📄 earth_science

Mutual Information assessment of the quality of the GRAV-D airborne gravity data and its compatibility with the NGS’ historical surface gravity database

This study demonstrates that Mutual Information serves as an effective diagnostic tool for assessing the internal consistency of GRAV-D airborne gravity data, evaluating its compatibility with historical surface and satellite datasets, and identifying the specific spectral contributions of airborne observations for future geoid modeling.

Original authors: Jarir Saleh, Yan Ming Wang, Kevin M Ahlgren, Ryan A Hardy, Jeffery A Johnson

Published 2026-07-09

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jarir Saleh, Yan Ming Wang, Kevin M Ahlgren, Ryan A Hardy, Jeffery A Johnson

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the Earth's gravity field as a giant, invisible blanket draped over the planet. To build a perfect map of this blanket (which scientists call a "geoid" and use to measure precise heights), we need to stitch together pieces of information from three different sources: satellites (seeing the big picture from space), ground surveys (measuring the details up close), and airplanes (flying in the middle to fill the gaps).

This paper is about a new way to check if the airplane measurements are good and if they fit well with the ground and satellite data. The authors use a mathematical tool called Mutual Information (MI).

Here is the simple breakdown of what they did and found:

1. What is "Mutual Information"?

Think of two people trying to describe the same painting.

Standard Correlation is like asking, "Do they agree on the color red?" It only checks for simple, straight-line agreements.
Mutual Information is like asking, "How much of the entire story of the painting do they share?" It doesn't care if the story is told in a straight line or a complex, twisting way. It measures how much one dataset "knows" about the other.

If the two datasets are perfect copies of each other, the Mutual Information is high. If one dataset is full of static noise, errors, or lies, the "shared story" gets shorter, and the Mutual Information drops.

2. Test #1: Does the Airplane Data Make Sense to Itself?

The researchers first compared the raw gravity data collected by the planes against a smoothed, computer-generated version of that same data (called a "downward-continued grid").

The Analogy: Imagine you take a photo of a landscape, then use a computer to smooth out the wrinkles in the photo. You then compare the original photo to the smoothed version. If they look almost identical, your camera was good. If the original photo is full of static or blurry spots, the two versions won't match up well.
The Result: The "match" was excellent almost everywhere, proving the GRAV-D airplane data is high quality overall. However, the MI map highlighted specific spots (mostly over mountains like the Rockies) where the match was weaker. This told the scientists exactly where the planes likely hit bumpy air, causing the sensors to get a little jittery.
Why it's better than old methods: Traditionally, they checked quality only where flight paths crossed each other (like checking a grid only at the intersections). This new method checks the entire flight path continuously, giving a complete picture of where the data is shaky.

3. Test #2: Do the Airplane Data and Old Ground Data Agree?

Next, they compared the new airplane data with the historical gravity data collected on the ground over the last century.

The Problem: Airplanes fly high; ground crews stand on the ground. Gravity changes as you move up or down, especially near tall mountains. Comparing them directly is like comparing a photo taken from a helicopter to a photo taken from the ground—they see different things because of the angle and height.
The Fix: The researchers used a "filter" (called isostatic anomalies) to remove the effects of the mountains and height differences, so they were comparing the "core" gravity signals.
The Result: Once they filtered out the height differences, the two datasets matched up very well in many places (like the East Coast). However, the MI map still showed "red zones" (low agreement) in places like Southern California, parts of the Southwest, and some ocean areas.
The Takeaway: These "red zones" tell us where the old ground data might be wrong, where the new airplane data might have errors, or where satellite measurements over shallow water are struggling. It gives scientists a map of exactly where they need to be careful when combining these datasets.

4. Test #3: What "Frequency" Does the Airplane Data Actually Provide?

Gravity data has different "frequencies" or wavelengths:

Satellites see the long, slow waves (the big hills and valleys of the gravity field).
Ground surveys see the short, sharp waves (the tiny bumps and rocks).
Airplanes are supposed to fill the middle.

The researchers tested which "wavelengths" the airplanes actually captured best.

The Analogy: Imagine a radio. The satellites are the AM station (long waves), and the ground surveys are the FM station (short waves). The airplane is supposed to be the perfect middle ground.
The Result: The MI analysis confirmed that airplanes are excellent at capturing the "middle" wavelengths (roughly 40 to 220 km long). However, when they tried to see if airplanes could capture the very shortest, sharpest details (like the ground surveys do), the "shared story" disappeared. The airplanes simply couldn't see those tiny details.
Conclusion: This confirms that we still need ground surveys for the finest details, but the airplanes are doing a perfect job filling the gap in the middle.

Summary

The authors didn't invent a new gravity sensor; they invented a new quality control flashlight. By using Mutual Information, they can:

Spot exactly where airplane flights got bumpy.
Find exactly where old ground maps disagree with new airplane maps.
Confirm exactly which part of the gravity spectrum the airplanes are best at measuring.

This helps scientists decide how much to trust each piece of data when they stitch them all together to create the perfect vertical map of the United States.

