with respect to calcite. Principally, two possible mechanisms can lead to supersaturation: (1) physical-chemical, through seasonal temperature effects on the solubility of carbon dioxide and calcite (i.e., the solubility of calcite decreases with increasing temperature); and (2) biogenic induction through assimilation of carbon dioxide by plankton blooms of photosynthesizing algae and bacteria in the phototrophic upper water column [2,9] with impact on the carbonate equilibria of the water, which varies with pH, alkalinity, and total dissolved carbon. Aside from that, cells of algae and cyanobacteria can act as surface catalysts for calcite precipitation well before supersaturation is reached [17,18]. In line with that, calcite precipitation events in lakes are recorded after peak phytoplankton blooms [13,16,17]. Calcite precipitation was found to intensify in relation with the trophic state (based on the concentration of dissolved P) from oligotrophic towards weakly eutrophic conditions, but became weaker towards hypereutrophic/polytrophic conditions because of the inhibition of the precipitation by increased P concentration [14–16]. The Northeast German Plain is a region dominated by many hardwater lakes [3]. Studies concerning these lakes showed that calcite precipitation is an important variable impacting on both the water quality and the ecology of these ecosystems [3,5,6]. Calcite precipitation reduces the nutrient concentration and, consequently, phytoplankton productivity and therefore is a natural protection mechanism of hardwater lakes against eutrophication [3,5]. The reduction of nutrient concentration (“self-cleaning”) is caused by the co-precipitation of soluble inorganic phosphorus and the flocculation of particles containing phosphorus, which are eventually transported to the sediment at the bottom of a respective lake [3,20]. In times of climate change, also the storage of CO2 in the sediments might be an important factor. A study at lake Breiter Luzin in Mecklenburg-Vorpommern, Germany, specified 2 sedimentation rates of 300 g CaCO3/m /day [3] and Koschel et al. estimate a calcite production and sedimentation of 150–900 ton/km2 per year for seven lakes in Mecklenburg-Vorpommern [6]. Whereas precipitated calcite resuspends to some extent, the majority sinks to the lake bottom [12] and is, without mixing, accumulated in calcite layers. Those calcite sediments have been used for the reconstruction of past precipitation events in other regions [20,21]. However, although the ecological importance of calcite precipitation is recognized, neither the number of lakes with calcite precipitation in the Northeast German Plain nor worldwide is known, because only individual lakes are monitored regularly [3,5,6]. Additionally, calcite precipitation varies both within and among lakes: intensity, frequency and duration of calcite precipitation events can vary from year to year [6]. Thus, calcite precipitation events may easily be missed during one-time observations or short-term monitoring of lakes. Here, remote sensing archives of optical satellite missions such as Landsat or Sentinel offer a great potential for a satellite-based long-term monitoring of lakes with high temporal and spatial resolution and for the synoptic monitoring of a larger region like the Northeast German Plain. However, only few studies have used remote sensing for the monitoring of calcite precipitation. In southwest Florida and Great Bahama Bank whitings have been monitored using medium-resolution MODIS imagery [22,23] and photographs from the NASA manned spacecraft program [24]. Two studies have been conducted on the spatial distribution of calcite precipitation within large lakes (the Great Lakes, and Lake Constance) with Landsat imagery [1,25] and Thiemann and Koschel classified calcite precipitation in 21 lakes in the Northeast German Plain using one hyperspectral airborne data set [2]. Water 2017, 9, 15 3 of 31

In our study, we exploit the multi-spectral long-term remote sensing archive of Landsat and test the applicability for the recently started Sentinel-2 for the long-term monitoring of calcite precipitation in the Northeast German Plain. In this context, the objectives of this study are:

• To develop a robust automated remote sensing-based approach for retrospective long-term multi-temporal calcite precipitation monitoring based on a multi-sensor remote sensing time series. • To characterize calcite precipitation in terms of frequency and duration to deepen the process understanding.

2. Study Area The lakes of the Northeast German Plain were formed during the late Weichselian glaciation [3]. We selected three representative study areas: Feldberg Lake District, the Klocksin Lake Chain, and Rheinsberg Lake Region which are located in the federal states of Mecklenburg-Western Pomerania and Brandenburg in Germany (Figure1). These regions are covered by the Landsat acquisition tiles 193023 and 194023. We chose three lakes with regular in situ measurements in the Feldberg Lake District (Feldberger Haussee, Breiter Luzin, and Schmaler Luzin) for the development of a calcite precipitation classiﬁcation approach and its validation. Then, the applicability has been tested on the other two regions, the Klocksin Lake Chain and Rheinsberg Lake Region.

Figure 1. Study area with three selected regions: Feldberg Lake District, the Klocksin Lake Chain, and Rheinsberg Lake Region. The gray lines illustrate the Landsat footprints of the acquisition tiles 193023 and 194023. The gray dashed line shows the footprint of Landsat 5, the solid gray line shows the footprint of Landsat 7, the dotted gray line the footprint of Landsat 8.

Water 2017, 9, 15 4 of 31

2.1. Feldberg Lake District Water 2017, 9, 15 4 of 31 Figure2 shows the Feldberg Lake District with its well-researched lakes: Feldberger Haussee (FH),2.1. Breiter Feldberg Luzin Lake (BL) District and Schmaler Luzin (SL). All three lakes are hardwater lakes with regular in situ measurements since 1998. The topography, morphology and limnological characteristics of the Figure 2 shows the Feldberg Lake District with its well‐researched lakes: Feldberger Haussee lakes(FH), are summarized Breiter Luzin in(BL) Table and 1Schmaler. Luzin (SL). All three lakes are hardwater lakes with regular in FHsitu potentiallymeasurements used since to 1998. be a The mesotrophic topography, lake, morphology but nutrient and limnological input by sewage characteristics (phosphorus of the and nitrogen)lakes andare summarized surface runoff in Table since 1. the 1960s until 1980 caused strong eutrophication [26]. In 1980, the sewage dischargeFH potentially was stoppedused to be decreasing a mesotrophic the external lake, but nutrient nutrient loadinginput by of sewage the lake (phosphorus by 90%. However, and becausenitrogen) of the and tremendous surface runoff amounts since the of 1960s nutrients until (especially1980 caused strong phosphorus) eutrophication stored [26]. in the In sediment1980, the the lake didsewage not discharge respond was with stopped a substantial decreasing improvement the external nutrient of water loading quality of the for lake decades by 90%. [27 However,]. Therefore, phosphorusbecause inactivationof the tremendous by treating amounts the of lake nutrients with poly-aluminum(especially phosphorus) chloride stored as a in flocculation the sediment agent the was implementedlake did not in Aprilrespond 2011. with The a substantial following drasticimprovement reduction of water of average quality totalfor decades phosphorus [27]. Therefore, concentration phosphorus inactivation by treating the lake with poly‐aluminum chloride as a flocculation agent in the mixed layer from 0.060 mg/L (2006–2010) to 0.017 mg/L (2011–2015) resulted in an improvement was implemented in April 2011. The following drastic reduction of average total phosphorus of theconcentration trophic status in the from mixed eutrophic layer from to mesotrophic 0.060 mg/L (2006–2010) [28,29]. to 0.017 mg/L (2011–2015) resulted in BLan improvement is located immediately of the trophic status downstream from eutrophic of FH. to BL mesotrophic is known [28,29]. for calcite precipitation [3,5,6]. Its LAWABL trophic is located index immediately of 1997 is downstream mesotrophic, of FH. and BL its is potential known for natural calcite trophyprecipitation is oligotrophic [3,5,6]. Its [30]. An unpublishedLAWA trophic study index of of the 1997 Leibniz-Institute is mesotrophic, of and Freshwater its potential Ecology natural & trophy Inland is Fisheriesoligotrophic (IGB) [30]. classified An BL inunpublished 2015 as mesotrophic. study of the Leibniz‐Institute of Freshwater Ecology & Inland Fisheries (IGB) classified SLBL in is 2015 potentially as mesotrophic. oligotrophic, but nutrient input lead to a moderate eutrophic state since the 1950s [30SL]. Sinceis potentially the 1980s oligotrophic, the catchment but nutrient was input restored lead by to a reducing moderate the eutrophic nutrient state input since from the the 1950s [30]. Since the 1980s the catchment was restored by reducing the nutrient input from the catchment [30], but only a lake restoration in 1996/1997 reduced the eutrophication significantly [12]. catchment [30], but only a lake restoration in 1996/1997 reduced the eutrophication significantly [12]. The lake was restored by artificially triggering calcite precipitation through in-depth aeration and The lake was restored by artificially triggering calcite precipitation through in‐depth aeration and the the addition of Ca(OH) in the hypolimnion which caused a significant decrease of total phosphorus addition of Ca(OH)2 in the hypolimnion which caused a significant decrease of total phosphorus contentcontent [30]. [30]. SL isSL classified is classified as as mesotrophic mesotrophic since since 1995, with with oligotrophic oligotrophic phases phases [30,31]. [30,31 ].

Figure 2. Study area of the Feldberg Lake District on 11 July 1999 (Landsat 7, RGB quasi‐true color) Figure 2. Study area of the Feldberg Lake District on 11 July 1999 (Landsat 7, RGB quasi-true color) with Feldberger Haussee (FH), Breiter Luzin (BL) and Schmaler Luzin (SL). BL has a distinct turquoise withcolor Feldberger whereas Haussee FH and (FH),SL are Breiter dark blue. Luzin All (BL)lakes and are framed Schmaler with Luzin white (SL). lines. BL The has sampling a distinct sites turquoise are colorillustrated whereas FHas white and SLtriangles. are dark blue. All lakes are framed with white lines. The sampling sites are illustrated as white triangles.

Water 2017, 9, 15 5 of 31

For FH, regular calcite precipitation events are documented since 1985. Nevertheless, their intensity (i.e., calcite concentration) clearly increased since the year 2006 and remained high ever since. That was the period when FH ﬁnally approached mesotrophic conditions. Koschel (1987) concludedWater that 2017, the9, 15 intensity of calcite precipitation may be highest in moderately nutrient-enriched5 of 31 hard water lakes, because photosynthesis is high enough to shift the lime-carbonic acid equilibrium to the carbonateFor side, FH, whileregular the calcite impact precipitation of factors events to impair are documented the growth since of calcite 1985. Nevertheless, crystals is minimal their [3]. intensity (i.e., calcite concentration) clearly increased since the year 2006 and remained high ever Tablesince. 1. ThatOverview was the of period morphology when FH and finally limnological approached characteristics mesotrophic of conditions. the lakes Koschel FH, BL, (1987) and SL. concluded that the intensity of calcite precipitation may be highest in moderately nutrient‐enriched The limnological characterization is based on the Bund/Länder-Arbeitsgemeinschaft Wasser (LAWA) hard water lakes, because photosynthesis is high enough to shift the lime‐carbonic acid equilibrium trophic index. to the carbonate side, while the impact of factors to impair the growth of calcite crystals is minimal [3].

Area Maximum Depth Mean Depth LAWA 1997 Trophic Reference State LakeTable 1. Overview of morphology and limnological characteristics of the lakes FH, BL, and SL. The (km2)[32] (m) [30] (m) [30] [30] [30] limnological characterization is based on the Bund/Länder‐Arbeitsgemeinschaft Wasser (LAWA) FHtrophic index. 1.29 12.5 4.9 Eutrophic Mesotrophic BL 3.41 58.3 22.3 Mesotrophic Oligotrophic Area Maximum Depth Mean Depth LAWA 1997 Trophic Reference State Lake SL (without (km2) [32] (m) [30] (m) [30] [30] [30] 0.84 22.5 12.2 Mesotrophic Oligotrophic Carwitzer Becken)FH 1.29 12.5 4.9 Eutrophic Mesotrophic BL 3.41 58.3 22.3 Mesotrophic Oligotrophic SL (without 0.84 22.5 12.2 Mesotrophic Oligotrophic 2.2. KlocksinCarwitzer Lake Becken) Chain The second case study area is the Klocksin Lake Chain with Flacher See (FS), Tiefer See (TS), 2.2. Klocksin Lake Chain Hofsee (HS), and Bergsee (BS) as shown in Figure3. There are no regular in situ measurements The second case study area is the Klocksin Lake Chain with Flacher See (FS), Tiefer See (TS), of CaCO concentrations in the lakes, but measurements in 1996 in TS, FS, and BG show high Ca Hofsee3 (HS), and Bergsee (BS) as shown in Figure 3. There are no regular in situ measurements of concentrations [30]. The sediment record of TS shows calcite layers in each year and in some years, CaCO3 concentrations in the lakes, but measurements in 1996 in TS, FS, and BG show high Ca even twoconcentrations sub layers [30]. can The be detectedsediment record during of thinTS shows section calcite inspection. layers in each The year calcite and in layers some years, of the years 1998, 1999,even 2003, two sub 2004, layers 2006, can 2007, be detected 2011, and during 2012 thin are section thinner inspection. than those The of calcite the otherlayers years of the inyears the period considered1998, in 1999, this 2003, study 2004, [33 2006,]. This 2007, hints 2011, at and shorter 2012 are or thinner less intensive than those calcite of the precipitation, other years in the but period dissolution of calciteconsidered particles in on this their study way [33]. through This thehints water at shorter column or mayless intensive also play calcite a role. precipitation, Analyses of but sediment dissolution of calcite particles on their way through the water column may also play a role. Analyses trap material (since 2012) indicate peak calcite precipitation either in May/June and August/September of sediment trap material (since 2012) indicate peak calcite precipitation either in May/June and or centeredAugust/September in July [34]. Theor centered known in topography, July [34]. The morphologyknown topography, and limnologicalmorphology and characteristics limnological of the lakes arecharacteristics summarized of the in Table lakes 2are. summarized in Table 2.

Figure 3. Study area of the Klocksin Lake Chain with Flacher See, Tiefer See, Hofsee and Bergsee on Figure 3. Study area of the Klocksin Lake Chain with Flacher See, Tiefer See, Hofsee and Bergsee on 11 July 1999 (Landsat 7, quasi‐true color RGB). All lakes appear dark and are framed with white lines. 11 July 1999 (Landsat 7, quasi-true color RGB). All lakes appear dark and are framed with white lines. Water 2017, 9, 15 6 of 31 Water 2017, 9, 15 6 of 31

Table 2. Overview of morphology and limnological characteristics of the lakes in the Klocksin Lake ChainChain..

