| 研究課題/領域番号 |
23K12463
|
|---|
| 研究機関 | 一橋大学 |
|---|
研究代表者 |
Wang Hongming 一橋大学, 社会科学高等研究院, 非常勤研究員 (20867048)
|
|---|
| 研究期間 (年度) |
2023-04-01 – 2025-03-31
|
| キーワード | climate change / mortality / machine learning |
|---|
| 研究実績の概要 |
The purpose of the research is to understand the impacts of climate change on mortality in Japan and to provide insights for policies mitigating the health impacts of climate change. So far, I have collected satellite re-analysis data and have processed the data to derive daily weather variations in temperature, humidity, precipitation, and wind speed across municipalities in Japan. Based on the weather data, I quantified the magnitude of climate change across Japan focusing on changes in the frequency and intensity of extreme weather events relative to their local climatology in 1951-1980. I show that over time, abnormally hot days with maximum temperature above the 90th percentile of local climatology increased in frequency across Japan, and the intensity of extreme heat above the historic normal has risen sharply in recent decades.
In addition to changes in univariate extremes in temperature and precipitation, climate change is also characterized by increases in compound extreme events such as prolonged heatwaves, hot-and-drought periods, and days with both high temperature and humidity. In light of their potentially devastating impacts on natural systems and human health, I quantify the magnitude of compound extreme events in Japan using respective indices developed in the climate literature. Building on the climate variables, in year 2024, I will explore the heterogeneous and non-linear relationship between mortality and climate drivers using machine learning.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
3: やや遅れている
理由
The research is slightly delayed relative to schedule for two reasons. First, I attempted to obtain individual-level mortality data covering the period 1972-2019 from the Ministry of Health, Labour and Welfare of Japan. I was not able to get access to the data from the Ministry. I later proceeded with using publicly available municipality-level mortality data covering the period 1980-2019 for my research. For this reason, the estimation sample covers the period 1980-2019 instead of the 1972-2019 period as initially outlined in the proposal. Second, there was an interruption in the computing service available in my office. I initially conducted exploratory data analysis using the Lenovo desktop housed in my office, but in early 2024, the desktop was moved to a separate office to be used exclusively by a different research team. I then purchased the Dell Precision 7865 desktop to continue my own work. However, I was not able to purchase additional memory capacity for the desktop using the FY2023 budget. The memory upgrade will be completed using budget from FY2024.
|
| 今後の研究の推進方策 |
Building on the climate variables quantified in year 2023, in year 2024, I employ machine learning to empirically characterize the relationship between mortality and climate change and to identify high-risk areas most vulnerable to the health impacts of climate change. Compared to linear regression models, machine learning has the advantage that it can non-parametrically quantify the mortality responses to a variety of climate drivers without imposing strong functional form assumptions such as linearity.
I apply several machine learning methods including penalized regressions, decision trees, and neural network, and use cross validation to select the best-performing model to reveal the non-linear relationship between mortality and climate drivers. To track the impacts of climate change over time, and to detect any meaningful adaptation that might mitigate the mortality consequences in more recent periods of climate change, I quantify the mortality response curves separately for 1980-1999 and 2000-2019. A flattened response curve to extreme heat in 2000-2019 is consistent with adaptation to the climate hazard in recent history. Across climate factors, I use the model to identify the leading hazards contributing to mortality, and across locations, I identify areas most vulnerable to climate-induced mortality due to increased exposure to such hazards. Due to difficulties in accessing individual mortality records that cover the period 1972-2019, I will focus on the period 1980-2019 where municipality-level mortality data is publicly available.
|
| 次年度使用額が生じた理由 |
I incurred balance in year 2023 because the purchase of Dell Precision 7865, which cost 823,669 JPY, was less than the 1,000,000 JPY budget allotted. After accounting for additional costs purchasing external hard drives and software, the 2023 balance is 150,322 JPY. I will combine the balance with my 2024 budget to purchase additional memory capacity for the Dell Precision 7865. This will upgrade the desktop's memory capacity from the current 16GB to a total of 640GB.
|
