Evidence of Climate Change Based on Lake Surface Temperature Trends in South Central Chile

: Lake temperature has proven to act as a good indicator of climate variability and change. Thus, a surface temperature analysis at different temporal scales is important, as this parameter inﬂuences the physical, chemical, and biological cycles of lakes. Here, we analyze monthly, seasonal, and annual surface temperature trends in south central Chilean lakes during the 2000–2016 period, using MODIS satellite imagery. To this end, 14 lakes with a surface area greater than 10 km 2 were examined. Results show that 12 of the 14 lakes presented a statistically signiﬁcant increase in surface temperature, with a rate of 0.10 ◦ C/decade (0.01 ◦ C/year) over the period. Furthermore, some of the lakes in the study present a signiﬁcant upward trend in surface temperature, especially in spring, summer, and winter. In general, a signiﬁcant increase in surface water temperature was found in lakes located at higher altitudes, such as Maule, Laja and Galletu é lakes. These results contribute to the provision of useful data on Chilean lakes for managers and policymakers.


Introduction
Inland water ecosystems provide multiple ecosystem services and are vital for human consumption, irrigation, sanitation, transportation, recreation, culture, and industry [1]. In recent decades, these ecosystems have experienced high stress from various human impacts as well as climate change [2,3]. Researchers around the world have evaluated lake surface water temperature (LSWT) trends and have found variable increases in water temperatures. In many cases, these changes have been attributed to global warming and increases in air temperatures [4][5][6][7][8]. Recently, Jane et al. [9], studied the deoxygenation of temperate lakes, using more than 45,000 oxygen and temperature profiles collected from nearly 400 lakes. The main results indicated that oxygen levels declined by 5.5% at the surface and that surface temperatures increased by 0.38 • C/decade [9]. These studies have mainly used data records from the Northern Hemisphere; there are fewer observational records or satellite-based studies related to LSWT trends in the Southern Hemisphere, such that they are less well understood than LSWT trends in Northern Hemisphere lakes.
Water temperature is a key factor in aquatic ecosystems, as it directly or indirectly regulates physicochemical processes and reactions that occur within them [10,11]. This abiotic factor sets constraints on the type of organisms that can exist in each ecosystem, as the biotic components of the environment, from microorganisms to larger animals, such as http://www.cr2.cl/access date 25 November of 2020). The central region is characterized as semi-arid, with average annual precipitation of 100-500 mm in the central valley, concentrated in the austral winter (June-August) [32]. It has a long dry season of 7-8 months with high global radiation [33], low relative humidity and high temperatures from September to April. Lakes with a surface area ≥ 10 km 2 were selected based on [6]. The 14 selected lakes present a temperate monomictic circulation pattern, with thermal stratification during summer [34][35][36][37]. At present, 11 of the lakes chosen in this study are oligotrophic, while Villarrica has been classified as meso-oligotrophic, and Vichuquén and Lanalhue, as eutrophic [38]. Geographical and morphometric characteristics that affect LSWT, such as, location, elevation, surface area, perimeter, volume and mean and maximum depth, are presented in Table 1 [39].  Lakes with a surface area ≥10 km 2 were selected based on [6]. The 14 selected lakes present a temperate monomictic circulation pattern, with thermal stratification during summer [34][35][36][37]. At present, 11 of the lakes chosen in this study are oligotrophic, while Villarrica has been classified as meso-oligotrophic, and Vichuquén and Lanalhue, as eutrophic [38]. Geographical and morphometric characteristics that affect LSWT, such as, location, elevation, surface area, perimeter, volume and mean and maximum depth, are presented in Table 1 [39].

In Situ Parameters
The lake surface temperature was obtained through the free web-based hydrometeorological service made available online (http://www.dga.cl/servicioshidrometeorologicos/ access) (accessed on 15 October 2020) by the Dirección General de Aguas (DGA) (accessed on 15 October 2020). This downloadable database is available for 16 lakes that are part of the Red Mínima de Lagos (RML), in which Laja, Lanalhue, Caburga, Villarrica, Calafquén, Panguipulli and Riñihue lakes are included, and measured seasonally for the 2000-2014 period, while for Riñihue lake, data is available until 2015 [25]. The database was validated by POCH Ambiental S.A. in the report Redefinición de la Red Mínima de Lagos [40]. Colico, Lleulleu and Vichuquén lakes were later incorporated into the RML in 2013 [25]. The surface temperature of all lakes was obtained in situ used a portable multiparameter device Hydrolab DS5x. This measurement was taken between 09:00 and 15:00 at a depth of 50 cm. . This study used the database of thermal infrared imagery with high spatial resolution (1 km) using a split-window algorithm designed for a wide variety of land cover types including inland water surfaces, satellite viewing angles, and atmospheric conditions from sensors aboard the TERRA satellite, specifically, version 6 of the MOD11A2_LST product from the Land Processes Distributed Active Archive Center (LP DAAC) available at: https://lpdaac.usgs.gov/dataset_discovery/ modis/modis_products_table, which was downloaded from the NASA Earth Observing System Data and Information System (EOSDIS) [41] (accessed on 10 October 2020).

