IRI-2020 vs GNSS-RO Indonesia

Scope status

This note synthesises the empirical comparison between IRI-2020 model predictions and GNSS Radio Occultation observations over the Indonesian region. It is a comparison note, not a standalone authority on either the model or the technique.

Boundary:

  • It does not restate the full IRI model specification.
  • It does not claim that the sampled days represent all Indonesian conditions.
  • It links empirical findings to SBAS integrity implications in the vault.

Comparison dimensions

Overall error summary (initial 5-day study)

MetricValueInterpretation
Bias66.72 TECUIRI underestimates TEC by 66.72 TECU on average
RMSE94.34 TECUTotal error magnitude
MAE67.01 TECUAverage absolute error
MAPE87.6%Mean percentage error
Correlation (r)0.4146Weak positive correlation
−0.7309Model performs worse than predicting the mean

Monthly variation

MonthBias (TECU)RMSE (TECU)
January83.7108.7
February70.291.9
March72.5113.2
April75.8102.3
May48.069.4

Altitude dependence

Errors are largest at lower occultation heights (near the F2 layer peak):

Altitude (km)Bias (TECU)
350–400126.0
400–45077.9
500–55047.9
550–60033.8

Spatial variation

RegionBias (TECU)RMSE (TECU)
Western (Sumatra/Java)98.0122.7
Central (Kalimantan)114.8148.2
Eastern (Sulawesi/Papua)106.6145.3

What this comparison implies

  1. Systematic underestimate, not random scatter — IRI-2020 consistently underpredicts TEC in the Indonesian equatorial region.
  2. Altitude sensitivity — The largest errors occur near the F2 peak height, suggesting the model’s vertical profile parameterization is misaligned with actual equatorial conditions.
  3. Spatial heterogeneity matters — Central Indonesia (Kalimantan) shows the largest errors, likely due to EIA crest proximity.
  4. Temporal gaps are integrity-relevant — No post-sunset (18–21 LT) profiles means the most challenging equatorial conditions are unobserved in this dataset.

Comparison to SBAS-relevant thresholds

QuantityTECUL1 equivalent (m)SBAS relevance
Median RO TEC46.97.6Typical delay
p95 RO TEC166.026.9Heavy delay, threat discovery
p99 RO TEC341.155.4Extreme delay, overbounding consideration
Maximum observed468.476.1Tail event, integrity penalty
IRI-2020 bias66.7210.8Systematic model error

Relationship to the vault

Parent domain

Sibling domain

Upstream notes

What this comparison does NOT justify

  • That IRI-2020 is globally invalid (findings are regional and conditional)
  • That GNSS-RO can replace ground GNSS for operational SBAS correction
  • That the sampled days represent all Indonesian conditions
  • That the post-sunset gap proves low risk (absence of data ≠ absence of risk)

Open questions

  • How do these errors translate to SBAS protection level inflation?
  • What additional ground-station density would be needed to bound these conditions operationally?
  • Would NeQuick or a physics-based model perform differently in the same region?

See also