**1. Introduction**

Phenotyping of crops is important to estimate biomass and the potential yield of new varieties of agricultural crops. Due to the increasing need to increase food production and improve the associated quality, it is important to optimize the yield, for which accurate estimation of biomass during the growing season is needed. In this context, phenotyping focuses on the characterization of morphological as well as physiological crop traits. Morphological parameters, such as plant height, stem diameter, leaf area or leaf area index (LAI), leaf angle, stalk length, and in-plant space [1], can be determined with LiDAR (light detection and ranging). Research on phenotyping using LiDAR often focusses on one specific crop, for example, wheat [2,3] or cotton [4].

Phenotyping of individual plants can be done in very high detail with LiDAR, analyzing complex phenotypical properties, such as leaf area, leaf width, and leaf angle [5,6]. For this, plants were put on a slowly turning platform while a fixed LiDAR instrument was used. A drawback of this setup is the low throughput of the scanning system while there is also a need to evaluate phenotypes under field conditions.

For high-throughput phenotyping, traits such as plant-height, LAI, and leaf cover fraction are determined directly in the field, using a LiDAR-based system mounted on a vehicle or RGB cameras mounted on a UAV (unmanned aerial vehicle) [3,4,7]. Tractor based LiDAR systems data have shown good correlation with in situ field measurements of plant height. Sun et al. [4] published an R2 of 0.98 for cotton plants, [2] published an R2 of 0.99 for wheat, and [7] showed an R2 of 0.90 for wheat. These studies show the capability of LiDAR to measure basic phenotypes such as plant height. However, LiDAR systems mounted on tractors can be unsuitable for labour-intensive crops such as rice or in orchards [8], for example, due to compaction of the soil [9]. A UAV equipped with a LiDAR system can overcome those limitations.

Earlier research on the relation between plant height and biomass was based on varying approaches for plant height measurements. Madec et al. [7] found an R<sup>2</sup> of 0.88 for the correlation between plant height and field-measured biomass, using a tractor based LiDAR system. Bendig et al. [10] used a structure from a motion (SfM) technique on UAV acquired imagery to derive plant height and found an R2 of 0.81 between field-measured height and SfM derived height.

In the last few years, LiDAR systems have been miniaturized, resulting in lower weights and reduced dimensions, and as a result, can be operated from UAVs. This development opens the way towards high throughput derived, more complex products like biomass and yield, thus, improving the speed and frequency at which these plant traits can be acquired in the field in an undisturbed way.

LiDAR-based biomass estimates of agricultural crops can be derived in different ways. Based on the Lambert-Beer LAI model of [11], the authors of [2] developed a biomass prediction model called the 3-Dimensional Profile Index (3DPI). Where the LIDAR 3D point cloud is divided into layers and for each layer, the fraction of points divided by the total amount of points is calculated. These layers are then summed, and the 3DPI values can be related to biomass with a linear function. As follow up [2] also proposed a voxel-based method (3DVI) to estimate the biomass of wheat. The 3DVI method divides the LIDAR 3D point cloud in voxels of equal size and calculates the ratio between the number of voxels containing points and the number of subdivisions in the horizontal plane. They showed that 3DVI could estimate wheat biomass accurately with an R<sup>2</sup> of 0.91, and for 3DPI, an R<sup>2</sup> of 0.93 was achieved. Jimenez et al. [2] used a tractor based LiDAR system. An alternative approach using airborne LiDAR was proposed by [12], who used Pearson's correlation analysis and structural equation modelling (SEM) to estimate plant height and LAI, which proved to be the best predictors of the biomass of maize (R<sup>2</sup> of 0.87).

The goal of this study was to investigate the potential of UAV-LiDAR for estimation of crop height and fresh weight biomass for three different agricultural crops. For this, the RIEGL RiCOPTER with a VUX-SYS LiDAR system was flown over fields with sugar beet, wheat, and potatoes, on a number of moments during the growing season. First, it was investigated how accurate this system can estimate crop height and biomass. Next, we answered the question if it is possible to create models that are generally applicable for different crops. Finally, we did experiments with different UAV flight patterns, altitudes, and speeds to investigate if this influences the LiDAR-derived plant height and biomass estimation. Optimization of the flight parameters was beyond the scope of this paper.

#### **2. Materials and Methods**

#### *2.1. Study Area*

Three fields with different crops, each covering approximately 2 ha, were selected at the experimental farm of Wageningen University, located just north of Wageningen in the Netherlands (Figure 1). The three types of arable crops in this study are sugar beet, winter wheat, and potatoes (Table 1), which were planted on a Placic Podzol [13]. The variation in elevation in the fields is very small with a height difference of less than 10 cm going from east to west in the area. Orientation of the main crop rows for all three crops is north–south. The plant density of winter wheat was 260 plants/m2. Planting distance for potato and sugar beet was 30 cm and 23 cm, respectively.

**Figure 1.** Left: Overview of fields within the study area: indicated are the three different fields and crops that are studied. The image was taken on 4 June 2018. The large inset shows the location of Wageningen in the Netherlands. Right: Impression of the three different crops used in this research. Potato and sugar beet images were taken on 5 June 2018. The winter wheat image was taken on 25 June 2018.

**Table 1.** General information about the three crops planted in the study area.

