Contents Shortcut
Main Menu Shortcut

SEA Fire and Haze EWS

Introduction

Haze from forest fires is among Southeast Asia’s most serious environmental problems. Severe forest fires in Indonesia occur only during years with anomalous low rainfall, causing severe haze in 1982, 1991, 1994, 1997 and 2006. However, measures to prevent these fires and mitigate their impacts remain limited by the absence of a long-lead early warning system (EWS). Severe burning conditions need to be forecast weeks to months in advance for any preventative actions to be effective. This indicates the clear need for a fire and haze EWS.

APCC is proud to present its developed fire and haze EWS which translates 3-month precipitation data into four fire danger ratings based on the relationship between the precipitation amount and CO2 emissions. It is our hope that this data will be utilized to prevent dangerous forest fires and haze.

SEA-Fire and Haze Early Warning System (FHEWS)

The results show the forecast summary for monthly precipitation and probability of forest fires in a selected region for August to October (ASO). The probabilistic forecast of forest fires is issued during April to July by considering the target period (ASO) and 6-month maximum lead-time.

  • The graph shows the graphical information for previous and current years by providing climatology (blue), observed (red), and multi-model ensemble (MME) forecast of precipitation (black). The boxplot in the figure shows the variations of the predicted values by individual models.
  • The bottom-left table indicates the precipitation forecast skill scores for each month according to different lead-times calculated based on the long-term (1983-2007) period. The performance measures include Temporal Correlation Coefficient (TCC), which can be used for continous variables and Accuracy and Heidke Skill Score (HSS), which in turn can be used for category forecasts. For calculating HSS, we equally divide the observed monthly precipitation into 3 categories (33.3% for each). HSS measures the fraction of correct forecasts after eliminating those forecasts which would be correct due purely to random chance (http://www.cawcr.gov.au/projects/verification).
  • The bottom-right table shows information on probabilistic forecast of forest fire during August to October using forecasted 3-month (ASO) total precipitation.

Methodology and Result

The overall procedures for developing the EWS prototype are as follows: 1) construct the statistical downscaling model to forecast monthly average precipitation levels for each region; 2) determine the number of categories and corresponding ranges for the fire danger rating system based on the relationship between total ASO precipitation amounts and CO2 emissions; and 3) forecast probabilistic fire danger ratings based on the predicted precipitation amount.

APCC has been collecting monthly precipitation data predicted by multiple dynamic models. Monthly prediction data regrided with 2.5°⨯2.5° resolution based on 11 individual Global Climate Models (GCM) (4 and 7 GCMs with 3-month and 6-month lead forecasts, respectively) were used for bias-correction using Simple Bias Correction (SBC) method. Table 1 shows the description of the 11 GCMs used in this EWS.

Table 1. List of seasonal forecast models used for the real-time EWS.
Name Modeling Centre Ensemble
Member
Lead time
(in month)
BCC Beijing Climate center Model 8 3
CWB Taiwan Central Weather Bureau 10 3
HMC Hydrometerorogical Center Russia 10 3
IRI_CA International Research Institute for Climate and Society 24 3
APCC APEC Climate Center 10 6
CMCC Centro Euro-Mediterraneo sui Cambiamenti Climatici (CMCC) 9 6
MSC Metrological Service in Canada 20 6
NASA National Aeronautics and Space Administration (NASA) 11 6
NCEP National Center for Environmental Prediction 20 6
PNU Pusan National University 5 6
POAMA Predictive Ocean Atmosphere Model for Australia ( BoM, and CSIRO) 33 6

The SBC method was used to adjust the monthly mean of predicted precipitation by correcting the monthly average of prediction using the average of observations for the same period. Bias correction between the region-average of forecast data and observation (APHRODITE) was conducted by adding anomalies in the forecasted data to the mean of the observation. Table 2 shows the results of the prediction models that were selected for each case (month and lead-time) in the Selatan region. The number of selected models decreased as the lead time increased.

Table 2. Selected model for forecasting month precipitation according to lead time in Selatan region.
Month Lead time
1 month 2 month 3 month 4 month 5 month 6 month
Jan NASA
Feb NASA
Mar NASA
CWB
MSC
NCEP
PNU
Apr
May BCC PNU
CWB
MSC MSC MSC APCC
Jun POAMA PNU MSC
Jul NASA
POAMA
APCC
CMCC
NCEP
CMCC
NCEP
POAMA
NCEP
POAMA
CMCC
POAMA
MSC NCEP
POAMA
Aug NASA
POAMA
APCC
CMCC
NCEP
IRI_CA
NASA
PNU
CMCC
NCEP
POAMA
APCC
MSC
IRI_CA
MSC
NCEP
POAMA
APCC
CMCC
NASA
MSC
CMCC
POAMA
APCC
NASA
NCEP
MSC
CMCC
NASA
NCEP
PNU
POAMA
MSC
NCEP
POAMA
NASA
Sep POAMA
NCEP
PNU
IRI_CA
NASA
NCEP
POAMA
IRI_CA
MSC
NCEP
POAMA
POAMA NCEP
POAMA
MSC
NCEP
POAMA
CMCC
Oct NASA
POAMA
CMCC
NCEP
PNU
NASA
PNU
CMCC
NCEP
POAMA
IRI_CA
HMC
NCEP
POAMA
CMCC
NASA
MSC
CMCC
POAMA
NASA
NCEP
PNU
NCEP NCEP
NASA
Nov POAMA
CMCC
NCEP
POAMA NCEP
POAMA
CMCC
CMCC
NCEP
CMCC
POAMA
APCC
CMCC
Dec

Finally, the bias-corrected seasonal precipitation forecasts were interpreted in terms of historical precipitation-fire relationships. APCC developed a fire danger rating with four criteria (low, moderate, high, and extreme) for the four provinces in Borneo Island. An analysis of the threshold levels for the study regions was conducted in order to translate the predicted precipitation amount to the fire danger ratings. If the amount of precipitation dips below the threshold level, this leads to a prediction of an increased risk for severe burning, carbon emissions, and transboundary haze. It is necessary to connect the forecasted precipitation to the possible EWS index based on region-specific threshold level. We used the relationship between region-average ASO precipitation amount and carbon emission amount which was derived from Global Fire Emissions Database (http://www.globalfiredata.org/). Figure 1 shows the time series of the 3-month accumulated monthly precipitation and carbon emission levels in Selatan region.

Time series of 3-month accumulated monthly precipitation and carbon emission levels in the Selatan region.

Figure 1. Time series of 3-month accumulated monthly precipitation and carbon emission levels in the Selatan region.

The TCC values were calculated using MME, which issues forecasts every month. As a result, more forecast information was used to calculate MME when the lead-time was shorter. For example, when we predict precipitation levels in August during the month of July based on 3-month lead-time data, all three prediction results (including 1-month lead prediction issued in July, 2-month lead prediction issued in June, and 3-month lead prediction issued in May) can be used to estimate MME. The summary results are provided in the bottom-left table of the forecast result.

TOP