Including shaded leaves in a sample affects the accuracy of remotely estimating foliar nitrogen. Issue 7 (3rd October 2019)

• Record Type: Journal Article
• Authors: Mutowo, Godfrey; Mutanga, Onisimo; Masocha, Mhosisi
• Is Part Of: GIScience & remote sensing. Volume 56:Issue 7(2019)
• Page Start: 1114
• Page End: 1127
• Publication Date: 2019-10-03
• Subjects: Nitrogen -- miombo woodlands -- remote sensing -- random forest -- multiple-levels; Geodesy -- Periodicals; Cartography -- Periodicals; Aerial photogrammetry -- Periodicals; Remote sensing -- Periodicals; 526.05
• DOI: 10.1080/15481603.2019.1620480
• Languages: English
• ISSNs: 1548-1603

Abstract

Abstract : The remote estimation of foliar nitrogen (N) has largely assumed that the spectral reflectance value measured on a remote sensing platform comes from only the sunlit leaves of the canopy. Whilst this could have been valid for leaf-level spectroscopy studies, the landscape level estimation of foliar N presents new challenges that need investigation. In addition, a growing interest in the application of broadband satellites in foliar N estimation has triggered the need to understand the confounding factors affecting the relationship between N and spectral reflectance. Field sampling criteria is therefore critical in obtaining representative foliar samples. However, the effect of using leaf samples drawn from different levels in the canopy on the accuracy of remotely estimating N is still poorly understood. Our study was carried out in the expansive miombo woodlands. A bootstrapped random forest regression technique, in the R environment, was used to predict foliar N from sentinel-2 broadband satellite remote sensing image. A weighted mean of foliar N was calculated by considering values from two levels in the canopy, from 0% to 100% at intervals of 10%. Our results showed that the most accurate model was the one where the mean N had equal weighting from both levels. Furthermore, there were significant (p < 0.05) differences amongst the root-mean-square errors (RMSE) of prediction of the models considered. We conclude that sampling leaves from two levels in the canopy improves the accuracy of remotely estimating N. This finding is significant in the canopy sampling for remotely estimating foliar N.