Area Area MaximumMaximum Depth Depth MeanMean Depth Depth LAWALAWA 1996 1996 TrophicTrophic Reference Reference State State Lake Lake 2 (km )(km [32]2 )[32] (m)(m) [30,35] [30,35 ] (m)(m) [30[30,36],36] [30[30]] FS FS1.25 1.2531.9 31.9 9.79.7 MesotrophicMesotrophic MesotrophicMesotrophic TS TS0.68 0.6862.5 62.5 18.518.5 MesotrophicMesotrophic OligotrophicOligotrophic HS HS0.39 0.3927 27 ‐ - MesotrophicMesotrophic ‐ - BS BS0.57 0.5715.0 15.0 6.46.4 MesotrophicMesotrophic MesotrophicMesotrophic

2.3. Rheinsberg Lake Region The third test area is the Rheinsberg Lake Region, with Stechlinsee Stechlinsee as as the the largest largest lake lake (Figure (Figure 44).). The LAWA trophictrophic statestate (1998)(1998) ofof StechlinseeStechlinsee isis oligotrophic,oligotrophic, which corresponds to the trophic reference state, and the lake has a lowlow phytoplanktonphytoplankton biomass [[37].37]. Stechlinsee is known for calcite precipitation [[5,6],5,6], and therethere waswas anan extraordinaryextraordinary intensiveintensive eventevent inin JulyJuly 20112011 [[28].28]. The northernnorthern and and southern southern parts parts of Nehmitzsee of Nehmitzsee have similar have chemical similar and chemical trophic characteristics:and trophic thecharacteristics: LAWA trophic the LAWA state in trophic 1997 classify state in both 1997 parts classify as mesotrophic, both parts as which mesotrophic, corresponds which to corresponds the trophic referenceto the trophic state reference of the lake. state Measurements of the lake. Measurements between March between and October March 2011and October showed 2011 constant showed low phytoplanktonconstant low phytoplankton of ≤0.5 mm3/L of ≤ biovolume.0.5 mm3/L biovolume. The topography,topography, morphology morphology and and limnological limnological characteristics characteristics of the lakes, of the if known, lakes, areif summarizedknown, are insummarized Table3. in Table 3.

Figure 4. Study area of the the Rheinsberg Rheinsberg Lake Region on on 11 11 July July 1999 1999 (Landsat (Landsat 7, 7, quasi quasi-true‐true color color RGB). RGB). All lakes appear dark and are framedframed withwith whitewhite lines.lines.

Water 2017, 9, 15 7 of 31

Table 3. Overview of morphology and limnological characteristics of the lakes in Rheinsberg Lake Region.

Maximum Area Mean Depth LAWA 1998 Trophic Reference Lake Depth (m) (km2)[32] (m) [37,38] [37,39–41] State [37] [35,37–39] Breutzensee 0.10 3.5 - Eutrophic - Dagowsee 0.20 9.5 5.0 Eutrophic - Gerlinsee 0.06 5.5 - Mesotrophic - Großer Glietzensee (Ost) 0.20 13.0 - Weakly eutrophic - Großer Krukowsee 0.25 13.0 - Mesotrophic - Kleiner Krukowsee 0.08 8.5 - Mesotrophic - Nehmitzsee (north) 1.00 18.6 6.79 Mesotrophic Mesotrophic Nehmitzsee (south) 0.64 18.6 6.79 Mesotrophic Mesotrophic Peetschsee 0.89 21,0 - Mesotrophic - Plötzensee 0.06 9.0 - Weakly eutrophic - Großer Glietzensee (West) 0.16 10.0 - Weakly eutrophic - Großer Boberowsee 0.18 9.5 - Eutrophic - Großer Pälitzsee 2.49 30 - Eutrophic Mesotrophic Kleiner Glietzensee 0.17 4.0 - Eutrophic - Menowsee 0.35 4.5 - Mesotrophic - Roofensee 0.56 19.0 - Weakly eutrophic - Stechlinsee 4.14 68.0 22.8 Oligotrophic Oligotrophic

3. Materials and Methods

3.1. Satellite Data and In Situ Data Archive The multi-temporal satellite remote sensing database comprises data from Landsat 5, 7 and 8 and for 2015 also Sentinel-2 data. The database covers a time span from 1998 to 2015. The repeat cycle of the Landsat satellites is 16 days, the one of Sentinel-2 10 days, but cloud coverage reduces the number of suitable satellite images. The Landsat archives cannot provide a continuous temporal coverage. However, during the years 2003 to 2006 and from 2013 on, high temporal coverage is provided due to the temporal overlap of at least 2 satellite missions. Thus, in this study, the number of suitable Landsat acquisitions varies between 2 and 20 data sets per year with time gaps between 1 day and 160 days between the acquisitions. In Table4, we list the bandwidths of the satellites Landsat 5, 7, 8, and Sentinel 2. The visible bands are blue, green, and red and the infrared wavelengths are near-infrared (NIR) and shortwave-infrared (SWIR) 1 and 2. The bands do not overlay perfectly and there are variations in the bandwidths between the sensors. With exception of the NIR band, the old sensors have broader bandwidths: for example, the blue bandwidth ranges between 70 nm to 65 nm, the green between 80 nm and 35 nm and the red between 60 nm and 30 nm.

Table 4. Overview of the bandwidth of the satellites Landsat 5, 7, 8, and Sentinel-2.

Band Width of Band (nm) Satellite Blue Green Red NIR SWIR 1 SWIR 2 Reference Landsat 5 450–520 520–600 630–690 760–900 1550–1750 2080–2350 [42] Landsat 7 450–515 525–605 630–690 775–900 1550–1750 2090–2350 [43] Landsat 8 450–515 525–600 630–680 845–885 1560–1660 2100–2300 [43] Sentinel-2 458–523 543–578 650–680 785–900 1565–1655 2100–2280 [44]

The Feldberg Lake District region is covered by 200 Landsat images (60 Landsat 5, 115 Landsat 7, 25 Landsat 8; tiles 193023 and 194023) and by two Sentinel-2 images in 2015. The Klocksin Lake Chain and Rheinsberg Lake Region are covered by additional three Landsat images (one Landsat 7 and two Landsat 8 images, all on tile 194023). The data archive is illustrated in Figure5. All the data sets were obtained in the form of orthorectiﬁed standard data products to reduce preprocessing efforts. The Landsat images are delivered by U.S. Geological Survey (USGS) as surface reﬂectance products (including atmospheric correction) [45]. The Sentinel-2 satellite images of region Water 2017, 9, 15 8 of 31

Water 2017, 9, 15 8 of 31 Feldberg Lake District were provided by ESA in processing level L1C [46]. We did the further preprocessingreflectance products (resampling (including and atmospheric atmospheric correction) correction) [45]. with The sen2cor Sentinel (version‐2 satellite 2.2.1) images and of Sentinel-2 region ToolboxFeldberg (version Lake 3.0)District provided were byprovided ESA [47 by,48 ].ESA The in spatial processing resolution level of L1C the [46]. data setsWe did ranges the from further 30 m forpreprocessing the Landsat sensors (resampling to 10 and m for atmospheric the Sentinel correction) 2 sensors. with sen2cor (version 2.2.1) and Sentinel‐2 ToolboxAs not (version all lakes 3.0) are provided completely by cloud-free,ESA [47,48]. the The further spatial preprocessing resolution of the included data sets the ranges cloud andfrom cloud 30 m for the Landsat sensors to 10 m for the Sentinel 2 sensors. shadow masking. If not noted differently, all processing was implemented and performed in the As not all lakes are completely cloud‐free, the further preprocessing included the cloud and free software R (version 3.2.2). The clouds and cloud shadow were removed based on the cloud and cloud shadow masking. If not noted differently, all processing was implemented and performed in cloud shadow classifications that are provided with the data. The cloud and cloud shadow masks of the free software R (version 3.2.2). The clouds and cloud shadow were removed based on the cloud Landsat mask generously, thus, at the Feldberg Lake District, we only use the cloud mask with a high and cloud shadow classifications that are provided with the data. The cloud and cloud shadow masks confidenceof Landsat and mask check generously, for cloud thus, shadows at the manually Feldberg to Lake keep District, the density we only of the use time the cloud series mask in accordance with a to thehigh in confidence situ data. Theand Sentinel-2check for cloud shadowshadows classification manually to failskeep over the lakes,density thus, of the the time images series have in to beaccordance checked manually. to the in situ data. The Sentinel‐2 cloud shadow classification fails over lakes, thus, the imagesSince have 1998, to there be checked are regular manually. water quality measurements at FH, BL and SL by the Leibniz-Institute of FreshwaterSince 1998, Ecology there are & Inlandregular Fisheries. water quality Besides measurements precipitated at FH, CaCO BL 3and, Chlorophyll SL by the Leibniz a (chl-a),‐ − 2− − 2− + + 2+ temperature,Institute of Freshwater pH, alkalinity, Ecology and & ionInland concentrations Fisheries. Besides (NO precipitated3 , SiO3 ,CaCO Cl ,3 SO, Chlorophyll4 , Na ,K a (chl, Mg‐a), , 2+ andtemperature, Ca ) are measured. pH, alkalinity, Based and on ion those concentrations parameters, (NO the3 CaCO−, SiO332−,saturation Cl−, SO42−, indexNa+, K (SI)+, Mg was2+, and calculated Ca2+) accordingare measured. to Debye-Hückel Based on those [49 ]parameters, using “WinIAP—Software the CaCO3 saturation for the index Calculation (SI) was calculated of Ion Activities according and Calciteto Debye Saturation‐Hückel Index” [49] using [50]. “WinIAP—Software The SI shows, depending for the on Calculation the trophic of state Ion andActivities the season, and Calcite if calcite precipitationSaturation isIndex” possible: [50]. theThe SI SI threshold shows, depending for calcite on precipitation the trophic in state oligotrophic and the season, lakes is if <5 calcite and in mesotrophicprecipitation lakes is possible: between the 5 and SI threshold 15. In eutrophic for calcite lakes, precipitation SI values ofin >15oligotrophic without calcitelakes is precipitation <5 and in aremesotrophic possible [6]. lakes In spring, between the 5 thresholds and 15. In eutrophic are generally lakes, higher SI values than of in >15 summer without due calcite to the precipitation inhibition of calciteare possible precipitation [6]. In by spring, phosphate the thresholds [6]. are generally higher than in summer due to the inhibition of calcite precipitation by phosphate [6]. The locations of measurement stations in the lakes are marked in Figure2. In FH, CaCO 3 is always The locations of measurement stations in the lakes are marked in Figure 2. In FH, CaCO3 is measured in the northern part of the lake (Figure2, ∆1) as the other parameters before 2011. Since 2011, always measured in the northern part of the lake (Figure 2, Δ1) as the other parameters before 2011. the other parameters are measured at another location further south (Figure2, ∆2). All parameters are Since 2011, the other parameters are measured at another location further south (Figure 2, Δ2). All measured in 0–5 m water depth and multiple measurements versus depth are averaged. parameters are measured in 0–5 m water depth and multiple measurements versus depth are averaged. On 68 days there are water quality measurements contemporary to Landsat images, but not at On 68 days there are water quality measurements contemporary to Landsat images, but not at everyevery time time all all three three lakes lakes are are sampled. sampled. “Contemporary” “Contemporary” in this this study study means means that that the the Landsat Landsat images images areare not not acquired acquired more more than than 3 3 days days before before andand notnot more than than 5 5 days days after after the the in in situ situ measurement. measurement. TheseThese thresholds thresholds are are set set under under the the assumption assumption thatthat calcite precipitation precipitation events events appear appear more more suddenly suddenly thanthan they they vanish. vanish.

FigureFigure 5. 5.Time Time series series ofof Landsat acquisitions acquisitions (1998–2015), (1998–2015), sorted sorted by tile by tile(cf. Figure (cf. Figure 1) and1) sensor. and sensor. “In “Insitu” situ” marks marks the the date date of of in in situ situ measurements measurements at Feldberg at Feldberg Lake Lake District. District.

3.2. Classification of Calcite Precipitation Using Satellite Imagery 3.2. Classiﬁcation of Calcite Precipitation Using Satellite Imagery The processing chain is of the classification is illustrated in Figure 6. TheInput processing data are chain the preprocessed is of the classiﬁcation Landsat and is Sentinel illustrated images. in Figure We manually6. digitized regions of interestInput (ROI) data arefor the the extraction preprocessed of lake Landsat spectra. and For Sentinel the three images. lakes of Feldberg We manually Lake District digitized we regions choose of interestup to (ROI)13 ROI for per the lakes extraction to avoid of gaps lake due spectra. to clouds, For the cloud three shadow lakes of and Feldberg the edge Lake of the District tile 194023 we choose to upkeep to 13 the ROI density per lakes of the to satellite avoid gaps data. due Thus, to clouds,those ROI cloud are shadowunevenly and distributed the edge in of the the lakes. tile 194023 At the to keepother the lakes density we selected of the satellite one ROI data. per lake Thus, and those the ROI ROI are are located unevenly in the distributed centers of inthe the lakes. lakes. All AtROI the other lakes we selected one ROI per lake and the ROI are located in the centers of the lakes. All ROI are

Water 2017, 9, 15 9 of 31

Water 2017, 9, 15 9 of 31 selectedare selected with distance with distance to islands to islands or shallow or shallow water water areas, areas, which which could could inﬂuence influence the the reﬂectance reflectance spectra. Dependingspectra. on Depending the lakes, on the the ROI lakes, are the differently ROI are differently sized and sized shaped. and shaped.

FigureFigure 6. Flowchart 6. Flowchart of theof the processing processing steps steps ofof LandsatLandsat surface surface reflectance reflectance data data for forthe themonitoring monitoring of of calcitecalcite precipitation. precipitation. The The blue blue boxes boxes contain contain inputinput data, data, the the gray gray boxes boxes illustrate illustrate the thedevelopment development of of the robustthe robust classification classification and and the the orange orange boxes boxes illustrateillustrate the the classification classification and and validation. validation.

