بررسي قدرت تفكيك مكاني مناسب داده‌هاي بارش روزانه پايگاه ERA-Interim در شمال شرق ايران

ERA-Interim

در اين پژوهش دقت 11 سري از داده‌هاي بازكاوي روزانه ERA-Interim مربوط به مركز اروپايي پيش‌بيني‌هاي ميان‌مدت ECMWF با تفكيك‌هاي مكاني 125/0 ، 25/0، 4/0، 5/0، 75/0، 1، 125/1، 5/1، 2، 5/2 و 3 درجه با داده‌هاي مشاهداتي 17 ايستگاه هواشناسي شمال شرق (خراسان شمالي، خراسان رضوي، خراسان جنوبي)، در يك دوره زماني 26 ساله (2015-1990) موردبررسي قرار گرفت. براي اين منظور از آماره‌هاي ضريب تبيين، شاخص توافق، اُريبي نسبي و NRMSE استفاده شد. نتايج نشان دادند كه بر اساس آماره NRMSE دقت تمامي تفكيك‌هاي مكاني موردمطالعه زير 10 قرار دارد كه در محدوده عالي مي باشد، اما سري داده‌هاي با تفكيك مكاني 4/0 درجه در سه نمايه آماري ديگر( ضريب تعيين،اريبي نسبي و ضريب توافق) بهترين عملكرد را دارا است و هماهنگي بهتري در مقايسه با ساير داده‌هاي ERA-Interim دارد و مي‌تواند براي تكميل خلاءهاي آماري و مناطق فاقد آمار مورداستفاده قرار گيرد. در سري داده‌هاي با تفكيك مكاني 4/0 درجه، بيشترين ضريب تبيين در بين ايستگاه‌هاي موردمطالعه در ايستگاه مشهد مشاهده شد كه ممكن است ناشي از بالا بودن بارش آن، پايين بودن تغييرپذيري بارش آن در مقايسه با ساير ايستگاه‌ها و نيز سابقه آماري بالاي اين ايستگاه و لحاظ داده‌هاي آن در سامانه داده‌گواري جهاني داده‌ها است كه به‌عنوان داده‌هاي پايه در آغازگري مدل پيش‌بيني عددي ECMWFاستفاده مي‌شود. كمترين مقدار نرمال شده جذر ميانگين مربع خطاها در ايستگاه‌هاي تقريباً پربارش شمالي منطقه شامل مشهد، بجنورد و قوچان و بيشترين خطاي مقدار نرمال شده جذر ميانگين مربع خطاها در ايستگاه‌هاي تقريباً خشك و كم بارش مشاهده شد. رفتار شاخص توافق كم و بيشتر شبيه NRMSE است. قدر مطلق باياس نسبي در تمامي ايستگاه‌هاي كمتر از 01/0 است كه بيشترين خطاي مربوطه در ايستگاه‌هايتقريباً كم بارش مركز و جنوب منطقه مشاهده شد. با توجه به يافته‌هاي اين پژوهش، داده‌هاي بازكاوي روزانه ERA-Interim با تفكيك مكاني 0.4 درجه مي‌توانند براي تكميل خلاءهاي آماري و مناطق فاقد آمار بارش در خراسان مورداستفاده قرار گيرد.

Introduction Climate model information usually is in large scales (> 100 km). It is often necessary to use downscaling methods to use this information. Downscaling can be defined as methods which interpret climate information in regional or local scale (10-100 km) from the large grid (>100km) GCMs (Fung et al. 2011). Two main methods of downscaling are dynamical and statistical downscaling. Statistical downscaling methods are less compute-intensive tasks which involve implementing local scale variables as a function of global climate model outputs (Chen et al. 2013). This paper predicts inflow into the Karaj dam reservoir using results of a Global climate model in different climate scenarios and downscaling methods (a total of 32 different runs). Material and methods Climate model and scenarios Intergovernmental Panel on Climate Change (IPCC) introduced four different greenhouse gas concentration (not emissions) trajectories (representative concentration pathways RCP) based on the amount of radiative forcing values at the end of the year 2100 namely RCP 2.6, RCP 4.5, RCP 6.0 and RCP 8.5 which all have been used in this study alongside CSIRO MK 3.6.0 Global Climate model (GCM). Statistical downscaling methods Based on choosing the predictors, observed data or historical GCMs simulation, downscaling methods (DSM) can be divided into two major groups namely Bias Correction (BC) and Change Factor (CF) (Ho, Stephenson et al. 2012, Wang, Ranasinghe et al. 2016). In BC based methods, it is assumed that the change between GCM data and observed data remain constant in time and the CF-based method, it is presumed that change in observed climate variable is same as changes in climate model data, and precipitation occurrence probability will remain constant. For a better estimation of future climate conditions and understand the effect of selecting different downscaling methods two different BC methods and two different CF methods have been explained and used in the study. Karaj dam watershed is located in the central part of Northern Iran in Tehran province between 51.05 and 51.60 degrees North and 35.88 and 36.18 East. It is one of the main supplies of urban and agricultural water demand of Tehran. The area of the watershed is about 846.5 km2 and its average height is about 2826 MASL. Twelve (12) weather stations which had data with proper length were selected over the region. Spatial downscaling was used by averaging 4 grid points near the station. The inflow was predicted for a future period (2013-2045) and compared with both observed and modeled data in the base period (1985-2012). Future temperature and precipitation in different DSM-RCPs have been plotted. Average monthly hydrograph and probability density functions of annual streamflow were compared. Runoff-precipitation simulation has been conducted using IHACRES software. Results The overall trend of temperature in different downscaling methods is rising and the uncertainty related to choosing DSM is more than choosing climate scenario. The range of changing temperature is wider in the CF method and choosing the overall DSM method (BC or CF) is more important than choosing sub-method (MB or VB). Unlike temperature precipitation changes in not similar in different scenarios and DSMs. Although in some scenarios precipitation increases, in others decreases. Despite small differences, it seems that the overall trend of streamflow in the CSIRO model is decreasing. Streamflow in April decreased significantly. The range of streamflow in the future is wider than historical observation and uncertainty especially in extreme events is higher. RCP 8.5 has the greatest streamflow range in the future which shows less reliability in predictions. Discussions For better performance of infrastructures in the future, climate changes and their effect on streamflow should be predicted. This paper investigated the effect of choosing different statistical downscaling methods and climate scenarios (RCP) on the future streamflow in the Karaj dam basin, Tehran, Iran. There are changes in flow pattern and in most scenarios, two peaks in April and May are recorded. Least annual average flow is for RCP 6.0 and the greatest annual average flow is for RCP 2.8. The study showed that for better understanding and prediction of future condition, different downscaling methods should be considered as well as different climate scenarios. Choosing whether the BC or CF method has more effect than MB or VB selection. The paper used different scenarios and methods to predict future streamflow. Probabilistic approach showed the importance of considering uncertainty in streamflow prediction and the possible range of future changes which may be used in defining the reservoir operating rule.