Hütt C., Bareth G.

University of Cologne, Institute of Geography, WG GIS & RS, Albertus-Magnus-Platz, 50923 Cologne, Germany

Abstract

Non-destructive monitoring of sward traits is of interest for grassland management. Remote sensing methods using sensors mounted on Unmanned Aerial Vehicles (UAVs) can provide timely and detailed information. Photogrammetric analysis of UAV-based image data can measure sward height which is used to estimate forage mass. In this contribution, we investigate the potential of 3D point clouds obtained with UAV-LiDAR in comparison to the established Structure from Motion and Multiview Stereopsis (SfM/MVS) analysis workflow based on UAV-derived image data to determine sward height. We (i) focussed on penetration depth in meadows of UAV-LiDAR, (ii) compared the results to SfM/MVS-derived sward height, and (iii) evaluated the results of the UAV approaches to RPM measurements.

Keywords: UAV, LiDAR, SfM, biomass, grass, forage mass, sward height

Introduction

The non-destructive estimation of sward height (SH) is beneficial for grassland management (Bareth, 2021). Sward height is known to provide a robust estimation of forage mass (Evans and Jones, 1958). Measurements of compressed SH using Rising Plate Meters (RPMs) is an established method to provide sward production rates (Sanderson et al, 2001). Bareth and Schellberg (2018) propose replacing RPM measurements with a remote sensing analysis workflow using UAV-derived images. By using Structure from Motion (SfM) and Multiview Stereopsis (MVS), multi-temporal Digital Surface Models (DSMs) can be created to provide SH. Bareth and Schellberg (2018) reported excellent performance of UAV-derived SH compared to manual RPM measurements (R² = 0.86) for six growth periods across three years. Similar results are reported by Viljanen et al. (2018), Grüner et al (2019) and Theau et al (2021). However, SFM/MVS data processing is computing-intensive, and several analysis steps are required. An improvement might be the direct measurement of SH using an active sensor, a UAV-mounted laserscanner (LiDAR). UAV-LiDARs are widely used in forestry to provide canopy height (Sankey et al, 2017) and are lately applied for crops and grasslands (Bates et al, 2021; Hütt et al., 2021; Maesano et al., 2020). The overall objective of this contribution is to investigate the performance of a UAV-LiDAR for forage mass estimation. Therefore, we (i) focused on penetration depth in meadows of UAV-LiDAR, (ii) compared the results to SfM/MVS-derived sward height, and (iii) evaluated the results of the UAV approaches with Rising Plate Meter (RPM) measurements.

Materials and methods

We conducted our study on a conventionally managed meadow field in the Wesermarsch, Northern Germany (53°24'05.3"N, 8°16'36.3"E). The farmer’s field is in the immediate neighbourhood of the Jade Bight, very close to the dike. The area is characterized by a temperate humid climate (Cfb) with a mean annual temperature of 10.0 °C and a mean annual precipitation of 833 mm (www.de.climate-data.org). The soils in the area are drained mires, and meadows and pastures are the typical land use for this dairy farm region. The investigated meadow is managed for three to four cuts for silage production annually. For data acquisition, we used a self-developed Real Time Kinematic (RTK-) RPM for precise georeferencing of the measurements of compressed SH. For UAV data acquisition, two different RTK systems were operated. (i) We conducted two UAV campaigns with a DJI Phantom 4 RTK (1” sensor, 20 MP) for RGB data acquisition on May 14^th and June 1^st 2021. (ii) A UAV-LiDAR was flown for one date, May 14^th, capturing 3D point clouds with a Riegl miniVUX-1UAV laserscanner mounted on a DJI Matrice 600 Pro. Agisoft Metashape and Esri ArcGIS pro were used to analyze the RGB image data. In Fig.1, the different systems are shown. As a result, we produced one Digital Surface Model (DSM) for each date while the DSM for June 1^st represents the ground model after the grass was cut. The difference between the two DSMs represents absolute sward height in cm (P4RTK-SH). The UAV-LiDAR data was analyzed with Lastools retrieving sward height directly from one flight on May 14^th (LiDAR-SH). Further regression analysis was performed in the R software.

