# Utilising unmanned aerial system (UAS) for precision agriculture and yield estimation of Tasmanian poppy

Iqbal, F ORCID: 0000-0003-3034-5707 2018 , 'Utilising unmanned aerial system (UAS) for precision agriculture and yield estimation of Tasmanian poppy', PhD thesis, University of Tasmania.

 Preview
PDF (Whole thesis)

| Preview

## Abstract

The state of Tasmania, Australia is the largest licit producer of poppy opium. The poppy industry in Tasmania cultivates poppies for morphine or thebaine as the main alkaloid product, and supplies up to 40% of the world’s opiates used for medication. The global pharmaceutical industry depends on reliable supply of opium poppy from Tasmania, thus, accurate prediction of opium production is important to fulfill national and international demands. Yield prediction well before crop maturity enables precision management of agronomic practices applied to the crop for enhancing the yield. Commonly, traditional direct methods are used to estimate poppy crop yield, whereby samples are collected from the field and appropriate laboratory analyses are performed. These methods are regarded as time-consuming, labour-intensive, costly, and can be biased as often samples do not represent the spatial variability in the field. Alternatively, remote sensing can provide a spatial continuous and efficient assessment of a poppy field. Using satellite imagery for obtaining spatiotemporal information about the crop’s phenological status during critical growth periods, however, is very challenging due to frequent cloud cover in Tasmania and due to limitations in spatial resolution of current satellite sensors. Data acquired with an unmanned aerial system (UAS) fills this gap by delivering timely high spatial resolution data. This study aims to investigate the potential of ultra-high spatial and temporal resolution UAS data in predicting poppy capsule volume and alkaloid yield in order to provide timely and reliable spatial information to the farmers and pharmaceutical industry for managing and monitoring crop growth and development prior to harvesting, ultimately improving yield potential.
The research presented in this thesis was conducted between 2014 and 2017. Studies in 2014 and 2015 were used to estimate the poppy capsule volume, while the 2016 and 2017 studies were used to predict the capsule volume prior to harvesting. In the capsule volume estimation study, UAS images acquired using a visible (RGB) camera were investigated to generate crop height estimates. This study provided a novel approach to estimate crop height based on a single UAS flight. UAS-derived plant height and field measured plant height were strongly correlated with R$$^2$$ values ranging from 0.93 to 0.97. The robustness of the proposed methodology was successfully tested on a second study site. Field-based experiments showed that plant height and poppy capsule volume were strongly related (R$$^2$$ 0.74), with relative error of 19.62%. The plant height generated from UAS data was used to estimate capsule volume, and it was concluded that plant height can be reliably estimated for poppy crops based on a single UAS flight and can be used to predict opium capsule volume at capsule formation stage.
The second study aimed to determine whether the combination of structural and spectral derivatives of UAS data could provide improvement in capsule volume prediction results using a random forest (RF) machine learning approach. Investigation was undertaken using multiple sensor dataset integration, field based spectroscopy data, UAS based multispectral data and crop height. Convolved field spectra were used to compute spectral vegetation indices (SVIs), and a combination of SVIs and plant height was used to train and test the random forest model. It was found that poppy capsule volume can be estimated using a combination of SVIs (NDVI, mSR, mTVI and RDVI) with RF regression and significantly higher prediction accuracy with an RMSE value of 15.60 cm$$^3$$ (10.27 %) based on training dataset and an RMSE value of 25.63 cm$$^3$$ (14.45 %) with validation dataset was observed. The proposed RF model provides improvement in capsule volume estimation as compared to that estimated using plant height only. The proposed RF model was successfully tested on UAS data acquired from two different experiment locations.
To predict the poppy capsule volume prior to harvesting, multi-temporal UAS data was collected using a 4-band Parrot Sequoia multispectral sensor. Data was collected from hook, flowering and capsule formation stages. This experiment investigated the potential to predict capsule volume of poppy crop based on multi-temporal SVIs and green vegetation fractional cover, using random forest regression and multi linear regression techniques. The combination of green vegetation fractional cover from flowering stage with spectral indices from hook stage (RDVI, SPVI and mTVI) provided optimal results with an R$$^2$$ of 0.88 and RMSE of 13.45 % using RF regression. Multi-temporal data of hook, flowering and capsule formation stage was also used to estimate the alkaloid yield in poppy capsule. It was found that the RF model with MSAVI, mSR, OSAVI, NDVI and EVI from capsule formation stage can provide optimal results to estimate thebaine with a relative error of 13.56 % to 22.36 % with training and validation dataset, respectively.
The findings in this thesis demonstrate that UAS based visible and multispectral imaging delivers valuable data for poppy crop yield estimation and prediction. The demonstrated results for plant height and yield open new possibilities in precision agriculture by capturing in-field variability and methods developed in this thesis provide tools to support operational decisions for site specific precision agriculture. The results achieved from methods developed in this thesis are robust and can be applied to other crops as well.