Estimation of oil palm’s leaf area index (LAI) using Unmanned Aerial Vehicle (UAV) images

Leaf Area Index (LAI) is a parameter commonly used to indicate oil palm growth and production. The destructive method is the standard method to estimate LAI. However, this method requires much effort and cost and can only be done at a specific period. Unmanned Aerial Vehicles (UAVs), which have been widely used for mapping and calculating trees in oil palm plantations, have the potential to estimate LAI values quickly, efficiently, and without disturbing the oil palm tree. This research was conducted to develop the model for predicting LAI based on UAV images. Data was collected from six plots of 12-year-old oil palm trees at Adolina Estate, Serdang Bedagai, North Sumatra, Indonesia. The research was conducted on six varieties released by IOPRI, consisting of DxP Yangambi, DxP Langkat, DxP Simalungun, DyP Dumpy, DxP LaMe, and DxP PPKS 540. The estimated canopy cover from the UAV images was employed as an independent factor (x) compared to the calculated LAI from the destructive method, which was used as a dependent factor (y). The results showed that the LAI estimation using UAV imagery followed a linear model with R-squared values ranging from 0.3874-0.9556. In conclusion, despite requiring further research, UAV images could be used as rapid tools to estimate oil palm LAI.


Introduction
The leaf area index, or LAI, is a measure of plant development that depicts the leaf area and the area covered by the plant canopy [1].In oil palm plantations, LAI is influenced by the number of fronds, leaf area, and palm trees per hectare [2,3].LAI is a crucial metric that affects crop respiration, transpiration, and photosynthesis.It is also a helpful indicator for determining yield levels [4].The amount of solar radiation that can be absorbed for photosynthesis depends on the canopy's overall area, which affects oil palm yield [5].The response of crops, particularly crop yield, to environmental circumstances can also be detected using LAI dynamics measurements [6,7].
LAI is determined by dividing the plant canopy's area by the land's total area [8].In oil palm, the mean area of each frond, the number of fronds per palm, and the planting density significantly impact LAI.Direct and non-direct methods for estimating LAI can be distinguished [9][10][11].The LI-3000C Portable Leaf Area Meter, LI-3100C Area Meter, and CI-202 Portable Laser Leaf Area Meter are equipment that can be used to manually measure leaf samples as part of the direct approach for determining LAI.The oil palm's frond will be destroyed by the direct method.It takes a lot of time and effort [11].Additionally, the equipment mentioned above is rather pricey, and not all oil palm plantations can afford it.
The indirect technique calculates LAI using additional factors like light transmission and contact numbers.This technique is more labor and time efficient because it does not harm plants and can be applied broadly.An example of an indirect LAI measurement technique is the allometric approach.LAI-2000, Tracing Radiation and Architecture of Canopies [TRAC], and Digital Hemispherical Photographs are three optical techniques that can quantify LAI indirectly [11].Several remote sensing technology studies have also been carried out to estimate LAI non-destructively quickly [6].Multispectral and hyperspectral images can be further processed to produce vegetative indices (VIs), which are proven qualified to estimate the LAI value of a particular vegetation [12][13][14].However, weather conditions, especially cloud cover, are a limiting factor in using satellite imagery to estimate LAI values, especially in oil palm plantations.
Unmanned Aerial Vehicles (UAVs) have great potential for monitoring LAI growth, especially in oil palm plantations in Indonesia.This technology has now been widely adopted by oil palm plantations, especially for mapping and oil palm tree counting.Furthermore, this technology can be maximized as the primary tool for monitoring plant growth, especially LAI [15].However, research on using UAVs to estimate LAI values in oil palm plantations still needs to be completed.
This research was conducted to calculate the LAI of oil palm plantations based on UAV images.Then, it was compared with the destructive manual (direct method).Furthermore, the results of comparing the two methods were used to develop an equation for calculating LAI for oil palm plantations based on UAVs.

Study Site
The study was carried out between October 2022 and March 2023.This study was conducted in the Adolina Estate in the Serdang Bedagai Regency of North Sumatra, Indonesia.The soil is composed of sandy clay loam.The daily relative humidity ranged from 73% to 95%, the daily solar radiation was 15 MJm -2 day -1 , and the windspeed was between 0-4.5 ms -1 .The average air temperature ranged from 23 to 34 o C. Adolina Estate has tropical rains.The peak rainy season often falls between April and October.However, the first three months' average rainfall was typically modest.

