The thesis deals with the determination of the movement patterns and the relation between landslide movement and landslide triggers for Kostanjek landslide. The Kostanjek landslide is the biggest landslide in the Republic of Croatia, activated in 1963. The determination of the landslide movement patterns and the relation between landslide movement and triggers was performed on data continuously measured by the Kostanjek landslide monitoring system. The integrated monitoring system consists of about 40 sensors that measure the landslide movement, the hydrological conditions and the landslide triggers. The relation between landslide movement and landslide triggers, determined empirically on the basis of continuous data series, represents a phenomenological landslide model. The Kostanjek landslide phenomenological model is defined on the basis of data collected over a monitoring period of two years, from January 2013 to January 2015. The Kostanjek landslide movement, interpreted as translational sliding along a fully developed sliding surface, consists of periods of faster displacement and periods of slower displacement or periods of rest. The beginning and the end of every period of faster or slower displacement is determined by statistical method of change point analysis. In total, five periods of faster displacement were determined in which more than 90% of cumulative landslide displacement occurred. The maximum measured velocity in the central part of landslide was 4.8 mm/day, while in the area near the landslide border velocities were two to three times lower. The total cumulative horizontal displacement, measured in the central part of landslide during the monitoring period, was 426.09 mm. Periods of faster displacement are the result of nine groundwater level rising periods, during which water levels rose from 0.19 m to 5.06 m. Statistical analysis also showed that the periods of groundwater level rising, detected in the piezometer in the central part of landslide, lasted from six to nine days, while the daily groundwater level changes were up to 0.87 m. Analyses revealed that the sliding velocity depends on groundwater level and that this relationship can be described with hysteresis. The groundwater level changes as well as the periods of faster displacement are result of extremely wet conditions during the monitoring period, which was determined according to historical data analysis from the meteorological station Zagreb-Grič. The result of correlation analysis between precipitation of different antecedent periods and groundwater level changes showed that precipitations usualy influence groundwater level over a seven days period. The relations between precipitation and groundwater level and groundwater level and landslide movement were statistically analyzed by multiple linear regression and random forest statistical methods. The purpose of development of these models is the prediction of groundwater level on the basis of precipitation data and the prediction of landslide movement based on the groundwater level data. In the process of the development of groundwater level prediction model, different types of precipitation parameters were used: different precipitation events, different anteceded precipitations, cumulative precipitations, effective precipitations, modified precipitations and influence of the season of the year. Modeling included influence of rainfall to groundwater level and influence of rainfall to groundwater level change rate. Independent variables, such as long lasting anteceded and effective precipitations and the time period from the last 50 mm precipitation have the greatest effect in all groundwater level prediction models. Groundwater level, groundwater level change rate for different periods, groundwater level change acceleration and the impact of the season of the year were used as independent variables for development of the landslide velocity prediction model. According to the analysis, independent variables such as groundwater level and seven-day groundwater level change rate have the greatest influence on the landslide movement prediction model. Model comparison, carried out on the basis of statistical indicators such as mean squared error of the model, correlation coefficient between measured and modeled values and the methods of k-fold cross validation and validation, demonstrated that the random forest model has a better prediction capability in both cases, i.e. for the groundwater level prediction and for the landslide velocity prediction. Verification of phenomenological model was performed for the period which encompasses the last 10 to 90 days of monitored period. Firstly, the groundwater levels were predicted from the precipitation data and secondly, the landslide velocities were predicted from the predicted groundwater levels, followed by calculation of cumulative landslide movements. Mean squared error and the difference between predicted and measured values of cumulative displacement showed that the error for the prediction period of 10 to 30 days is acceptable. Relative error of cumulative displacement prediction depends upon duration of the prediction period and on the cumulative displacement during the prediction period. It is possible to increase the prediction capability of landslide movement by further development of the phenomenological landslide model derived by random forest method, on the basis of longer time data series of the Kostanjek landslide monitoring. On the basis of the fact that there are no sufficient data for the development of a reliable physically based model of the Kostanjek landslide, the developed phenomenological model of the Kostanjek landslide movement is the most appropriate model for predictions for the purpose of establishing an early warning system.