Short-term droughts forecast using Markov chain model in Victoria, Australia

• Published in: Theoretical and applied climatology 2017-07, Vol.129 (1-2), p.445-457
• Main Authors: Rahmat, Siti Nazahiyah, Jayasuriya, Niranjali, Bhuiyan, Muhammed A.
• Format: Article
• Language: English
• Subjects: Analysis; Aquatic Pollution; Atmospheric precipitations; Atmospheric Protection/Air Quality Control/Air Pollution; Atmospheric Sciences; Chains; Climate science; Climatology; Clusters; Drought; Drought characteristics; Droughts; Earth and Environmental Science; Earth Sciences; Long-term planning; Markov analysis; Markov chains; Markov processes; Mathematical models; Methods; Monitoring; Monitoring systems; Original Paper; Precipitation; Precipitation (Meteorology); Predictions; Probability theory; Risk; Risk management; Short-term planning; Stations; Strategic planning (Business); Time; Warning; Waste Water Technology; Water Management; Water Pollution Control; Weather; Weather forecasting
• ISSN: 0177-798X; 1434-4483
• DOI: 10.1007/s00704-016-1785-y

A comprehensive risk management strategy for dealing with drought should include both short-term and long-term planning. The objective of this paper is to present an early warning method to forecast drought using the Standardised Precipitation Index (SPI) and a non-homogeneous Markov chain model. A model such as this is useful for short-term planning. The developed method has been used to forecast droughts at a number of meteorological monitoring stations that have been regionalised into six (6) homogenous clusters with similar drought characteristics based on SPI. The non-homogeneous Markov chain model was used to estimate drought probabilities and drought predictions up to 3 months ahead. The drought severity classes defined using the SPI were computed at a 12-month time scale. The drought probabilities and the predictions were computed for six clusters that depict similar drought characteristics in Victoria, Australia. Overall, the drought severity class predicted was quite similar for all the clusters, with the non-drought class probabilities ranging from 49 to 57 %. For all clusters, the near normal class had a probability of occurrence varying from 27 to 38 %. For the more moderate and severe classes, the probabilities ranged from 2 to 13 % and 3 to 1 %, respectively. The developed model predicted drought situations 1 month ahead reasonably well. However, 2 and 3 months ahead predictions should be used with caution until the models are developed further.