Drought Forecasting with Time Series and Machine Learning Approaches

2017-12-08
EVKAYA, ÖMER OZAN
YOZGATLIGİL, CEYLAN
KESTEL, AYŞE SEVTAP
As a main reason of undesired agricultural, economic and environmental damages, drought is one of the most important stochastic natural hazard having certain features. In order to manage the impacts of drought, more than 100 drought indices have been proposed for both monitoring and forecasting purposes [1], [3]. For different types of droughts, these indices have been used to understand the effects of dry periods including meteorological, agricultural and hydrological droughts in many distinct locations. In this respect, the future projections of drought indices allow the decision makers to assess certain risks of dry periods beforehand. In addition to the use of classical time series techniques for understanding the upcoming droughts, machine learning methods might be effective alternatives for forecasting the future events based on relevant drought index [2]. This study aims to identify the benefits of various methods for forecasting the future dry seasons with widely known drought indices. For that purpose, Standardized Precipitation Index (SPI), Standardized Precipitation Evapotranspiration Index (SPEI) and Reconnaissance Drought Index (RDI) have been considered over different time scales (3, 6, 9 months) to represent drought in Kulu weather station, Konya. The considered drought indices were used for forecasting the future period using both time series prediction tools and machine learning techniques. The forecast results of all methods with respect to different drought indices were examined with the data set of 1950-2010 for Kulu station. The potential benefits and limitations of various methods and drought indices were discussed in detail
Citation Formats
Ö. O. EVKAYA, C. YOZGATLIGİL, and A. S. KESTEL, “Drought Forecasting with Time Series and Machine Learning Approaches,” presented at the 10.International Statistics Congress (ISC2017), (6 - 08 Aralık 2017), Ankara, Türkiye, 2017, Accessed: 00, 2021. [Online]. Available: https://hdl.handle.net/11511/79067.