^{1}

^{2}

^{3}

^{2}

^{*}

The predictability of pan evaporation and air temperature in the southern part of the Dead-Sea region (Sdom) was investigated according to two approaches, prediction by mesoscale models and with the aid of synoptic classification. First, the predicted temperature, wind speed and relative humidity that directly affect the evaporation are obtained from the WRF mesoscale model predictions. Predictions according to multilinear regression equations and a Penman-Monteith approach were also validated against observations in Sdom. The WRF model predicts the temperature reasonably well. However, the wind speed and relative humidity predictions were found to be very poor. The unique approach in this paper is employing a semi-objective synoptic systems classification according to the global GFS model. Relationships were defined between the 19 Eastern Mediterranean’s (EM) synoptic systems and the Sdom evaporation, temperature, wind speed and relative humidity. A monthly evaluation was performed for each of the systems and the semi-objective prediction was verified by the semi-objective classification. Since some synoptic systems affect the evaporation and temperature similarly, the 19 synoptic systems were grouped into seven clusters, each containing systems with similar evaporation and temperature records. This method has yielded a significant improvement in the daily prediction of evaporation and temperature. Semi-objective definitions for the synoptic systems were performed for the ranges of 12 - 132 hours. The synoptic system approach succeeded in the prediction of the evaporation and temperature changes in Sdom for a few days in advance. The predictability skill for the 12 hour forecast achieved about 80% of success, dropping to 70% at 36 hours. For 60 to 132 hours the prediction stabilized at a skill of 60%.The method presented here is a new attempt to predict meteorological parameters by using a synoptic classification approach in the Dead-Sea area where even high-resolution mesoscale modeling forecasts are not very successful.

The Dead-Sea (DS) is a unique place on Earth for several reasons. It is located at the lowest spot of the earth’s surface (−429 m), its water is the densest and saltiest of any natural water body, and staying there can help cure some illnesses [

The evaporation in the southern Dead-Sea (Sdom station) is an important factor in the local climate and for the local potash industry. The evaporation rate in this area is very high in respect to other places on Earth. The concentrated salts of the lake and the arid climate help the DS factories to manufacture potash, bromine, magnesium and other minerals through the natural high evaporation which makes the production process efficient. Hence, the short-range (1 - 4 d) prediction of the evaporation rate and the temperature are important for the planning of the production processes. Knowledge of the daily change of evaporation rate several days ahead can therefore improve the production process by a more efficient management of the DS water flow from the north to the evaporation ponds.

Several earlier works deal with the evaporation in Sdom. [

Several researches deal with the synoptic system change in the Eastern Mediterranean (EM). [

Several attempts were made to predict evaporation with different methods [

Several earlier studies used an approach similar to ours by using similar methods based on the [

The uniqueness of this study is the employment of a semi-operational classification of the synoptic systems for several days of weather prediction over the very complex DS topography. This approach was adopted following the examination of alternative real-time prediction methods to be described next.

Multi linear regression equations were tested for Sdom during 1964-2009 (10,679 days). Many options were checked including combinations of hourly values of wind, temperature and humidity. The highest correlation was reached with daily averages of the temperature, wind speed and relative humidity as the independent variables at three hours (6:00, 12:00, and 18:00 UTC). The pan evaporation is the dependent variable in the regression equation, and the correlation achieved reached R = 0.89. A similar correlation value (R = 0.88) was found by the Penman-Monteith’s approach tested for the 1964-2005 period. For the Penman-Monteith’s calculations algorithms developed by [

The same was done for the sensible heat flux which was calculated as a residual of observed energy balance terms and latent heat flux estimate from the Penman-Monteith approach. The study shows that evaporation calculated through the Penman-Monteith approach in arid regions can in general be reproduced through simulations.

