Prediction of leaf area index using thermal infrared data acquired by UAS over a mixed temperate forest

• Published in: International journal of applied earth observation and geoinformation Vol. 114; p. 103049
• Main Authors: Stobbelaar, Philip, Neinavaz, Elnaz, Nyktas, Panagiotis
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2022; Elsevier
• Subjects: Land surface emissivity; Land surface temperature; Leaf area index; Thermal infrared; Unmanned aerial system; Unmanned aerial vehicle; Vegetation indices; Thermal infrared; Unmanned aerial vehicle; Unmanned aerial system; Vegetation indices; Land surface emissivity; Land surface temperature; Leaf area index
• ISSN: 1569-8432; 1872-826X
• DOI: 10.1016/j.jag.2022.103049

•LAI prediction accuracy improves by integrating remote sensing VNIR and TIR data.•The relationship between LAI and LST is found to be insignificant.•LSE has a positive correlation with LAI.•Accurate measurement of the percentage of vegetation cover is crucial for LSE retrieval accuracy.•LAI prediction accuracy is higher with LST retrieved from a lower flight altitude. The leaf area index (LAI) is a crucial biophysical variable for remote sensing vegetation studies. LAI estimation through remote sensing data has mostly been investigated using visible and near-infrared (0.4–1.3 μm, VNIR) and Shortwave Infrared (1.4–3 μm, SWIR) data. However, Thermal Infrared (3–14 μm, TIR) data for LAI retrieval has rarely been explored. This study aims to predict LAI by integrating VNIR and TIR data from Unmanned Aerial Systems (UAS) in a mixed temperate forest, the Haagse Bos, Enschede, the Netherlands. The VNIR and TIR images were acquired in September 2020, in conjunction with fieldwork to collect LAI in situ data for 30 plots. TIR images were acquired at two heights (i.e., 85 m and 120 m above ground) to investigate the effect of flight height on the LAI prediction accuracy by means of UAS data. Land Surface Temperature (LST) and Land Surface Emissivity (LSE) were computed and extracted from the collected images. LAI was estimated using seven vegetation indices and Partial Least Squares Regression (PLSR). LAI prediction accuracy using VNIR reflectance spectra was compared to the accuracy achieved by integrating VNIR data with LST or LSE applying vegetation indices as well as PLSR. Among the applied vegetation indices, the Reduced Simple Ratio (RSR) gained the highest prediction accuracy using VNIR data (R2 = 0.5815, RMSE = 0.6972); the prediction accuracy was not improved when LST was integrated with VNIR data but increased when LSE was included (RSR: R2 = 0.7458, RMSE = 0.5081). However, when LST from 85 m altitude and VNIR data was applied as an input using PLSR (R2 = 0.5565, RMSECV = 0.7998), the LAI prediction accuracy was slightly increased compared to when only VNIR data was used (R2 = 0.4452, RMSECV = 0.8668). Integrating VNIR data with LSE significantly improved the LAI retrieval accuracy (R2 = 0.7907, RMSECV = 0.8351). These findings corroborate prior research indicating that combining LSE with VNIR data can increase the prediction accuracy of LAI. However, LST was found to be ineffective for this purpose. The relationship between LAI and LSE should be the subject of more investigation through various approachesto bridge the existingscientific gap.