Search

Chromophoric dissolved organic matter (CDOM) is crucial in the biogeochemical cycle and carbon cycle of aquatic environments. However, in inland waters, remotely sensed estimates of CDOM remain challenging due to the low optical signal of CDOM and complex optical conditions. Therefore, developing efficient, practical and robust models to estimate CDOM absorption coefficient in inland waters is essential for successful water environment monitoring and management. We examined and improved different machine learning algorithms using extensive CDOM measurements and Landsat 8 images covering different trophic states to develop the robust CDOM estimation model. The algorithms were evaluated via 111 Landsat 8 images and 1708 field measurements covering CDOM light absorption coefficient a(254) from 2.64 to 34.04 m−1. Overall, the four machine learning algorithms achieved more than 70% accuracy for CDOM absorption coefficient estimation. Based on model training, validation and the application on Landsat 8 OLI images, we found that the Gaussian process regression (GPR) had higher stability and estimation accuracy (R2 = 0.74, mean relative error (MRE) = 22.2%) than the other models. The estimation accuracy and MRE were R2 = 0.75 and MRE = 22.5% for backpropagation (BP) neural network, R2 = 0.71 and MRE = 24.4% for random forest regression (RFR) and R2 = 0.71 and MRE = 24.4% for support vector regression (SVR). In contrast, the best three empirical models had estimation accuracies of R2 less than 0.56. The model accuracies applied to Landsat images of Lake Qiandaohu (oligo-mesotrophic state) were better than those of Lake Taihu (eutrophic state) because of the more complex optical conditions in eutrophic lakes. Therefore, machine learning algorithms have great potential for CDOM monitoring in inland waters based on large datasets. Our study demonstrates that machine learning algorithms are available to map CDOM spatial-temporal patterns in inland waters.

Aquatic vegetation serves many important ecological and socioeconomic functions in lake ecosystems. The presence of floating algae poses difficulties for accurately estimating the distribution of aquatic vegetation in eutrophic lakes. We present an approach to map the distribution of aquatic vegetation in Lake Taihu (a large, shallow eutrophic lake in China) and reduce the influence of floating algae on aquatic vegetation mapping. Our approach involved a frequency analysis over a 2003–2013 time series of the floating algal index (FAI) based on moderate-resolution imaging spectroradiometer (MODIS) data. Three phenological periods were defined based on the vegetation presence frequency (VPF) and the growth of algae and aquatic vegetation: December and January composed the period of wintering aquatic vegetation; February and March composed the period of prolonged coexistence of algal blooms and wintering aquatic vegetation; and June to October was the peak period of the coexistence of algal blooms and aquatic vegetation. By comparing and analyzing the satellite-derived aquatic vegetation distribution and 244 in situ measurements made in 2013, we established a FAI threshold of −0.025 and VPF thresholds of 0.55, 0.45 and 0.85 for the three phenological periods. We validated the accuracy of our approach by comparing the results between the satellite-derived maps and the in situ results obtained from 2008–2012. The overall classification accuracy was 87%, 81%, 77%, 88% and 73% in the five years from 2008–2012, respectively. We then applied the approach to the MODIS images from 2003–2013 and obtained the total area of the aquatic vegetation, which varied from 265.94 km2 in 2007 to 503.38 km2 in 2008, with an average area of 359.62 ± 69.20 km2 over the 11 years. Our findings suggest that (1) the proposed approach can be used to map the distribution of aquatic vegetation in eutrophic algae-rich waters and (2) dramatic changes occurred in the distribution of aquatic vegetation in Lake Taihu during the 11-year study.

The Secchi disk depth (ZSD, m) has been used globally for many decades to represent water clarity and an index of water quality and eutrophication. In recent studies, a new theory and model were developed for ZSD, which enabled its semi-analytical remote sensing from the measurement of water color. Although excellent performance was reported for measurements in both oceanic and coastal waters, its reliability for highly turbid inland waters is still unknown. In this study, we extend this model and its evaluation to such environments. In particular, because the accuracy of the inherent optical properties (IOPs) derived from remote sensing reflectance (Rrs, sr−1) plays a key role in determining the reliability of estimated ZSD, we first evaluated a few quasi-analytical algorithms (QAA) specifically tuned for turbid inland waters and determined the one (QAATI) that performed the best in such environments. For the absorption coefficient at 443 nm (a(443), m−1) ranging from ~0.2 to 12.5 m−1, it is found that the QAATI-derived absorption coefficients agree well with field measurements (r2 > 0.85, and mean absolute percentage difference (MAPD) smaller than ~39%). Furthermore, with QAATI-derived IOPs, the MAPD was less than 25% between the estimated and field-measured ZSD (r2 > 0.67, ZSD in a range of 0.1−1.7 m). Furthermore, using matchup data between Rrs from the Medium Resolution Imaging Spectrometer (MERIS) and in-situ ZSD, a similar performance in the estimation of ZSD from remote sensing was obtained (r2 = 0.73, MAPD = 37%, ZSD in a range of 0.1−0.9 m). Based on such performances, we are confident to apply the ZSD remote sensing scheme to MERIS measurements to characterize the spatial and temporal variations of ZSD in Lake Taihu during the period of 2003−2011.