Based on the ROI we extracted the reflectance values of the satellite images. The reflectance Basedvalues are on evaluated: the ROI we if the extracted standard the deviation reflectance in any valuesband is ofhigh the (>100) satellite or if more images. than The 10% reflectanceof the valuesROI are consist evaluated: of not applicable if the standard (NA) values, deviation the next in anyROI bandis tested is or, high if no (>100) next ROI or if is more available, than the 10% of the ROIreflectance consist values of not of applicable the lake are (NA) set NA. values, Then, the the next extracted ROI is reflectance tested or, values if no next of each ROI band is available, are the reflectanceaveraged to values get the ofmean the reflectance lake are set spectrum NA. Then, of the the lake. extracted reflectance values of each band are averagedThe to get next the step mean is the reflectance calculation spectrum and evaluation of the lake.of the spectral indices based on the extracted Themean next reflectance. step is theThe calculation spectral indices and are evaluation described of in the Table spectral 5. indices based on the extracted mean reflectance. The spectral indices are described in Table5. Table 5. Overview of the spectral indices.

Index Name Table 5.AbbreviationOverview of the spectral indices.Formula Reference Ratio of the reflectance (Ref.) values of Ref Ref Ratio RG Ratio RG [51] band red and green Ref Ref Index Name Abbreviation Formula Reference Ref Ref Normalized difference water index NDWI NDWI [52] Ratio of the reflectance (Ref.) values of RefRefredRe− Reffgreen Ratio RG Ratio RG = + [51] bandModified red and normalized green difference RefRefRered fRefSWIR1green MNDWI MNDWI [53] water index RefRef Re− ReffSWIR1 Normalized difference water index NDWI NDWI = green NIR [52] “Area Blue Green Red” as the triangular Refgreen+ RefNIR Area BGR 0.5482 ∗ Ref 560 ∗ Modifiedarea in normalizedthe reflectance difference values of blue, Refgreen−Ref(SWIR1) MNDWIArea BGR RefMNDWI 655 ∗= Ref 560 ∗ Ref [54][ 53] watergreen index and red (using the central Refgreen+Ref(SWIR1) 655 ∗ Ref 480 ∗ Ref wavelength of Landsat 8 for all sensors) “Area Blue Green Red” as the triangular Area BGR = 0.5(Re482f∗ Refgreen + areaNormalized in the reflectance absorption values feature of blue, NAFD 1 AreaNAFD BGR 560 ∗ Refred + 655Re∗fRefblue − 560 ∗ Refblue − [55][ 54] greendepth and of red red (using the central __ 655 ∗ Ref −/Area480 ∗ Ref ) wavelength of Landsat 8 for all sensors) green ,,red Ratio of bands of an unknown lake and __ Ratio RL Ratio RL This work Normalizeda dark reference absorption lake e.g., feature lake SL Refred _ _ NAFD NAFD = 1 − Areagreen,red,NIR [55] depth of red Refcontinumline_of_red

Ratio of bands of an unknown lake and − Ratio RL Ratio RL = Refunknown_lake Refreference_lake This work a dark reference lake e.g., lake SL Refunknown_lake+ Refreference_lake Water 2017, 9, 15 10 of 31

The evaluation of the spectral indices is based on the visual classification of “good quality images” of lake FH: we checked visually the image quality and marked images with a low quality. Reasons for low image quality are large cloud coverage, ice and low incidence angles of the sun in winter. Of the 200 images of Feldberg Lake District 114 have a good image quality. Then, greenish-turquoise colored lakes in a quasi-true color Red Green Blue (RGB) image are classified visually as lakes with calcite precipitation. Finally, each index is separated in two groups, dates with and without calcite precipitation, and the groups are compared. The validation of the satellite-derived classification results of FH, BL and SL is based on the in situ measured CaCO3 concentrations using confusion matrices. Therefore, the dates with in situ CaCO3 concentrations were classified as “calcite precipitation” and “no calcite precipitation”. Additionally, we validate the results based on a visual classification in quasi-true color images. At Feldberg Lake District the visual classification itself is validated using the in situ CaCO3 concentrations. In the other two lake regions without in situ measurements, the satellite-derived calcite precipitation events are only validated with the visual classification in quasi-true color images. Instead of confusion matrices for every lake, we summarize all results of each region and show each region in one confusion matrix. The confusion matrices compare true results, in this study the grouped in situ measurements or visual classification results, with predicted results, in this study the satellite-derived classification results. True positive (TP) and true negative (TN) are accurate classification results, where predicted and true results are equal. False positive (FP) are dates in which the true results show no calcite precipitation, but the satellite-derived calcite precipitation shows calcite precipitation. It means the classification overestimated the number of calcite precipitation. False negative (FN) are dates in which the true results shows calcite precipitation, but the satellite-derived calcite precipitation shows no calcite precipitation. It means the classification underestimated the number of calcite precipitation. The accuracy is calculated by dividing the sum of TP and TN by the sum of TP, TN, FP and FN. Relevant for the validation with confusion matrices is the number of Landsat images with contemporary in situ measurements, the threshold for the grouping of the in situ CaCO3 concentrations on dates with “calcite precipitation” and “no calcite precipitation”, and the relation of dates with calcite precipitation to dates without. FH has 46 dates with Landsat images and field measurements and their CaCO3 concentration ranges from 0 to 3.52 mg/L. BL has 48 dates with Landsat images and field measurements. The CaCO3 concentration ranges from 0.05 to 2.96 mg/L. SL has 31 dates with Landsat images and field measurements and their CaCO3 concentration is always low between 0 and 0.47 mg/L. The number of dates with calcite precipitation varies depending on the CaCO3 that is considered as calcite precipitation. A previous study found that without optical tools, calcite precipitation in the open water is only visible during high calcite concentrations with >1 mg/L [3]. Thus, first, we consider all dates with CaCO3 concentration ≥1 mg/L as dates with calcite precipitation. Then, we lower the threshold by 0.1 mg/L steps down to ≥0.5 CaCO3 mg/L, because we suspect a higher sensitivity of the satellite images to calcite precipitation. As algal blooms are potentially a source for misclassification, we also analyze the lake spectra with high chl-a concentration ≥20 µg/L. The occurrence of considerable algal blooms are related to an chl-a concentration of at least 20 µg/L [56].

4. Results

4.1. In Situ Measurements

The time series of CaCO3 and SI of the three lakes is illustrated in Figure7. FH had low CaCO 3 concentrations of <1 mg/L before 2006, but exceeded 1 mg/L ten times between 2006 and 2015. BL’s CaCO3 concentrations exceeded 1 mg/L twelve times between 1998 and 2015 and SL remained always <0.5 mg/L CaCO3. The SI values of the three lakes range between 0.7 and 13.7 between 2000 and 2015. Based on their SI values and trophic states, in SL and BL calcite precipitation could have occurred during the monitoring period. The trophic state of FH changes during the monitoring period: before Water 2017, 9, 15 11 of 31

2011 with eutrophic condition no calcite precipitation is possible, after 2011 the SI values indicate possible calciteWater 2017 precipitation, 9, 15 events. The chl-a concentration of SL ranges from 1 to 10 µg/L11 of 31 (average: 3 µg/L) and of BL from 1 to 18 µg/L (average: 3 µg/L). FH has high variation in its chl-a concentration: possible calcite precipitation events. The chl‐a concentration of SL ranges from 1 to 10 μg/L (average: between3 1998 μg/L) and and of 2002 BL from its chl-a1 to 18 μ concentrationg/L (average: 3 μ rangesg/L). FH fromhas high 4 variation to 21 µg/L in its (average:chl‐a concentration: 9 µg/L), then betweenbetween 2005 and 1998 28 and March 2002 2011its chl it‐a rangesconcentration from ranges 5 to 53 fromµg/L 4 to (average: 21 μg/L (average: 19 µg/L). 9 μg/L), After then this chl-a maximum,between the concentration 2005 and 28 March declines 2011 again it ranges to 2–17 fromµ 5g/L to 53 (average: μg/L (average: 8 µg/L). 19 μ Ong/L). twelve After this dates chl FH’s‐a chl-a maximum, the concentration declines again to 2–17 μg/L (average: 8 μg/L). On twelve dates FH’s concentrationWater 2017 exceeds, 9, 15 20 µg/L and thereof, ﬁve dates exceed 30 µg/L (23 March 2005,1119 of 31 April 2005, chl‐a concentration exceeds 20 μg/L and thereof, five dates exceed 30 μg/L (23 March 2005, 19 April 30 May 2008, 30 March 2009, and 28 March 2011). 2005,possible 30 May calcite 2008, precipitation 30 March 2009,events. and The 28 chl March‐a concentration 2011). of SL ranges from 1 to 10 μg/L (average: 3 μg/L) and of BL from 1 to 18 μg/L (average: 3 μg/L). FH has high variation in its chl‐a concentration: between 1998 and 2002 its chl‐a concentration ranges from 4 to 21 μg/L (average: 9 μg/L), then between 2005 and 28 March 2011 it ranges from 5 to 53 μg/L (average: 19 μg/L). After this chl‐a maximum, the concentration declines again to 2–17 μg/L (average: 8 μg/L). On twelve dates FH’s chl‐a concentration exceeds 20 μg/L and thereof, five dates exceed 30 μg/L (23 March 2005, 19 April 2005, 30 May 2008, 30 March 2009, and 28 March 2011).

Figure 7. Time series (1998 to 2015) of (a) in situ measured CaCO3 concentrations (mg/L); and (b) the Figure 7. Time series (1998 to 2015) of (a) in situ measured CaCO3 concentrations (mg/L); and (b) the calculated CaCO3 saturation index (SI) in FH, BL, SL lakes. calculated CaCO3 saturation index (SI) in FH, BL, SL lakes. 4.2. Calcite Precipitation Visible in Lake Reflectance 4.2. Calcite PrecipitationThe methodological Visible indevelopments Lake Reflectance aim at the automated multi‐temporal mapping of calcite Theprecipitation methodologicalFigure 7. based Time serieson developments satellite (1998 to remote 2015) of sensing aim(a) in situ at time themeasured series automated CaCO data.3 concentrationsThus, multi-temporal the approach (mg/L); andneeds mapping (b) tothe be able of calcite to identifycalculated calcite CaCO precipitation3 saturation indexoccurring (SI) in atFH, different BL, SL lakes. times during the analyzed time span, whereas precipitationthe determination based on satellite of the time remote of calcite sensing precipitation time series occurrence data. Thus, depends the on approach the length needs of the time to be able to identify calciteperiod4.2. Calcite between precipitation Precipitation two subsequent occurringVisible in imagesLake at Reflectance different contained times in the remote during sensing the analyzed time series time database. span, whereas the determinationFigureThe of themethodological 8 compares time of calcite two developments quasi precipitation‐true aimcolor at Landsat occurrencethe automated RGB dependsimages, multi‐ temporalone on with the mapping length[28] and ofof the thecalcite other time period betweenwithout twoprecipitation subsequent calcite based precipitation, images on satellite containedwith remote photos sensing from in the timecontemporary remote series data. sensing field Thus, campaigns. time the approach series Figure database. needs 9 shows to be ablea time Figureseriesto8 identify compares of five calcite Landsat two precipitation quasi-trueRGB images occurring color in which Landsatat different BL has RGB differenttimes images, during CaCO the one3 concentrations.analyzed with [ time28] span, and The the whereasaccording other without meanthe determination reflectance spectra of the and time the of calculation calcite precipitation of Area BGR occurrence are shown depends in Figure on the 10. lengthOn 12 ofJuly the 1999 time the calcite precipitation, with photos from contemporary field campaigns. Figure9 shows a time series CaCOperiod3 concentration between two subsequentof BL was very images high contained with 2.00 in mg/Lthe remote and the sensing lake istime opaque series turquoisedatabase. colored. of five LandsatUntil 15Figure RGBSeptember 8 imagescompares 1999 in thetwo which CaCO quasi3 ‐ BLconcentrationstrue has color different Landsat decreased RGB CaCO images,to3 0.84concentrations. mg/L,one with but in[28] the and TheLandsat the according other image mean reflectancethewithout spectracolor changecalcite and precipitation, is the still calculationclearly with visible. photos of On Area from 11 October contemporary BGR are1999 shown the field calcite campaigns. in precipitation Figure Figure 10. 9diminished Onshows 12 a time July and 1999 the CaCOseries3 concentrationof five Landsat was RGB very images low inwith which 0.09 BLmg/L has and different BL appears CaCO3 dark concentrations. again. The Theonly according exceptions CaCO3 concentration of BL was very high with 2.00 mg/L and the lake is opaque turquoise colored. aremean lakes reflectance with separated spectra lakeand thebasins calculation with narrow of Area passages: BGR are shownOn 13 Septemberin Figure 10. 1999 On 12the July northeastern 1999 the Until 15 September 1999 the CaCO3 concentrations decreased to 0.84 mg/L, but in the Landsat image basinCaCO of3 BLconcentration is darker than of BL the was main very basin high (Figure with 2.00 9d). mg/L and the lake is opaque turquoise colored. the color changeUntil 15 September is still clearly 1999 the visible. CaCO3 concentrations On 11 October decreased 1999 to the 0.84 calcite mg/L, but precipitation in the Landsat diminished image and CaCO3 concentrationthe color change was is still very clearly low visible. with 0.09On 11 mg/L October and 1999 BL the appears calcite precipitation dark again. diminished The only and exceptions are lakes withCaCO separated3 concentration lake was basins very low with with narrow 0.09 mg/L passages: and BL appears On 13dark September again. The only 1999 exceptions the northeastern are lakes with separated lake basins with narrow passages: On 13 September 1999 the northeastern basin of BLbasin is darker of BL is thandarker the than main the main basin basin (Figure (Figure9 9d).d).

Figure 8. Two quasi‐true color Landsat RGB of Stechlinsee with photos showing the water surface taken from a boat: (a) August 2011, showing the lake with calcite precipitation; and (b) September 2015, with clear water. The photos were taken: (a) two days before; and (b) nine days after the Landsat acquisition.

