In this study, we combined machine learning and remote sensing techniques to estimate the value of chlorophyll-a concentration in a freshwater ecosystem in the South American continent (lake in Southern Chile). In a previous study, nine artificial intelligence (AI) algorithms were tested to predict water quality data from measurements during monitoring campaigns. In this study, in addition to field data (Case A), meteorological variables (Case B) and satellite data (Case C) were used to predict chlorophyll-a in Lake Llanquihue. The models used were SARIMAX, LSTM, and RNN, all of which showed generally good statistics for the prediction of the chlorophyll-a variable. Model validation metrics showed that all three models effectively predicted chlorophyll as an indicator of the presence of algae in water bodies. Coefficient of determination values ranging from 0.64 to 0.93 were obtained, with the LSTM model showing the best statistics in any of the cases tested. The LSTM model generally performed well across most stations, with lower values for MSE (<0.260 (μg/L)²), RMSE (<0.510 ug/L), MaxError (<0.730 μg/L), and MAE (<0.442 μg/L). This model, which combines machine learning and remote sensing techniques, is applicable to other Chilean and world lakes that have similar characteristics. In addition, it is a starting point for decision-makers in the protection and conservation of water resource quality.

Surface water storage in floodplains and wetlands is poorly known from regional to global scales, in spite of its importance in the hydrological and the carbon balances, as the wet areas are an important water compartment which delays water transfer, modifies the sediment transport through sedimentation and erosion processes, and are a source for greenhouse gases. Remote sensing is a powerful tool for monitoring temporal variations in both the extent, level, and volume, of water using the synergy between satellite images and radar altimetry. Estimating water levels over flooded area using radar altimetry observation is difficult. In this study, an unsupervised classification approach is applied on the radar altimetry backscattering coefficients to discriminate between flooded and non-flooded areas in the Cuvette Centrale of Congo. Good detection of water (open water, permanent and seasonal inundation) is above 0.9 using radar altimetry backscattering from ENVISAT and Jason-2. Based on these results, the time series of water levels were automatically produced. They exhibit temporal variations in good agreement with the hydrological regime of the Cuvette Centrale. Comparisons against a manually generated time series of water levels from the same missions at the same locations show a very good agreement between the two processes (i.e., RMSE ≤ 0.25 m in more than 80%/90% of the cases and R ≥ 0.95 in more than 95%/75% of the cases for ENVISAT and Jason-2, respectively). The use of the time series of water levels over rivers and wetlands improves the spatial pattern of the annual amplitude of water storage in the Cuvette Centrale. It also leads to a decrease by a factor of four for the surface water estimates in this area, compared with a case where only time series over rivers are considered.

The floods are an annual phenomenon on the Pacific Coast of Ecuador and can become devastating during El Niño years, especially in the Guayas watershed (32,300 km²), the largest drainage basin of the South American western side of the Andes. As limited information on flood extent in this basin is available, this study presents a monitoring of the spatio-temporal dynamics of floods in the Guayas Basin, between 2005 and 2008, using a change detection method applied to ENVISAT ASAR Global Monitoring SAR images acquired at a spatial resolution of 1 km. The method is composed of three steps. First, a supervised classification was performed to identify pixels of open water present in the Guayas Basin. Then, the separability of their radar signature from signatures of other classes was determined during the four dry seasons from 2005 to 2008. In the end, standardized anomalies of backscattering coefficient were computed during the four wet seasons of the study period to detect changes between dry and wet seasons. Different thresholds were tested to identify the flooded areas in the watershed using external information from the Dartmouth Flood Observatory. A value of −2.30 ± 0.05 was found suitable to estimate the number of inundated pixels and limit the number of false detection (below 10%). Using this threshold, monthly maps of inundation were estimated during the wet season (December to May) from 2004 to 2008. The most frequently inundated areas were found to be located along the Babahoyo River, a tributary in the east of the basin. Large interannual variability in the flood extent is observed at the flood peak (from 50 to 580 km²), consistent with the rainfall in the Guayas watershed during the study period.

