Validation Technique For Simple Operational Use (Indramayu, Sumedang, And Majalengka Districts Case Of 2003) Teknik Validasi untuk Penggunaan Operasi Sederhana (Kasus Kabupaten Indramayu, Sumedang, dan Majalengka Tahun 2003)

The model output of rainfall total prediction has to be validated before being applied to the operational use. After understanding the accuracy of this rainfall total prediction output, one has to make decision whether applying it in the field or not. This depends upon the value of accuracy as well. Validation technique for simple operational use can be made by applying Pearson's correlation coefficient (r), Root Mean Square Error (RMSE), and spatial rainfall defferentiation values (ΔRR). The Wavelet Transformation Technique (WTT) for providing rainfall total prediction output of rainy and transition seasons 2003 had been applied to the domain of interest Indramayu, Sumedang, and Majalengka districts. Results show that during January 2003 (rainy season respectively) r = 0.65 with RMSE = 296 mm and 75.39% spatial disagreement area; but within April 2003 (transition season respectively) r = 0.49 with RMSE = 152 mm and 43.55% spatial disagreement area. Topography condition has play a role to the rainfall deferenciation values, especially for Lee-ward location. This is described by higher differentiation values of rainfall total prediction after reaching the top of elevation above mean sea level.


INTRODUCTION
Rainfall total prediction using miscellaneous models are commonly used in operational purposes. These models can be based on statistical approach, dynamical approach, or even both statistical and dynamical approaches at once (Swarinoto 2001). Each approach has its advantages and disadvantages of output quality. This is depending upon the purposes of prediction goals. Somehow ones are very difficult to compare the results taken from each model output. Factually, models are usually developed in order to reduce the complexity of systems. A system is an organized array of individual elements and parts forming as a unit. Consequently, models are pointing out of 'a small-scale or miniature representation of something' (Webster 1990). Using a model means to simplify the system. Factually, a system has complex variables and parameters. Here, models will use mathematical equations as necessary.
Commonly, models are run and based on the using of machines or computers to process the data. The use of machines to provide rainfall prediction is meant to help human to compute miscellaneous climate parameters in a fast manner. In this case, human has limitation to compute huge data in order to provide a longer time of rainfall prediction. According to the experiment, machines have a better capability to provide longer period of prediction comparing to the human capability ( Figure 1).
As usual, models are based on applying mathematical equations. In order to make easier computation procedure, there are often used some assumptions in these mathematical equations. This condition will imply to the results of prediction output. It is no doubt that some errors will occur in the models output. In simply way, ones will meet errors in every results of rainfall total prediction. The important thing is how to evaluate the models results by investigating the model output in a simple way. This is the goal from this presented paper. Figure 1 The capability of machines in numerical weather prediction comparing to human capability (Lee 2006).
For operational use, it is very important to utilize an easy way to evaluate the result of rainfall total prediction. Because operational purposes need a rapid way to know the result of evaluation.
This condition has relationship with the application of models in operational use. If the models have no accuracy enough, so consequently that these models will not be used in operational usage for providing rainfall total prediction. Because the prediction output will have a lot of risks to be applied in the field.
Step of Prediction  Figure 2 The flow of steps to provide prediction for operational use.
In order to offer the rainfall total prediction (models) outputs, one has to pay attention to the flow of steps to provide prediction especially for operational use chart ( Figure 3). The steps can be described as follow: (1) Starting from the data availability, the quality of data should be paid attention well. No missing data is preferred for data processing. If there is any missing data, one will fulfill firstly before entering into data processing step. Miscellaneous ways are able to be used to insert or replace missing data values. Having long enough data series such 20 years data, are also necessary in order to get realistic results closed to the field condition.
(2) Method of approaching is necessary to be designated in line with the goal of data processing.
Here, one is able to choose whether statistical, dynamical, or both statistical and dynamical approaches at once.
(3) Based on data processing, the rainfall total prediction outputs can be catered. One should pay attention to these outputs. Because of using the model output as simplified of the complex systems, these ouputs consider having errors or biases.
(4) Evaluation process of rainfall total model outputs have to be examined. This step is done in order to understand the level of accuracy from rainfall total model outputs. In this step, errors or biases from the rainfall total model outputs can be computed.
(5) Based on step (4), the level of errors or biases coming from the rainfall total prediction outputs can be computed. These information will be used to decide the possibility of rainfall total model outputs to be applied in the fields. If there is still weak in accuracy of the rainfall total prediction outputs, one will take the models to the step of (4).
(6) Rainfall total prediction or model outputs which have a appropriate level of errors or biases, can be applied for operational use. But the step of (4) should continuously be done in order to maintain the quality of prediction outputs.

