Close Menu

Student Research: Evaluating LAI Estimation Using Time-Lapse Game Camera Images

Megan Smith and Grant Casady 
Whitworth University Department of Biology, Spokane


Seasonal fluctuation in vegetation can be described by phenological metrics, such as start of season, peak of season, and end of season. Previously, canopy leaf area has been measured using leaf area index (LAI) meters (Korhonen and Heikkinen 2009). Using LAI meters to characterize phenology over time can be very costly and time-consuming. Instead, digital cameras could theoretically be used to take photos of the forest canopy and measure changes in closure. Before using cameras in such a way, the relationship between leaf area estimates and camera images must be established and a protocol for taking and processing photos must be developed. This study attempts to address these issues with the following objectives:

  1. To determine whether there is a significant relationship between LI-COR LAI 2200 measurements and digital camera images of the forest canopy
  2. To develop a methodology for acquiring and processing canopy images

With regard to these objectives, we predicted that a significant relationship exists between LAI measurements and canopy closure estimates gained from canopy images.

Materials and Methods

In order to compare measurements between the LI-COR LAI 2200 and commercial digital cameras, measurements were taken with both instruments at 35 survey sites in a mixed conifer forest in northeast Washington. Moultrie I-60 game cameras were mounted 1.7m off the ground on a PVC stand and aimed at the forest canopy. Cameras were set to take pictures once an hour for three days. At the same height and location as the game cameras, readings were taken with the Li-COR LAI 2200 using a tall canopy technique (LI-COR Biosciences 2009).

(Figure 1, right: Game camera images before (top) and after (bottom) determination of sky and canopy pixels.)

In the lab, game camera images and the LAI 2200 data were analyzed. From the game cameras, the best set of hourly images from 5:00 am to 9:00 pm was selected. An algorithm was developed to classify each pixel of these 8-bit images as either sky or canopy (see equation below), and used to assess the total percent of the image covered by canopy (Fig. 1). 

Sky = blue > 100 and (blue x 1.1) > red
Canopy = blue ≤ 100 or (blue x 1.1) ≤ red 

This was done with and without a circular mask applied to the images in order to emulate the readings taken by the circular LAI sensor (Fig. 2). Both the masked and unmasked estimates of canopy cover for each hour of daylight were recorded and compared to LAI estimates as measured using the LAI 2200. The LAI 2200 has five ring-shaped sensors (Fig. 3), and LAI estimates can be made using any combination of these five readings. In this study, LAI was found using various subsets of the five rings (Fig. 3).

The four estimators of LAI were then compared to the masked and unmasked estimates of canopy closure for each hour of daylight. R2 values were calculated for each hour, and linear regression was performed to assess the degree to which LAI could be predicted as a function of game camera canopy cover estimates. Accuracy assessment was performed to characterize the ability of the algorithm to assess canopy closure from game camera images.

Figure 2: Game Camera image with a circular mask.

Figure 3.


We found that photos without a circular mask had the best correlation with LAI readings using only the inner four sensor readings (R2 = 0.71). It was also determined that on average, photos taken at 3:00 pm gave the highest correlation to LAI readings (R2 = 0.71), and photos taken at noon or twilight (9:00 pm) or gave the poorest correlations (R2 = 0.59). In general, images taken in the hours just before or after 12:00 pm gave better results than those taken at other times (Fig. 4). Simple linear regression showed that the prediction of leaf area index as a function of canopy closure estimates was highly significant (p < 0.001) (Fig. 5). An accuracy assessment of five game camera images indicated that on average the overall accuracy of canopy to non-canopy pixel classification was 90.6%.

Discussion and Conclusion

This study found that photos taken in the hours just before and after noon gave the best correlations with LI-COR LAI 2200 measurements. This is reasonable, considering that these hours provide the most light in the canopy while avoiding direct sunlight, as seen in many 12:00 pm photos. It was found that although 9:00 pm photos gave very good estimates from canopy images, these pictures had very poor correlations with LAI readings. In photos from other hours, thin canopy cover was often underestimated due to glare from the sun. In contrast, the twilight images successfully captured these branches. It is possible that these thin layers of canopy do not have significant influence on LAI measurements, but have a considerable impact on canopy closure estimates if included. Future research could determine whether the exclusion of thin layer canopy elements significantly improves the LAI to camera image correlation.

Figure 4: Rvalues indicating the variation in LAI explained using game camera images across the hours of the day.

It was also found that 53° leaf area estimates and images without a circular mask had good correlation. This is reasonable, considering the Moultrie I-60 game cameras have a field of view of 52°. Unmasked images are larger than their masked counterparts, increasing the likelihood that there would be more overlap between the areas captured by the camera and the LAI meter.

Though we addressed some aspects of this method of monitoring forest canopies, other factors have yet to be explored. Future research could investigate the influence of tree species composition, weather conditions, or canopy height on the correlation between photos and LAI readings


We would like to thank Mackenzie Grow and Lindsey LaShaw for their assistance in collecting field data, and we appreciate the Verbrugge family for the use of their land for our survey sites. Funds for this project were provided by Whitworth University's STEM Faculty Research Program.

Figure 5: A scatterplot showing the relationship between LAI and canopy closure for game camera images collected at 3 p.m. without having a circular mask applied.