Unmanned Aerial Vehicle (UAV)-Based Remote Sensing for Crop Phenotyping Sanaz Shafian 1, Nithya Rajan 1, Yeyin Shi 2, John Valasek 3 & Jeff Olsenholler 4 1 Dept. of Soil and Crop Sciences; 2 Dept. of Biological and Agricultural Engineering; 3 Dept. of Aerospace Engineering; 4 Dept. of Geography; Texas A&M University, College Station, TX
Agriculture Challenges World population projected to reach 9.3 Billion in 2050 Food production must increase by 70% by 2050, in spite of: Limited availability of arable lands Climate change impact Needs to increase and optimize agricultural production Future of agriculture = plant breeding and phenotyping
Phenotyping Challenges Comprehensive assessment of complex plant traits Growth Development Yield Traditional phenotyping Labor intensive Time consuming Late discovery High-throughput phenotyping Automated data collection Non-destructive Low cost
What is UAV? Air Vehicle Ground Control Station Antennas
Texas A&M UAVs Fixed-wing aircrafts Anaconda PrecisionHawk Lancaster Rotary-wing aircraft TurboAce X88
Fixed-wing aircrafts UAV used in the Study Anaconda (John Valasek) Anaconda fixed-wing Wingspan (m) 2.0 Maximum payload (kg) 4.0 Battery (mah number) 5,000 2 Endurance w/ payload (min) 20 (45 max) Airspeed (m/s) 15 Flying altitude in study (m) 120 (400 ft)* * Our COA permits 600 ft
Sentek GEMS multispectral camera Spectral sensitivity (nm) Blue: 399 501 Green: 502 604 Red: 558 672 NIR: 742 878 Frame rate (s/frame) 1.4 Shutter type Dynamic range Exposure time (ms) Global shutter 8-bit Fixed but varied from flight to flight Imaging sensor pixel resolution (megapixel) 1.2 (1280 960) Weight (g) 170 Ground sampling distance (cm) Typical flying altitude (m) Sensor 6.5 120
Payload Sentek GEMS multispectral camera system integrated: Global positioning system (GPS) Inertial measurement unit (IMU) Sensors onboard image storage capability
Winter Wheat ~ 4.6 acres 2340 Plots 1230 Varieties Study Site Bio-energy crops ~ 0.1 acres 36 Plots 12 Varieties
UAV Image Pre-processing Images were collected with 75% forward overlap and 60% side overlap: 422 total images each for RGB and NIR (<300 was useful) 1.9 GB of input imagery Pix4Dmapper software Image ortho-mosaicking 414 MB of output imagery
Raw Images June 10 June 18 June 28 July 23 February 26 March 26 April 9 April 26
UAV Image Post-processing ENVI image analysis software Atmospheric calibration Radiometric calibration Convert digital numbers (DN) to reflectance (ρ) Normalized difference vegetation index(ndvi)
Field Data Collection LAI measurements (LI-COR 2200C) Overhead photos (Canon camera) Dry biomass (bio-energy crops) No yield data
NDVI Maps February 26 March 26
NDVI Maps April 9 April 26
NDVI Maps June 10 June 18
NDVI & Standard Deviation Duster & TAM 401 have the lowest standard deviations Strongly stable in time and space Fannini has the highest standard deviations
NDVI vs. LAI NDVI is highly correlated with LAI NDVI saturates at LAI >2.5
NDVI vs. f c NDVI is highly sensitive to f c < 0.8 NDVI becomes less sensitive to f c >0.8
NDVI vs. Biomass Strong exponential relationships between NDVI and dry matter biomass (R² = 0.73)
UAVs provide for: Summary & Conclusions Good spatial resolution for small plot research Automated or semi-automated operation Cost effective use of technology Spatial-temporal correlation between NDVI vs. LAI and f c UAV-based NDVI is useful for crop growth monitoring and modeling UAV-based NDVI has potential for estimating dry biomass for bio-energy crops More phenological stages should be targeted for increasing the accuracy Future directions: Sensor-integration (multi-sensor analysis) for crop phenotyping
Acknowledgment Texas A&M AGRILIFE Resrach Amir Ibrahim Dr. Russell W. Jessup Bryan Simoneaux
Questions?