Pre-Processing of MODIS Images
For the pre-processing, daytime images (local time for the MODIS satellite overpass is approximately 10:30 a.m.) were reprojected to WSG84 19S using R Studio, which was achieved through an original sinusoidal reprojection (R Development Core Team 2016) [42]. In total, from a spatial resolution of 1 km and a temporal resolution of 8 days, 46 samples were attained for 8 days. The original scenes in HDF format were converted to raster GeoTIFF format and the pixels contaminated by cloud cover were replaced by null values and then visual inspection was used to remove cloud-contaminated images. Consequently, there are large LST spatial-temporal gaps over the studied lakes especially during winter due to the high percentage of cloud cover. We used images with less than 9% cloud cover. Monthly mean values from the pixel centroids for each lake were obtained to produce LSWT the time_series for the study area and period.

Statistical Analysis 2.4.1. Linear Fit between MODIS and In Situ Data
For the validation process, a least squares linear fit was applied to find the relationship between the surface water temperature acquired by processing MODIS images and in situ data. Errors in satellite derived LSWT may arise from instrument noise and drift, sun glint, residual cloud contamination (e.g., thin cirrus), misspecification of atmospheric attenuation and surface emissivity effects [43]. There are two important components of the skin-water temperature differences: the cool skin and the warm layer. The cool skin is always present at the air-water interface and refers to a systematic cool bias of the water skin temperature (0.1-0.6 • C) compared to that of the water less than 1 cm below [44]. Data was analysis in Remote Sens. 2021, 13, 4535 5 of 13 three temporal resolutions: monthly, seasonally, and annually measured temperatures. To quantify estimation errors, basic statistical analysis such as the coefficient of determination (R 2 ), root mean square error (RMSE) and mean absolute error (MAE) were used.

Trend Analysis of Time Series
To estimate LSWT trends in the time series (2000-2016), the Mann-Kendall nonparametric test was applied [45,46]. The Mann-Kendall test is a statistical test widely used for the analysis of trend in climatology and in hydrologic time series. One benefit of this test is that the data need not conform to any distribution [47].
Trend were tested using the slope estimator based on Kendall's Tau (τ) coefficient [48], and to assess the point of change of the series, the Pettitt test was applied [49]. The Pettitt test identifies whether the historical temperature series are homogenous. It is a non-parametric test that does not require a hypothesis on data distribution. The Pettitt test is adapted from the Mann-Whitney test and based on a range that identifies the moment of a transition (rupture) in the series (Pettitt 1979). The test consists of cutting the main series of N elements into two sets at each time t between 1 and N-1. The main series has a break at time t if the two sub-series have different distributions. This approach has been suggested by the World Meteorological Organization (WMO) of the United Nations [50], for analyzing trends in climatological and hydrological timeseries, and has been applied to an array of studies addressing climate change and variability [46,[51][52][53]. Statistical analysis was carried out using R software (R package version 1.1.0. on https: //cran.r-project.org/package=trend (accessed on 20 December 2020) [42,54]. Finally, all tests shown in this paper were considered statistically significant at the 5% level.

