Abstarct: Water bodies are known as an effective factor on the human environment and on the other living organism. Hence, investigating the quality of water bodies is one of the most important issues in the environmental researches. For this aim, awareness of water quality parameters is necessary and inevitable. Due to some difficulties in measuring these parameters in different regions, and nowadays due to the effects of various parameters, researches try to estimate these parameters by machine learning and advanced statistical methods. According to the development of space science, researchers have used spectral radiance data. Considering the measurement error of these data, and the impact of atmosphere on the spectral radiance data, noise has always been an inseparable part of this type of data. So, finding a method for appropriate modeling under noise conditions is necessary to to convert the spectral radiance data to water quality data. In this thesis, not only the Random Forest (RF) and Support Vector Machine (SVM) are introduced, but also the performance of these methods for estimation of water quality parameters from spectral radiance data is evaluated. According to the results, utilizing of RF and SVM methods for modeling of chlorophy and pigment using NOMAD and SeaBAM data, respectively and R_rs(λ)/R_rs(555) as input variables lead to appropriate results. In the estimation of chlorophyll-a concentration using NOMAD databaxse, SVM method lead to minimum error of MPAE among three methods RF, SVM and Active Learning Method (ALM)) under different noise levels. In the estimation of pigment concentration using SeaBAM databaxse, the RF and SVM methods lead to smaller RMSE than ALM, Artificial Neural Networks (ANN) and some of the empirical algorithms. In general, for MOMO databaxse RF and SVM methods presented higher accuracy than ANN method for the estimation of water quality parameters. In general, according to the accuracy of SVM and RF and their calculation costs, it can be implied that the performance of RF is better than SVM Method in this study.