NUPO Data Research

Science Center Objects

Data research activities at the NUPO focus on developing new processing techniques and workflows, ensuring sensor calibration of UAS mounted sensors, and utilizing the higher resolution UAS acquired data to generate novel and statistically better types of leading-edge geospatial products.


Graphic showing platform heights and associated data resolution

Meeting the challenges of the various mandated missions over the 500 million acres of surface land DOI manages requires access to remotely sensed data over vast lands, including areas that are remote and potentially dangerous to access. Satellite imagery has proven to be a vital resource through Landsat and other satellite-based missions, but it has limitations related to resolution and orbital dependencies. Sensors aboard manned aircraft can, and frequently do, provide large area coverage with high resolution imagery, but increasing costs and potential safety risks can limit their availability and utility. And although DOI scientists can obtain high accuracy in situ data based on ground measurements and sample collections, the scale of public lands can make this approach unrealistic.

The raw images acquired by sensors mounted on UAS, including full-motion video (FMV) and still frame photography, can produce a wide variety of high-resolution geospatial products for scientists to use in performing studies and various types of analyses. Combining built-in Global Navigation Satellite System (GNSS) data with the raw imagery provides an average geospatial accuracy of 1-3 meters, and if additional ground control data is added by ground control points (GCP) accuracies can be improved to centimeters. Low-cost UAS and compatible sensor payloads also provide a much safer data collection tool and bridge the data gap between expensive manned aircraft image collection and time-consuming on-the-ground efforts.


Raw image data acquired from sensors mounted on sUAS must be converted into orthophotos using a process called orthorectification before it can be used for geospatial analysis or as source data to produce geospatial products. Orthorectification removes the effects of topography (surface relief) and compensates for any sensor tilt or distortions in the raw data to produce a distortion free aerial photograph with a completely uniform scale called an orthophoto. An orthomosaic, frequently used as a base map, can then be created by combining a series of orthophotos into a seamless image.

Orthomosaic of the Northeast Indiana study site

Orthomosaic of the Northeast Indiana study site generated from Ricoh GR natural color images

Traditionally orthorectification required knowledge of the distortions associated with a specific camera lens system (i.e. sensor) to perform the high-quality calibration required to support the precise and accurate extraction of topography (e.g., terrain and 3D surfaces) and/or planimetric features (e.g., road centerlines, streams, or vegetation boundaries) from stereoscopic imagery. Today this can be performed by using the GPS data acquired during the sUAS flight to provide precise geolocation information to calibrate the imagery in structure from motion (SfM) software that applies a multi-view solution to empirically model the lens distortion for any radial ground lens (effectively generating a high-quality in-situ calibration). In other words, NUPO researchers can use the GPS information captured during a low-altitude sUAS flight with imagery acquired from a mounted low-cost digital single-lens reflex (SLR) or any sensor that uses a radial ground lens, to produce high-resolution orthophotos with ground sample distances of less than six inches. And while most digital SLR and point-and-shoot cameras capture natural color (RGB) images, one of the most common types of imagery used to create base maps, the low-cost sensors available today can readily acquire a variety of different data types including natural color, thermal, and multispectral.

Thermal orthophotos are generated from images taken by a thermal camera, such as the FLIR Vue Pro R, that captures non-contact temperature measurements of surfaces as photographs. Orthophotos generated from raw 16-bit radiometrically calibrated thermal imagery can result in a geospatial raster dataset where each pixel location has an associated absolute surface temperature. And if absolute temperature is not required, relative temperature orthophotos can be generated from histogram-stretched JPGs with various color palettes such as WhiteHot, BlackHot, etc. UAS acquired low-altitude thermal imagery generally produces products that provide ground sample distances of less than 15 cm and can be used to support wild fire monitoring, search and rescue, solar panel inspections, and water temperature monitoring.

Color infrared orthophotos are made from imagery acquired from visible and near infrared sensors that detect the red (near infrared edge of the electromagnetic spectrum) centered around 690-720 nm near infrared, green, and blue. Early research missions used a natural color camera modified with a notch filter that blocks the low to mid red-light range resulting in a sensor that detects the red (near infrared edge of the electromagnetic spectrum) centered around 710-740 nm, green, and blue as a low-cost method of capturing near-infrared imagery. Today a MicaSense RedEdge camera is frequently used to capture source imagery in the visible, red edge or near-infrared wavelength range. Orthophotos made from near infrared images are a valuable resource in vegetation analysis and support the generation of Normalized Difference Vegetation Index (NDVI).

