A Lifetime of Exploration: Data, Hiking, and Everything In Between

This is our second blog from one of our IDL Fellows. As I mentioned in the preface to the first entry (Unlocking the Secrets of Our Atmosphere: A Deep Dive into Solar Aureole Research - NV5 Geospatial), The IDL Fellows program is our way of supporting passionate retired IDL users who may need support to continue their work with IDL. There is continual innovation behind IDL, and this program is one of the many ways we hear about the new innovative ways people are using it on a regular basis. As you can see, this is certainly the case with Robert Velthuizen! If you are retired and interested in becoming an IDL Fellow and sharing your research through a blog post, feel free to reach out to me to see if you qualify for the program.

Exploring the Wonders of Data

My name is Robert Velthuizen. I’m a retired scientist who has spent a lifetime creating masterpieces with Interactive Data Language (IDL®). Even in retirement, my passion for data continues to flourish, and I’m excited to share some of my adventures with you.

From Pocket Calculators to Powerful Programming

I learned programming on a Texas Instrument TI-58: a weighty pocket calculator with a few kilobytes of memory – quite advanced for its time. I quickly progressed to using Fortran on a Cyber mainframe computer, that I squeezed to do remarkable jobs such as a complete membership administration program for my rowing club. Over the years, I’ve explored a whole host of different programming languages and environments, but none have captivated me quite like IDL. It opened a new world of possibilities, interactively prototyping code and interfacing with numerous specialized applications.

My graduate work was built almost entirely on IDL. I used it to create an environment to manage Magnetic Resonance Images (MRI) of brain tumors, release a host of image processing and machine learning methods, and report the accuracy. A follow-up project with my student Gloria Beyer built on this work to automatically determine a brain tumor radiation volume and interface with a planning system. The results were remarkable: automatic delineation results are as good as the outlines generated tediously by the expert radiation oncologist.

Figure 1: Rendering of the two radiation plans. Showing overlap in orange.

Figure 2: Representative skin colors generated using the "mountain method".

One of my interesting endeavors was exploring “subtractive clustering”, also known as the “Mountain Method”. I found the original method hard to use due to the lack of guidance on setting various parameters and the excessive time required for execution with MRI data. So, I implemented the method in IDL, generalized it up-to 7dimensional data, developed recommended parameter settings, and used a grid to build the “Mountain”. I used the method for a study into the variations and variability of skin color measurement devices, revealing significant variations, and inaccuracies in cheaper devices.

Adventures in Hiking and Data Tracking

Even in retirement, my affinity with data continues to inspire questions. I hike almost every day; and I hike a lot. Just for kicks, I keep track of where I go. My iPhone has three different tracking applications for my hikes, and surprisingly these apps often show significant differences in distance and climb measurements for the same hikes. Values may differ by as much as 14% in distance and 40% in climb. The discrepancy led me to question their accuracy and delve into the world of GPS tracking. By comparing my iPhone tracks with maps created by the United States Geological Survey (USGS), I’ve been able to uncover fascinating insights and complexities of GPS tracking the impact of environmental factors.

Figure 3: Ground-truth trail (white) and iPhone trace by walking both directions (red).

Figure 4: After a 21-mile hike in O'ahu.

The “Body Hypothesis” and GPS Accuracy

One hypothesis I explored was the effect of the iPhone’s position on my body on GPS accuracy. By switching sides of the pocket the iPhone was in while hiking the same trail in opposite directions, I observed a noticeable difference in the recorded tracks. This experiment suggested that the GPS signals traveling through the body can affect the accuracy of the measurements. My findings also challenged the common belief that single-frequency GPS has an accuracy of 30 meters, as my data indicated much lower noise levels.

The Experiment in Action

Ground Truth can be extracted from maps surveyed and created by USGS. Many trails have been mapped which allows a comparison with the tracks generated by iPhone apps. An example is: While the red iPhone traces roughly follow the actual trail, going North trace differs so much from the reverse direction. Moreover, the gap is not constant, and there are variations in the distance with the true trail.

GPS tracking is complex with incredibly small signals (differences in light travel times) from many moving satellites (possibly complemented with cell phone tower locations). In the case of hiking, the actual route is influenced by obstructions such as rocks and trees, ice and puddles, and “ground truth” can be elusive. Hence to develop a full understanding of the GPS tracks, much data and clever analysis is required, including interfacing with various applications. For my questions, IDL came to the rescue again, allowing quick programming to do file format conversions and visualize and quantify sources of variations.

Figure 5: Data from a short (415m) out-an-back on narrow trail.

One idea for the observation above is that, having one’s iPhone in a pocket, some GPS satellite signals need to travel through the body which has a different speed-of-light than space or air. This body hypothesis is easily tested: when going back on a trail, switch the side the iPhone is on to keep the GPS satellite locations the same with respect to the body and the phone. The gap should be less when using the opposite pocket comparing to keeping the phone in the same pocket. Using data from a short (415 m) out-an-back on a narrow trail gave the result show in Figure 5.

The tracks were recorded using a Fitbit app, and the gaps were calculated using geodetic distances (WGS 84 model). The effect is clearly visible in the graphs; and statistically highly significant (p < 10_-20). Also useful to note is that the single-frequency GPS is often said to have an accuracy of 30 meters, which is shown here is not random noise: gap distances indicate that noise levels are much, much less.

My curiosity pertained to the traces recorded while hiking, but to fully understand the behavior of the iPhone GPS, repeated experiments that are controlled, for example, for time of day because the GPS satellites orbit every 12 hours, exact locations (such as USGS markers or survey data), and iPhone position and orientation. One can also think about variations between iPhones, tracing apps, and even GPS hardware. The tools I could develop quickly in IDL, such as the file conversions, geodetic distance calculation, and trace matching can easily be applied to the analysis of any such experiment. The results could be used to improve the tracing apps and real-time locations, for example to give more timely warnings that one has strayed from an intended path.

Contributing to the Community

Beyond satisfying my own curiosity, my work with IDL allows me to contribute to my community. I’ve provided my township with maps of previously unmapped trails and local parks, which will soon be available online. It’s incredibly rewarding to see how my passion for data can have practical applications and benefit others. I hope my story inspires others to dive into the world of data and discover the endless possibilities it offers.