University of Twente Student Theses


Prediction of leaf area index using the integration of the thermal infrared with visible and near-infrared data acquired with an UAV for a mixed forest

Stobbelaar, Philip (2021) Prediction of leaf area index using the integration of the thermal infrared with visible and near-infrared data acquired with an UAV for a mixed forest.

[img] PDF
Link to full-text:
(only accessible for UT students and staff)
Abstract:The leaf area index (LAI) is one of the most important biophysical variables, providing crucial information about vegetation and its processes. The LAI is invaluable for climate and biodiversity studies, being part of the essential climate variables (ECVs) as well as the remote-sensing-enabled essential biodiversity variables (RS-enabled EBVs), as it is needed for many environmental models and can be of use for stakeholders like agriculturalists, foresters, and ecologists. Due to its importance, it is essential to be able to accurately measure or predict the LAI. The most common methods make use of remote sensing, and so far, LAI has been successfully estimated using spectral reflectance in the visible and near-infrared (0.4 – 1.3 μm, VNIR) and the short wave infrared (1.4 – 3 μm, SWIR) domains. However, the possibility of using the characteristics of vegetation in the thermal infrared (3 – 14 μm, TIR) for LAI estimation has not been researched as much. With studies under laboratory settings proving that the TIR data can be of interest when predicting LAI on the canopy level, there is a need for research in other environments and different settings. In this study, the prediction of LAI with the integration of VNIR and TIR data was investigated for a mixed forest, the Haagse Bos located in the North of Enschede, the Netherlands. During the field campaign in September 2021, in-situ LAI measurements were carried out. Simultaneously, VNIR and TIR images were captured by means of an unmanned aerial system (UAS). To assess the capabilities of integrating TIR with VNIR data for LAI prediction, land surface temperature (LST) and land surface emissivity (LSE) were calculated. For analysis using LST, data from two different separate flight heights of the UAS (85 m and 120 m) were used to assess the effect of altitude on the LAI prediction accuracy. LSE was calculated using the normalised difference vegetation index (NDVI) threshold method, which makes use of emissivity values for vegetation and bare soil, the NDVI, and the percentage of vegetation cover (PV). PV was taken from two different approaches, from in-situ data as well as from a canopy height model (CHM). The LAI prediction analysis was done by examining the relationship of LAI with nine different vegetation indices as well as with the use of partial least squares regression (PLSR). Analysis using vegetation indices as well as using PLSR was done by comparing the LAI prediction accuracy obtained using only VNIR reflectance spectra and when integrating the VNIR reflectance spectra with either LST or LSE. The highest prediction accuracy obtained between LAI and VNIR data was (R² = 0.5815, RMSE = 0.6972) using reduced simple ratio (RSR) vegetation index. Prediction accuracy of LAI was not improved with the integration of LST and VNIR data using vegetation indices; however, it was increased when VNIR data integrated with LSE (RSR: R² = 0.7458, RMSE = 0.5081). The best result was obtained with LST integration with the VNIR reflectance spectra using PLSR with LST from the 85 m altitude (R2 = 0.5565, RMSECV = 0.7998). However, the integration of VNIR data with LSE significantly improved the results using PLSR (R² = 0.7907, RMSECV = 0.8351). This indicates that LST is not beneficial for LAI prediction when integrating with reflectance spectra when using vegetation indices or PLSR. Additional results also showed that differentiation of plots by dominant species could improve LAI prediction accuracy with PLSR, especially when integrating LSE with VNIR data. Th study confirms outcomes of previous research, stating that information of the TIR domain shows promising results when integrating with reflectance data to predict LAI. An important finding is that while LST is not suitable for improving the LAI prediction accuracy, LSE seems to be a beneficial addition. Consequently, further research should aim to increase the knowledge on the relationship between LAI and LSE, focusing on different approaches and different environments to fill the existing scientific gap.
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