Point Clouds and 3D Models

Point clouds are a set of geographic data points in a three-dimensional coordinate system that typically represented X, Y and Z and are an invaluable resource for a variety of geographic applications that evaluate and monitor landscape change. Point clouds vary from sparse to dense and can be derived by using structure from motion (SfM) techniques on aerial imagery or collected by Light Detection And Ranging (LiDAR) scanners. True-color point clouds can also be generated by processing standard imagery in SfM software or combining standard imagery with LiDAR data.

Closeup of the 3D model of Devils Tower in Wyoming

Closeup of the 3D model that was generated for Devils Tower in Wyoming

Highly accurate 3D or physical models can be generated by overlying natural color orthophotos or other types of georeferenced orthophotos onto the point cloud data. These realistic 3D models can be used to support computer simulations, display as a two-dimensional image via 3D rendering, or physical model creation by printing from a 3D printer.


Contours over the Piute Valley in southern California

Contours over the Piute Valley in southern California derived on a ground sample distance (GSD) of 1.4 inch

Contour lines (or contours) are a series of joined points of equal elevation (height) above a given level, such as height above mean sea level. Elevation values derived from an orthomosaic is an ideal data source for generating contour lines. The contour interval used when generating the lines represents the elevation difference between successive contours and can be illustrated as a contour map which are effective tools for terrain visualizations showing valleys and hills, and the steepness of slopes.

Elevation Models (DEMs, DSMs, DTMs)

A digital elevation model (DEM) is a digital dataset of bare surface elevations (z) at horizontal (x, y) coordinates and can be accurately derived from ground positions sampled at regularly spaced horizontal intervals by standard commercial off-the-shelf cameras mounted on a UAS. Once generated these high-resolution DEMs can also be used as a low-cost option for generating accurate volumetric measurements.

Digital Surface Model of the West Fork Mine in Missouri

Digital Surface Model of the West Fork Mine in Missouri generated from high-resolution imagery (5-10 cm pixel size) and elevation data (6-10 cm vertical and 2-4 cm horizontal resolution)

Digital surface models (DSMs) are a form of DEM that contains reflective surface elevations of natural terrain features in addition to vegetation and cultural features such as buildings and roads. Point clouds generated from UAS-mounted LiDAR sensors, which calculate elevation values from both the tops of surfaces and the bare ground, can be used to generate highly accurate DSMs. LiDAR point cloud data can also be used to generate digital terrain models (DTMs) by removing the elevation signals of features such as vegetation and buildings, leaving only the elevation of the terrain or ground.

Extracted Features

Extracted bird locations at the Chase Lake National Wildlife Refuge

Extracted bird locations at the Chase Lake National Wildlife Refuge; pelican nests (red), cormorant nests (blue), gull/snowy egret non-nesting (black)

Feature extraction automates the process of recognizing spatial and spectral patterns within an image and outlining or classifying those features into a newly defined dataset. A high-resolution orthomosaic generated from imagery collected on low-altitude UAS flights, provides an ideal method for accurately identifying small-scale (~1 m) to larger-scale features. Features may then be measured in area, length or count with feature counts providing a valuable tool to support species identification, population studies, and other aspects of resource management.

Normalized Difference Vegetation Index (NDVI)

NDVI over the Sycan River in the Klamath Basin in Oregon

Normalized Difference Vegetation Index (NDVI) derived from an orthomosaic of near infrared imagery taken from approximately 400 feet AGL over the Sycan River in the Klamath Basin in Oregon

Orthomosaics made from multispectral imagery with bands in the red and near infrared range are a valuable resource in vegetation analysis and support the generation of Normalized Difference Vegetation Index (NDVI) maps. Normalized Difference Vegetation Index (NDVI) calculations processed against near infrared data creates a standardized index utilizing the amount of infrared light reflected from a plant. The ratio between reflected infrared light to reflected red light has a strong correlation to the health of the plants imaged where values closer to 1 are healthy vegetation, and values closer to -1 are soils. In other words, the bright red display of the color ramp indicates healthy or highly reflective plants and the blue color indicates the lower reflectivity and possibly less healthy vegetation. Other spectral indices can be calculated using orthomosaics for various applications, such as the Normalized Burn Ratio for assessing burn severity.


NUPO data processing techniques can be found at Other Resources