Validated Results for the Relationship between MODIS LSWT and In Situ LSWT
The MODIS-derived one-meter below surface temperature is essential to validate the MODIS-derived skin temperature against the one-meter below surface temperature from the in situ measurements. Satellite infrared sensors during completely cloud-free conditions only observed the temperature from the immediate surface or "skin" of the water rather than the surface temperature as measured from the in situ in monitoring campaign. The results of the validation between daily MODIS-derived skin temperature and the one-meter below surface temperature measure during satellite overpass in 2000-2016 are presented here. Although not all the lakes had an equal number of in situ data, the results show a high correlation between MODIS LSWT and in situ LSWT, with an R 2 coefficient ranging from 0.85 to 0.94 for six of the 14 lakes analyzed in this study ( Table 2). The best correlations were obtained for Villarrica lake, with R 2 = 0.94. There are currently no studies of LSWT using satellite imagery for most of the lakes. However, [55] and [56] studied the temporal variation of water characteristics of Panguipulli lake using Landsat 5 TM+, Landsat 7 ETM+ and Landsat 8 OLI/TIRS for surface water temperature and other parameters. The LSWT for Panguipulli lake found in this study are similar (R 2 = 0.86, RMSE = 1.61 • C and n = 43), as [55] achieved a R 2 = 0.86, although with the RMSE = 2.77 • C and n = 21. The results suggest that the processing of MODIS LSWT images and a larger dataset (possibly n = 30 to n = 40), and thus a greater temporal resolution, might be a better alternative for future observations. Furthermore, the results obtained from the MODIS LSWT data in this study expressed in root mean square error (RMSE) were between 1.07 and 1.88 • C. These results are similar to those obtained by Oesch et al. [54], who used AVHRR and MODIS to estimate LSWT in Swiss lakes (RMSE = 0.90-1.60 • C) and Moukomla and Blenken 2016 [57] for Great Lakes in North America (RMSE = 1.24-2.06 • C). Another study [27] using AVHRR, and multiple NOAA satellites achieved an RMSE < 1.50 • C.

Annual Trend Analysis of MODIS LSWT Timeseries
The trend analysis of the annual MODIS LSWT timeseries detected that only six of the 14 lakes present a significative increase (p < 0.05) (Figure 2). Particular attention should be paid to remote mountain lakes, as they are sensitive recorders of global change, and temperatures in these regions are increasing faster than in adjacent lowland sites [58]. The results of this study are yet another observation of this pattern, as Maule and Laja lakes, which are found at higher altitudes (2166 m a.s.l. and 1390 m a.s.l., respectively) had higher warming rates (0.010 • C/year) than lakes found at lower altitudes. Colico, Caburga and Villarrica lakes, with altitudes between 230 m a.s.l. and 505 m a.s.l., presented an approximate warming rate of 0.004 • C/year. Schneider et al. [5] used ATSR and SST data derived from the AVHRR sensor to observe the rapid warming of inland water bodies globally. The results of this study are similar to those obtained through other methods, including studies using data only available for the Northern Hemisphere.

Monthly Trend Analysis of MODIS LSWT Time Series
At a monthly scale, significant warming trends in most lakes were found in January, which corresponds to summer in the Southern Hemisphere (Table 3). Warming rates were between 0.007 • C/year to 0.016 • C/year, although Lanalhue and Huilipilún Lakes presented an increasing warming trend, these results were not significant (p-values > 0.05) and were therefore excluded from the results. It bears mentioning that evident warming trends in LSWT in January are consistent with studies of climate change in Chile. New climate trends in Chile are already evident, mainly manifested in changes in rainfall and temperatures across the country. According to [59,60], changes in temperature present upward trends on the ocean and the coast, while there are downward trends in the central valley and the Andes Mountains. A recent study by Vuille et al. [58], identified a similar contrast, highlighting a significant warming trend at inland sites, which is generalized in spring, summer, and autumn in recent decades. As with the annual results, a positive trend in LSWT was observed, particularly in lakes located at higher altitudes (Maule and Laja lakes). Contrasting results were observed in lakes at lower altitudes, consistent with the coastal cooling pattern reported by Chilean climatic studies [59,61,62]. Statistically significant results for the change point were obtained for the month of January for Vichuquén, Maule and Laja lakes (change point in 2011), while Galletué Lake reached the change point in 2007. According to the Pettitt test, change point in Villarrica Lake occurred in 2011. However, p-values obtained are >0.069. Nevertheless, it is interesting that around that time, much attention was given to the lake due to an increased frequency of algal blooms [63,64], the main causes of which were attributed to anthropogenic factors that influenced the water quality of the lake, which resulted in Decree 19 (2013) to protect the environmental quality of the lakes waters [65]. Recent studies indicate that warming trends in lakes, due to global warming, result in increasing oxygen loss that lead to higher phosphorous release from sediments; increasing oxygen depletion in deeper zones of lakes with thermal stratification patters [9]. The same article reported that some lakes present rising oxygen concentration near the surface with increased temperatures, particularly in lakes that have undergone nutrient enrichment from agriculture and urbanization, resulting in algal growth. As both nitrogen and phosphorous are important drivers of cyanobacterial blooms in terms of abundance and dominance, it is possible that the combined effect of higher temperatures and land use/change surrounding the lake is causing blooms. Most studies have focused on the role of nutrient loadings due to human and agricultural activities in water bodies [66]. Based on the results of this study, it can be concluded that the increase in LSWT has contributed to the potentially toxic cyanobacterial blooms in Villarrica Lake. The possible direct links between LSWT and land use and cyanobacterial blooms, need further research.

