Climate Projection Over Indonesia Based on the Total Fossil Fuel Co2 Emission Prediction Using the Box-jenkins Arima Model (Proyeksi Iklim Wilayah Indonesia Berdasarkan Prakiraan Emisi Co2 Dari Penggunaan Bahan Bakar Fosil Menggunakan Model ...

This paper mainly discusses about the development of estimation models raising the rate of gas emissions of carbon dioxide (CO2) as the main parameters of global warming in Indonesia. This is important to remember not many comprehensive scientific study which shows that the impact of global warming has actually experienced by Indonesia. Using Box-Jenkins method and the stage of identification, assessment, and testing, then the best prediction model obtained for the above data, the model of ARIMA (8,1,3). This means that the predicted value for the next year depending on the data before and 8 years 3 years earlier error. In the validation data with predicted results, the MAD (Mean Absolute Deviation) is relatively high. However, the pattern of results followed the pattern predicted almost the original data with a correlation value of 99%. Based on this result, we can estimate the climate projection over Indonesia, especially during 2012-2014.


INTRODUCTION
During the pre-industrial era the atmospheric carbon dioxide concentration has been stable (IPCC 1996). This stability is due to the equilibrium situation when the global carbon dioxide absorption rate of about 220 GtC/a carbon to cold ocean water and growing biomass is balanced by an emission of 220 GtC/a from warm ocean water and decomposing biomass. When the global mean temperature has been high, the equilibrium has changed towards a slightly higher atmospheric carbon dioxide concentration, probably because of decreased solubility of carbon dioxide in the warmer ocean water (Ahlbeck 2000).
When carbon dioxide is emitted from fossil fuels, cement production, or deforestation, the increased partial pressure of carbon dioxide in the atmosphere will force an increase of the absorption rate and thus a net sink flow of carbon to the backmixed surface layer of the oceans and to the biosphere.As we know in 1992, the Intergovernmental Panel on Climate Change (IPCC), presented a group of emission scenarios for different greenhouse gases. A "mid-range" emission scenario was called IS92a. In reality, we can see that the increase rate of atmospheric carbon dioxide has, despite the substantial increase of carbon dioxide emissions, remained on a very stable level during the recent 30 years. In fact, the airborne fraction, or the portion of the yearly emissions that stays in the atmosphere, has decreased from 52% in the year 1970 to 39% today. The IPCC model using IS92a implies however a nearly constant future airborne fraction.
Although, is not included in the list of countries as the largest contributor to global warming, but with the forest fires which occurred almost throughout the year, especially in the dry season length (as in 1982 and 1997), estimated there were about 2.5 billion tons of CO 2 that we contribute to global warming.In this paper, we mainly concern on the projection of the total fossil fuel of carbon dioxide (CO 2 ) emission over Indonesia based on the Box Jenkins ARIMA model analysis. The steps analysis to get that the best model prediction of that data will be discussed in this paper.

MATERIALS AND METHODS
The main data used in this study is the CO 2 emission taken from Indonesian territory that are downloaded from the web-side http://cdiac.ornl.gov/ftp/trends/emissions/ido.dat. From this web-site address, then the set of numbers obtained as follows (Table 1). The data is then in-plot in the form of time-series to be investigated the variations with time. The Complete data were calculated from 1889 to 2004 (about 115 years observation). Since that data is relatively long to be shown (Table 1).  2000  99728  22237  58348  2001  98331  15821  57194  2002  113285  21410  60159  2003  111345  22216  63969  2004  103170  17363  68378 Source: http://cdiac.ornl.gov/ftp/trends/emissions/ido.dat Please note here, we applied the Box-Jenkins method with the following steps, namely: identification, assessment and testing before the application of the model itself.

Identification of Model
The first step that we need to do is we need to check if the data is stationery or no. If the data

Eksponential decrease
Cutoff at lag to-q Eksponential decrease with start lag to -p

PACF
Cutof f pada lag kep Eksponential decrease Eksponensial decrease with start lag to -q

Suspect of Model Parameters
To help choose the type of tentative (temporary), using the results of the analysis and partial autocorrelation with a certain lag length. After the model the time series had been identified, the next step is to suspect the model parameters are based on least square criteria. There are two basic ways to obtain these parameters: a. By way of experimentation (trial and error) that is testing several different values and selecting a value (or set of values, if there are more than one parameter to be estimated) that minimizes the sum of squares residual value / value of the error (sum of squared residuals ).
b. Iterative improvement of selecting initial estimates and then let the computer programs are watched by iterative forecasting (Makridakis, 1999).

Validation Model
After the ARIMA model is determined, the next step is to conduct diagnostic tests to test the feasibility of the model and suggest improvements if necessary. One way that can be done is by analyzing the error (residual). In other words, examining the difference (difference) between observation data and model output. Error value (error) that remains after matching is ARIMA model, expected only a random disturbance. Therefore, if the plot function and autocorrelation partial of error values have been obtained, is expected to: a. There was no significant autocorrelation.
b. There was no significant partial autocorrelation.
The second is to study the statistical sampling of the optimum solution to see whether the model can still be simplified. Statistical assumptions underlying the general model of ARIMA that gave some statistics that should be calculated after the values measured optimum coefficients. For example, for each coefficient / parameter values that are obtained will be calculated so that the error sum of squares error value. Coefficient value is selected that has the smallest squared error values. Error values can be obtained from (Makridakis, 1999).

Forecasting Model
The next step is to forecast (forecasting) if the model is suitable. The next step is to forecast (forecasting) if the model is suitable.

Identification of Data
The data used to make this prediction model is data on the CO2 emissions of Indonesia since 1889 to 2004. In this study analysis, we applied the ARIMA (Autoregressive Integrated Moving Average), because it involves time series data, thus obtained a model that describes the time series data.
Stationery test needs to be done before the creation of models for forecasting in time series data requires that data must be stationary. The number of time series data distinction will become the order of d values in the model used ARIMA. A stationary data when said average value and variance are constant over time. Is not stationary data need to be modified (made the distinction) to generate stationary data. Here is a plot autocorrelation function (ACF), and partial autocorrelation function (PACF) as shown in Figure 1 and 2 below. We present also for the PACF and the first distinction at Figure 3 and 4, respectively.

The Estimated and Validation Model
Through the ACF and PACF plot of the original data is performed first distinction, while the model is determined CO2 emissions data period 1889 to 2004. From the ACF plot (Figure 3-2) and PACF ( Figure 3-3) obtained information that the CO 2 emissions ACF lag signnifikan at 1,2,3,4,5. While CO 2 significant PACF at lag 1 and 2. Thus while the model of the data plot is a mixture of CO 2 emissions from autoregressive, the first distinction, and moving averages or ARIMA model (p, 1, q). With the pvalue is 1 and 2 while the value of q selected 1, 2, 3, 4, and 5. Next is an estimate of the lag-lag is to get the best model. After establishing the identification of the model temporarily, then the parameters AR and MA should be established.

SUMMARY
Based on the above results it can be concluded that the best predictor model for the Total Fossil Fuel CO 2 Emissions over Indonesia is ARIMA (8,1,3). This means that the predicted value for the next year depending on the data before and 8 years 3 years earlier error. In the validation data with predicted results, the MAD (Mean Absolute Deviation) is relatively high. However, the pattern of results followed the pattern predicted almost the original data with a correlation value of 99%. Based on this result, we can estimate the climate projection over Indonesia, especially during 2012-2014.