# The precision of JEDI

**URL:** <https://forums.jcsda.org/t/the-precision-of-jedi/706>\
**Category:** JEDI\
**Created:** [September 25, 2025, 8:44pm UTC](https://forums.jcsda.org/t/the-precision-of-jedi/706 "2025-09-25T20:44:23Z")\
**Posts on this page:** 7\
**Page:** 1

<div class="post-metadata">

**Author:** ![JiaWANG](https://avatars.discourse-cdn.com/v4/letter/j/f17d59/32.png) [@JiaWANG](https://forums.jcsda.org/u/JiaWANG)\
**Post date:** [September 25, 2025, 8:44pm UTC](https://forums.jcsda.org/t/the-precision-of-jedi/706/1 "2025-09-25T20:44:23Z")

</div>

In the JEDI tutorial, it stated that “Note that the tutorial test cases are designed with single-precision MPAS model test dataset. Therefore, we pre-compiled mpas-bundle code with single-precision mode“.

How to compile the MPAS-JEDI bundle with a single-precision mode?

When I have an already compiled MPAS-JEDI, how to tell whether it is compiled in a single-precision mode or a double-precision mode?

---

<div class="post-metadata">

**Author:** ![JiaWANG](https://avatars.discourse-cdn.com/v4/letter/j/f17d59/32.png) [@JiaWANG](https://forums.jcsda.org/u/JiaWANG)\
**Post date:** [September 25, 2025, 9:02pm UTC](https://forums.jcsda.org/t/the-precision-of-jedi/706/2 "2025-09-25T21:02:40Z")

</div>

I found out that there is a line in the **CMakeLists.txt** to control the precision:

**set(MPAS\_DOUBLE\_PRECISION “ON” CACHE STRING “MPAS-Model: Use double precision 64-bit Floating point.”)**

On a relative but separate note, does the precision affect the data assimilation impacts?

---

<div class="post-metadata">

**Author:** ![byoung-joo](https://yyz1.discourse-cdn.com/flex031/user_avatar/forums.jcsda.org/byoung-joo/32/126_2.png) [@byoung-joo](https://forums.jcsda.org/u/byoung-joo)\
**Post date:** [September 25, 2025, 9:48pm UTC](https://forums.jcsda.org/t/the-precision-of-jedi/706/3 "2025-09-25T21:48:31Z")

</div>

Hi @JiaWANG ,

> On a relative but separate note, does the precision affect the data assimilation impacts?

I don’t expect that the use of single precision MPAS model will affect the DA result very much.  
The data structure of mpas-jedi fields inherit from MPAS-Model’s pool data structure. So the precision of MPAS-Model build will also determine the precision of mpas-jedi fields data structure. This leads to the less memory requirement for mpas-jedi application and smaller mpas/mpas-jedi NetCDF file size.

Thank you,  
BJ

---

<div class="post-metadata">

**Author:** ![JiaWANG](https://avatars.discourse-cdn.com/v4/letter/j/f17d59/32.png) [@JiaWANG](https://forums.jcsda.org/u/JiaWANG)\
**Post date:** [September 25, 2025, 10:01pm UTC](https://forums.jcsda.org/t/the-precision-of-jedi/706/4 "2025-09-25T22:01:40Z")

</div>

Thanks BJ. Understand that the precision needs to be consistent between MPAS and JEDI.

May I ask that in your group’s research work, which precision do you use?

---

<div class="post-metadata">

**Author:** ![byoung-joo](https://yyz1.discourse-cdn.com/flex031/user_avatar/forums.jcsda.org/byoung-joo/32/126_2.png) [@byoung-joo](https://forums.jcsda.org/u/byoung-joo)\
**Post date:** [September 25, 2025, 10:22pm UTC](https://forums.jcsda.org/t/the-precision-of-jedi/706/5 "2025-09-25T22:22:56Z")

</div>

The precision of MPAS Model and MPAS-JEDI does not need to be the same. It is not a requirement. Also note that the (most) DA procedures, which happen in the generic JEDI components, is still done with a double precision.

In MMM, we are mostly using the single precision build of MPAS-Model for our experiment.

Thank you,  
BJ

---

<div class="post-metadata">

**Author:** ![JiaWANG](https://avatars.discourse-cdn.com/v4/letter/j/f17d59/32.png) [@JiaWANG](https://forums.jcsda.org/u/JiaWANG)\
**Post date:** [September 25, 2025, 11:14pm UTC](https://forums.jcsda.org/t/the-precision-of-jedi/706/6 "2025-09-25T23:14:00Z")

</div>

Then a double precision MPAS can generate a single precision DA input file, which will be read by a single precision JEDI?

Thanks for the clarification about the double-precision internal procedures.

Have you done internal comparisons of DA performance between the single versus double precision?

---

<div class="post-metadata">

**Author:** ![byoung-joo](https://yyz1.discourse-cdn.com/flex031/user_avatar/forums.jcsda.org/byoung-joo/32/126_2.png) [@byoung-joo](https://forums.jcsda.org/u/byoung-joo)\
**Post date:** [September 30, 2025, 3:26pm UTC](https://forums.jcsda.org/t/the-precision-of-jedi/706/7 "2025-09-30T15:26:53Z")

</div>

> Then a double precision MPAS can generate a single precision DA input file, which will be read by a single precision JEDI?

Yes, you can find the related information from [Configuring Model Input and Output — MPAS Atmosphere documentation](https://www2.mmm.ucar.edu/projects/mpas/site/documentation/users_guide/configuring_io.html)

> Have you done internal comparisons of DA performance between the single versus double precision?

I don’t remember if we’ve done this before.

Thank you,  
BJ