BRN 9-3 - Flipbook - Page 21
Predicting Landslides
Predicting land movement is
important to housing developers, to
insurance companies, and to the
average Joe. All have an interest in
knowing when the land they have an
interest in might suddenly slide
downhill, or when the hill above
might become their, or someone
elseÕs, new yard.
If you walk or drive through the Black
Range following a big rainstorm (our
deÞnition: anything over one
hundredth of an inch) you are likely to
see evidence that it can happen here.
Rocks, some large, may be in the
road; low water crossings may be
covered in rock and sand a foot or
more deep; a portion of the road may
be gone; and road cuts may have
calved onto the highway. Be glad you
were not here when a good part of
the mountain decided to relocate.
At the top of the following page there
is a description of a geologic unit,
IPme, which decided to relocate some
time during the Tertiary, ~35 million
years ago. The rock, of course, is
much older, being from the Carboniferous. It just decided to relocate
later in life. The geologic map detail
just below the IPme description is
from the lower left (southwestern)
corner of the Hillsboro geologic quadrangle map. It is a huge chuck of the
earth. It is more than a simple yard
replacement.
Given our history in this regard, Jon
Barnes performed some of his
geology masterÕs thesis work here.
Conceptualizing, and building, a
system of inexpensive motion
detectors which could be linked
together using a wireless neural
network, he was able to evaluate
earth movement across three axis.
Output from one of the ÒrunsÓ and
some of the equipment is shown at
the center of the following page.
Direct observation from systems like
this can provide (generally short timescale) warning of earth slippage and
report on that activity from remote
areas. They can provide some
warning and identify areas in need of
emergency resources much more
quickly than other systems.
They are typically put into place
where other methodologies have
identiÞed the potential for landslides.
In general, such predictions are based
on models of varying sophistication
and/or efÞcacy.
Liu et al. noted that ÒLandslide risk is
traditionally predicted by processbased models with detailed assessments or point-scale attribute-based
machine learning (ML) models with
Þrst- or second-order features (e.g.,
slope and curvature) as inputs.Ó (See
ÒThe value of terrain pattern, highresolution data and ensemble
modeling for landslide susceptibility
predictionÓ, Journal of Geophysical
Research: Machine Learning and
Computation, 25 September 2025.)
They hypothesized Òthat terrain
patterns might contain useful higherorder information that could be
extracted, via computer vision ML
models, to elevate prediction
performance beyond that achievable
with attribute-based models.Ó They
used a variety of input types,
including imagery, to train the model
and found that ÒModel performance
improved with Þner data resolution,
reaching optimal performance at
10-m resolution, and variables such as
rainfall, land cover, and soil moisture
had the greatest impact on landslide
susceptibility.Ó
Along these lines, Barnes described
his project as follows.
Resilient Arduino Sensor
Network for Slope
Monitoring: Ground Truth
Validation
The efÞcacy of AI models predicting
slope failure from satellite and aerial
imagery hinges entirely on accurate
ground truthing. This project
introduces a sub-$100 low-cost,
distributed sensor network built for
this Þeld validation; a complete sixnode deployment, including solar
power and batteries, is estimated to
cost roughly $500 total. Each node
uses an Arduino Nano and a HiLetgo
MPU-6050 6-axis module to precisely
measure tilt and subtle angle
changes, with all nodes connected
wirelessly via LoRa radios. Given
typical hillside conditions with light
tree cover, this six-node network is
capable of reliably covering a
hillslope area spanning approximately
20
2 to 3 square kilometers (0.8 to 1.2
square miles).
Communications and Range
The communication system relies on
REYAX RYLR998 LoRa 915 MHz
modules, which are fundamental to
establishing the robust Logical Tree
Topology. While effective range can
reach several kilometers in open air,
realistically, the system is designed to
achieve reliable ranges of 1 to 3
kilometers (0.6 to 2 miles) in hillside
environments with light tree cover,
though this drops to below 1
kilometer in dense forest. Crucially,
the Arduinos utilize standardized AT
Commands to manage the LoRa
modules, enabling the rapid
reconÞguration process that drives
the failover logic when a primary
communication link fails.
Flexible Deployment and Data
Redundancy
The six-node system offers ßexible
deployment for targeted monitoring.
It can be placed in a long line across
the top of a hillslope to monitor a
speciÞc fracture line, or arranged
going down the hillslope with two
sensors placed at the top, two in the
middle, and two at the bottom to
proÞle movement across different
elevations. The network's base
physical layout consists of a main
stem of three sequential nodes with a
single "leaf" sensor node branching
off each point, with the main hub at a
corner. For redundancy, all sensor
data is saved locally to micro SD cards
on each node, and the main hub saves
all incoming telemetry to its own
micro SD card, ensuring a complete,
redundant long-term record.
Logical Topology and
Resilience
The network's ultimate resilience is
achieved through its underlying
Logical Tree Topology and integrated
failover logic. Sensor nodes report
hierarchically, and if a primary
connection fails due to a hardware or
path issue, the logic automatically and
dynamically reconÞgures the node to
report to an alternate parent. A
tabletop prototype demonstrates this
resilience: simulating a slope
movement and an intentional