The multi linear regression and the Penman-Monteith methods were tested for each of the seasons, and the correlations were similar. The simple seasons’ definition was adopted, i.e. December-February, March-May, June-August and September-November for winter, spring, summer and autumn, respectively. The new synoptic definition as suggested by [

These two methods, i.e. multi linear regression and the PM, were tested in real-time by employing the WRF model predictions (horizontal interval of 1.3 km and vertical resolution of 31 sigma layers) for the temperature, wind speed and relative humidity in the nearest grid-point to Sdom. The model was validated against Sdom observations for a period of 29 days from Jan-Mar 2010, and the range of the forecast was 24 - 72 hours. Correlation results (

[

Forecast (hours) | T (˚C) R^{2 } | T (˚C) R^{ } | T Sig at | RH (%) R^{2 } | RH (%) R | RH Sig at | WS (m/s) R^{2 } | WS (m/s) R | WS Sig at |
---|---|---|---|---|---|---|---|---|---|

24 | 0.74 | 0.86 | 0.005 | 0.01 | 0.1 | No sig | 0.39 | 0.62 | 0.005 |

48 | 0.7 | 0.83 | 0.005 | 0.16 | 0.4 | 0.025 | 0.12 | 0.34 | 0.05 |

72 | 0.69 | 0.83 | 0.005 | 0.02 | 0.15 | No sig | 0.04 | 0.21 | No sig |

was defined as semi-objective because the 426 daily synoptic systems where defined by a group of meteorological experts, and served as the “training data-base”. The minimal Euclidean distance found between any other day and each of these manually classified days has determined the synoptic system of that specific day.

Sdom station is located in the southern part of the DS (31˚1'N, 35˚23'E) at an altitude of 390 m below the sea level. This station is operating since 1951, except for the period 1978-1981.The observations of pan evaporation (mm/day), temperature (˚C), wind speed (m/s) and relative humidity (%) were measured for the period 1964-2006. The pan evaporation is measured once a day, while the temperature, wind speed and relative humidity are measured three times daily (6:00, 12:00, and 18:00 UTC). Evaporation is measured with a standard US class A pan with 121 cm diameter, 25.5 cm deep, resting on an open wooden platform set on the ground and protected by the standard wire net screen of the Israel Meteorological Service. Each variable includes averages of data from the three aforementioned times. The data series of evaporation is the longest when compared to the other variables. The reason for the different sizes of data vectors is the fact that some of the days are missing one or more observation times (of the three daily) or because one of the variables is missing.

The synoptic systems are influencing the EM climate in general, including the DS region. According to the daily synoptic systems classification by [

systems were tested as well (Appendixes 2-7). We choose to show the RST example because of its complexity. The same statistics are shown in

The predictability of the evaporation with the aid of the synoptic systems was tested on the basis of the past, employing the NCEP/NCAR reanalysis. The data was downloaded from the reanalysis site (http://www.cdc.noaa.gov/). The reforecast dataset was based on the 1998 version of the NCEP Medium Range Forecast (MRF) model run with a ~200 km horizontal interval with 28 vertical sigma levels [

This case is defined when the predicted synoptic system exactly fits the analyzed one.

This case is defined when the predicted system fits the cluster of synoptic system it belongs to. For instance, the major cluster belongs to the summer. This is because of the similar patterns of the PTW, PTM, PTD and HW synoptic systems. In addition, the cluster must have a similar influence on the average daily pan evaporation and temperature. The PT summer systems constitute a very large cluster along with HW. On the other hand, RST systems do not carry similar average evaporation and temperature characteristics. Therefore, they are not considered as a cluster, even if there carry similar synoptic pattern. The improvement of the ability to predict evaporation in Sdom according to ‘cluster accurate forecast’ is shown in

In this third case the forecast of the predicted synoptic system did fit to neither the analyzed system nor to that of the cluster which it belongs to.

Cluster | Synoptic systems | Min evap (mm/d) | Max evap (mm/d) | Min temp (˚C) | Max temp (˚C) | 12 h (%) | 36 h (%) | 60 h (%) | 84 h (%) | 108 h (%) | 132 h (%) | Dominant season |
---|---|---|---|---|---|---|---|---|---|---|---|---|

1 | PT-W, PT-M, PT-D, Hw, CLn-S | 14.03 | 14.91 | 32.92 | 34.53 | 19 | 23 | 25 | 25 | 25 | 27 | Sum/Spr |

2 | RSTe, Hw, Hc | 7.7 | 8.06 | 22.54 | 23.2 | 5 | 7 | 7 | 6 | 8 | 7 | Aut/Wint/Spr |

3 | RSTc, Hn | 7.66 | 8.16 | 21.26 | 22.52 | 1 | 2 | 2 | 2 | 1 | 1 | Aut/Wint/Spr |

The skill was calculated as follows: The skill for the first order accurate forecast was defined as the number of days that were classified as first order accurate forecast divided by the total number (1435 days). The skill for the cluster accurate forecast was defined as number of days classified as cluster accurate forecast divided by the total number (1435 days). The skill for inaccurate forecast was defined similarly.