Results and discussion

As mentioned before, numerous studies have shown that SfM/MVS analysis provides precise canopy data to derive SH data. Therefore, we compare via regression analysis the P4RTK-SH against the LiDAR-SH. The results are shown in Fig.2. The R² between the two SH-datasets is 0.62 but shows significantly higher SHs for the P4RTK-SH data than for the LiDAR-SH data over the complete data range. It appears that the UAV-LiDAR used was not able of penetrating the complete grass canopy to derive ground points for precise sward height estimation.

The potential of the two different methods to estimate forage mass based on deriving spatial SH data is presented in Fig.3. For this analysis, we plotted the results of the two methods against the manual RTK-RPM measurements. In Fig.3(a), the results of LiDAR-SH are shown, resulting in a moderate R² of 0.58. After the investigation of the penetration depth, this moderate performance was expected. In Fig.3(b), the performance of the P4RTK-SH is shown. As expected and documented by several studies (Bareth and Schellberg, 2018; Viljanen et al, 2018), the established SfM/MVS analysis workflow for UAV-derived RGB data results in a R² of 0.74 having a trendline inclination of approx. 1. In this study, the UAV-LiDAR does not perform as well as the established SfM-MVS analysis workflow to derive SH data for estimating forage mass. However, we investigated the UAV-LiDAR using only one date to derive absolute SH, while for the P4RTK-SH we used two dates, before and after the cut.

Conclusion

We conclude that penetration depth of UAV-LiDAR is limited in grassland canopies, and one acquisition date is not sufficient to derive absolute SH in cm. Therefore, we propose investigating UAV-LiDAR using two dates, before and after cut, to improve the method. Finally, we have to extend our analysis on a larger data set for multiple growths in a year and multiple years.

Acknowledgement

This research was conducted within the BMBF-funded GreenGrass project (Grant number 031B0734F)

References

Bareth G. (2021) Towards an informed grassland farming - Sensors, platforms and Algorithms. Grassland Sci. in Europe 26, 11-20. (www.europeangrassland.org/fileadmin/documents/Infos/Printed_Matter/Proceedings/EGF2021.pdf)

Bareth G. and Schellberg J. (2018) Replacing manual rising plate meter measurements with low-cost UAV-derived sward height data in grasslands for spatial monitoring. PFG J Photogramm. RS Geoinf. 86, 157–168.

Bates, J. S., Montzka, C., Schmidt, M., & Jonard, F. (2021). Estimating canopy density parameters time-series for winter wheat using UAS Mounted LiDAR. Remote Sensing, 13(4), 710.

Evans R.A. and Jones M.B. (1958) Plant height times ground cover versus clipped samples for estimating forage

production. Agronomy Journal 50 (9), 504-506

Grüner E., Wachendorf M. and Astor T. (2020) The potential of UAV-borne spectral and textural information for predicting aboveground biomass and N fixation in legume-grass mixtures. Plos one, 15 (6), e0234703.

Hütt C., Bolten A., Hohlmann B., Komainda M., Lussem U., Isselstein J., Bareth G. (2021) First results of applying UAV laser scanning to a cattle grazing experiment. Grassland Science in Europe 26, 135-137.

Maesano, M., Khoury, S., Nakhle, F., Firrincieli, A., Gay, A., Tauro, F., & Harfouche, A. (2020). UAV-Based LiDAR for High-Throughput Determination of Plant Height and Above-Ground Biomass of the Bioenergy Grass Arundo donax. Remote Sensing, 12(20), 3464.

Sanderson M.A., Rotz C.A., Fultz S.W., Rayburn E.B. (2001) Estimating forage mass with a commercial

capacitance meter, rising plate meter, and pasture ruler. Agronomy Journal 93(6), 1281–1286.

Sankey, T., Donager, J., McVay, J., & Sankey, J. B. (2017). UAV lidar and hyperspectral fusion for forest monitoring in the southwestern USA. Remote Sensing of Environment, 195, 30-43.

Théau, J., Lauzier-Hudon, É., Aubé, L., & Devillers, N. (2021). Estimation of forage biomass and vegetation cover in grasslands using UAV imagery. Plos one, 16(1), e0245784.

Viljanen N., Honkavaara E., Näsi R., Hakala T., Niemeläinen O. and Kaivosoja J. (2018) A novel machine learning method for estimating biomass of grass swards using a photogrammetric canopy height model, images and vegetation indices captured by a drone. Agriculture 8(5):70.