Figure 8. Two quasi‐true color Landsat RGB of Stechlinsee with photos showing the water surface taken Figure 8. Twofrom quasi-truea boat: (a) August color 2011, Landsat showing RGB the oflake Stechlinsee with calcite precipitation; with photos and showing (b) September the water 2015, with surface taken from a boat:clear (a water.) August The photos 2011, were showing taken: ( thea) two lake days with before; calcite and (b) precipitation; nine days after the and Landsat (b) September acquisition. 2015, with

clear water. The photos were taken: (a) two days before; and (b) nine days after the Landsat acquisition. Water 2017, 9, 15 12 of 31 Water 2017, 9, 15 12 of 31

Figure 9. Quasi‐true color RGB Landsat 7 images of the Feldberg Lake District. The extent is the same Figure 9. Quasi-true color RGB Landsat 7 images of the Feldberg Lake District. The extent is the as in Figure 2. All lakes are framed with lines and the positions of in situ measurements are marked same as in Figure2. All lakes are framed with lines and the positions of in situ measurements are with white triangles. The orange rectangle shows the ROI that was used for the extraction of the marked with white triangles. The orange rectangle shows the ROI that was used for the extraction of reflectance spectra. BL is turquoise colored on 11 July 1999 (a) due to calcite precipitation. In the the reﬂectance spectra. BL is turquoise colored on 11 July 1999 (a) due to calcite precipitation. In the following, calcite precipitation diminishes from 3 August 1999 to 13 September 1999 (b–d), and on 15 following, calcite precipitation diminishes from 3 August 1999 to 13 September 1999 (b–d), and on October 1999 (e) BL appears dark again. 15 October 1999 (e) BL appears dark again.

The precipitated CaCO3 particles cause a decrease in Secchi depth and an increase of the reflectanceThe precipitated [2,4]. According CaCO 3toparticles Thiemann cause and Koschel a decrease the inadditive Secchi effect depth of calcite and an precipitation increase of is the reflectanceuniform in [2 the,4]. visible According (RGB) toand Thiemann near‐infrared and wavelength Koschel the range, additive so that effect spectral of calcite characteristic precipitation like is uniformabsorption in thebands visible and (RGB)reflection and maxima near-infrared are kept wavelength [2]. Figure range,10 illustrates so that the spectral enhanced characteristic reflectance like absorptionvalues mainly bands between and reflection blue and maxima NIR caused are by kept calcite [2]. precipitation. Figure 10 illustrates However, the the enhanced increase varies reflectance by valueswavelength: mainly betweenthe green blue band and shows NIR the caused strongest by calcite increase precipitation. and has the However, maximum the reflectance increase values. varies by wavelength:The reference the greenreflectance band spectrum shows the without strongest calcite increase precipitation and has theon maximum15 October reflectance 1999 has values.low Thereflectance reference reflectancevalues with spectrum a maximum without blue calcite band. precipitation Even though on the 15 October reflectance 1999 spectra has low of reflectance calcite valuesprecipitation with a maximum show higher blue NIR band. and EvenSWIR though reflectance, the reflectancean analysis spectraabove 800 of nm calcite is not precipitation recommended show as the absorption of clear water superimposes the effect of water components [1,2,11,25]. Figure 11 higher NIR and SWIR reflectance, an analysis above 800 nm is not recommended as the absorption of illustrates the variation of lake reflectance spectra with and without calcite precipitation. In the clear water superimposes the effect of water components [1,2,11,25]. Figure 11 illustrates the variation boxplots all mean reflectance spectra of good quality images of BL, SL and FH are combined. With of lake reflectance spectra with and without calcite precipitation. In the boxplots all mean reflectance low quality images, the ranges would be even higher. In Figure 11 the mean reflectance values of spectra of good quality images of BL, SL and FH are combined. With low quality images, the ranges NIR, SWIR 1, and SWIR 2 do not indicate regular increase of the reflectance, as it could be suspected wouldby the be selected even higher. reflectance In Figure spectra 11 in the Figure mean 10. reflectance values of NIR, SWIR 1, and SWIR 2 do not indicateThere regular are increasetwelve dates of the with reflectance, high chl‐ asa concentration it could be suspected ≥20 μg/L by at FH. the selectedOf the twelve reflectance dates spectrawith inhigh Figure chl 10‐a. concentration, only two show a green color: on 12 April 2005 FH appears greenish in the quasiThere‐true are color twelve RGB datesLandsat with 12 April high 2005 chl-a and concentration on 1 June 2008≥20 FHµ appearsg/L at FH.green Of bright. the twelve However, dates with1 June high 2008 chl-a has concentration, in addition to only its high two chl show‐a concentration a green color: a calcite on 12 precipitation April 2005 FH event appears with greenisha high inCaCO the quasi-true3 concentration color RGBof 3.4 Landsat mg/L. The 12 Apriltwo lake 2005 spectra and on of 1 JuneFH with 2008 greenish/green FH appears green color bright.are However,characterized1 June by 2008 a steeperhas in increase addition from to its blue high to chl-a green concentration reflectance and a calcite a (small) precipitation peak in green. event The with a highreflectance CaCO values3 concentration of red and ofNIR 3.4 are mg/L. equally The high. two lake spectra of FH with greenish/green color are characterized by a steeper increase from blue to green reflectance and a (small) peak in green. The reflectance values of red and NIR are equally high.

Water 2017, 9, 15 13 of 31 Water 2017, 9, 15 13 of 31

Water 2017, 9, 15 13 of 31

Figure 10. Reflectance spectra of BL with calcite precipitation (11 July 1999), diminishing calcite FigureFigure 10. 10.Reﬂectance Reflectance spectra spectra ofof BLBL withwith calcite precipitation (11 (11 July July 1999), 1999), diminishing diminishing calcite calcite precipitation (3 August 1999 to 13 September 1999), and without calcite precipitation (15 October precipitationprecipitation (3 August(3 August 1999 1999 to 13to September13 September 1999), 1999), and and without without calcite calcite precipitation precipitation (15 (15 October October 1999). 1999). The ROI that has been used for the extraction and calculation of the mean reflectance spectra The1999). ROI The that ROI has that been has used been for used the extraction for the extraction and calculation and calculation of the meanof the reﬂectancemean reflectance spectra spectra of BL is of BL is marked in Figure 7. The transparent triangles and colored numbers illustrate the “Area BGR”. markedof BL is in marked Figure 7in. TheFigure transparent 7. The transparent triangles triangles and colored and colored numbers numbers illustrate illustrate the “Area the “Area BGR”. BGR”.

FigureFigure 11. 11.Reflectance Reflectance values values in in all all spectral spectral bandsbands of Breiter Luzin, Luzin, Schmaler Schmaler Luzin Luzin and and Feldberger Feldberger FigureHausseeHaussee 11. of Reflectance goodof good quality quality values images. images. in Allall threespectral All lakesthree bands are lakes visually of Breiterare classifiedvisually Luzin, classified asSchmaler turquoise as Luzin (calciteturquoise and precipitation) Feldberger (calcite Hausseeorprecipitation) dark (no of calcite good or dark precipitation).quality (no calciteimages. precipitation). All three lakes are visually classified as turquoise (calcite precipitation) or dark (no calcite precipitation). 4.3. Best Performing Spectral Indices 4.3. Best Performing Spectral Indices 4.3. BestThe Performing evaluation Spectral of the Indices green reflectance and the spatial indices of FH is illustrated in Figure 12. The evaluation of the green reflectance and the spatial indices of FH is illustrated in Figure 12. AreaThe BGR evaluation has clearly of thethe best green distinction reflectance between and thethe spatialtwo cases indices with a of 74% FH increase is illustrated from 3rd in quantileFigure 12. Area BGR has clearly the best distinction between the two cases with a 74% increase from 3rd quantile Areaof “no BGR calcite has clearly precipitation” the best todistinction the 1st quantile between of the the “yes two calcite cases withprecipitation” a 74% increase boxplot. from The 3rd next quantile best of “no calcite precipitation” to the 1st quantile of the “yes calcite precipitation” boxplot. The next3 best of “nois the calcite reflectance precipitation” of green bandto the with 1st quantile 19% increase. of the We “yes selected calcite aprecipitation” conservative thresholdboxplot. The 13 × next 10 asbest is the reflectance of green band with 19% increase. We selected a conservative threshold 13 × 103 as is thethe maximumreflectance value of green of dates band without with 19% calcite increase. precipitation We selected of Area a BGR. conservative threshold 13 × 103 as the maximum value of dates without calcite precipitation of Area BGR. the maximum value of dates without calcite precipitation of Area BGR.

Water 2017, 9, 15 14 of 31 Water 2017, 9, 15 14 of 31 Water 2017, 9, 15 14 of 31

Figure 12. Boxplots of the reflectance of green and the indices of FH of good quality images. The FigureFigure 12. BoxplotsBoxplots ofof the the reflectance reflectance of of green green and and the indicesthe indices of FH of of FH good of qualitygood quality images. images. The indices The indices are separated into dates with calcite precipitation (“Yes”) and without calcite precipitation indicesare separated are separated into dates into with dates calcite with precipitationcalcite precipitation (“Yes”) (“Yes”) and without and without calcite precipitationcalcite precipitation (“No”) (“No”) based on the visual classification of FH. Index “Area BGR” has the best separation of the two (“No”)based on based the visualon the classification visual classification of FH. Indexof FH. “Area Index BGR” “Area has BGR” the has best the separation best separation of the two of the classes: two classes: the red line marks the conservative threshold 13 × 103 as the maximum value of dates without classes:the red the line red marks line marks the conservative the conservative threshold threshold 13 × 1310 ×3 10as3 as the the maximum maximum value value of of datesdates without calcite precipitation. calcitecalcite precipitation. The calculation of Ratio RL based on SL failed as soon as no reflectance values of the lake could The calculation of Ratio RL based on SL failed as soon as no reflectance values of the lake could be extracted,The calculation because of the Ratio area RL of based the onlake SL was failed masked as soon out as noin reflectancethe satellite values images of thedue lake to couldcloud be extracted, because the area of the lake was masked out in the satellite images due to cloud becoverage. extracted, It was because also thenot area possible of the to lake derive was a masked universal out reference in the satellite spectrum, images because due to of cloud the variation coverage. coverage. It was also not possible to derive a universal reference spectrum, because of the variation Itof wasreflectance also not values possible of to SL. derive Figure a universal 13 illustrates reference the variation spectrum, of because the mean of spectra the variation of SL, of grouped reflectance by of reflectance values of SL. Figure 13 illustrates the variation of the mean spectra of SL, grouped by valuesthe satellites. of SL. FigureThere 13are illustrates large variations the variation within of the meanbands, spectra especially of SL, in grouped NIR. Additionally, by the satellites. the the satellites. There are large variations within the bands, especially in NIR. Additionally, the Therereflectance are large values variations between within the satellites the bands, vary: especially between in blue NIR. and Additionally, NIR, Landsat the 8 reflectance has significantly values reflectance values between the satellites vary: between blue and NIR, Landsat 8 has significantly betweenlower reflectance the satellites values vary: than between Landsat blue 5, andand NIR, the reflectance Landsat 8 has values significantly of Landsat lower 7 range reflectance between values the lower reflectance values than Landsat 5, and the reflectance values of Landsat 7 range between the thantwo other Landsat satellites. 5, and the reflectance values of Landsat 7 range between the two other satellites. two other satellites.

Figure 13. Variation of the mean lake spectra of lakelake SLSL betweenbetween 19981998 andand 2015.2015. The spectra are Figure 13. Variation of the mean lake spectra of lake SL between 1998 and 2015. The spectra are grouped by the Landsat satellites. grouped by the Landsat satellites.

Water 2017, 9, 15 15 of 31

4.4. Validation of Landsat-Derived Calcite Precipitation The satellite-derived calcite precipitation of each region was validated using confusion matrices. Table6 illustrated the number of accurate classification (TP and TF) and misclassified (FN and FP) results at Feldberg Lake District in comparison to in situ measurements of CaCO3 concentration. The classification accuracy depends on the choice of threshold in CaCO3 concentration: the higher the CaCO3 threshold is set, the higher is the number of FP results and the smaller is the number of FN results. The best accuracy of 0.88 has the comparison with the ≥0.7 mg/L CaCO3. In that case, FN is six (three at FH and three at BL) and the number of FP is nine (all at BL). The nine false positive dates have (slightly) increased CaCO3 concentrations between 0.41 and 0.69 mg/L.

Table 6. Confusion matrices of Landsat-derived calcite precipitation with in situ measurements at Feldberg Lake District.

CaCO3 Concentration (mg/L) TP TN FN FP Sum Accuracy ≥1 84 21 4 16 125 0.84 ≥0.9 82 24 6 13 125 0.848 ≥0.8 82 26 6 11 125 0.864 ≥0.7 82 28 6 9 125 0.88 ≥0.6 78 31 10 6 125 0.872 ≥0.5 71 33 17 4 125 0.832

The accuracy of the visual classification is 0.85 using the threshold ≥0.7 mg/L CaCO3: the sum is 124, with 75 TP, 31 TN, three FN and 16 FP results. Table7 illustrates the accuracies of Feldberg Lake District, Klocksin Lake Chain and Rheinsberg Lake Region using confusion matrices with visual classifications. The accuracy in the Feldberg Lake District is high with 0.94, but with 28 FN and four FP classification results. The FN results occur at all three lakes (FH: 14, BL: 9, and SL: 5), the FN results only at FH (3) and BL (1). The accuracy in the Klocksin Lake Chain is very high with 0.99, with only two FP (at FS) and two FN classification results (at FS and HS). The accuracy in Rheinsberg Lake Region is also high with 0.97; however, there are 54 false negative results. The FN results are two times at Breutzensee and kleiner Glietzensee, four times at Dagowsee and Großer Boberowsee, six times at Stechlinsee, eight times at Großer Pälitzsee and Roofensee, and 20 times at Menowsee. Even though several calcite precipitation events are missed, the lakes are still classified as lakes with calcite precipitation at other dates during our monitoring period. Extraordinary is the bright green color of Kleiner Glietzensee in March 2014 in the quasi-true color Landsat. Those two dates have been classified as calcite precipitation visually and automatically via Area BGR, but there is no in situ data as evidence available.

Table 7. Confusion matrices of Landsat-derived calcite precipitation with visual classiﬁcations at Feldberg Lake District, Klocksin Lake Chain, and Rheinsberg Lake Region.