Domain of Interest
The domain of interest in this study is used of 3 districts in West Java Province. These districts are namely Indramayu, Sumedang, and Majalengka. These districts located in Java Island. The condition of the domain of interest ( Figure 2). These districts are chosen in order to examine as well the role of topographic condition. Commonly cases, rainfall total pattern has interrelationship with topographic condition. Especially in Southern part of the domain of interest surrounding higher topography, annual rainfall total seems significantly higher than other locations. There is more than 3.000 mm per year. These occur in Sumedang and Majalengka districts respectively.  Seasonally rainfall total for rainy season within the domain of interest ( Figure 6). This map is based on 1981-2000 data series. During rainy season, seasonally rainfall total 500-1.000 mm per season occurs mostly in Indramayu district. Especially in both Sumedang and Majalengka districts, seasonnally rainfall total of 1.500-2.000 mm per season happens. Surrounding higher topography, seasonally rainfall total is recorded more than 2.000 mm per season in both later districts. Figure 6 Seasonally rainfall within Indramayu, Sumedang, and Majalengka districts during rainfall season (November-April).

Monthly Rainfall Total
Based on 1971-2000 monthly rainfall total data series of seasonal zones (ZOM), monthly rainfall total within the domain of interest . ZOM was designed by using clustered data processing (BMG 2003). Certain ZOM will have similar pattern in such as: (1) (BMG 2008).
Accordingly the threshold of monthly rainfall total of 150 mm per month (BMG 2008), the domain of interest experiences commonly dry season in May-October as mentioned also in

Prevailing Winds
The streamlines of prevailing wind for the level of 850 mb during Southern Hemisphere summer in Indonesia region is shown in Figure 8. The low level westerly has relationship with the rainy season in Southern Hemisphere of Indonesia region.

Cases Study
In order to know the accuracy of rainfall total prediction using WTT over the domain of interest, herewith January 2003 and April 2003 cases were chosen to be discussed. January has represented the maximum rainfall total occurrences (Figure 7). Meanwhile, April has depicted as transition season before entering dry season from rainy season over the domain of interest. In this month, factually, the rainfall still has significant probability to occur over here. This condition has been pictured using 1971-2000 data records as shown in (3).
Validation technique for simple operational use is going to be applied for both January 2003 and April 2003 cases. The rainfall total model output has been considered to be applied during both of rainy and transition season within the domain of interest.

DATA
Monthly rainfall total data taken from 3 districts namely Indramayu, Sumedang, and Majalengka are used in data processing. These data as mentioned above are taken from 36 rainfall stations from Indramayu district, 12 rainfall stations from Sumedang district, and 12 rainfall stations as well from Majalengka district. Firstly, the data records are in the format of daily rainfall total. This format should be arranged into monthly rainfall total format.
The series of data are used from 1981-2002 to be computed for training period purposes.
This means that the process of models calibration is being taken during this certain period. Then, the year of 2003 will be used for rainfall prediction case of Indramayu, Sumedang, and Majalengka districts. The process of validation will be taken place in this year. The cases of study will be January 2003 for representing rainfall season case and April 2003 for the first transition season case in the domain of interest.

