GEOSPATIALLY ENABLED AIRBORNE RAPID RESPONSE SYSTEMS: DESIGN, CALIBRATION AND PERFORMANCE ANALYSIS

The National Oceanic and Atmospheric Administration (NOAA) has been responsible for enabling The United States rapid responders during the hurricane season. Over decades of experience in enabling Rapid Response nationwide resulted in NOAA identifying the detailed engineering requirements for designing the next generation geospatially enabled airborne rapid response systems for aerial survey image capture, georeferencing, and high precision mapping. This paper is focused on presenting the state of the art, new trends, and the performance analysis of NOAA’s DSS version 6 for Rapid Response which development and testing took place in 2020/2021.


INTRODUCTION
For several decades, the National Oceanic and Atmospheric Administration (NOAA) has assisted with recovery efforts following a variety of natural disasters, including earthquakes, tsunamis, and hurricanes. In the early days, NOAA used aerial film cameras to survey areas affected by these events, which meant key information about the extent of damage to structures and the environment would not be known for days. Since then, NOAA has been researching and utilizing emerging technologies to produce geospatially accurate information as quick as possible to better enable rapid repose efforts (c.f., Brennan and Lander, 1991;White and Aslaksen, 2006).
As a result of these efforts, NOAA pioneered the use of fully digital directly georeferenced aerial surveys for Rapid Response (c.f., White and Aslaksen, 2006).

NOAA'S RAPID RESPONSE SYSTEM DESCRIPTION
From their years of experience in collecting and providing Rapid Response imagery nationwide and performing aerial surveys of the national shoreline, NOAA identified the detailed engineering requirements for designing the next generation geospatially enabled airborne mapping system that could be used optimally for collecting both rapid response imagery and traditional imagery for shoreline mapping.
The system design, development, and testing took place during 2020 and 2021 which was logistically challenging due to many imposed restrictions caused by the Covid-19 pandemic.
NOAA's DSS V6 system design includes: two PhaseOne RGB 150 Megapixel cameras, two PhaseOne NIR 100 Megapixel cameras, an Applanix POS AV 610 direct georeferencing system, and a flight management system. The system comprises two camera pairs with one RGB and one NIR cameras in each pair.
A unique and novel feature of the system is that each camera pair can be moved by the operator into a nadir or an oblique position during flight. For rapid response missions, both pairs are flown in oblique mode enabling a very large footprint to be collected for efficiency, and with large viewing angle to capture the sides of buildings to better view damage. For traditional shoreline mapping projects, either pair of the cameras is moved to the nadir mode to collect 4 band stereo imagery.
This unique feature of the DSS V6 means a single camera system can be utilized to meet all of NOAA's imagery mandates thus greatly increasing efficiency. Figure 1 shows the system aircraft installation onboard NOAA's King Air Aircraft as well as system operation during airborne surveys.  Figure 2 shows the multi-camera part of NOAA's system installed onboard NOAA's aircraft. Figure 3 shows the cameras during operation. The arrow shown in Figure 3 points in the flight direction. Additionally, Figure 3 shows that the first camera pair in the flight direction (shown on the left-hand side of Figure 3) was adjusted to the nadir position while the second camera pair was adjusted to the oblique position by the operator. Figure 4 shows NOAA's online Rapid Response Portal which is accessible by rapid responders around the entire nation. Figure 4 shows hurricane IDA imagery as an example. The oblique positions for both camera pairs are tilted by 22.5 degrees from the vertical axis. This oblique angle has been one of the parameters that NOAA carefully considered during the system design phase to accommodate the requirements for both Rapid Response and Coastal Zone mapping. Table 1 lists the RGB camera dimensions while Table 2 lists the NIR camera dimensions. Table 3 lists the Trimble Applanix POS AV 610 accuracy specifications.
Please note that the photogrammetric geometry of the RGB camera when coupled with a 60% overlap allows for an excellent Base-To-Height ration suitable for high precision mapping using POS AV 610 in a pure Direct Georeferencing mode.    The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLIII-B1-2022 XXIV ISPRS Congress (2022 edition), 6-11 June 2022, Nice, France