Region TP TN FN FP Sum Accuracy Feldberg Lake District 429 115 28 4 576 0.94 Klocksin Lake Chain 408 29 2 2 441 0.99 Rheinsberg Lake Region 1862 52 54 0 1968 0.97

4.5. Frequency and Duration of Landsat-Derived Calcite Precipitation Based on the Landsat classiﬁcation results, we analyzed the frequency and duration of calcite precipitation. The results for each region are illustrated in Figure 14. A table that lists the classiﬁcation results of all lakes and dates can be found in the AppendixA (Table A1). Water 2017, 9, 15 16 of 31 Water 2017, 9, 15 16 of 31

Figure 14. Frequency and duration of Landsat‐derived calcite precipitation events at the study areas Figure 14. Frequency and duration of Landsat-derived calcite precipitation events at the study areas (1998–2015, cf. Appendix A). Images with calcite precipitation are illustrated as bars. (a) The Feldberg (1998–2015,Lake cf. District Appendix with FH,A). BL, Images and SL; with (b) the calcite Klocksin precipitation Lake Chain arewith illustrated Flacher See as (FS), bars. Tiefer ( a See) The (TS), Feldberg Lake Districtand Hofsee with (HS); FH, BL,and ( andc) the SL; Rheinsberg (b) the Klocksin Lake Region Lake with Chain Menowsee with (MS), Flacher Roofensee See (FS), (RS), Tiefer Kleiner See (TS), and HofseeKrukowsee (HS); and(KK), ( cKleiner) the Rheinsberg Glietzensee Lake(KG), RegionDagowsee with (DS), Menowsee Stechlinsee (MS),(SS), Großer Roofensee Boberowsee (RS), Kleiner Krukowsee(GB), (KK), Breutzensee Kleiner (BS) Glietzensee and Großer (KG), Paelitzersee Dagowsee (GP). (DS), When Stechlinsee more than (SS), one Großer lake shows Boberowsee calcite (GB), Breutzenseeprecipitation (BS) and at Großerthe same Paelitzersee date, the bars (GP).of the Whenlakes are more stacked. than one lake shows calcite precipitation at the4.5.1. same Feldberg date, the Lake bars District of the lakes are stacked. In all lakes in Feldberg Lake District calcite precipitation events are detected (Figure 14a). All 4.5.1. Feldberg Lake District events occur between May and end of September. In FH regular calcite precipitation first occurred in In2005. all lakes Between in Feldberg2005 and Lake2010 FH District had regular calcite calcite precipitation precipitation events events. are In detected 2011 no (Figurecalcite 14a). All eventsprecipitation occur between occurred, May but and in the end following of September. three years In FHregular regular calcite calcite precipitation precipitation reoccurred. first In occurred 2015, there was only a single calcite precipitation event. BL has calcite precipitation every year except in 2005.in Between2001, but this 2005 year and only 2010 had two FH Landsat had regular acquisitions calcite on 13 precipitation May 2001 and events.20 October In 2001. 2011 At noSL calcite precipitationone event occurred, was detected but in on the the following Landsat image three on years 13 August regular 2014. calcite precipitation reoccurred. In 2015, there was onlyFor the a single calculation calcite of precipitationdurations, we excluded event. BLevents has which calcite are precipitation only detected everyin one yearLandsat except in 2001, butimage, this yearso called only “single had two events”. Landsat The acquisitionsmaximum duration on 13 at May FH 2001is then and 32 20days, October with an 2001. average At SL one event wasduration detected of 24 on days the (standard Landsat deviation: image on 11 days). 13 August The maximum 2014. duration at BL is then 96 days, with an average duration of 57 days (standard deviation: 26 days). An example for a single event is FH in For the calculation of durations, we excluded events which are only detected in one Landsat 2013: calcite precipitation was classified on 30 July 2003, but the acquisitions 16 days earlier and 16 image, sodays called later both “single show events”. a dark lake The without maximum calcite durationprecipitation. at FHThis isLandsat then‐ 32derived days, classification with an average durationresults of 24 is days equal (standard to the visual deviation: classification. 11 days). Generally, The FH maximum and BL show duration “start–stop–new–start–stop” at BL is then 96 days, with an averagetemporal duration pattern: of 57 Acquisitions days (standard with deviation:Landsat‐derived 26 days). calcite An precipitation example for are a interrupted single event by is FH in 2013: calciteacquisition precipitation without calcite was classified precipitation, on 30 in July2013 in 2003, FH, butand the2014 acquisitions for both lakes. 16 However, days earlier the visual and 16 days later bothclassification show a dark differs lake from without the Landsat calcite‐derived precipitation. “start–stop–new–start–stop” This Landsat-derived at FH classificationin 2014: The two results is of the five dates at FH between 4 July 2014 and 13 August 2014 that are classified as dates without equal to the visual classification. Generally, FH and BL show “start–stop–new–start–stop” temporal calcite precipitation are classified visually as calcite precipitation and as bad quality images. pattern: Acquisitions with Landsat-derived calcite precipitation are interrupted by acquisition without calcite precipitation,4.5.2. Klocksin Lake in 2013 Chain in FH, and 2014 for both lakes. However, the visual classification differs from the Landsat-derivedIn three of the four “start–stop–new–start–stop” lakes of the Klocksin Lake Chain, at FH calcite in 2014: precipitation The two events of the are five detected dates at FH between(Figure 4 July 14b). 2014 At and BS 13no Augustcalcite precipitation 2014 that areevents classified were derived as dates from without the Landsat calcite‐images precipitation in the are classifiedmonitoring visually period. as calcite At HS precipitation one event was and classified as bad on quality 7 August images. 2015. For FS calcite precipitation events are detected on individual Landsat images in 1999, 2003, 2007 4.5.2. Klocksinand 2014. Lake In 2013 Chain there are three consecutive Landsat‐derived events on 8 July 2013, 9 July 2013, and

In three of the four lakes of the Klocksin Lake Chain, calcite precipitation events are detected (Figure 14b). At BS no calcite precipitation events were derived from the Landsat-images in the monitoring period. At HS one event was classiﬁed on 7 August 2015. For FS calcite precipitation events are detected on individual Landsat images in 1999, 2003, 2007 and 2014. In 2013 there are three consecutive Landsat-derived events on 8 July 2013, 9 July 2013, and Water 2017, 9, 15 17 of 31

48 days later, on 26 August 2013. TS shows the most frequent calcite precipitation events (in 11 of the 18 monitored years), whereas sediment analyses show either one or two thinner calcite layers Water 2017, 9, 15 17 of 31 every year. The selection of events in at least two consecutive acquisitions leaves seven years with long-lasting48 days calcite later, on precipitation 26 August 2013. events: TS shows the the maximum most frequent duration calcite precipitation is 56 days, events the average (in 11 of isthe 31 days (standard18 monitored deviation: years), 17 days). whereas sediment analyses show either one or two thinner calcite layers every year. The selection of events in at least two consecutive acquisitions leaves seven years with long‐ 4.5.3. Rheinsberglasting calcite Lake precipitation Region events: the maximum duration is 56 days, the average is 31 days (standard deviation: 17 days). In nine of the 17 lakes in Rheinsberg Lake Region calcite precipitation is derived from Landsat images4.5.3. (Figure Rheinsberg 14c). Lake The Region lakes without calcite precipitation are Peetschsee, Großer Glietzensee (Ost), GroßerIn nine Glietzensee of the 17 lakes (West), in Rheinsberg Großer Krukowsee,Lake Region calcite Nehmitzsee precipitation (north is derived and south),from Landsat Plötzensee, and Gerlinsee.images (Figure In each 14c). one The image lakes without Großer calcite Boberowsee precipitation and are Kleiner Peetschsee, Krukowsee Großer Glietzensee are classified (Ost), as lakes with calciteGroßer precipitaion. Glietzensee Breutzensee,(West), Großer Dagowsee, Krukowsee, and Nehmitzsee Stechlinsee (north have and each south), two acquisitionsPlötzensee, and that show calcite precipitationGerlinsee. In each events. one image Only Großer Kleiner Boberowsee Glietzensee, and Kleiner Roofensee, Krukowsee Menowsee, are classified and as Großer lakes with Pälitzsee calcite precipitaion. Breutzensee, Dagowsee, and Stechlinsee have each two acquisitions that show show frequent calcite precipitation events on six to 18 dates. However, because of the high number calcite precipitation events. Only Kleiner Glietzensee, Roofensee, Menowsee, and Großer Pälitzsee of FN results,show frequent we renounced calcite precipitation the calculation events on of durations.six to 18 dates. Extraordinary However, because are twoof the calcite high number precipitation events classificationsof FN results, we in renounced March at the Kleiner calculation Glietzensee of durations. (13 Extraordinary March 2014 andare two 30 calcite March precipitation 2014). events classifications in March at Kleiner Glietzensee (13 March 2014 and 30 March 2014). 4.6. Sentinel-2-Derived Calcite Precipitation in the Feldberg Lake District 4.6. Sentinel‐2‐Derived Calcite Precipitation in the Feldberg Lake District Finally, we tested the Area BGR classification approach on a Sentinel-2 data set. Figure 15 Finally, we tested the Area BGR classification approach on a Sentinel‐2 data set. Figure 15 illustrates the two existing Sentinel-2 images in 2015 and the contemporary Landsat images, together illustrates the two existing Sentinel‐2 images in 2015 and the contemporary Landsat images, together with theirwith lake their reflectance lake reflectance spectra. spectra.

Figure 15. Comparison of two contemporary acquired quasi‐true color RGB Landsat 8 and Sentinel‐2 Figure 15. Comparison of two contemporary acquired quasi-true color RGB Landsat 8 and Sentinel-2 images of Feldberg Lake District (left and middle) and their according mean spectra (right side). In images of(a), Feldberg the Landsat Lake image District was acquired (left and on middle) 3 August and 2015; their and the according Sentinel meanimage spectraon 7 August (right 2015, side). in In (a), the Landsat(b) both image images was are acquiredacquired on on 23 3 August August 2015. 2015; The orange and the rectangles Sentinel in imageFH, BL and on 7SL August illustrate 2015, the in (b) both imagesROI for are the acquired extraction on of 23 the August mean reflectance 2015. The spectra. orange Landsat rectangles spectra in FH,are illustrated BL and SL as illustratesolid lines, the ROI for the extractionSentinel spectra of the as meandotted reflectancelines. spectra. Landsat spectra are illustrated as solid lines, Sentinel spectra as dotted lines. On 7 August 2015 BH has calcite precipitation, on 23 August 2015 BL and FH. The lakes with calcite precipitation are in the Sentinel‐2 images also characterized by a higher reflectance in Oncomparison 7 August 2015 to the BH dark has lakes calcite without precipitation, calcite precipitation. on 23 August On both 2015 dates BL the and Sentinel FH. The‐2 images lakes have with calcite precipitationlower reflectance are in the values Sentinel-2 in the images visible and also NIR characterized wavelength range by a higherthan Landsat reflectance 8. The deviation in comparison is to the darklargest lakes in without the NIR calcite band. precipitation. On both dates the Sentinel-2 images have lower reflectance values in the visible and NIR wavelength range than Landsat 8. The deviation is largest in the NIR band. The classification of the Sentinel-2 images via threshold 13 × 103 of Area BGR shows the following results: On 3 August 2015, BL is classified as calcite precipitation with an Area BGR of 22.8 × 103. Water 2017, 9, 15 18 of 31

FH (Area BGR: 7.0 × 103) and SL (5.5 × 103) are classified as lake without calcite precipitation. On 23 August 2015, BL (27.8 × 103) and FH (18.4 × 103) are classified as calcite precipitation, whereas SL (8 × 103) is classified as lake without calcite precipitation. The classification results of the Sentinel-2-derived classifications are equal to the Landsat-derived classification results and the confusion matrices of the Sentinel-2-derived classifications with the Landsat-derived classification and with the visual classification show a perfect accuracy of 1.00.

5. Discussion

5.1. Visibility of Calcite Precipitation in Multi-Spectral Satellite Imagery (Landsat and Sentinel-2) The atmospheric correction is the most complex and error-prone processing step in water remote sensing as atmospheric correction models try to remove a large noise (atmosphere) from the small signal of water. However, in time series analysis atmospheric correction is essential for the comparison of different dates and sensors. In this study, we use the Landsat archive (Landsat 5, 7 and 8) and Sentinel-2 imagery, but the focus is the fast and easy applicability of satellite images for the monitoring of calcite precipitation. Thus, we ordered the satellite images in their highest processing level, including atmospheric correction, or used the state-of-the-art processor (e.g., sen2cor for the atmospheric correction) recommended by the provider of the satellite data. However, the models for the atmospheric correction differ for the different sensors and cause significant variation in the lake spectra (cf. Figure 13). Despite the significant variation in the lake spectra, calcite precipitation events are clearly visible in multispectral Landsat and Sentinel-2 images. Whereas lakes without calcite precipitation appear black in quasi-true color RGB images, high CaCO3 concentrations cause an additive effect to the spectra, especially in the green band, resulting in a turquoise color. In addition to calcite precipitation, the lake spectra can be influenced by other suspended minerals, yellow substances (“Gelbstoff”), and phytoplankton (chl-a concentration). Other suspended mineral particles can increase the reflectance similar to calcite precipitation, but sediment entry is negligible in this study area because of dense vegetation cover, low topography and slow flow velocities that cannot carry sediments [2]. In other study areas, however, sediment entry might distort the monitoring of calcite precipitation. Another possible source of error are shallow water areas where the lake bottom can be seen or where wind can resuspend sediments that cause high particle concentrations in the open water [2]. Thus, the lake spectra for the analysis are extracted in deep areas of the lakes to avoid misclassifications. Yellow substances (“Gelbstoff”) in the water absorb in ultraviolet and blue [2,10,11], however, as colored dissolved organic matter (CDOM) was not measured in situ, we cannot estimate its influence on our lake spectra. Whereas lakes with calcite precipitation are mostly turquoise, on some dates, the lake color appears more greenish than turquoise or is even bright green color (e.g., FH on 1 June 2008). Those green colors can be explained by (a mixture of calcite precipitation and) high chl-a concentrations: Phytoplankton scatters diffusely within the algal biomass (additive effect to the spectra), but also absorbs in blue and red [10,56,57]. Lake spectra with high chl-a are characterized by an peak around 700 nm (red edge) [56]. This peak cannot be detected using Landsat imagery because of the missing red-edge band of the Landsat sensors [42,43]. The new Sentinel-2 mission has a red-edge band and first tests with Sentinel-2 images showed its potential for the estimation of chl-a [58]. Thus, we expect that Sentinel-2 will be used in future for additional distinction between calcite precipitation and algal blooms. Even though in this study, the limited amount of Sentinel-2 data and a lack of contemporary in situ measurements hindered a further analysis. Thus, in this study calcite precipitation and phytoplankton blooms cannot be distinguished clearly. The potential risk of misclassification in the lakes in Feldberg Lake District or Klocksin Lake Chain is very low because the lakes are mostly mesotrophic and considerable algal blooms are most likely in eutrophic lakes with chl-a concentration above 20 µg/L [56]). In Rheinsberg Lake Region, Water 2017, 9, 15 19 of 31 several lakes are eutrophic and, whereas calcite precipitation events also occur in eutrophic lakes (cf. Feldberger Haussee), occasional misclassifications of algal blooms cannot be excluded (e.g., March 2014 at Kleiner Glietzensee).

