Advanced workflows for UAV-based crop height estimation using structure from motion (SfM) point clouds

Authors: CHATTERJEE, SUMANTRA - Texas A&M Agrilife; SAPKOTA, BALA - Texas A&M Agrilife; BAATH, GURJINDER - Texas A&M Agrilife; Flynn, Kyle; Smith, Douglas

Submitted to: Remote Sensing Applications: Society and Environment
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/11/2025
Publication Date: 12/12/2025
Citation: Chatterjee, S., Sapkota, B.R., Baath, G.S., Flynn, K.C., Smith, D.R. 2025. Advanced workflows for UAV-based crop height estimation using structure from motion (SfM) point clouds. Remote Sensing Applications: Society and Environment. https://doi.org/10.1016/j.rsase.2025.101828.
DOI: https://doi.org/10.1016/j.rsase.2025.101828

Interpretive Summary: Crop height is an important biophysical parameter, having close correlations with biomass, yield, etc. However, in-field measurements of crop heights can be labor intense and expensive. With recent advancements in unmanned aerial vehicle (UAV) technologies, UAV based remote sensing estimations of crop biophysical factors are becoming a popular alternative. This research focuses on developing workflows of estimating crop height using UAV generated point clouds. Though the most popular procedure is estimating surface elevation once in absence and presence of crops, this research focuses on eliminating necessity of the UAV flight in absence of crops. The experiment was conducted during the corn growing season across of 2022 and 2023 in Temple, TX. Three workflows were tested: (i) CHMB: where UAVs were flown in absence of plants, and the surface generated was used as the digital terrain model (DTM); (ii) CHMS: where a Sentinal -1A derived surface, from the dormant season, was used as the DTM; and (iii) CHML: where the DTM was created by fitting a 2.5 dimensional (2.5D) surface to a group of points selected by passing a low pass filter through a moving window. The results indicated that CHML was a far more efficient method, for estimating crop height, when compare to any other method. The R2 of the best fit models (when field observations of crop heights were fitted with CHML estimations) were consistently higher than (~0.90) than estimations made using CHMB or CHMS. The slopes of the best fit lines for CHML for all such analysis remained close to unity, indicating higher efficiencies among estimations. Thus, the CHML method has potential to be an efficient estimation of crop height in combination with only one UAV flight.

Technical Abstract: Crop height is an important factor in understanding plant growth and yield, but measuring it directly in the field can be difficult and expensive. With advancements in drone (UAV) technology, scientists are now using drones to estimate crop height more efficiently. This study focused on improving these methods by reducing the need for extra drone flights before planting. The research took place during the 2022 and 2023 corn seasons in Temple, TX, where three approaches were tested. The first method (CHMB) required a drone to fly before crops grew, creating a base map of the land. The second method (CHMS) used satellite data from the dormant season instead of a drone-generated map. The third method (CHML) created a base map by selecting points from drone data and applying a special filtering technique. The results showed that the CHML approach was the most efficient and accurate, consistently producing better crop height estimates than the other methods. The correlation between real measurements and drone predictions was strong, suggesting that CHML could be a reliable way to measure crop height with fewer drone flights, saving time and resources while maintaining accuracy.