The analysis of drought indices forecast using neural networks

Bc. Anna Morozova

Master's thesis

Abstract:

Drought forecasting is a critical component of drought risk management. The presented thesis analyses the forecasting drought indices with Artificial Neural Network (ANN). The indices used are the Standardized Precipitation Index (SPI), the Standardized Precipitation Evaporation Index (SPEI) and the Standardized Soil Moisture Index (SSI). Tested neural network was multilayer perceptron with two hidden layers and was trained using backpropagation algorithm. I used the data obtained from 13 meteorological stations, located in different parts of Czech Republic. The records were from the period from 1.1.1982 to 1.12.2015. For all three indices, a number of different models with the lead time of 1 to 12 months have been tested out. The best models have the R2 values from 0,83 to 0,98, it is also show that forecasts of SSI were superior. The result of drought indices forecast, explained by the values of four model performance indices, such as Mean Error (ME), Mean Square Error (MSE), Root Mean Square Error (RMSE) and coefficient of determination (R2). Artificial Neural Network model show excellent results for forecasting drought.

Keywords

drought index, neural network, forecast, drought

Jazyk práce: English

Datum vytvoření / odevzdání či podání práce: 18. 4. 2017

Obhajoba závěrečné práce

• Vedúci: doc. Ing. Petr Máca, Ph.D.

Czech University of Life Sciences Prague

Faculty of Environmental Sciences

Master programme / odbor:
Landscape Engineering / Water in the Landscape