Technical Summary: Mutual Information Assessment of GRAV-D Airborne Gravity Data

Problem Statement
The National Geodetic Survey's (NGS) GRAV-D project has acquired over 2.2 million kilometers of airborne gravity data to support the redefinition of the American Vertical Datum. While this dataset offers high spatial consistency, it faces challenges inherent to continental-scale airborne gravimetry: residual track-dependent errors (biases and tilts), stochastic noise, and the complexities of downward continuation. Furthermore, integrating this modern airborne data with the heterogeneous historical surface gravity database (compiled from various instruments and standards over decades) and satellite-derived models requires robust methods to assess data quality, internal consistency, and compatibility. Traditional quality control metrics, such as crossover analysis, are limited by their sparsity (existing only at flight-line intersections) and their reliance on linear assumptions. There is a need for a diagnostic tool capable of providing a continuous, geographically resolved, and nonlinear measure of similarity between these disparate datasets to guide future geoid modeling and data weighting strategies.

Methodology
This study employs Mutual Information (MI), an information-theoretic metric, as a nonlinear measure of similarity between datasets. Unlike correlation coefficients, which detect only linear or monotonic relationships, MI quantifies the total statistical dependence (linear and nonlinear) between two random variables. The authors utilize the Kraskov et al. (2004) nonparametric estimator based on K-nearest-neighbor (KNN) statistics with $K=1$ to estimate MI from finite samples without assuming specific probability distributions.

The analysis is conducted in three distinct phases:

Internal Consistency: MI is computed between ~17.5 million GRAV-D airborne gravity disturbance observations and a downward-continued (DWC) grid derived from the same data at a constant altitude of 6,000 m. A circular neighborhood of 60 km radius is used to sample the data, excluding points with altitude differences greater than 200 m.
Dataset Compatibility: MI is evaluated between GRAV-D airborne free-air gravity anomalies and a historical surface gravity grid (merged with satellite altimetry over oceans). To isolate data inconsistencies from physical gravity field variations caused by elevation differences, the analysis is repeated using isostatic anomalies, which act as a low-pass filter to attenuate short-wavelength signals associated with topography and shallow density contrasts.
Spectral Contribution: To determine the spectral role of airborne gravity, MI is computed between GRAV-D disturbances and synthetic gravity disturbances generated from specific spectral bands of the satellite-only model GOCO06s (degrees 2–180) and the terrestrial-based model EGM2008 (degrees 181–2160). This analysis focuses on a test region covering Wyoming and surrounding states.

Key Results

Internal Consistency: The MI map between airborne observations and the DWC grid shows uniformly high values (average ~4.5 bits) across the survey area, confirming the overall quality and internal consistency of the GRAV-D dataset. Regions of reduced MI correlate broadly with areas of high crossover errors and mountainous terrain (e.g., the Rockies), where atmospheric turbulence introduces noise and track-dependent errors. However, unlike crossover errors, MI provides a continuous spatial assessment rather than sparse point estimates.
Compatibility with Historical Data: Initial comparisons between airborne and historical surface anomalies showed reduced MI in mountainous regions, largely due to strong gravity gradients affecting datasets at different elevations. After applying isostatic filtering, MI values improved significantly in these areas. Persistent low-MI regions remain in the western/southwestern US, parts of the Atlantic coastal plain, and specific offshore areas (e.g., south of Puerto Rico and offshore Southern California). These discrepancies likely indicate a combination of residual airborne errors, inaccuracies in legacy terrestrial data, or limitations in altimetry-derived gravity in shallow waters.
Spectral Role:
- Long Wavelengths: High MI values between GRAV-D and GOCO06s (degrees 2–180) confirm strong agreement with satellite-derived long-wavelength gravity, even without explicit bias adjustment.
- Medium Wavelengths: Comparisons with EGM2008 (degrees 181–1000) yield high MI values, indicating that airborne gravity provides substantial information in this spectral band, bridging the gap between satellite and terrestrial data.
- Short Wavelengths: MI values drop significantly for the 1001–2160 degree band, confirming that airborne gravity contributes little independent information at very short wavelengths, where terrestrial observations remain essential. The transition appears gradual rather than sharply defined.

Significance and Claims
The authors posit that Mutual Information serves as an effective, nonlinear diagnostic tool that complements traditional geodetic analysis. The study claims that MI offers three primary advantages:

Continuous Spatial Resolution: It provides a geographically complete measure of data consistency, overcoming the sparsity limitations of crossover analysis.
Nonlinear Sensitivity: It detects a broader range of dependencies and inconsistencies, including those arising from complex processing artifacts or nonlinear noise characteristics, without requiring explicit functional models.
Spectral and Compatibility Insight: It provides a spatially resolved framework for assessing the compatibility of heterogeneous datasets (airborne, terrestrial, satellite) and identifying the specific spectral ranges where airborne data offers unique, non-redundant information.

The paper concludes that MI maps provide a quantitative basis for determining relative data weights in future geoid determination. While the spectral findings align with previous spherical-harmonic degree-variance studies, the MI approach offers a new, straightforward method to visualize these relationships geographically. The authors emphasize that the tool is particularly useful for identifying regions where data discrepancies exist, thereby guiding further investigation into data quality and combination strategies.

1. What is "Mutual Information"?

2. Test #1: Does the Airplane Data Make Sense to Itself?

3. Test #2: Do the Airplane Data and Old Ground Data Agree?

4. Test #3: What "Frequency" Does the Airplane Data Actually Provide?

Summary

More like this