Determining Biophysical and Biochemical Attributes of Pasture Using High Resolution Remote Sensing Techniques

Authored by MSc. Gustavo Togeiro de Alckmin. Gustavo is a PhD candidate at the University of Tasmania (Australia) and at Wageningen University (the Netherlands), he is also a past 2018 ASD Goetz Instrument Program Winner. 

Grasses are the cornerstone of most forage-based livestock production systems. The possibility of transforming a non-digestible resource into high-value, nutritive products is an important asset to ensure a secure and sustainable global food production system.

Regardless of the laymen consensus, pasture production is a highly dynamic activity. However, measurement and monitoring of grass growth is usually a time-consuming, tedious activity, and possibly, prone to inaccuracies. Alternatively, the deployment of remote sensing techniques could substantially automate and enhance this activity, providing quantitative (dry matter per area) and qualitative (feed quality) measurements at an intra-paddock-scale.

Given such goals, I have been funded by Dairy Australia, and am currently enrolled in a joint-doctoral program at the University of Tasmania (Australia) and Wageningen University (the Netherlands). The main goal of the PhD is to compare current grazing-management methods (mostly relying on canopy height observations) against spectra-based models. Additionally, the research aims to scrutinize the use of Unmanned Aerial Systems (multispectral and hyperspectral imaging sensors) as a feasible and more accurate alternative for a farm operation.

Ideally, this work could inform farmers about the limitations of each method as well as provide the necessary information for sensor design at different scales (handheld, low-level flight or satellite). Given the detailed nature of reflectance data, several statistical techniques (commonly named “machine learning“) could boost the quality of predictions.

Data provided by hyperspectral measurements are the optimal starting point to find relations between spectra and properties of interest. Alternative coarser multispectral systems will, most likely, reduce the level of information measured (Figure 1); thus, restricting the search-space where meaningful relations can be tested.

As the research spans across two different countries and several sites and seasons, results can validate whether if findings generalize. Such would be a strong indicative of how robust and reliable these methods are. In short, the research-work can be the basis of a robust and reliable custom-made sensor (most likely mounted on a drone) to better serve the needs of pasture managers.

The experimental design consisted of 30 plots, under five different fertilization regimes and three growth intervals, to achieve a wide gradient of pasture quantity and quality. Specifically, for the pasture species of interest (perennial ryegrass), the number of fully developed leaves is an important morphological proxy for feed quality. When reaching three fully developed leaves, the eldest starts to senesce (die-off) while a newer leaf starts to develop. Consequently, the total value of the pasture decays, given that leaves became more mature and senescent material is now part of the canopy. The experimental design capitalized on this natural cycle, having growth intervals that would allow for plots with one, two or three fully developed leaves.

During the past November (spring), the ASD FieldSpec® spectroradiometer was employed in three data collection campaigns, as well as to measure calibration targets employed in the validation of the radiometric calibration of drone-mounted sensors. Additionally, spectral measurements taken at handheld levels will be employed as a benchmark for the radiometric quality of low-level flight imagery. While collecting hyperspectral data (Figure 2), a multispectral (drone) imagery data set was also acquired allowing us to validate the radiometric accuracy of the imagery set (Figure 3 & Figure 4). Pasture samples (ground truth) were dried, weighed and sent to the lab for feed analysis.

Findings and models, however, are only as good as the input data in which they are based on. The timeless axiom “garbage in, garbage out” holds out regardless of the analysis performed. To that end, the availability of the ASD FieldSpec was invaluable. The instrument is known as the “gold-standard” for field spectroscopy, a well-deserved reputation; the reflectance measurements that we collected with the instrument prove to be stable and, consequently, reliable.

A good way to depict the ASD FieldSpec’s measurement stability is to illustrate the in-field quality assurance protocol which was in place for data collection; basically, after measuring each plot, a white reference (i.e. Spectralon®) measurement was taken. This activity served two purposes: to check the stability of the sensor (which could drift due to temperature fluctuations), or to identify changes in atmospheric conditions. Figure 5 presents two white reference spectra (orange and blue lines, primary axis) collected after finalizing the measurements at each plot (time interval of around 3-5 minutes): both spectra are virtually indistinguishable. The average difference between both spectra is equal to 0.013 (grey line, secondary axis).

Additionally, when collecting spectral data, the experimental protocol required five repeated measurements of any target (including white references). That way, measurements taken in sequence could detect any sort of abnormality on the instrument’s performance. As an illustration, the average standard deviation for a sequence of five spectralon measurements was 0.003733, which is negligible (Figure 6). It should be highlighted that both these numbers would be even smaller if the noisy spectral regions were masked.

The ASD FieldSpec is a versatile instrument; there are a high number of ASD accessories that can be employed and paired with the spectroradiometer to respond to different research questions. One of the ASD accessories which was particularly useful for this research application was the ASD Contact Probe and calibrated light source provided to verify reflectance values of the reference panels (Fig 3), which were then employed for the purpose of radiometric calibration of the drone imagery.

Research results are now being analyzed and are to be published in 2019.