Nov. 30, 2021
Nov. 16, 2021
Precise and portable scientific devices that can non-destructively and rapidly record complex data in the field are becoming integral to forest research. These tools have miniaturized sophisticated technology for simultaneous data collection and data analysis. Find out how the CI-110 Plant Canopy Imager helped forest scientists achieve a breakthrough in producing accurate predictive models for remote imagery that can now be used in forest productivity estimations.
To predict and estimate forest growth, forest productivity, carbon cycle, and energy balance, scientists need to know the Leaf Area Index (LAI) and light extinction coefficient.
The Leaf Area Index is the total one-sided area of leaves per unit area of ground. It is necessary to measure LAI, as it is an important indicator of the amount of photosynthesis, transpiration, respiration, rainfall interception, and carbon absorption. In a forest, LAI depends on the tree species and their stage of growth, site conditions, season, and also management practices.
The light extinction coefficient (k) depicts how much light penetrates through the canopy and how it decreases towards the ground.
Due to their influence on individual tree and forest productivity, these two factors determine carbon capture and energy balance. Therefore, to predict forest productivity, scientists use the parameters in models to analyze spectra collected through remote satellite imagery.
Scientists prefer using remotely sensed imagery to gather information about the forest, as direct measurements are difficult and have to be restricted to small areas.
To create these models, scientists need a set of physical measurements to establish a correlation with imagery in the calibration and validation stages of modeling.
Indian scientists, Srinet, Nandy, and Patel decided to use the nonlinear machine learning model, Random Forest, to predict forest productivity for a Tropical Moist Deciduous forest in central India.
The prominent forest species were Shorea robusta, Mallotus philippensis, Ehretia laevis, Cassia fistula, and Tectona grandis.
To create models, scientists need known values of predictors; in this case, spectra and responses or Leaf Area Index and k are used.
There are several methods to determine leaf area index, directly or indirectly. Direct methods are destructive, as vegetation has to be harvested to measure all the leaves in a given area.
It is also possible to find an allometric relationship between leaf area estimated from litter traps with canopy spread. Another indirect method tries to estimate LAI using diameter at breast height and the canopy diameter. The indirect methods are not accurate, and all the methods are time consuming and difficult for use on large trees.
Moreover, all these methods are also species and season dependant and do not give a picture of the composite overlapping forest canopy of all trees. Therefore, the scientists decided to look for a scientific plant canopy analyzer to find the leaf area of the forest.
Light extinction coefficient (k) depends on leaf area index, distribution of leaves, and the zenith and azimuth angles of incident solar radiation.
Usually, k is calculated by numerical deductions, as forest canopies are complex and manual measurement of all the factors influencing k is difficult. Therefore, the value of k for different forests is unknown and the same constant is used for all terrestrial forests.
However, this practice can severely compromise the accuracy of a model, which is meant for a specific forest type.
Since manual estimation is difficult, the only option the scientist had was using a tool. Ideally, the scientists needed a single tool that could measure all the necessary parameters, LAI and the photosynthetically active radiation (PAR), to estimate k for the sal forests.
After considering the pros and cons of several plant canopy analyzers on the market, based on reports from previous research studies, the scientists decided to use the CI-110 Plant Canopy Imager. The tool can measure LAI, PAR, and also k.
In all, the scientists collected data from 50 sites, where data from 35 sites were used for training and testing the Random Forest model, and data from 15 sites were used as the test set to validate the model. The scientists decided to use the systematic sampling method for all measurements.
The scientists used the steps shown in Figure 1 to calculate Leaf Area Index and k, and then use them for training and validation of the Random Forest Model.
To calculate the light extinction coefficient, the scientists measured the incident PAR above the canopy (Io) and below canopy (I) at nine locations within a 30 x 30 m site. The center coordinates of the sample plots were also recorded using a handheld GPS.
Using Beer-Lambert's law, the researchers calculated k:
I = Io e-kLAI
k= -1/ LAI (ln I/ Io)
Initially, the scientists calculated the canopy gap fraction using the equation:
Canopy gap fraction = I/ Io
Having a plant canopy analyzer that could measure Photosynthetically Active Radiation (PAR), Leaf Area Index (LAI), and light extinction coefficient (k) was extremely useful for the experiment. Moreover, the CI-110, manufactured by CID Bio-Science Inc, is light and easy to use in the field.
The CI-110 Plant Canopy Imager took images of the canopy using the self-leveling hemispherical 150o fisheye lens. The tool did not need any manual image masking. The device also has a trigger for delayed image release to ensure high quality photos.
It has a long handle on which 24 PAR sensors are placed to simultaneously record the solar radiation coming through sun flecks.
By using the CI-110 Plant Canopy Imager, the scientist got the desired accuracy of data readings for I and Io measurements, which was 5 μmol m−2 s−1.
LAI measurement by the CI-110 followed the Gap Fraction Method using integrated software. The device was also designed to measure leaf inclination angle and calculate extinction coefficients.
The measuring time for an individual reading is very rapid and takes only a second.
The readings, as well as images, are shown on the digital screen. The images and calculations can also be stored in the device and transferred to the computer easily for further analysis.
The scientist had used imagery from Landsat 8 Operational Land Imager. The images were reflections from six spectra, from which 21 spectral indices were extracted.
Field data were collected from 13 random sites. Based on this data, scientists could classify the vegetation into 17 types. Of these, 13 vegetation types were used in the experiment.
The spatial distribution of Leaf Area Index and k was predicted successfully using the Random Forest, as shown in Figure 2. The accuracy of prediction for the spatial distribution of Leaf Area Index and k was R2 of 0.79 and 0.77, respectively. The relationship between Leaf Area Index and k was inverse, following Beer Lambert’s Law; see Figure 3.
The bands of light that best predicted the two parameters were short-wave infrared bands 1 and 2 and moisture stress index (MSI). Normalized Difference Moisture Index (NDMI) was used in the model for Leaf Area Index, and the Normalized Difference Vegetation Index (NDVI) was used for estimating k.
It is now possible to get k for even small patches of different vegetation types, so the estimation of small-scale forest productivity and the carbon cycle is accurate.
In the absence of tools like the CI-110 Plant Canopy Imager, it would not be possible to calculate difficult, but important parameters such as the light extinction coefficient. Moreover, having forest-specific values of it can produce efficient predictive models. These models, in turn, can be used to quantify forest productivity and carbon capture easily even by a layperson, in efforts to mitigate climate change.
|CI-110 Plant Canopy Imager|
Science Writer, CID Bio-Science
Ph.D. Ecology and Environmental Science, B.Sc Agriculture
Feature image courtesy of Nicholas A. Tonelli
Srinet, R., Nandy, S., & Patel, N.R. (2019). Estimating leaf area index and light extinction coefficient using Random Forest regression algorithm in a tropical moist deciduous forest, India. Ecological Informatics, 52, 94-102. https://doi.org/10.1016/j.ecoinf.2019.05.008
1-360-833-8835 ext. 217