Introduction

With most grib packing methods, the grid values are usually stored as scaled integers. The grib format allows the scaled integers to have up to 254 bits (simple packing). This is an unreasonably high precision as wgrib2 converts the data to a single-precision floating point number which typically has only 25 bits of precision. To speed up the encoding and decoding processing, wgrib2 uses 32 bit integers which limits the maximum size of scaled integers to 25 after accounting for the limits of the packing and unpacking routines.

The complex-2 and complex-3 compression schemes also store scaled integers but different. The complex-1 stores the scaled intger values, so it is limited to a 25 bit integers. Hewever, complex-2 stores the delta (the difference from previously defined grid point). The magnitude of delta is limited by the original series but the delta can be either positive or negative. So under worst case, the deltas need one more bit to store the scaled integers. So if you scale the grid point values to 24 bits, you will have a scaled deltas being 25 bits. The complex-3 packing stores the deltas of the delta series. This is good for smooth fields, but for the worst case could cause two extra bits to store the original grid data. So at worst case, the data would need to be scaled at 23 bits.

The extra bits needed for complex-2 and complex-3 are very rare. It was first encountered in 11/2024 which is about a decade after the complex packing was introduced. In 11/2024, code was added to the complex packing to detect the increase in bits needed and reduce the precision if it cause the precision to be greater than 25 bits.

Grib fields are usually stored as a scaled integers, and usually you don't need 25 bits of precision. For example, temperature to the nearest 0.1 degree, means that the a temperature range of 50 C can be stored as 500 descrete values and only 7 bits are needed to store the temperature to the nearest 0.1 degree.

The problem is that people can set the scaling factors to inappropriate values, using 25 bits of precision will was (disk) space. Consequently wgrib2 has a resetable limit of the precision of the scaled integers allowed. If you exceed the limit, the scaling factors are change so that the scaled integers are limited.

The -set_grib_max_bits option sets the maximum number of bits that the scaled integers can have when encoding data into grib. The value should never be greater than 25 as that is the limit for wgrib2 encoding. However, some grib packages may not support 25 bit precision in decoding, so you may want to limit the precision to 24 to remain compatible with other software packages.

The -set_grib_max_bits option does not set the binary precision of the grib output but sets the maximum possible precision of the numbers.

Usage

-set_grib_max_bits N
  N = maximum number of bits used to encode data
  the default value is 16, N = 1..25

Comments

The -set_grib_max_bits option does not affect grib data written using the -grib option because the -grib option does not encode data but copies the input grib message.

See alse: -set_scaling -set_bin_prec -grib_max_bits