5.2. Classification and Validation of Calcite Precipitation Using Multi-Spectral Satellite Imagery Even though previous studies highlighted the patchiness of calcite precipitation in large lakes (area >20 km2)[1,2,25], the lake colors in our study area are homogeneous (cf. Figures2 and9) as the smaller size of the lakes in northeastern Germany supports mixing processes. The only exceptions are different lake colors in separated basins of lakes (cf. BL in Figures9d and 15). Some heterogeneity within the water bodies is smoothed out by the 30-m resolution of Landsat, but there are also artifacts from the atmospheric correction of Landsat 8 images that cause some variation [59]. Thus, in this study, a classification based on mean lake spectra was chosen to minimize the variation of the lake spectra. In other regions with higher heterogeneity classifications on pixel level might be preferable. Thiemann and Koschel proposed a classification of calcite concentration based on 800 nm in hyperspectral images and considered more than 3% reflectance at 800 nm as calcite precipitation [2]. Although Landsat images do not have an 800 nm band, red is on average 655 nm and NIR is at 865 nm and a comparison with the spectra of BL (Figure 10) shows that only the strongest calcite precipitation on 11 July 1999 meets their criteria for calcite precipitation. At the other dates the reflectance of red and NIR are below 3% reflectance, even though in situ measurements showed increased CaCO3 concentrations. Therefore, we tested several spectral indices for the classification of calcite precipitation. Best results were achieved with a classification based on the triangular area between the blue, green and red band in the spectra, the “Area BGR”. If the Area BGR value is ≥13 × 103, the lake is classified as lake with calcite precipitation. This threshold is suitable for Landsat imagery and Sentinel-2 despite their differences in bandwidth and atmospheric correction. Before this study, it was unknown which calcite concentrations were detectable from space. Thus, we compared our Landsat-derived classifications with different in situ CaCO3 concentrations. Without optical aid a previous study suggested the limit for the visual detection of calcite precipitation is 1 mg/L [3]. Our classification approach was able to detect CaCO3 concentrations ≥0.7 mg/L with an accuracy of 0.88. A higher threshold in the CaCO3 concentration increases the number of FP classification results, whereas a lower CaCO3 concentration increases the number of missed calcite precipitation events (=FN). The accuracy of the automatic classification is hereby also slightly better than the accuracy of the according visual classification of the quasi-true color RGB Landsat images (0.85). The accuracy of the Landsat-derived classifications is decreased by FN and FP results: Here, FN results can be explained by the conservative threshold that is optimized for the correct classification of lakes without calcite precipitation: Thus, dates with only a slight color change are missed by the threshold. However, the six missed calcite precipitation events all have CaCO3 concentrations between 0.95 and 1.65 mg/L which should have resulted in a visible color change. The visual classification of those dates confirms that four of the six dates indeed showed no or only a slight color change. The other two images have image quality problems and biased lake reflectance spectra because of cloud coverage. The lack in color change despite calcite precipitation might be explained by the heterogeneity of CaCO3 concentrations within the lake or by time gaps of one day between in situ measurement and Landsat images. Thiemann and Koschel already noted the varying CaCO3 concentrations within lakes in northeastern Germany and the possible problems for validation [2]. Despite careful work, measurement and transmission errors cannot be excluded for certain. The nine FP classifications are related to already increased calcite concentrations between 0.41 and 0.69 mg/L and resulting in a turquoise color of the lakes. This applies specifically in the visual classification because of the high sensitivity of the human eye and thus a more sensitive visual classification. Because of the time gaps between in situ measurements and Landsat acquisition, it could also be that the concentration during the Landsat acquisition has already increased. An additional validation of single dates using the SI failed because of: (a) its short term changes as soon as Water 2017, 9, 15 20 of 31

crystallization of CaCO3 starts; and (b) its limitations in case of bio-induced calcite precipitation [17,18] as the occurrence of crystal nucleus, e.g., in form of bacteria [59,60], is equally important for calcite precipitation. Considering its SI FH has before 2011 not the potential for calcite precipitation, but in situ CaCO3 measurements and the analysis of Landsat show clear calcite precipitation, whereas SL has high SI values since 2000 and therefore the potential for calcite precipitation, but still calcite precipitation events are very rare. After the calibration and validation at Feldberg Lake District, the classificaion approach was applied in two other regions without in situ measurements. Here, the validation is based on visual classifications of the lakes. The accuracies are very high (0.99 and 0.97), but whereas Feldberg Lake District has a ratio of calcite precipitation events to normal lake conditions of 1:4, the ratio at Klocksin Lake Chain is already 1:14 and in Rheinsberg Lake Region even higher with 1:36. Thus, the accuracy is biased and in relation to the number of calcite precipitation events, the number of missed calcite precipitations (FN) is high in Rheinsberg Lake Region. The FN results can here also be explained by the conservative classification threshold in combination with the high sensitivity of the human eye and thus a more sensitive visual classification. The accuracy based on the visual classification at Feldberg Lake District is 0.94. The lower accuracy at Feldberg Lake District in comparison to the other regions is caused by the less strict cloud removal. Especially, haze and missed cloud pixels hinder the accurate classification (FN results). Whereas FN and FP classification results (e.g., FN results at lake Menowsee) distort the analysis of frequencies and durations, they do not affect the determination of the total area of lakes with calcite precipitation in this study. In comparison to the visual classification, FP are very rare, whereas all FN results occur at lakes that are, at other times, classified as lakes with calcite precipitation. Overall, calcite precipitation events are detected in 15 of 24 lakes in the study areas and the total lake area of lakes with calcite precipitation is approximately 17 km2. Thus, our study supports that calcite precipitation is a common phenomenon in the hardwater lakes, but it also emphasized that the durations and frequencies strongly vary so that each lake must be monitored individually.

5.3. Time Series: Frequency and Duration of Calcite Precipitation Remote sensing data and especially, the large Landsat image archive, enable a long-term monitoring of large areas. However, for an accurate monitoring of the duration and frequency of calcite precipitation, a high acquisition density is required. The Landsat archives cannot provide a continuous temporal coverage, due to their limited repetition rate of 16 days and acquisition gaps by cloud coverage. Whereas in some years with overlapping satellite missions the number of acquisitions is high, other years have only very few acquisitions, e.g., 2001 (cf. Figure5). Since 2012, Landsat 7 and Landsat 8 are operating together providing a doubled repetition rate of eight days. However, cloud coverage still reduces the coverage significantly. For example, in 2012 the monitoring of the lakes in the Klocksin Lake Chain is only very irregular, because of the removal of cloud and cloud shadow (cf. Table A1). Thus, for a regular monitoring of calcite precipitation at selected lakes a less strict cloud and cloud shadow removal should be considered, even if this comes along with a higher number of misclassification because of cloud and haze (cf. classification accuracy of Feldberg Lake District). Earth observation with the Sentinel-2 satellite(s) will further increase the data density with its higher time resolution and repetition rate of 10 days. However, so far, only two Sentinel-2a images cover the study area, so that the full potential of Sentinel-2 could not yet be studied. Calcite precipitation is known to be a spatial and temporal very variable process and it is linked to different factors, e.g., trophic state and occurrences of bacteria of the lakes. This causes changing calcite precipitation patterns in time and different calcite precipitation patterns even for adjacent lakes, e.g., in the Feldberg Lake District: even though SL was known for calcite precipitation before 1998, there were no calcite precipitation detected with in situ measurements after the lake restoration in 1996/1997. Whereas the in situ measurements between 1998 and 2015 show CaCO3 concentrations <0.5 mg/L, the Landsat classification shows a clear calcite precipitation event on 13 August 2014 Water 2017, 9, 15 21 of 31 and misses further five visually classified calcite precipitation events. At FH, no calcite precipitation occurred before 2003. This fits to the in situ measurements: The long-term record (1985–2015) of mean seasonal (May–September) CaCO3 concentration of FH indicates low values of <0.5 mg/L until 2005. Afterwards a substantial increase was observed. However, the reasons are not entirely clear. Koschel et al. (1983) have concluded that calcite precipitation might be most intensive in mesotrophic lakes [60]. By 2005 the seasonal (May–September) total phosphorus concentration of the mixed layer had dropped to 0.046 mg/L which is still in the eutrophic range but relatively close to mesotrophic conditions [61]. In 2011, when a second restoration measure has been carried out, calcite precipitation did not occur, while in 2012 a few calcite precipitation events and in 2013 several calcite precipitation events have been observed at FH. In 2014, the duration and frequency of calcite precipitation events is reduced again and in 2015 only one event has been observed. These variations can be explained by the complex process with competing inhibiting factors and supporting factors for calcite precipitation. During eutrophication phases, excess phosphorus has an inhibiting effect on calcite precipitation [12] so that artificial removal of phosphorous through poly-aluminum chloride might increase calcite precipitation. A follow-up monitoring will show if the restoration measures at FH have the same result (absence of calcite precipitation) as the restoration measures at SL. The three lakes with the most calcite precipitation events are FH, BL, and TS. Their average durations are 24, 57, and 31 days (overall average: 37 day), but all three lakes show start–stop–start–stop patterns. However, the comparison of those patterns with the visual classifications reveals that most of those short-term variations are caused by misclassifications (e.g., FN results at FH between 4 July 2014 and 13 August 2014). Visually validated start–stop–start patterns, e.g., at FH between 24 July 2013 and 26 August 2013, substantiates the results of previous studies that showed calcite precipitation with lower calcite concentrations before and after the main event and non-steady, periodic variations [5]. Whereas the validation in Section 5.2 discusses the quality of the Landsat-derived calcite precipitation monitoring, it is still unknown, how many calcite precipitation events are missed in times without suitable satellite images or in situ measurements. Therefore, we included sediment analyses at TS to discuss the detection rate of calcite precipitation via remote sensing: The Landsat-derived monitoring at lakes in the Klocksin Lake Chain shows calcite precipitation events at TS at 11 of 18 years, whereas sediment analyses show calcite layers every year. A running monitoring study at TS using sediment traps shows high calcite deposition between May and September. The years 2001, 2007, 2010, and 2012 have only one or two Landsat acquisitions during May and September, thus calcite precipitation events were most likely missed. The years 1998, 2004, and 2011 have four, five, and nine acquisitions. However, the acquisitions are not equally distributed with maximum gaps of 55 days (2004 and 2011) to 89 days (1998) days between the acquisitions and thus, still calcite events could have happened in times without Landsat acquisitions. Especially, as the sediment layers in all of the missed years are extraordinary thin [33], which indicates either short-term or weak calcite precipitation events. At region Rheinsberg Lake Region, eight of the 17 lakes show at least one calcite precipitation event, but the frequency of events is probably higher than our analyses shows, because the validation based on a visual classification reveals approximately as many calcite precipitation events as FN classifications. On the other hand, five lakes are eutrophic, thus, algal bloom may also occur, either contemporary to high CaCO3 concentrations or independent of calcite precipitation. In this region the calcite precipitation on 20 August 2011 at the oligotrophic Stechlinsee has to be highlighted. This calcite precipitation is caused by storm “Otto” in July 2011, which caused the mixing of cyanobacteria populations from 7 m to 8 m depth into the surface water [28]. The resulting increase of the photosynthesis activity caused an increase of the CaCO3 saturation index and lead to intensive calcite precipitation, still clearly visible in the Landsat image on 20 August 2011.

6. Conclusions In this study, we tested the potential of the Landsat archive and Sentinel-2 for the classification and monitoring of calcite precipitation in lakes. Calcite precipitation due to increased CaCO3 concentrations Water 2017, 9, 15 22 of 31 cause an additive effect to the lake reflectance spectra, especially in the green band, resulting in quasi-true color RGB images in a turquoise color. Thus, we classify calcite precipitation events in lakes based on the calculation of the triangular area between blue, green and red in the mean lake spectra (Area BGR). We chose a conservative threshold, based on the comparison of visually turquoise and dark lake values of one lake (FH) for which a long-term in situ data archive of CaCO3 concentrations is available. Overall, our study area covers 24 lakes. The classification results of FH, BL, and SL are validated with in situ measurements of calcite precipitation, for TS with sediment core data. We detected calcite precipitation with CaCO3 concentrations ≥0.7 mg/L in the Feldberg Lake district with a good accuracy of 0.88. Our classification is here better than expected; a previous study suggested a limit of >1 mg/L. The analysis of the false classified events showed that at some dates the lakes do not show a change of color even though the contemporary CaCO3 concentration is high (FN classifications), whereas other dates with only slightly increased CaCO3 values have a change of color (FP classifications). Important to consider is the time gap between in situ measurement and Landsat acquisition as well as a possible heterogeneous distribution of CaCO3 in the lake. In a next step, we tested the monitoring approach on 21 other lakes in the regions Klocksin Lake Chain and Rheinsberg Lake Region and validated the classification results based on a visual inspection of the Landsat data. Whereas FN results are relatively frequent in comparison to the number of detection calcite precipitation events, the overall accuracy in these two regions is still >0.97. Our study shows that 15 of the 24 lakes covering a total area of approximately 17 km2 have at least one calcite precipitation event in the observation period. The frequency of calcite precipitation events varies between one detection and regular detections nearly every year. The durations of calcite precipitation events also vary between the lakes, but are for the lakes with regular calcite precipitation (FH, BL, and TS) in average 37 days. The time series for Feldberg Lake District is denser than the one of Klocksin Lake Chain and Rheinsberg Lake Region, but has also a higher risk of misclassifications due to haze and remaining cloud pixels. However, still the effect of lake trophy restoration measures on calcite precipitation can be shown at SL and FH. The high number of missed calcite precipitation events (=FN results), together with gaps in Landsat time series (e.g., 2001), reduces the accuracy of frequency and duration monitoring. For example, the comparison with sediment data at TS shows that calcite precipitation events have been missed in some years due to low image density in the critical time periods (May to September). In future the image density will increase by acquisitions of Sentinel-2a and coming Sentinel-2b. We tested the application of Area BGR classification method to Sentinel-2 and even although the sensors and the atmospheric correction differ, the classification approach is transferable. Another great potential of Sentinel-2 comes with its red-edge band: We expect that Sentinel-2 can also be used in future for the distinction between algal blooms and calcite precipitation. Our results emphasized the variety of the lakes and the need to monitor each lake individually. This is due to the complex processes of calcite precipitation, which are influenced by a number of factors including lake trophic state, algae composition and activity, human measures and climate. Using the large Landsat archive and Sentinel-2 imagery, we now can provide an algorithm for monitoring calcite precipitation in lakes in the entire Northeast German Plain. This is an essential prerequisite, in combination with geochemical analyzes, to investigate the role of permanent CO2 storage in form of calcite in this region.

