Stochastic modeling of hydrological and meteorological drought in the South-West region of Bangladesh

Abstract

Drought is a hindering natural disaster that may affect every aspect of living life. It is believed that the trend of the recent climate change has triggered this disaster more than past. In this study, therefore, meteorological drought and ground water drought (hydrological drought) has been assessed to understand the historical scenario of water deficiency. A more recent and comparatively better performing Reconnaissance Drought Index (RDI) is used to assess meteorological drought. While a threshold level approach is used to derive the groundwater drought scenario. RDI is also used as a climate change indicator in the current study. While the trend in RDI index is calculated using nonparametric Mann-Kendall trend test, fluctuation of the trend is identified using Sequential Mann-Kendall test. Future forecasting drought events are done using ARIMA (Autoregressive Integrated Moving Average) and ANN (Artificial Neural Network) technique. Seasonal RDI shows that a fluctuation of the wet and dry cycle is common in most of the cases. The study area experienced several moderate but few extreme droughts in three months scale varying spatially and temporally. Six-month RDI reveals that Khulna and Mongla station experienced extreme droughts in 1992-93 hydrologic year in the reference period of April to September. In contrast, in the reference period October to March, only Jessore experienced severe droughts in 1976-77 and 2011-12. On the other hand, annual RDI shows that 1992-93 is an all dry condition for the whole study area varying from mild to extreme condition. However, it is observed that seasonal fluctuation of drought parameters reduces the number of drought events in annual scale. Trends in initial RDI index is used as the climate change indicator and found that most of the negative trends have started from after the ’90s. Results of the trend analysis indicate that there is a change in either precipitation or Potential Evapotranspiration. Future forecasting of RDI index is performed using ARIMA and BPNN (Backpropagation Neural Network). Results show that ANN outperformed ARIMA in future forecasting. BPNN forecast up to 3 times ahead with a reasonable accuracy while ARIMA to 1 ahead. Groundwater drought in the study area reveals that drought events are mostly confined to the February to June. Spatial distribution of water deficit shows that Satkhira has most water deficit than other areas. Normalized water deficit is found greater than 2m for several wells in the Kalaroa Upozilla of Satkhira district.