The CRU JRA (Japanese reanalysis) data is a replacement to the CRU NCEP dataset, CRU JRA data follows the style of Nicolas Viovy's original dataset rather than that which is available from UCAR.

The CRU JRA dataset is based on the JRA-55 reanalysis dataset and aligned where appropriate with the CRU TS dataset version 4.08 (1901-2023).

All JRA variables are regridded from their native TL319 Gaussian grid to the CRU regular 0.5° x 0.5° grid, using the g2fsh spherical harmonics routine from NCL (NCAR Command Language), based on the 'Spherepack' code. The exception is precipitation, which is regridded using ESMF 'nearest neighbour': all other algorithms tried exhibited unwanted artifacts.

The JRA-55 reanalysis dataset starts in 1958. The years 1901-1957 are constructed using randomly-selected years between 1958 and 1967. Where alignment with CRU TS occurs, the relevant CRU TS data is used.

Of the ten variables listed above, the last four do not have analogs in the CRU TS dataset. These are simply regridded, masked for land only, and output as CRUJRA. The other six are aligned with CRU TS as follows:

TMP is aligned with CRU TS TMP. A monthly mean for the JRA data is
calculated and compared with the equivalent CRU TS mean. The difference
between the means is added to every JRA value.

---

TMAX and TMIN are aligned with CRUJRA TMP and CRU TS DTR. Firstly, at
each time step, the TMAX-TMP-TMIN triplets are checked and adjusted so
that TMAX is always >= TMP, and TMIN is always <= TMP. This triplet
alignment is prioritised above DTR alignment. Secondly, monthly JRA DTR
is calculated by first establishing the daily maxima and minima (max and
min of the subdaily values in TMAX and TMIN respectively), then monthly
maxima and minima, (means of the daily DTR values), giving JRA monthly
DTR. This is compared with CRU TS DTR and the fractional difference
(factor) calculated as (CRU TS DTR) / (JRA monthly DTR). This factor is
then used to adjust the DTR of each pair of subdaily TMAX and TMIN
values, though not if the triplet alignment would be broken.

---

PRE is aligned with CRU TS PRE and WET (rain day counts). Firstly, the
monthly total precipitation is calculated for JRA and compared to CRU TS
PRE; an adjustment factor is acquired (crupre/jrapre) and all values
adjusted. Precipitation amounts are now aligned at a monthly level, and
this alignment is prioritised above WET alignment. Secondly, the number
of rain days is calculated for JRA: a day is declared wet if the total
precipitation is equal to, or exceeds, 0.1mm (the same threshold as CRU
TS WET). If JRA has more wet days than CRU TS, then the driest of those
are reduced to a random amount below 0.1 (an adjustment factor is
calculated and applied to each time step, to preserve the subdaily
distribution). If JRA has fewer wet days than CRU TS, then sufficient
dry days are set to a random amount equal to or closely above 0.1mm,
again using an adjustment factor to preserve the subdaily distribution.
Where wet day alignment threatens precipitation alignment, the process
is abandoned and the cell/month reverts to the previously-aligned
precip version. Exception handling is very complicated and cannot be
summarised here.

---

SPFH is aligned with CRU TS VAP. VAP is converted to SPFH, and JRA mean
monthly SPFH is calculated. The fractional difference (factor) is
calculated as (CRU TS SPFH) / (JRA monthly SPFH), this factor is then
applied to the JRA subdaily humidity values.

---

DSWRF is aligned with CRU TS CLD. CLD is converted to shortwave
radiation, and JRA mean monthly DSWRF is calculated. The fractional
difference (factor) is calculated as (CRU TS SWR) / (JRA monthly DSWRF),
this factor is then applied to the JRA subdaily radiation values.

---

Where appropriate, CRUJRA values are kept within physically-appropriate
constraints (such as negative precipitation), which could result from
regridding as well as adjustments.