Acknowledgments: This study was funded by the “Helmholtz Association of German Research Centres Initiative—Networking Fund for funding a Helmholtz Virtual Institute” (VH-VI-415). Author Contributions: Iris Heine developed the methodological framework, performed programming, conducted the analysis and wrote the article; Peter Kasprzak and Ulrike Kienel provided in situ data for and contributed their expert knowledge on the study areas; Birgit Heim supported the analysis of the lake spectra based on her experience of water remote sensing; and Achim Brauer, Birgit Kleinschmit and Sibylle Itzerott were involved in formulating the research questions and contributing to critical discussions. All authors were involved in the general paper review. Conﬂicts of Interest: The authors declare no conﬂict of interest. Water 2017, 9, 15 23 of 31

Appendix A

Table A1. Calcite precipitation based on the Landsat time series (1998–2015) using the threshold of the area BGR. Lakes with calcite precipitation events are marked with “1” and highlighted in gray. Dark lakes without calcite precipitation are marked with “0”. Blank cells are the dates without data.

26-March-1998 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 13-May-1998 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 20-May-1998 0 1 0 0 0 0 0 0 1998 29-May-1998 0 100000000 0000000000000 05-June-1998 0 10000000000000000000000 21-June-1998 0 10 00000000000000000 1 02-September-1998 0 0 0 0 0 11-July-1999 0 1000000000000000000 1 0 0 0 03-August-1999 0 1 0 0 0 1 10000000000 00 1 0 0 0 04-September-1999 0 10000000000000000000 1 0 0 1999 13-September-1999 0 100000000 0000000000 1 0 1 29-September-1999 0 0 0 0 0 15-October-1999 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 19-January-2000 0 0 0 0 0 0 0 0 0 0 27-February-2000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 24-April-2000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 17-May-2000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2000 02-June-2000 0 1 0000000 000 14-August-2000 0 1 0 0 0 10000 0000000000 1 0 0 22-September-2000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 01-October-2000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 02-November-2000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 13-May-2001 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2001 20-October-2001 0 0 0 0 0 0 0 0 0 0 0 0 0 0 29-March-2002 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 05-April-2002 0 0 0 2002 21-April-2002 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 01-June-2002 0 1 0 0 0 1 0 0 0 0 0 20-August-2002 0 1 0 0 0 1000000000000000 1 0 0 Water 2017, 9, 15 24 of 31

Table A1. Cont.

23-March-2003 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 17-April-2003 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 28-June-2003 0 100000000 0000000000000 14-July-2003 0 1 0 0 0 0 1 0 0 30-July-2003 1 1 0 0 0 10000 000000000000 1 06-August-2003 0 1 0 0 0 1 1 0 0 0 0 0 0 0 0 0 1 1 0 2003 07-August-2003 0 10 000000 00 000 1 1 0 1 31-August-2003 0 0 0 0 0 0 0 0 0 07-September-2003 0 0 0 0 0 0 1 0 0 0 0 0 16-September-2003 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 02-October-2003 0 0 0 0 0 0 17-October-2003 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 18-October-2003 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 18-April-2004 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 29-May-2004 0 10000000000000000000000 23-July-2004 0 1000000000 000000000000 31-July-2004 0 1 0 0 0 0 0 2004 01-August-2004 0 10000000000000000000000 09-August-2004 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 10-September-2004 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 11-October-2004 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 21-March-2005 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 28- March -2005 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 21-April-2005 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 16-May-2005 0 0 0 0 0 0 09-June-2005 0 10 0 0000000 00 16-June-2005 0 1 0 0 0 0 24-June-2005 0 1 0 0 2005 25-June-2005 0 1 0 03-July-2005 0 1 0 0 0 0 0 10-July-2005 0 1000000000000 000000 1 0 0 11-July-2005 0 10 0000000 0 000 0 04-August-2005 1 10 0 00000000 20-August-2005 1 1 0 0 0 1 0 0 0 0 0 0 0 0 05-September-2005 1 100000000 0000000000 1 0 1 Water 2017, 9, 15 25 of 31

Table A1. Cont.

06-October-2005 0 0 0 0 0 0 0 0 0 0 0 0 07-October-2005 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 14-October-2005 0 0 0 0 15-October-2005 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 30-October-2005 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 31-October-2005 0 0 0 0 0 0 0 0 0 0 0 0 0 0 17-April-2006 0 0 0 0 0 0 0 0 03-May-2006 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10-May-2006 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 11-June-2006 0 10000000000000000000000 12-June-2006 0 10 0 00000 00 000 000 06-July-2006 0 100000000 0000000000000 13-July-2006 1 1 0 0 1 0 0 0 0 0 0 0 0 21-July-2006 1 1 0 0 0 0 0 0 0 1 2006 22-July-2006 1 1 0 0 0 10000 000000000 1 1 0 0 14-August-2006 1 1 0 0 0 0 0 0 0 0 15-September-2006 0 10000000000000000000000 24-September-2006 1000000000 0000000000 1 0 0 01-October-2006 0 0 0 0 0 0 0 0 09-October-2006 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10-October-2006 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 17-October-2006 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 03-November-2006 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 27-March-2007 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 12-April-2007 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 28-April-2007 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2007 06-May-2007 0 0 0 0 0 0 0 0 0 16-July-2007 0 1 0 0 0 0 100000000000000 1 0 0 18-August-2007 1 1 0 11-September-2007 0 0 0 0 0 0 20-March-2008 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 21-April-2008 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2008 07-May-2008 100000000 0 00 0 01-June-2008 1 10 0000000000 00 000 0 Water 2017, 9, 15 26 of 31

Table A1. Cont.

Year Date Hofsee Bergsee Gerlinsee Tiefer See Dagowsee Roofensee Peetschsee Plötzensee Menowsee Flacher See Stechlinsee Breutzensee Breiter Luzin Schmaler Luzin Großer Pälitzsee Nehmitzsee north Nehmitzsee South Großer Krukowsee Kleiner Krukowsee Feldberger Haussee Kleiner Glietzensee Großer Boberowsee Großer Glietzensee (Ost) Großer Glietzensee (West) 08-June-2008 1 100000 000000 0 17-June-2008 0 1 0 0 0 0 03-July-2008 0 1 0 0 0 1 000 00 00 000 000 26-July-2008 0 100000 0000 0 23-October-2008 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 01-April-2009 0 0 0 0 0 0 0 0 0 0 0 17-April-2009 0 0 0 0 0 0 0 0 0 0 24-April-2009 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10-May-2009 0 0 0 0 0 0 0 0 0 0 0 19-May-2009 0 0 0 0 0 0 0 0 30-July-2009 1 0 1 0 0 2009 06-August-2009 1 1 0 07-August-2009 1 1 0 0 0 0 0 1 0 23-August-2009 1 1 0 1 0 0 0 0 0 0 0 0 30-August-2009 1 1 0 1 08-September-2009 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 16-September-2009 0 0 0 0 10000 000 00000000 24-September-2009 0 0 0 0 20-April-2010 0 0 0 0 0 0 0 0 0 0 0 29-May-2010 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 30-June-2010 0 0 0 0 0 0 0 2010 09-July-2010 1 10 000000000 0 1 0 16-July-2010 1 1 0 0 0 0 0 0 0 0 0 11-Sptember-2010 0 0 0 0 0 0 0 0 0 0 13-October-2010 0 0 0 0 0 0 0 0 0 0 0 0 0 0 30-March-2011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 22-April-2011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 23-April-2011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 30-April-2011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 01-May-2011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2011 08-May-2011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 09-May-2011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 02-June-2011 0 100000000 0000000000000 10-June-2011 0 1 0 0 0 0 0 0 0 0 0 27-July-2011 0 100000 000 00 Water 2017, 9, 15 27 of 31

Table A1. Cont.

Year Date Hofsee Bergsee Gerlinsee Tiefer See Dagowsee Roofensee Peetschsee Plötzensee Menowsee Flacher See Stechlinsee Breutzensee Breiter Luzin Schmaler Luzin Großer Pälitzsee Nehmitzsee north Nehmitzsee South Großer Krukowsee Kleiner Krukowsee Feldberger Haussee Kleiner Glietzensee Großer Boberowsee Großer Glietzensee (Ost) Großer Glietzensee (West) 20-August-2011 0 1 0 0 0 0 1 0 0 0 0 0 06-September-2011 0 0 0 0 0 0 0 0 0 29-September-2011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 30-September-2011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 15-October-2011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 16-October-2011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 23-October-2011 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 16-April-2012 0 0 0 0 0 0 02-May-2012 0 0 0 0 27-May-2012 1 1 0 0 0 0 0 12-June-2012 0 0 0 2012 19-June-2012 0 0 0 0 0 0 0 30-July-2012 1 0 15-August-2012 1 1 0 0 02-October-2012 0 0 0 0 0 0 0 0 18-October-2012 0 0 0 0 0 0 0 0 0 0 0 0 20-April-2013 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 28-April-2013 0 0 0 0 0 0 0 05-May-2013 0 0 0 0 0 0 0 0 0 0 06-May-2013 1 0 0 0 0 0 0 13-May-2013 1 0 0 0 06-June-2013 0 0 0 0 0 0 0 0 0 0 0 0 0 07-June-2013 0 1 0 0 0 0 0 0 0 10000000000000 23-June-2013 1 2013 08-July-2013 1 0 0 0 1 0 0 0 0 0 0 0 09-July-2013 1 1 0 0 0 0 1000 000000000000 1 16-July-2013 1 10 0 0000 0000 1 0 0 17-July-2013 1 0 0 0 0 0 24-July-2013 1 1 0 0 0 1 0 0 0 0 0 0 0 0 1 02-August-2013 0 1 0 0 0 1 000 0000000000000 26-August-2013 1 1 0 0 0 1 1 00 000 0 00 00 29-October-2013 0 0 0 0 0 13-November-2013 0 0 0 0 0 0 0 0 0 0 0 25-January-2014 0 0 0 0 0 0 0 0 2014 13-March-2014 0 0 0 0 0 0 0 0 10000000000000 Water 2017, 9, 15 28 of 31

Table A1. Cont.

Year Date Hofsee Bergsee Gerlinsee Tiefer See Dagowsee Roofensee Peetschsee Plötzensee Menowsee Flacher See Stechlinsee Breutzensee Breiter Luzin Schmaler Luzin Großer Pälitzsee Nehmitzsee north Nehmitzsee South Großer Krukowsee Kleiner Krukowsee Feldberger Haussee Kleiner Glietzensee Großer Boberowsee Großer Glietzensee (Ost) Großer Glietzensee (West) 30-March-2014 0 0 0 0 0 1 0 0 0 0 0 0 0 01-May-2014 0 0 0 0 1 0 0 0 0 0 0 0 0 17-May-2014 0 0 0 0 10-June-2014 0 1 0 0 0 1 0 18-June-2014 100 0000000 00 0 03-July-2014 1 1 0 0 0 0 0 0 1 0 0 04-July-2014 1 1 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 11-July-2014 0 1 0 19-July-2014 1 1 0 0 0 0 1000 000000000 1 1 0 1 27-July-2014 0 1 0 0 0 0 0 0 0 0 13-August-2014 1 1 1 0 0 1 0 0 1 0000000000 1 0 0 05-September-2014 0 1000000000000000000 000 06-September-2014 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 08-October-2014 0 0 0 0 0 0 0 0 0 0 0 08-March-2015 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 17-March-2015 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10-April-2015 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 04-June-2015 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 05-June-2015 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 12-June-2015 0 0 0 1 0 0 0 0 0 0 0 0 13-June-2015 0 1 0 0 28-June-2015 0 1 0 0 0 0 0 0 0 29-June-2015 0 1 0 0 0 0 0 0 06-July-2015 0 1 0 0 0 0 2015 07-July-2015 0 1 0 1 000 0000 0000000 07-August-2015 0 1 0 1 1 00000000000000 1 0 1 15-August-2015 0 1 0 0 0 0 23-August-2015 1 1 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 1 0 1 08-September-2015 0 0 0 0 1 0 0 0 0 0 0 0 0 17-September-2015 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 03-October-2015 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 10-October-2015 0 0 0 0 11-October-2015 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 26-October-2015 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 27-October-2015 0 0 0 0 0 0 0 0 0 0 0 0 0 0 Water 2017, 9, 15 29 of 31

References

1. Strong, A.; Eadie, B.J. Satellite observations of calcium carbonate precipitations in the Great Lakes. Limnol. Oceanogr. 1978, 23, 877–887. [CrossRef] 2. Thiemann, S.; Koschel, R. Erfassung des räumlichen Verteilungsmusters von Kalkfällung mit Fernerkundungsdaten. Wasser Boden 2001, 53, 25–28. 3. Koschel, R.; Proft, G.; Raidt, H. Autochthone Kalkfällung in Hartwasserseen der Mecklenburger Seenplatte. Limnologica 1987, 18, 317–338.