DIRECT GEOREFERENCING
Direct Georeferencing (DG) is the process of computing the ground coordinates for any ground point that appears in an airborne image using inertial/GNSS integrated systems.
DG is a scientific concept that is applicable to any airborne or terrestrial camera once integrated with a GNSS/Inertial system. Schwarz et al (1993) pioneered DG in the Academia for airborne and terrestrial applications. Successful academic research was published in the 1990s when DG was integrated with many airborne sensors including film cameras (c.f., Škaloud, et al, 1996) or digital cameras (c.f., Mostafa and Schwarz, 2000).
In the commercial environment, Reid, et al (1998) introduced the first professional-grade commercial product to the airborne mapping profession using an IMU-integrated DGPS (c.f., Hutton et al, 1997). Mostafa (2003) presented the design, development, and performance of the DSS. Mostafa and Hutton (2005) demonstrated the ability of the DSS to operate in a pure DG mode. Ip and Mostafa (2006) demonstrated that DSS aerial surveys for rapid response operating in a pure DG mode produce near realtime high-quality mapping products, especially when data processing workflows properly addressed the balance between the desired geospatial data resolution, accuracy, and processing time.
NOAA has been using the DSS for Rapid response in a pure Direct Georeferencing mode for over a decade. On the other hand, NOAA's scientists and engineers designed and implemented a seamless and sophisticated data processing workflow that integrates the raw imagery, raw inertial-GNSS data, and the national elevation models to produce high precision orthophotos in near real time without aerial triangulation or ground control.
NOAA's workflow integrates a number of industry standard software packages, open-source software utilities and homemade software tools. The above has been streamlined in order to allow for mass production of high precision maps in near real time. NOAA devised the workflow necessary to do nadir/oblique image integration with inertial-GNSS data while incorporating national elevation models to produce quality orthomosaics without ground control in a pure direct georeferencing environment as shown in Figure 5.

CALIBRATION, TESTING, AND ACCURACY ASSESSMENT
During system development, a number of test flights were flown over two of NOAA's test fields in Lakeland, and Gainesville, Florida for system testing, characterisation, validation, performance analysis, and accuracy assessment purposes.
High accuracy ground control was established in NOAA's test fields to be used as check points for independent accuracy checks at different steps of NOAA's processing workflow.
One of the processing workflow aspects is system calibration including boresight, camera, and lever arms. A number of requirements were devised for accurate system calibration during data acquisition and processing including the flight pattern, and the calibration processing mechanism.
Dual-Altitude flights have been proven successful in providing high precision simultaneous camera and boresight calibration (c.f., Mostafa, 2003, Ip et al, 2006and Casella et al, 2006.) This is due to the fact that a dual-altitude flight containing N-S and E-W flight line establish a strong 3D geometrical network that enables overcoming the mathematical correlation betwee boresight and camera interior orientation.
As a result, a number of dual-altitude flights were flown for system calibration and performance analysis purposes over NOAA's test fields in Lakeland and Gainesville, Florida. One of these flights (flown on August 5 th , 2021) is shown in Figure 6.   Figure 7 shows the positioning accuracy while Figure 8 shows the orientation accuracy. Figure 9 shows the forward/reverse separation which is a recommended QC plot to inspect in POSPac to assess the difference between the trajectory processed forward versus backward in time. The consistency between the forward and reverse solutions indicates that the absolute accuracy of the trajectory is in balance with the RMS of the forward/reverse separation.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLIII-B1-2022 XXIV ISPRS Congress (2022 edition), 6-11 June 2022, Nice, France In this flight, for example, the forward/reverse separation RMS is about 1 cm for easting and northing and about 2 cm in height, while the maximum error occurred at the height component of 5.5 cm as shown in Figure 9, which is within the RMS system specifications listed for in Table 3. These results are typical nowadays with PP-RTX and the Trimble survey grade GNSS systems that include all satellites and frequencies from GPS, GLONASS, BeiDuo, and Galileo satellite positioning and navigation systems.  Figure 10 shows the selected flight lines used to calibrate the system from the Aug 5 th , 2021 test flight data.
A total of twenty four check points were established by NOAA in its test field in Lakeland, Florida represented by triangles in Figure 10).
These check points were used to assess the system accuracy during system calibration.  Since the system has four cameras and one IMU, it was necessary to assess the rigidity of mounting the cameras and the IMU.
This is due to the fact that the boresight angles between each camera and the IMU are always assumed constant during data processing. Therefore, making sure that boresight does not change between flights was one of the testing parameters.
In order to assess the boresight calibration stability, and repeatability from different flights, three flights were used. These flights were flown over the same test field in Lakeland, Florida on August 5 th , 6 th , and 13 th , 2021, respectively.
The calibrated values of the three boresight components Tx, Ty, and Tz for the three flights are sown in Figure 12, Figure 13, and Figure 14, respectively for the second camera pair RGB2/NIR2, as an example. The variation of the boresight angles in the three flights has been computed and it is presented in the form of a standard deviation for each boresight component for both RGB2 and NIR2 cameras (of the second camera pair). The computed standard deviations are listed in Table 4.
These results demonstrate that the system boresight calibration computed from three different flights is consistent and repeatable with minimal differences (noise) that is well within the POS AV 610 system specifications.

