University of Twente Student Theses


Modeling grassland traits using UAV-based RGB and multispectral images.

Umar, Zeeshan (2020) Modeling grassland traits using UAV-based RGB and multispectral images.

Link to full-text:
(only accessible for UT students and staff)
Abstract:Dry matter predictions provides farmers with important information about the nutritional quality of the grasslands. The assessment of spatial distribution of dry matter in grass provides farmers with useful feed information for dairy animals. However the existing remote sensing near real-time techniques, which involves unmanned aerial vehicle (UAVs) mounted with high resolution multispectral sensors to compute the multispectral vegetation indices (MI) is expensive. To address this problem, there is a need of using cost effective Red, Blue, Green based Vegetation indices techniques (RGBVI) for the prediction of dry matter. Therefore, the main objective of this research is to compare the dry matter prediction for both RGBVI and multispectral vegetation indices (MI) acquired via a UAV. The sequoia sensor time-series VI (i.e. RGBVI & MI) dataset are compared using linear and non-linear random forest regression model. This research investigate the impact of time-series dataset of multiple UAV flight to further optimize the model prediction performance. The linear regression model with selected metrics (i.e. RMSE, r-square, MAE) perform better for Kieftenweg using the RGBVI dataset compare to the Vonderweg. The non-linear model using the same metrics perform better for Vonderweg using the RGBVI dataset compare to the Kieftenweg. However, the two models perform almost similar for types of VIs (i.e. RGBVI and MI) combined datasets for two grasslands. The VI’s calculated from the timeseries dataset were also ranked to analyse them for their importance to prediction results. The last two flights means and cumulative values of VIs were ranked the highest based on their contribution to the final results of prediction of dry matter for Kieftenweg using the RGBVI datasets. The commutative and mean of last two flight based VIs were ranked second highest for Vonderweg and two fields combined dataset of RGBVI. The spatial distribution of the dry matter were also analysed using the surrounding pixels around the ground observation. The Kieftenweg and Vonderweg predictions maps displayed predictions result within a certain range beyond which the model couldn’t predict. The present study finds out the RGBVI can produce better results for the individual grasslands. However, the RGBVI produce almost similar to MI for the two fields combined dataset.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Spatial Engineering MSc (60962)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page