4. Weidemann, A.D.; Bannister, T.T.; Effler, S.W.; Johnson, D.L. Particulate and optical properties during CaCO3 precipitation in Otisco Lake. Limnol. Oceanogr. 1985, 30, 1078–1083. [CrossRef] 5. Proft, G. Die pelagische Calcitfällung und der Carbonatgehalt von Sedimenten pleistozäner Seen. Acta Hydrochim. Hydrobiol. 1984, 12, 321–326. [CrossRef] 6. Koschel, R.; Kasprzak, P.; Schreiber, A. Kalzitfällung und Nahrungskettenmanipulation. Lauf. Semin. 1997, 3/97, 61–76. 7. Kempke, S.; Schick, R.; Rinke, K.; Rothhaupt, K. Biogene Calcitfällung im Bodensee-Prozessverständnis und Modellierung. WasserWirtschaft 2008, 98, 31–33. 8. Hamilton, S.K.; Bruesewitz, D.A.; Horst, G.P.; Weed, D.B.; Sarnelle, O. Biogenic calcite–phosphorus precipitation as a negative feedback to lake eutrophication. Can. J. Fish. Aquat. Sci. 2009, 66, 343–350. [CrossRef] 9. Kelts, K.; Hsü, K. Chapter 9: Freshwater carbonate sedimentation. In Lakes—Chemistry Geology Physics; Springer: New York, NY, USA, 1978; pp. 295–323. 10. Kronberg, P. Fernerkundung der Erde. Grundlagen und Methoden des Remote Sensing in der Geologie; Enke Verlag: Stuttgart, Germany, 1985. 11. Fan, C. Spectral Analysis of Water Reflectance for Hyperspectral Remote Sensing of Water Quailty in Estuarine Water. Water 2014, 2, 19–27. [CrossRef] 12. Dittrich, M.; Koschel, R. Interactions between calcite precipitation (natural and artificial) and phosphorus cycle in the hardwater lake. Hydrobiologia 2002, 469, 49–57. [CrossRef] 13. Brunskill, G.J. Fayetteville Green Lake, New York. II. Precipitation and sedimentation of calcite in a meromictic lake with laminated sediments. Limnol. Oceanogr. 1969, 14, 830–847. [CrossRef] 14. Gal, J.Y.; Bollinger, J.C.; Tolosa, H.; Gache, N. Calcium carbonate solubility: A reappraisal of scale formation and inhibition. Talanta 1996, 43, 1497–1509. [CrossRef] 15. House, W. Inhibition of calcite crystal growth by inorganic phosphate. J. Colloid Interface Sci. 1987, 119, 505–511. [CrossRef] 16. Koschel, R. Structure and function of pelagic calcite precipitation in lake ecosystems. Verh. Int. Ver. Limnol. 1997, 26, 343–349. 17. Stabel, H.-H. Calcite precipitation in Lake Constance: Chemical equilibrium, sedimentation, and nucleation by algae. Limnol. Oceanogr. 1986, 31, 1081–1094. [CrossRef] 18. Thompson, J.B. Cyanobacterial Precipitation of Gypsum, Calcite and Magnesitre From Natural Lake Water. Geology 1990, 18, 995–998. [CrossRef] 19. Romero, L.; Camacho, A.; Vicente, E.; Miracle, M.R. Sedimentation patterns of photosynthetic bacteria based on pigment markers in meromictic Lake La Cruz (Spain): Paleolimnological implications. J. Paleolimnol. 2006, 35, 167–177. [CrossRef] 20. De Vicente, I.; Cattaneo, K.; Cruz-Pizarro, L.; Brauer, A.; Guilizzoni, P. Sedimentary phosphate fractions related to calcite precipitation in an eutrophic hardwater lake (Lake Alserio, northern Italy). J. Paleolimnol. 2006, 35, 55–64. [CrossRef] 21. Romero-Viana, L.; Julià, R.; Camacho, A.; Vicente, E.; Miracle, M.R. Climate signal in varve thickness: Lake la Cruz (Spain), a case study. J. Paleolimnol. 2008, 40, 703–714. [CrossRef] 22. Long, J.; Hu, C.; Robbins, L. Whiting events in SW Florida coastal waters: A case study using MODIS medium-resolution data. Remote Sens. Lett. 2014, 5, 539–547. [CrossRef] 23. Dierssen, H.M.; Zimmerman, R.C.; Burdige, D.J. Optics and remote sensing of Bahamian carbonate sediment whitings and potential relationship to wind-driven Langmuir circulation. Biogeosci. Discuss. 2008, 5, 4777–4811. [CrossRef] 24. Robbins, L.L.; Tao, Y.; Evans, C.A. Temporal and spatial distribution of whitings on Great Bahama Bank and a new lime mud budget. Geology 1997, 25, 947–950. [CrossRef] Water 2017, 9, 15 30 of 31

25. Pulvermüller, A.G.; Kleiner, J.; Mauser, W. Calcite patchiness in Lake Constance as viewed by LANDSAT-TM. Aquat. Sci. 1995, 57, 338–349. [CrossRef] 26. Koschel, R.; Haubold, G.; Kasprzak, P.; Küchler, L.; Proft, G.; Ronneberger, D. Eine limnologische Zustandsanalyse des Feldberger Hausssees. Acta Hydrochim. Hydrobiol. 1981, 9, 255–279. [CrossRef] 27. Kasprzak, P.; Koschel, R.; Krienitz, L.; Gonsiorczyk, T.; Anwand, K.; Laude, U.; Wysujack, K.; Brach, H.; Mehner, T. Reduction of nutrient loading, planktivore removal and piscivore stocking as tools in water quality management: The Feldberger Haussee biomanipulation project. Limnologica 2003, 33, 190–204. [CrossRef] 28. Leibniz-Institute of Freshwater Ecology & Inland Fisheries (IGB). Unpublished Data, 2016. 29. Nixdorf, B.; Rücker, J.; Dolman, A.M.; Wiedner, C.; Hilt, S.; Kasprzak, P.; Köhler, A.; van de Klaus, W.; Sandrock, S.; Scharf, E.-M. Prozessverständnis als Grundlage für die Gewässerbewirtschaftung—Fallbeispiele für Limitation, Konkurrenz, Gewässerstruktur und Nahrungsnetzsteuerung. Korresp. Wasserwirtsch. 2013, 6, 693–701. 30. Nixdorf, B.; Hemm, M.; Hoffmann, A.; Richter, P. Dokumentation von Zustand und Entwicklung der Wichtigsten Seen Deutschlands (Teil 2); Umweltbundesamt: Dessau-Roßlau, Germany, 2004. 31. Bachor, A.; Carstens, M.; Klitzsch, S.; Korczynski, I.; Lemke, G.; Mathes, J.; Müller, J.; Schenk, M.; Seefeldt, O.; Schöppe, C.; et al. Gewässergütebericht Mecklenburg-Vorpommern 2003/2004/2005/2006: Ergebnisse der Güteüberwachung der Fließ-, Stand- und Küstengewässer und des Grundwassers in Mecklenburg-Vorpommern; Landesamt für Umwelt, Naturschutz und Geologie: Güstrow, Germany, 2006. 32. Geodaten des Bundesamt für Kartographie und Geodäsie. Digitales Basis-Landschaftsmodell (Basis-DLM). Bundesamt für Kartographie und Geodäsie: Frankfurt, Germany, 2011. Available online: https://www.bkg.bund.de/DE/Produkte-und-Services/Shop-und-Downloads/DigitaleGeodaten/ Landschaftsmodelle/Deutschland/DLMDeutschland_cont.html (accessed on 27 June 2011). 33. Kienel, U.; Dulski, P.; Ott, F.; Lorenz, S.; Brauer, A. Recently induced anoxia leading to the preservation of seasonal laminae in two NE-German lakes. J. Paleolimnol. 2013, 50, 535–544. [CrossRef] 34. Kienel, U.; Kirillin, G.; Brademann, B.; Plessen, B.; Lampe, R.; Brauer, A. Effects of the mixing duration in spring on diatom deposition in the deep Lake Tiefer See, NE Germany. J. Paleolimnol. 2016, 1–13. [CrossRef] 35. Müller, J. GAIA-MV: Tiefenkarten der Seen in Mecklenburg-Vorpommern. Available online: http://www. geodaten-mv.de/ (accessed on 21 September 2016). 36. Spieß, H.-J. Ergebnisse der Untersuchungen submerser Makrophyten in mesotroph-eutrophen Seen Mecklenburg-Vorpommerns. Rostock. Meeresbiol. Beitr. 2004, 13, 73–84. 37. Nixdorf, B.; Hemm, M.; Hoffmann, A.; Richter, P. Dokumentation von Zustand und Entwicklung der Wichtigsten Seen Deutschlands (Teil 5); Umweltbundesamt: Dessau-Roßlau, Germany, 2004. 38. Mietz, O.; Arp, W.; Riemer, A.; Vitinghoff, H.; Henker, H.; Wöbbecke, K.; Gabrysch, M.; Dahm, J.; Pausch, S.; Psille, D.; et al. Seenkataster Brandenburg; Bericht des Projektes; Die Seen im Land Brandenburg: Seddin, Germany, 1995. 39. Koschel, R. Leitbilder eines integrierten Seen- und Landschaftsschutzes. In Stechlin-Forum; Umweltstiftung WWF-Deutschland: Brandenburg, Germany, 1998; pp. 53–62. 40. Kabus, T.; Wiehle, I. Die Armleuchteralgen (Characeae) in ausgewählten Seen des Naturparks Stechlin-Ruppiner Land (Brandenburg, Deutschland). Ergebnisse der Untersuchungen außerhalb von FFH- und Naturschutzgebieten. Rostock. Meeresbiol. Beitr. 2013, 24, 63–74. 41. Kerstin, W.; Enviteam, B. Badegewässerproﬁl Nach Artikel 6 der Richtlinie 2006/7/EG und § 6 der Verordnung über die Qualität und die Bewirtschaftung der Badegewässer vom 06.02.2008 (BbgBadV); Ministerium der Justiz und für Europa und Verbraucherschutz des Landes Brandenburg: Potsdam, Germany, 2008. 42. National Aeronautics and Space Administration (NASA). Landsat 7 Science Data Users Handbook; NASA: Washington, DC, USA, 2011. 43. Irons, J.R.; Dwyer, J.L.; Barsi, J.A. The next Landsat satellite: The Landsat Data Continuity Mission. Remote Sens. Environ. 2012, 122, 11–21. [CrossRef] 44. Drusch, M.; Del Bello, U.; Carlier, S.; Colin, O.; Fernandez, V.; Gascon, F.; Hoersch, B.; Isola, C.; Laberinti, P.; Martimort, P.; et al. Sentinel-2: ESA’s Optical High-Resolution Mission for GMES Operational Services. Remote Sens. Environ. 2012, 120, 25–36. [CrossRef] 45. U.S. Department of the Interior; U.S. Geological Survey. USGS Website: Earthexplorer. Available online: http://earthexplorer.usgs.gov/ (accessed on 24 February 2016). Water 2017, 9, 15 31 of 31

46. European Union. European Space Agency (ESA) for Earth observation Sentinels Scientific Data Hub. Available online: https://scihub.copernicus.eu/dhus/#/home (accessed on 6 June 2016). 47. European Space Agency (ESA) for Earth Observation. Sen2Cor Toolbox. Available online: http://step.esa. int/main/third-party-plugins-2/sen2cor/ (accessed on 6 June 2016). 48. European Space Agency (ESA) for Earth Observation. Sentinel-2 Toolbox. Available online: http://step.esa. int/main/toolboxes/sentinel-2-toolbox/ (accessed on 6 June 2016). 49. Hepperl, D.; Krienitz, L. Phacotus lenticularis (Chlamydomonadales, Phacotaceae) zoospores require external supersaturation of calcium carbonate for calcification in culture. J. Phycol. 1997, 33, 415–424. [CrossRef] 50. Hepperle, D. WinIAP—Software for the Calculation of Ion Activities and Calcite Saturation Index. SequentiX—Digital DNA Processing. Available online: http://www.sequentix.de (accessed on 29 April 2016). 51. Tucker, C.J. Red and Photographic Infrared Linear Combinations for Monitoring Veqetation. Remote Sens. Environ. 1979, 8, 127–150. [CrossRef] 52. McFeeters, S.K. The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. Int. J. Remote Sens. 1996, 17, 1425–1432. [CrossRef] 53. Xu, H. Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. Int. J. Remote Sens. 2006, 27, 3025–3033. [CrossRef] 54. Braden, B. The Surveyor’s Area Formula. Coll. Math. J. 1986, 17, 326–337. [CrossRef] 55. Curran, P.J.; Dungan, J.L.; Peterson, D.L. Estimating the foliar biochemical concentration of leaves with reflectance spectrometry: Testing the Kokaly and Clark methodologies. Remote Sens. Environ. 2001, 76, 349–359. [CrossRef] 56. Thiemann, S.; Kaufmann, H. Determination of chlorophyll content and trophic state of lakes using field spectrometer and IRS-1C satellite data in the Mecklenburg Lake District, Germany. Remote Sens. Environ. 2000, 73, 227–235. [CrossRef] 57. Schultz, G.; Engman, E. Remote Sensing in Hydrology and Water Management; Springer: Berlin/Heidelberg, Germany, 2000. 58. Toming, K.; Kutser, T.; Laas, A.; Sepp, M.; Paavel, B.; Noges, T. First experiences in mapping lakewater quality parameters with sentinel-2 MSI imagery. Remote Sens. 2016, 8, 1–14. [CrossRef] 59. U.S. Department of the Interior; U.S. Geological Survey. Product Guide—Provisional Landsat 8 Surface Reflectance Product; Version 2.1.; U.S. Geological Survey: Sioux Falls, SD, USA, 2016. Available online: https://landsat.usgs.gov/landsat-surface-reflectance-high-level-data-products (accessed on 17 August 2016). 60. Koschel, R.; Benndorf, J.; Proft, G.; Recknagel, F. Calcite precipitation a a natural control mechnism of eutrophication. Arch. Hydrobiol. 2011, 98, 340–408. 61. Riedmüller, U.; Hoehn, E.; Mischke, U. Trophieklassifikation von Seen. Richtlinie zur Ermittlung des Trophie-Index nach LAWA für natürliche Seen, Baggerseen, Talsperren und Speicherseen; Bund/Länderarbeitsgemeinschaft Wasser (LAWA), Ed.; Kulturbuch-Verlag: Berlin, Germany, 2014.

© 2017 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).