Monthly Trend Analysis of MODIS LSWT Time Series
At a monthly scale, significant warming trends in most lakes were found in January, which corresponds to summer in the Southern Hemisphere (Table 3). Warming rates were between 0.007 °C/year to 0.016 °C/year, although Lanalhue and Huilipilún Lakes presented an increasing warming trend, these results were not significant (p-values > 0.05) and were therefore excluded from the results. It bears mentioning that evident warming trends in LSWT in January are consistent with studies of climate change in Chile. New climate trends in Chile are already evident, mainly manifested in changes in rainfall and temperatures across the country. According to [59,60], changes in temperature present upward trends on the ocean and the coast, while there are downward trends in the central valley and the Andes Mountains. A recent study by Vuille et al. [58], identified a similar

Seasonal Trend Analysis of MODIS LSWT Timeseries
At the seasonal scale, no significant trends were observed; however, significant trends in temperature data in spring for Maule and Laja lakes, summer for Laja Lake, and winter for Vichuquén and Caburga lakes were observed, with Caburga Lake presenting a warming rate between 0.003 • C/year and 0.020 • C/year (Figure 3). In Europe, multiple studies confirmed the late spring/summer warming of European lakes [67], central European lakes [68], and lakes south of the Alps [69]. In addition, regional studies have evidenced the rapid warming of the Great Lakes of North America and lakes in Europe using satellite derived LSWT, such as [5] and [27]. O'Reilly et al. [6] used a combination of in situ and satellite observations to estimate the long-term LSWT trends and found an average summer increase rate of 0.03-0.04 • C/year. Using a different approach [4,68,69] reported warming at similar rates using in situ LSWT data for European lakes and the Great Lakes of North America. A more recent study by Jane et al. [9] achieved similar results, estimating rates of 0.036 • C/year using in situ data on surface temperatures of lakes worldwide.
Although most studies have shown that lakes are warming during summer [6,70,71], it is no less important that lakes are showing some warming during winter. In this context, the increasing trend found in Vichuquén and Caburga during winter could suggest that these systems tend to reduce the extent of the mixing period or extend their thermal stratification period. According to Straile et al. [72], a reduction of the mixing period could alter the transport of dissolved oxygen and nutrients through the water column. In addition, this winter increase could have a positive and/or negative effect on habitat availability for aquatic species. Positive because it would favor the emergence of cyanobacteria [64], and negative because stenothermal species would tend to migrate towards more favorable thermal conditions that allow them to survive [73,74].
Break points or positive direction change (increase in temperature) were reached for Vichuquén, Maule, Laja, Galletué, Caburga and Villarrica lakes. Additionally, at the seasonal time scale, Laja Lake reached a statistically significant change point in 2007. Maule Lake experienced another change point during spring 2009, while Vichuquén and Caburga lakes reached one during the winters of 2005 and 2011, respectively ( Table 4).
The rate obtained indicate that the seasonal trends are similar to those found in other studies, but they present lower rates of change, as mean summer values obtained by O'Reilly et al. [6], 0.030-0.040 ( • C/year), were almost double those found in this study. However, this difference might be due to the limited set of historical data on the studied lakes.
alter the transport of dissolved oxygen and nutrients through the water column. In addition, this winter increase could have a positive and/or negative effect on habitat availability for aquatic species. Positive because it would favor the emergence of cyanobacteria [64], and negative because stenothermal species would tend to migrate towards more favorable thermal conditions that allow them to survive [73,74].

Conclusions
The aim of this investigation was to analyze monthly, seasonal, and annual surface temperature trends in 14 south central Chilean lakes during the 2000-2016 period using MODIS satellite imagery. The results of this study suggest that the processing of MODIS LSWT images is appropriate and show excellent agreement with in situ LSWT, making it a viable alternative for future observations of lakes >10 km 2 . Furthermore, 12 of the 14 lakes presented a statistically significant increase in surface temperature, with a rate of 0.10 • C/decade (0.01 • C/year) over the study period. At a seasonal scale, some of the lakes in the study area present a significant upward trend in surface temperatures, especially in spring, summer, and winter. In general, significant increase in surface water temperatures are found in lakes located at higher altitudes, such as Maule and Laja lakes. The increase in surface temperature and the change point obtained by the Pettitt test are consistent with site observations and an increased frequency of potentially toxic cyanobacterial blooms in Villarrica Lake. This suggests direct links between LSWT, dissolved oxygen at the surface of the lake and land use/change.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author.