Exploring the prediction potential of the average daily pan evaporation in the Dead-Sea (Sdom) according to the daily synoptic systems is the primary goal here. Our aim is to investigate the predictability of the daily evaporation at Sdom through large scale predictions of the semi-objective synoptic class employing the previously calculated relationships of these classes with the daily evaporation. The connection between the synoptic systems and the local meteorological variables in Sdom is very important for this kind of prediction. The large climatological dataset shows that we need to analyze each of the 19 EM synoptic systems separately.

In order to predict the pan evaporation changes on a daily scale in Sdom, we examined the daily averages of the evaporation and explained the results with other variables (daily average of temperature, wind speed and relative humidity). The calculations were conducted for each synoptic system for each month for a 42-year period (1964-2006). The standard deviation is also included in the climatological data. The monthly scale reveals some interesting connections between the synoptic systems and the other variables at Sdom.

The semi-objective prediction of the synoptic systems (synoptic class) was verified by a semi-objective classification for the daily synoptic systems in the period of 2006-2009. The prediction skill of the synoptic systems is shown in

The first order forecast skill decreased dramatically with time limited by the capabilities of the global model. For 12 hours the skill value was ~56% and from 60 hours up to 132 hours it was stabilized at about 20% of skill value. Because of this poor success we used a different approach by dividing the synoptic systems to clusters. The cluster forecast skill increased up to 36 hours because of the similar conditions of the synoptic systems which achieve a better prediction than the first order accurate forecast. From 36 hours onward the skill of the cluster was quite stable. In this approach the total accurate forecast. We predict synoptic system and according to the synoptic system, we know the daily pan evaporation from our large statistics (1964-2006)) is the total sum of first order accurate forecast and that of the cluster synoptic systems. The 12-h total skill was above 80%. After 36 hours the skill dropped to ~70%, while for 60 - 132-h it was about ~60%. According to the results, the cluster approach significantly improved the skill. There are seven groups of clusters (three of them are represented in

In order to understand these low values of skills we decided to verify four variables (H, T, U, and V) at 1000 hPa in three spots over one year (2007). The verification was conducted according to the reanalysis. The total number of the investigated days was 360. The first spot is located on the Turkey Mountains (37.5˚N, 30˚E). The topographic area makes it more difficult for the global models to predict the variables accurately. The second spot is located south to the Cyprus Island (32.5˚N, 32.5˚E). This spot represents the Sea and there are less mesoscale influences in this area. The third spot is located near the coastline of Israel (32.5˚N, 35˚E). The Sea/Land breeze is a significant mesoscale meteorological phenomenon which global models have difficulties to predict. Tables 3-5 (or Figures 7-9) show the verification results for the Turkey, Mediterranean Sea and the Coastline Spot respectively. According to the verification in all of the three spots (see Appendix 8), the T and H variables have a very high skill. The model has difficulties in predicting the U and V wind variables in all of the three spots. From the verification we can deduce that a model with very coarse resolution (~200 km) has a very low prediction skill of the winds variables for 36 hours (Tables 3-5 or Figures 7-9). The verification explains the low values of skill for the first order accurate forecast even after 12 hours of prediction (

37.5˚N, 30˚E | 12 h | 36 h | 60 h | 84 h | 108 | 132 |
---|---|---|---|---|---|---|

H | 0.9768 | 0.8264 | 0.6537 | 0.5686 | 0.5382 | 0.5176 |

T | 0.9957 | 0.9707 | 0.9373 | 0.9146 | 0.8803 | 0.8677 |

U | 0.8107 | 0.4028 | 0.0989 | 0.0133 | 0.0203 | −0.0166 |

V | 0.9578 | 0.5268 | 0.2202 | 0.1540 | 0.0932 | 0.0172 |

32.5˚N, 32.5˚E | 12 h | 36 h | 60 h | 84 h | 108 h | 132 h |
---|---|---|---|---|---|---|