Method of Validation
In order to understand the agreement phase between prediction output and observation data, Where: x = rainfall prediction output of WTT (mm), y = observation data taken from the rainfall station, and r = Pearson's correlation coefficient values. The agreement phase between rainfall prediction and rainfall observation will have two conditions such as: 1. The Same Phase, where rainfall prediction has the same direction with observation data. Figure 8. Pearson's correlation coefficient values should be positive. Figure 8 The same direction between rainfall prediction output (red) and observation data (blue).

Illustration of this condition is shown in
2. The Different Phase, where rainfall prediction has opposite direction with observation data. Illustration of this condition is shown in Figure 9. Pearson's correlation coefficient values should be negative. Figure 9 The opposite direction between rainfall prediction output (red) and observation data (blue).
In order to know the value of error between rainfall prediction output and observation data, ones is able to use the Root Mean Square Error (RMSE) value (Wilks 1995). Mathematical equation of RMSE is written as follow: Where: F = rainfall prediction output of WTT (mm), O = observation data (mm), n = total data used, and RMSE = root mean square error (mm).
If the rainfall prediction output has the same value with observation data so that the RMSE value will be zero (0). This is the best RSME value where imply the best performance of rainfall prediction output. In a simply peaking, bigger RMSE value means worse of rainfall prediction output performance. Spatial information of the rainfall prediction output performance can be ∆RR value should be computed for every station data. Then, ∆RR value has been plotted spatially using Geo-graphical Information System (GIS) software (Prahasta 2005a). Intersection Mode in overlay processing is used in this case for providing spatial information (Prahasta 2005b). Mathematical equation for this condition is written as follow:

January 2003 Case
Results of rainfall total prediction output using WTT    This condition is pictured especially in Majalengka and Sumedang districts. Lesser rainfall total during this month is observed in Indramayu district. This situation is pictured the agreement between rainfall total prediction and rainfall total observation data.
Rainfall total differentiation between rainfall prediction and rainfall observation data in January 2003 within the domain of interest using overlay mode ( Figure 12). Using rainfall total tolerance value of 100 mm per month during rainy season, 24.61% of total area of the domain of interest shows a well agreement between rainfall total prediction and rainfall total observation   Overlay results both rainfall total prediction and rainfall total observation data within the domain of interest for April 2003 ( Figure 15). Using rainfall total difference tolerance value of 100 mm per month for the first transition season, 56.45% of total area of the domain of interest shows a well agreement between rainfall total prediction and rainfall total observation data (green). The rest of area (43.55%) of the domain of interest seems disagreement (blue). During this transition season, Pearson's correlation coefficient between both of them is gotten r = 0.49 and RMSE = 152 mm per month.

The Role of Topography
Topography condition within the domain of interest ( Figure 16). The cross section for meridional (north-south) direction is chosen. There are 7 rainfall stations available ( Figure 17).
Topography condition seems to increase southward. After reaching around 500 meter above mean sea level, topography condition decreases southward. This location behind the hill is called Lee-ward site.   Figure 18 The value of ∆RR within domain of interest in relation to the meridional topography cross section.
The value of ∆RR seems to increase especially during the peak of rainy season ( Figure 18).
Other months are also showing the same pattern. This condition has relationship with decreasing of topography for location behind the hill. In this case, WTT output could not meet the purposes of prediction due to the topographic condition.

CONCLUSSION
The WTT had been applied to proceed rainfall total of both prediction and observation data. Results show that during rainy season of 2003 Pearson's correlation coefficient is noted r = 0.65, RMSE = 296 mm per month, and area of agreement both rainfall total prediction and rainfall total observation data 24.61% of total the domain of interest. In the mean time, during the first transition season of 2003 Pearson's correlation coefficient is gotten r = 0.49, RMSE = 152 mm per month, and area of agreement between rainfall total prediction and rainfall total observation data 56.45% of total the domain of interest.
The impact of topography condition is implied to the rainfall total differentiation (∆RR) between rainfall total prediction and rainfall total observation data. The ∆RR increased after reaching the Lee-ward site location.