Institute of Geophysics, University of TehranJournal of the Earth and Space Physics2538-371X40120140421Detection of climate change effect on meteorological droughts in northwest of IranDetection of climate change effect on meteorological droughts in northwest of Iran1671843670410.22059/jesphys.2014.36704FAMahdiGhamghamiPh.D. Student of Agro-Meteorology, University of Tehran, IranNozarGhahremanAssociate Professor, Department of Irrigation and Reclamation Engineering, University of Tehran, IranSomayehHejabiPh.D. Student of Agro-Meteorology, University of Tehran, IranJournal Article20130716Climate change that the human faces is a somewhat unavoidable phenomenon. Successful management of water resources needs recognition and perception of climate change in order to cope with water scarcity. The water scarcity is created by natural forcings such as drought, which is affected by regional climate. In other words, variation of climate variables as a result of climate change leads to variations in drought severity and frequency. Since climate change scenarios are based on assumption of increasing, decreasing or non-significant trend in climatic means, It is expected that the effects of these assumptions would be reflected in the prediction of meteorological phenomena like
drought. In Markov analysis, these variations are determined as change in transfer probability function values or shift in drought severity class, which are both important in management decisions. For instance, by increasing the temperature or decreasing the rainfall it is expected that occurrence of a drought event under certain conditions would be more probable. In this study, the outputs of three General Circulation Models (GCMs) namely; ECHO-G, CGCM3T63 HADCM3 under three climate change scenarios were downscaled using a non-parametric approach for simulation of rainfall and temperature series during 2011-2040 in northwest of Iran. This downscaling approach is combination of two techniques i.e. Kernel probability density function estimator (KDE) and Strategic Re-sampling method by which predicted variations of GCM outputs are extended and transformed to generated time series of a given future period. In KDE method, A probability density function is defined with center value of i<sup>th</sup> observation from series (<em>x<sub>i</sub> , i=1,...,n</em>). Contribution of each observation in estimation of probability density function of <em>i</em><sub>th</sub> observation is estimated by this Kernel function. The main parameter of this function is the bandwidth which is, by mathematical definition, a distance on x-axis in which the function variation is insignificant. Firstly a random normal kernel is selected and its average is considered as the base vector. Selection probability of each vector is 1/n. Then by calculation of cumulative probability and comparison with a random number between 0 and 1, one of the normal kernels is selected for rest of the simulations.
The strategic re-sampling method uses a rule for generating series with specific feature such as increasing frequency of warmer or more rainy days. The criteria for such features are selected by the user based on the outputs of GCMs. Considering its semi random nature, this approach cannot be used alone for regional climate change simulations and should be combined with a weather generator such that the applied rule should be run on observed or historical series. Then, the outputs are feed in weather generator for generating a completely random series coincide with climatic scenario. After simulation of climate, Reconnaissance Drought Index (RDI) was used for monitoring drought during two periods 1971-2000 and 2011 to 2040 in northwest of Iran. This index uses the ratio of precipitation and evapotranspiration (calculated by Thorntwait method), hence as the index becomes smaller, more severe would be the drought. Thus, the necessary variables for RDI estimation are monthly mean temperature and total rainfall. For RDI calculation, firstly, the precipitation (prec) and potential evapotranspiration (PET) are calculated cumulatively with determination of the moving window value, and then, RDI values are obtained as logarithm of cumulative prec to PET ratio. Four classes are considered for RDI including: normal class (larger than -1), moderately drought class (-1 to -1.5), severe drought class (-1.5 to -2) and very severe drought (less than -2).
Taking into account the length of the dry periods in the arid regions of the country, the Reconnaissance Drought Index in 6-month timescale was used for drought monitoring. Markov analysis was applied for calculation of transfer probability and corresponding drought severity classes with three steps forward to assess the sequences of climatic assumptions on management early warnings. Behavior of a Markov model is determined by a series of probabilities in transition from one state to another namely transition probabilities. These probabilities may vary by climate change. The first-order Markov chain model was employed to predict drought condition up to 3-step ahead. This model was fitted on the RDI series at all stations of interest, and it was identified that it can represent the probabilistic behavior of drought over northwest of Iran.
Research findings are presented in three parts of downscaling method implementation, RDI monitoring and Markov analysis. The weather generator model was successful for simulation of monthly normals including means and standard deviation. Also, strategic re-sampling technique as aligning method was successful for simulation of deviations from normal. Drought monitoring with RDI showed a water tension resonance in second 15 years of 1971-2000 periods. Likewise, in part of Markov analysis, findings of this study revealed that under conditions of increasing temperature and decreasing rainfall, as the worst case, the effect of climate change on meteorological drought would appear as the class shift, and in most of the study stations under this scenario, increased duration of extremely drought (class 4) was forecasted, even 2 steps ahead, which is important in water resource management.