H | 0.9812 | 0.8031 | 0.6445 | 0.5682 | 0.5534 | 0.5669 |

T | 0.9913 | 0.9235 | 0.8539 | 0.8084 | 0.7933 | 0.7924 |

U | 0.9287 | 0.4771 | 0.1897 | 0.0461 | 0.0229 | 0.0502 |

V | 0.9078 | 0.4078 | 0.1446 | 0.1274 | 0.1123 | 0.0226 |

32.5˚N, 35˚E | 12 h | 36 h | 60 h | 84 h | 108 h | 132 h |
---|---|---|---|---|---|---|

H | 0.9770 | 0.8218 | 0.6877 | 0.6310 | 0.6203 | 0.6370 |

T | 0.9935 | 0.9390 | 0.8781 | 0.8419 | 0.8141 | 0.8073 |

U | 0.9217 | 0.4212 | 0.1370 | 0.0388 | 0.0217 | 0.0917 |

V | 0.8734 | 0.3718 | 0.1185 | 0.0396 | 0.0041 | -0.0097 |

Using the WRF model with a very high resolution to predict meteorological variables of Sdom in the southern DS did not yield good enough forecasts for the surface wind and humidity which were crucial for the evaporation prediction. The complex topography in the DS causes difficulties even to a high-resolution mesoscale model with ~1 km grid interval. Therefore, the synoptic systems approach was employed here. First, we have found the influence of the synoptic systems on the evaporation, temperature, wind speed and relative humidity in Sdom based on a monthly scale. The semi-objective prediction (reforecast) of synoptic systems was verified by the reanalysis during 2006-2009. From previous study we know that some of the synoptic systems are very difficult to predict. For example, it is difficult to accurately predict the evaporation when the axis of the Red Sea- trough is located over the EM. Another example is the depth of a Persian trough (weak, medium or high). Therefore, the method of clustering the synoptic systems into several groups was adopted. This new method works through assuming that the influence of the different synoptic systems in each class on the average daily evaporation and temperature is similar. This was entitled as the “Cluster accurate forecast” and it has improved the skill of the prediction. The predictions were divided into three groups: the first order accurate forecast, the cluster accurate forecast and the inaccurate forecast. The sum of the “first order accurate forecasts” and the “cluster forecast” provides the total accuracy. The success after 12 hours was about 80%. After 36 hours the success dropped to ~70%. And, from 60 - 132 h the total accurate forecast was stabilized at ~60%.

In order to explain the low values of skill in the “first order accurate forecast” (after 12 hours the skill was approximately equal to ~56% and approximately equal to ~20% after 132 hours) three points in the region based on the GFS grids forecasts were evaluated. In these points the verification was tested for 2007. The model can predict the temperature and the geopotential height variables for these three points very well. However, the model cannot predict the wind variables (U and V) sufficiently. Therefore, it can be concluded that the coarse resolution is the reason for the low values of skills in the “first order accurate forecast”.

Employing the new version of the GFS model with a better resolution may improve the results. However, in this case, we will need to classify the synoptic systems again according to a new version with new grids points. In addition, collection of climatology data from Sdom from 2007 until present will improve the connection between the synoptic systems and the average meteorological variables (evaporation, temperature, wind speed and relative humidity) of Sdom. Moreover, achievement of better results in mesoscale models outputs, similar to the one we used, can lead to more successful ways to predict (of predicting) the evaporation. The method presented here is a first attempt to predict meteorological parameters by using a synoptic classification approach in the DS area where even high-resolution mesoscale modeling forecasts are not very successful.

We wish to thank the Dead-Sea Works that supported this research. The German Helmholtz Association is gratefully acknowledged for partly funding the finalization of this project within the Virtual Institute DESERVE (Dead Sea Research Venue) under contract number VH-VI-527. Thanks also to the Israel Meteorological Service (IMS) for supplying the observations of Sdom station. We thank Barry Lynn for supplying us with results from the WRF model.

EyalIlotoviz,HaimShafir,PhilippGasch,PinhasAlpert, (2015) Semi Operational Prediction of the Dead Sea Evaporation—A Synoptic Systems Approach. Journal of Water Resource and Protection,07,1058-1074. doi: 10.4236/jwarp.2015.713087