"Process Screening" is a platform that can evaluate multiple processes at once. It calculates the process capability and control chart indexes for each process, allowing you to efficiently screen important processes.
Did you know that this platform can draw a "shift graph" like the one shown below? This graph allows you to visually consider when a shift (described below) occurred in which process. You can also drill down to analyze any process you are concerned about in detail.
In this blog, we will consider temperature data from all over Japan as a kind of "process" and consider when sudden increases in temperature (shifts) occurred, as well as the differences in the extent of increases at each observation point.
Analysis target
- Target area: Regional and local meteorological observatories in Japan (58 locations with data)
- Target period: 1995 to 2024 (30 years) - data from June to August (summer) of each year
- Indicator: Calculate the three-month average of daily average temperature, maximum temperature, and minimum temperature, and investigate annual trends.
Data source: Japan Meteorological Agency website
What is "Shift"?
A shift is a sudden change in value in time series data. In process control, it refers to a phenomenon in which a characteristic value changes suddenly due to a quality change or other reason.
For example, in the graph below showing changes in average temperature, it is believed that sudden upward shifts occurred in 2004, 2010, and 2023 (green lines).
*This analysis only considers upward shifts and does not take downward shifts into account.
I would like to examine the shifts for each target area, but this time I am targeting 58 weather stations, each with three indicators: average temperature, maximum temperature, and minimum temperature, so it would take a lot of time to look at the graphs for each one.
Therefore, we will use "process screening" to consider temperature shifts.
Detecting shifts using "process screening"
Select [Analysis] > [Quality and Process] > [Process Screening] and follow the steps below to detect shifts.
If you select Maximum upward shift in the Shift detection option, you can output a table like the one below.
"Maximum upward shift" and "Year of upward shift" show "the year in which the largest upward shift occurred" during the target period and "the upward shift at that time."
The shift calculated by the JMP algorithm* is divided by the within-group sigma (an index representing the variability of the data) and displayed as the "maximum upward shift."
That is, the larger the value shown for "Maximum upward shift" is, the larger the shift is relative to the variability of the data.
* In JMP, the magnitude of the shift is calculated for data that has been smoothed and outliers removed, not for raw data.
Using the shift graph
The "shift graph" introduced at the beginning is a graph that visualizes the occurrence of an "upward shift."
Shift graph when the shift threshold is set to 2
The graph below is a shift graph plotting the areas where there was an upward shift of 2 or more, and the temperatures (average, maximum, and minimum).
For example, the average temperature at the Matsue Meteorological Observatory, at the top, is expected to shift upward by more than 2 in 2023.
シフトの閾値を2にしたときのグラフGraph when shift threshold is set to 2
Looking at the graph, the most upward shifts are occurring in 2023.The next most common year was 2010.
Shift graph when the shift threshold is set to 3
If we narrow the upward shift to 3 or more, we get a plot like this:
The plot is all centered around the year 2023.
By using a "drift graph" to determine when a shift occurred, it is possible to investigate in detail the specific time when the shift occurred and the progression.
*We will not go into detail about "drift" here.
Analysis results and discussion
This shows a bar graph (number of locations) tallying the years when the largest shifts occurred for the 58 observation regions.
The average temperature, maximum temperature, and minimum temperature are all concentrated in the last three years, from 2022 to 2024, highlighting the clear trend of global warming in recent years.
The maximum upward shift in average temperature at each location is shown color-coded on the map.
We can see that eastern Japan tends to experience a larger upward shift than western Japan.
Although not shown here, maximum and minimum temperatures also show similar trends.
summary
- Using JMP's Process Screening platform, we attempted to detect temperature shifts at various locations in Japan.
- The optional "shift graph" was used to visualize the times when large upward shifts in temperature were occurring.
These analyses show that the largest shifts are likely to occur between 2022 and 2024, with larger shifts in eastern Japan than in western Japan.
As someone living in eastern Japan, I have felt that the summer heat over the past few years has been "unbelievable," and the data backs this up.
*This blog is based on part of the content presented at Discovery Summit Japan in 2024. The presentation materials can be downloaded from the following page.
Discovery Summit Japan 2024 - Handouts - JMP User Community
⇒ Download C3_Naohiro-MASUKAWA.pdf
by Naohiro Masukawa (JMP Japan)
Naohiro Masukawa - JMP User Community