SYSTEM ACCURACY ASSESSMENT
NOAA's DSS V6 is used in two different applications; namely: Coastal Zone Mapping and Rapid Response. The DSS V6 accuracy requirements are higher for coastal zone mapping than Rapid Response since rapid response data typically has the balance of resolution, accuracy, and the required time to generate useful mapping products.
In order to assess the system accuracy, a number of parameters needs to be taken into consideration, including: • System accuracy for each of the four cameras • System accuracy for each camera position (nadir versus oblique) • Accuracy repeatability for different flights • System absolute versus relative accuracy Therefore, a number of flights were used to assess the abovementioned parameters. These flights were planned in such a way, that the system was set to the nadir position. Some other flights were flown while the system was set to the oblique position. Additionally, in some flights, the system operator moved the system from the nadir to the oblique position and then back to the nadir position during flight.
The same 24 check points shown in Figure 10 were used to assess the system accuracy in a number of flights. CalQC was used for this accuracy assessment exercise. The check point accuracy is shown in Figure 15 which demonstrates that the X and Y check point RMS for all flights is about one pixel on average, and that the Z check point RMS ranges from 2 to 4 pixels. In each data set, one of the check points was turned into a ground control during processing in order to calibrate a global datum shift and remove it from the data. NOAA's Rapid Response DSS V6 system is designed to be used to survey disaster areas in any imaging configuration including single or multiple flight lines to cover the desired area. Therefore, the accuracy assessment was done again using only one image strip at a time for each flight. Figure 16 shows the RMS of check points for all individual single image strips in all of the selected flights in both nadir and oblique camera positions. The results are similar to those when using image blocks to assess accuracy.
NOAA's Rapid Response ortho generation workflow uses pure directly georeferenced imagery together with the national elevation model and camera models as shown in Figure 17.
The workflow starts with importing the POS AV 610 data together with the calibrated boresight angles and the measured lever arms in POSPac MMS. Upon processing the POS AV 610 data, an Exterior Orientation (EO) file is directly generated without the need for a base station which is then directly imported in NOAA's Rapid Response ortho generation workflow along with the camera model and the national elevation data in order to directly produce orthomosaics.
No Aerial Triangulation is done in the process. No ground control is used in the process. This process flow guarantees utmost efficiency and the highest productivity for rapid response. In other words, NOAA's ortho generation workflow uses the latest multi-sensor system technology to generate high quality mapping-grade orthomosaics in near real-time without Aerial Triangulation or ground control. Using NOAA's Rapid Response ortho generation workflow shown Figure 17, the orthomosaics were generated for a number of flights. The accuracy of these orthomosaics was assessed against the same 24 check point in Lakeland, Florida shown in Figure 10. In many high precision airborne mapping applications, professional mappers prefer to extract a Digital Surface Model (DSM) using the imagery in order to produce the highest possible accuracy DSM (that typically leverages the existing dense tie points) for ortho generation. However, NOAA's Rapid Response ortho generation software processing workflow uses the national elevation model in order to save the DSM extraction time since time is of an essence in Rapid Response.

SUMMARY AND CONCLUSIONS
NOAA's next generation geospatially enabled airborne rapid response system is presented in this paper. NOAA's DSS V6 system design, development, calibration, and testing took place during 2020 and 2021.
The system design includes: two PhaseOne RGB 150 Megapixel cameras, two PhaseOne NIR 100 Megapixel cameras, an Applanix POS AV 610 direct georeferencing system, and a flight management system. The system comprises two camera pairs with one RGB and one NIR cameras in each pair. Each camera pair can be set to either a nadir or an oblique position by the operator at any time during flight.
Several test flights were flown over NOAA's Lakeland and Gainesville, Florida test fields. Both test fields have a wellestablished network of ground control that were used as check points for accuracy analysis purposes.
Many test flights have been used to calibrate and characterise the system as well as to assess its mapping geometric accuracy. Boresight calibration using the dual-altitude flights including N-S and E-W flight lines was done for multiple flights. The results of three of these test flights are presented in this paper. The boresight calibration accuracy proved to be accurate enough for high precision mapping purposes. Comparing the boresight results for the three flights used here showed that their differences are well within the POS AV 610 system accuracy specifications.
Accuracy assessment took place for a number of imaging configurations including image blocks and single strips. The final check point accuracy consistently showed a one-pixel accuracy for X & Y and 2 -4 pixels for Z in both cases where imagery was used in a block configuration or in a single image strip.
This comparison confirmed that the use of NOAA's DSS V6 for rapid response using any imaging configuration will result in a consistent accuracy when flying one or more image strips to survey a disaster area. NOAA's Rapid Response ortho generation workflow has been used to generate orthomosaics in a pure direct georeferencing environment without Aerial Triangulation or Ground Control to generate an Orthomosaic of the full flight flown on August 5 th , 2021.
The ground coordinates of all check points that appeared in the orthomosaic were computed form the airborne data and then compared to the check point land-surveyed coordinates.