This study aims to develop and implement a methodology for retrieving bio-optical parameters in a lagoon located in the Biobío region, South-Central Chile, by analyzing time series of Landsat-8 OLI satellite images. The bio-optical parameters, i.e., chlorophyll-a (Chl-a, in mg·m⁻³) and turbidity (in NTU) were measured in situ during a satellite overpass to minimize the impact of atmospheric distortions. To calibrate the satellite images, various atmospheric correction methods (including ACOLITE, C2RCC, iCOR, and LaSRC) were evaluated during the image preprocessing phase. Spectral signatures obtained from the scenes for each atmospheric correction method were then compared with spectral signatures acquired in situ on the water surface. In short, the ACOLITE model emerged as the best fit for the calibration process, reaching R² values of 0.88 and 0.79 for Chl-a and turbidity, respectively. This underlies the importance of using inversion models, when processing water surfaces, to mitigate errors due to aerosols and the sun-glint effect. Subsequently, reflectance data derived from the ACOLITE model were used to establish correlations between various spectral indices and the in situ data. The empirical retrieval models (based on band combinations) yielding superior performance, with higher R² values, were subjected to a rigorous statistical validation and optimization by applying a bootstrapping approach. From this process the green chlorophyll index (GCI) was selected as the optimal choice for constructing the Chl-a retrieval model, reaching an R² of 0.88, while the red + NIR spectral index achieved the highest R² value (0.79) for turbidity analysis, although in the last case, it was necessary to incorporate data from several seasons for an adequate model training. Our analysis covered a broad spectrum of dates, seasons, and years, which allowed us to search deeper into the evolution of the trophic state associated with the lake. We identified a striking eight-year period (2014–2022) characterized by a decline in Chl-a concentration in the lake, possibly attributable to governmental measures in the region for the protection and conservation of the lake. Additionally, the OLI imagery showed a spatial pattern varying from higher Chl-a values in the northern zone compared to the southern zone, probably due to the heat island effect of the northern urban areas. The results of this study suggest a positive effect of recent local regulations and serve as the basis for the creation of a modern monitoring system that enhances traditional point-based methods, offering a holistic view of the ongoing processes within the lake.

The watershed hydrologic conditions in the Madre de Dios (MDD) Basin in the Peruvian Amazon have been irreversibly impacted by deforestation and changes in land cover. These changes have also had detrimental effects on the geomorphology, water quality, and aquatic habitat within the basin. However, there is a scarcity of hydrological modeling studies in this area, primarily due to the limited availability of hydrometeorological data. The primary objective of this study was to examine how deforestation impacts the hydrological conditions in the MDD Basin. By implementing the Soil and Water Assessment Tool (SWAT) model, this study determined that replacing 12% of the evergreen broadleaf forest area with bare land resulted in a significant increase in surface runoff, by 38% monthly, a 1% annual reduction of evapotranspiration, and an average monthly streamflow increase of 12%. Changes in spatial patterns reveal that the primary impacted watershed is the Inambari River subbasin, a significant tributary of the Madre de Dios River. This area experiences an annual average surge of 187% in surface runoff generation while witnessing an annual average reduction of 8% in evapotranspiration. These findings have important implications, as they can contribute to instances of flooding and extreme inundation events, which have already occurred in the MDD region.

In this study, we employ in situ, meteorological, and remote sensing data to estimate chlorophyll-a concentration at different depths in a South American freshwater ecosystem, focusing specifically on a lake in southern Chile known as Lake Maihue. For our analysis, we explored four different scenarios using three deep learning and traditional statistical models. These scenarios involved using field data (Scenario 1), meteorological variables (Scenario 2), and satellite data (Scenarios 3.1 and 3.2) to predict chlorophyll-a levels in Lake Maihue at three different depths (0, 15, and 30 m). Our choice of models included SARIMAX, DGLM, and LSTM, all of which showed promising statistical performance in predicting chlorophyll-a concentrations in this lake. Validation metrics for these models indicated their effectiveness in predicting chlorophyll levels, which serve as valuable indicators of the presence of algae in the water body. The coefficient of determination values ranged from 0.30 to 0.98, with the DGLM model showing the most favorable statistics in all scenarios tested. It is worth noting that the LSTM model yielded comparatively lower metrics, mainly due to the limitations of the available training data. The models employed, which use traditional statistical and machine learning models and meteorological and remote sensing data, have great potential for application in lakes in Chile and the rest of the world with similar characteristics. In addition, these results constitute a fundamental resource for decision-makers involved in the protection and conservation of water resource quality.