ارزيابي اثر نمايه چند متغيره انسو بر بارش زمستانه خراسان شمالي

Evaluate the MEI Effect on Winter Precipitation In Northern Khorasan

از آنجا كه كشاورزي در خراسان شمالي از اهميت و جايگاه ويژه اي برخوردار است. پيش بيني فصلي بارش مي تواند تاثير بسيار مهمي در تابستان و پاييز در پيش بيني بارش زمستانه استان خراسان شمالي با به كار بردن نرم افزارCPT(Climate predictability tool) مي‌باشد. بدين منظور از مدل تحليل هم بستگي متعارف CCA (Canonical correlation analysis) و رگرسيون خطي چند گانه در نرم افزار CPT استفاده شده است. سري هاي زماني فصلي نمايه MEI به عنوان پيشگوكننده و بارش زمستانه دوره 19862008 هفده ايستگاه خراسان شمالي به عنوان پيشگو شونده در نظر گرفته شده است. در روش تحليل هم بستگي متعارف به منظور كاهش تعداد متغيرهاي پيشگوكننده از روش متعامد تجربي EOF (Empirical orthogonal function) استفاده شد و 5 مؤلفه اصلي كه 89% از كل واريانس مجموعه داده ها را شرح مي دهند، انتخاب گرديد. نتايج به دست آمده از دو مدل مذكور نشان مي دهند كه بين بارش زمستان و نمايه MEI در فصل بهار همبستگي ضعيفي وجود دارد و نمايه پاييز MEI همبستگي قوي تري با بارش زمستان خراسان شمالي دارد و بيشترين همبستگي از ايستگاه تازه قلعه و كمترين همبستگي از ايستگاه منگلي به دست آمد. منفي بودن همبستگي نشان دهنده اين است كه با افزايش نمايه MEI بارش زمستان كاهش مي يابد و برعكس. بارش ها در تمامي ايستگاه ها نسبت به سال 2008 كه خشكسالي به وقوع پيوسته بود، افزايش داشتند كه مشاهدات نيز اين افزايش بارش را تاييد مي كنند. اختلاف بين داده هاي خروجي مدل ها و بارش مشاهده شده نشان دهنده اين است كه فقط با تعيين فاز MEI نمي توان بي هنجاري بارش زمستان را از نظر علامت و  شدت پيش بيني نمود.

Introduction North Khorasan province is in north east of IRAN and has different climates. West side has cold semiarid, east side has cold arid and north part has cold semiwet. Agriculture has special situation so seasonal prediction is very important. If precipitation predictions show above normal or below normal, agriculture should take a right decision for type of farming, dry farming or water farming. They are many studies about climate prediction in IRAN some of them are about effects of sea surface temperature of Atlantic, Pacific and Indian Ocean on IRAN precipitation and the others are about the effects of teleconnections on climate prediction. These studies concluded that there is a relationship between changes of SST and precipitation fluctuations. Present study use CCA and MLR model of CPT to predict winter precipitation and evaluate the effect of seasonal MEI on North Khorasan winter precipitation. Materials and Methods Precipitation Data At this study the monthly precipitation data (Jan, Feb, and March) of 17 synoptic and rain gauge stations of North Khorasan during 19862008 is used and then the mean of winter precipitation is calculated. MEI data The MEI data (time series 12 month) is used of NOAA data bank from 1986 to 2008 and then the seasonal MEI data is calculated in excel and SPSS. MEI is computed on the six main observed variables over the tropical Pacific. These six variables are: sealevel pressure (P), zonal (U) and meridional (V) components of the surface wind, sea surface temperature (S), surface air temperature (A), and total cloudiness fraction of the sky (C). Reconstruction and Normal test data The Ratio Method is used to construct and complete the monthly precipitation data and then normal test data by using JMP4 software is done. The results show that yearly, winter and autumn data are normal. CCA Model Various dynamical and statistical models are used to predict seasonal and climate prediction. The most popular method and model that is used for seasonal prediction is CCA. IRI[1] released the Climate Predictability Tool (CPT) that provides a Windows package for constructing a seasonal climate forecast model, performing model validation, and producing forecasts given updated data at 2002. Although the software is specifically tailored for these applications, it can be used in more general settings to perform canonical correlation analysis (CCA), principal components regression (PCR), or multiple linear regressions (MLR) on any data, and for any application. For this study we use the last version of CPT (11.10) CCA and MLR model. Two data sets are required by CPT. the first data set contains the "X variables" here are seasonal MEI data from 1986 to 2008 and these variables are sometimes called "predictors", "independent variables". The Y variables are sometimes called "predictands", "dependent variables" here are winter precipitation (19862008). At first we choose spring MEI data as predictor and winter precipitation as predictant and by using CCA model we study the effects of spring MEI on winter precipitation and then choose summer and autumn MEI data and again implement the model. Model assumes a linear relationship between the predictor, x, and the predictand, y:                                      y=β0+β1x                                                 (1) Where β0 and β1 are regression constant and regression coefficient or the “slope”. Correlation coefficient is a widely used measure of the strength of linear association between the predictor and the predictand.   Where sx and sy are the standard deviations of x and y, respectively. The numerator in Eq. (2) is related to the covariance by a factor of n, and will be positive if positive anomalies in both the predictor and the predictand tend to occur in corresponding cases, and will be negative if opposite anomalies tend to occur. At this study negative correlations refer to winter precipitation will decrease by increasing MEI. The model expresses The value of a predictand variable as a linear function of one or more predictor variables and an error term.            (3) Here  is regression constant,  is coefficient on the kth predictor, k is total number of predictors,  is predictand in year i and  is error term. Model Validation After implementing the model by choosing cross validation method, Forecast performance scores and graphics can be obtained for the crossvalidated forecasts. The performance window for an individual series provides a variety of forecast performance scores divided into those based on continuous measures, and those based on measures in which the observations, and in some cases the forecasts as well, are divided into three categories. The continuous forecast measures calculated are: Pearson correlation, spearman correlation, Mean squared error, Root mean squared error (RMSE) … and the categorical forecast measures are: Hit score, ROC area. Jajarm and Mokhaberat stations have the minimum RMSE that means their forecasts accuracy is high. Hit rate versus false alarm rate plots are also provided (ROC curve), which indicate how well the models forecast winter precipitation. The perfect prediction system would have a hit rate of 1.0 and a false alarm rate of 0.0. Winter precipitation forecasts with summer and autumn MEI have better results. Results The first five EOF modes can explain 89% of total precipitation variance. The model forecast winter precipitation for every station from beginning of duration (1986) so every station has time series of observations and hindcast. Most stations have positive correlation between winter precipitation and spring MEI and negative correlation between winter precipitation and summer and autumn MEI. The maximum correlation belongs to Langar winter precipitation and autumn MEI in CCA Model and Tazehgale in MLR Model and the minimum correlation is for Bolgan. Conclusion With regard to effect of teleconnection on precipitation, the relationship between MEI and precipitation is evaluated. The MEI time series are predictors and winter precipitation are predictants at CCA and MLR model. The results of this study are very important for North Khorasan agriculture. Comparison of model forecasts and precipitation observations of 2009 show us that we haven’t skillful forecast of seasonal precipitation by only use MEI and other teleconnections effect on seasonal precipitation, too.   1. IRI, INTERNATIONAL RESEARCH INSTITUE FOR CLIMATE PREDICTION