Research design
Six varieties plots were observed in this study, namely DyP Dumpy, DxP PPKS 540, DxP Langkat, DxP Simalungun, DxP Lame, and DxP Yangambi.Each variety has a specific morphological performance (see Table 1).All plots were planted in 2010.The six plots were grown in the same environmental conditions and applied relatively similar agronomic practices.The total trees at each plot were 101, 147, 124, 152, 152, and 155, respectively.The number of sample for manual LAI measurement in each plot was five trees.The selected trees are healthy and represent the morphological performance of the oil palm trees in the plot.
LAI was manually calculated by taking the 17th frond on each sample tree.Frond number 17 was chosen because of its location in the middle of the oil palm canopy and had stable nutritional content [17].Furthermore, measurements were made on the number of leaflets, the leaflet's width and length, the number of fronds, and the number of trees per hectare.Manual calculation of LAI followed the equation below: 3 LA estimated from the equation below: where nf is the number of fronds, nt is the number of trees, LA is the total leaf area, b is the correction factor (oil palm age of 1-2 years = 0.512; 4-7 years = 0.529, >8 years = 0.573), nl is the number of leaflets, l is the average length of leaflets, w is the average width leaflets.Parallelly with manual LAI calculation, DJI Phantom 4 Pro captured images of the oil palm canopy.The flight path was created using the Drone Deploy before image acquisition.The flying altitude was 150 meters above sea level, and the image was captured every 25 m.The total aerial photos obtained were 224 photos from the front and side overlap of 80%.

Data analysis
The flowchart of data analysis in this study is presented in Figure 1.The data used were destructive sampling and aerial photo data from the drone.Calculating LAI based on the destructive method was based on equation no. 1. Agisof Photoscan and ArcGIS processed drone images to obtain a canopy coverage area for each sample.Canopy cover refers to the extent of tree or shrub coverage observed from an aerial perspective, encompassing the leaves, branches, and trunk [18].Therefore, the canopy cover in this study was presented in m 2 area units.The image data processing began by combining 224 aerial drone photos.Next, the drone's images were aligned and built densely with medium quality, followed by building mesh, creating texture and orthomosaic.The results of the images merger were then converted in *tif file extension to be processed in ArcGIS (Figure 2).The oil palm canopy images were digitized for each sample tree (Figure 3).This digitization was used to create polygons and determine the canopy coverage area of each sample.The canopy coverage was used as the basis for linear regression to determine the LAI equation for oil palm based on the UAV image.The estimated canopy cover from the UAV images was employed as an independent factor (x) compared to the calculated LAI from the destructive method, which was used as a dependent factor (y).If there are outlier data, the data was removed to obtain a better R-squared value.The minimum data used in the regression analysis is 3 data.

Results and discussions
In general, there was a positive correlation between canopy coverage and leaf area index.It was relatively similar to previous research [19].Canopy cover and LAI provide information about the density and extent of vegetation in a specific area.Areas with greater canopy coverage tend to have a higher leaf area index, indicating a denser and more extensive vegetation cover.Conversely, areas with lower canopy coverage generally have a lower leaf area index, suggesting less dense vegetation.
Canopy Coverage (m 2 ) The results of linear regression analysis between canopy coverage based on UAV images and LAI based on destructive methods are presented in Figure 4.The R-squared values vary from 0.3874 to 0.9556.The highest R-squared value was obtained in the DxP Simalungun variety, while the lowest was observed in the DxP Langkat variety.Previous research on wheat has mentioned that canopy structure dramatically influences the LAI estimation model [20].Regarding this study, the variation in R-squared values is closely related to the morphological characteristics of each variety's canopy.In this study, varieties with more compact canopies, namely DxP Langkat and DxP LaMe, were more Leaf Area Index challenging to model.However, the variety with excessively long fronds, DyP Dumpy, also has a lower R-squared value.Furthermore, the low R-squared value of several varieties is also caused by the lack of variance in the morphology of the oil palm canopy.Data collection at different oil palm ages and locations is also needed to improve model performance.Therefore, further research is needed to develop a more accurate model for each variety.

Conclusion
The calculation of canopy cover based on UAV images is directly proportional to the calculation of LAI using the destructive method.Among the six tested varieties, the canopy cover-based LAI models derived from UAV images with high R-squared values (>0.75) are DxP Yangambi, DxP PPKS 540, and DxP Simalungun.On the other hand, the varieties DxP LaMe and DxP Langkat, which have more compact canopies, as well as DyP, which has a broader canopy, tend to be more challenging to model.Based on this research, despite the need for further research, UAV images could be used as rapid tools to estimate oil palm LAI.

Figure 2 .
Figure 2. The UAV's image of all plots after merged.

Figure 3 .
Figure 3. Results of canopy digitation.The greenish yellow color is the DxP Langkat variety, the golden yellow color is the DyP Dumpy variety, the dark green is the DxP Simalungun variety, the light green is the DxP LaMe variety, the orange is the DxP PPKS 540 variety, and the red is the DxP Yangambi variety.

Figure 4 .
Figure 4.The results of regression analysis between UAV-based canopy cover and destructively measured leaf area index were obtained for the varieties DxP Yangambi, DxP LaMe, DxP Langkat, DyP Dumpy, DxP PPKS 540, and DxP Simalungun.In the case of the DxP Langkat variety, two outlier data points were removed, while for DyP Dumpy, one outlier data point was removed.Additionally, for DxP PPKS 540, two outlier data points were removed, and for DxP Simalungun, one outlier data point was removed.