Simulating chlorophyll a in dam reservoirs using remote sensing and data-driven approaches

Chlorophyll A is used as a Indicator to measure the amount of algae growing in water, which can be applied to classify the nutritional status of water bodies. Accordingly, the concentration of chlorophyll a is considered a key indicator for water quality. In this research, an attempt has been made to estimate the amount of chlorophyll A in Sardasht dam reservoir using remote sensing techniques and data driven models. The data categories was used in this study: the first part is the measured values of chlorophyll A in Sardasht dam reservoir, and the second part includes 7 stages of measurement at 10 points with different coordinates in the reservoir of the dam and is related to March 2016 to June 2018. The second part of the data which used in data driven models, was extracted from the Sentinel-2 satellite images. The information of different bands was extracted from Sentinel-2 images and based on the band values, the criteria for measuring the amount of chlorophyll A were calculated and provided to the data driven models for training. In this research, three data driven models XGBoost, M5 and MARS were used to estimate the amount of chlorophyll A. In this study, 9 chlorophyll A estimation equations were considered as input of data driven models and the measured logarithm value of chlorophyll A was considered as output. 80% of the available data was used to train and the remaining 20% was used to verify the effectiveness of the used models. Based on the input data, the M5 model tree has divided the problem space into 5 parts and presented a linear equation for each part. Based on the structure provided by M5 and MARS algorithms, blue, red and green band combinations as well as infrared and red have a high impact on the models provided by these two algorithms. The results obtained from XGBoost algorithm show the importance of blue, red and green band combinations on the presented results. Based on this, the combination of blue, red and green bands has been used in all three algorithms as the most important or one of the most important input variables to calculate chlorophyll A. The coefficients of determination for three models XGBoost, M5 and MARS was calculated as 0.61, 0.49 and 0.31, respectively. The value of Nash-Sutcliffe coefficient for XGBoost, M5 and MARS models was calculated as 0.54, 0.47 and 0.27, respectively, which shows that the results of XGBoost and M5 models are favorable. The results show that the XGBoost and M5 models provided more accurate results than the MARS model. The use of Taylor's diagram also shows the close efficiency of the XGBoost and M5 models in calculating the amount of chlorophyll A. The spatial distribution of chlorophyll A in the Sardasht dam reservoir shows that the lack of information used has caused differences between the measured and calculated values in limited areas. The spatial distribution of chlorophyll A in the Sardasht dam reservoir shows that the limited data used and the lack of complete temporal compatibility of the Sentinel-2 images with the measurement data in the dam reservoir has caused differences between the measured and calculated values in limited areas. The use of a large number of data in the reservoirs of different dams, the use of various data driven models and applying images from other sensors can provide a suitable tool for the managers of the reservoirs so that they can evaluate the water quality more accurately.