연구보고서
- 저자
- 정유란 박사
- 작성일
- 2019.06.13
- 조회
- 361
- 요약
- 목차
The purpose of this research was to apply seamless statistically downscaled climate forecast data to water management and agriculture to the 2012–2013 drought in Thailand and build a risk assessment framework.
Capitalizing on Boreal Summer Intraseasonal Oscillation (BSISO) forecasts and Multi-Model Ensemble (MME) seasonal prediction provided by the APCC, we investigated the mechanisms of rainfall in Thailand. We also developed procedures for the collection, quality control, and downscaling of multi-range climate information such as sub-seasonal to seasonal (S2S) data, seasonal forecasts, and climate change scenarios for use in drought, agriculture, and water resources management.
We evaluated the usability of the satellite and meteorological reanalysis data on ungauged watersheds, considering that there were difficulties in drought analysis after 2013 due mainly to the large quantities of missing data. The missing values were filled in using the correlation between observed and MERRA2 data which showed the best results at the daily, monthly, yearly, and 30 year average monthly time scales. The quality-controlled data were then used for drought analysis.
A Simple Bias Correction (SBC)-based method was used to downscale the ECMWF (European Center for Medium-Range Weather Forecasts) S2S data against that from weather stations. In terms of seasonal forecasts, SBC, Climate Index Regression (CIR), Moving Window Regression (MWR), and Observation-based Moving Window Regression (MWR-Obs) modules were used to downscale precipitation and temperature for Northeast Thailand based on the 11 individual models collected for the APCC’s MME. In terms of climate change scenarios the APCC integrated Modeling Solution (AIMS) platform was used to downscale 29 Global Climate Models based on the Simple Quantile Mapping (SQM) and Spatial Disaggregation and Quantile Delta Mapping (SDQDM) methods. SQM was selected for further analyses using the reproducibility test against the historical period. The performance of the downscaled forecast data that was developed in this project, including S2S, was evaluated in the first year.
Two approaches to predict drought in Thailand were developed and modified throughout two year project. One is based on machine learning using the standardized precipitation index (SPI) and the other is based on the effective drought index (EDI). The drought prediction model based on machine learning for ungauged areas was modified so that S2S data could be used for predictions with a two to month lead-time in addition to one month predictions. When compared to the SPI3 prediction results from the bias-corrected seasonal forecast data, the latter was outperformed only for the one month predictions. Nevertheless, combining S2S and seasonal forecast data with different time scales for application in a drought prediction model is a novel approach worth following up in further studies. Machine learning-based drought prediction models with finer time scales may be more appropriate for use with S2S forecast data.
We also developed a system to forecast droughts with a six month lead-time using downscaled precipitation forecasts and the EDI. A system to forecast area-averaged drought over a target region was developed and its prediction skill was evaluated in the first year. In the second year, a system was developed to forecast the spatial distribution of drought conditions over a target area based on the downscaled precipitation data at a station level. The system was applied to the 2012–2013 droughts in Northeast Thailand but showed some limitation in predicting the spatial distribution of drought at its beginning and end.
While the six month forecast for the Lam Takhong dam inflow was assessed during the first year, changes in dam inflow according to changes in the climate were evaluated in the second year. According to the analysis of variations in the monthly average dam inflow and maximum daily inflow in each month, the dam inflow increased mainly during the July–October period and the uncertainty range was high during this period. Both the total and daily maximum inflow increased for all future periods in all Representative Concentration Pathway (RCP) scenarios; RCP8.5 indicated a higher rate of increase compared to RCP4.5. The increase in available water due to the increase in dam inflow can be considered for drought risk management in Northeast Thailand.
We evaluated whether the climate prediction information on various temporal scales could be used to predict potential rice yield. Rice yield potential were estimated based on seasonal precipitation forecasts and S2S precipitation forecasts using correlation analysis and probability density function analysis. In the estimate that used the seasonal forecast precipitation, shorter lead-times had higher correlations than those with longer lead-times. When the S2S precipitation forecast was used, potential rice yield predictions were highly correlated when predicted with a lead-time of approximately 15 days. When the two forecast climate datasets were combined, the correlation of the potential rice yields predicted using forecasts with lead-time of one month (L1) were higher than those with a lead-time two months (L2), but they were not significant.
The date of the onset of the rainy season is a critical factor for determining planting dates, and high temperatures and a lack of sunshine during the heading time and the grain-filling period, which affect yield, could be components of the risk to potential rice yield during the growing season.
BSISO forecasts can be utilized to predict the onset date three to four weeks in advance. It is also possible to obtain information such as the amount of precipitation necessary to determine the planting date using the ECMWF S2S 46 day precipitation forecast. It is also expected that the combination of S2S temperature, seasonal temperature forecasts with a one month lead-time, or temperature forecasts on an individual time scale could provide data on the temperature at which rice yield could fail (e.g., 39 °C) during the heading and grain-filling periods. Therefore, it was determined that it would be possible to derive the climate exposure (e.g., drought) risk components for each stage of rice growth and development during the growing season (e.g., planting phase or maturity stage) from sub-seasonal and seasonal forecasts.
Finally, we developed a framework for drought risk assessment in agriculture and water resources management using multi-range climate data for Thailand based on the Hyogo Framework for Action (HFA) and the Sendai Framework for Disaster Risk Reduction. The concept of climate exposure–sensitivity, which is used mainly in vulnerability assessments for climate change adaptation, was considered a component of risk assessment framework. The components of the drought risk index were defined as functions of hazard, exposure, and vulnerability. The Integrated Water Supply Risk Index (IWSRI) and the Integrated Rice Potential Yield Index (IRPYI), which could affect dam inflow and agricultural yield, were developed for agricultural and water resources drought risk assessment, respectively. This will be useful to predict drought risk in Thailand using multi-range climate data.
In addition, the results of evaluating the predictability of potential rice yield conducted in this study are based only on case studies, and there is still much room for the improvement. However, this study is expected to be a good prototype for the application of seasonal forecast climate data to predicting potential rice yield since the climate prediction skill is being improved continuously and there are various spatiotemporal downscaling methods being developed.