Climate change that the human faces is a somewhat unavoidable phenomenon. Successful management of water resources needs recognition and perception of climate change in order to cope with water scarcity. The water scarcity is created by natural forcings such as drought, which is affected by regional climate. In other words, variation of climate variables as a result of climate change leads to variations in drought severity and frequency. Since climate change scenarios are based on assumption of increasing, decreasing or non-significant trend in climatic means, It is expected that the effects of these assumptions would be reflected in the prediction of meteorological phenomena like
drought. In Markov analysis, these variations are determined as change in transfer probability function values or shift in drought severity class, which are both important in management decisions. For instance, by increasing the temperature or decreasing the rainfall it is expected that occurrence of a drought event under certain conditions would be more probable. In this study, the outputs of three General Circulation Models (GCMs) namely; ECHO-G, CGCM3T63 HADCM3 under three climate change scenarios were downscaled using a non-parametric approach for simulation of rainfall and temperature series during 2011-2040 in northwest of Iran. This downscaling approach is combination of two techniques i.e. Kernel probability density function estimator (KDE) and Strategic Re-sampling method by which predicted variations of GCM outputs are extended and transformed to generated time series of a given future period. In KDE method, A probability density function is defined with center value of i<sup>th</sup> observation from series (<em>x<sub>i</sub> , i=1,...,n</em>). Contribution of each observation in estimation of probability density function of <em>i</em><sub>th</sub> observation is estimated by this Kernel function. The main parameter of this function is the bandwidth which is, by mathematical definition, a distance on x-axis in which the function variation is insignificant. Firstly a random normal kernel is selected and its average is considered as the base vector. Selection probability of each vector is 1/n. Then by calculation of cumulative probability and comparison with a random number between 0 and 1, one of the normal kernels is selected for rest of the simulations.
The strategic re-sampling method uses a rule for generating series with specific feature such as increasing frequency of warmer or more rainy days. The criteria for such features are selected by the user based on the outputs of GCMs. Considering its semi random nature, this approach cannot be used alone for regional climate change simulations and should be combined with a weather generator such that the applied rule should be run on observed or historical series. Then, the outputs are feed in weather generator for generating a completely random series coincide with climatic scenario. After simulation of climate, Reconnaissance Drought Index (RDI) was used for monitoring drought during two periods 1971-2000 and 2011 to 2040 in northwest of Iran. This index uses the ratio of precipitation and evapotranspiration (calculated by Thorntwait method), hence as the index becomes smaller, more severe would be the drought. Thus, the necessary variables for RDI estimation are monthly mean temperature and total rainfall. For RDI calculation, firstly, the precipitation (prec) and potential evapotranspiration (PET) are calculated cumulatively with determination of the moving window value, and then, RDI values are obtained as logarithm of cumulative prec to PET ratio. Four classes are considered for RDI including: normal class (larger than -1), moderately drought class (-1 to -1.5), severe drought class (-1.5 to -2) and very severe drought (less than -2).
Taking into account the length of the dry periods in the arid regions of the country, the Reconnaissance Drought Index in 6-month timescale was used for drought monitoring. Markov analysis was applied for calculation of transfer probability and corresponding drought severity classes with three steps forward to assess the sequences of climatic assumptions on management early warnings. Behavior of a Markov model is determined by a series of probabilities in transition from one state to another namely transition probabilities. These probabilities may vary by climate change. The first-order Markov chain model was employed to predict drought condition up to 3-step ahead. This model was fitted on the RDI series at all stations of interest, and it was identified that it can represent the probabilistic behavior of drought over northwest of Iran.
Research findings are presented in three parts of downscaling method implementation, RDI monitoring and Markov analysis. The weather generator model was successful for simulation of monthly normals including means and standard deviation. Also, strategic re-sampling technique as aligning method was successful for simulation of deviations from normal. Drought monitoring with RDI showed a water tension resonance in second 15 years of 1971-2000 periods. Likewise, in part of Markov analysis, findings of this study revealed that under conditions of increasing temperature and decreasing rainfall, as the worst case, the effect of climate change on meteorological drought would appear as the class shift, and in most of the study stations under this scenario, increased duration of extremely drought (class 4) was forecasted, even 2 steps ahead, which is important in water resource management.
https://jesphys.ut.ac.ir/article_36704_93cf68bfe4d34a3463e382c5313aee51.pdf