Mapping canopy nitrogen in European forests using remote sensing and environmental variables with the random forests method

REMOTE SENSING OF ENVIRONMENT

卷:247

Article

Loozen, Yasmina^1,2  Rebel, Karin T.¹  de Jong, Steven M.²  Lu, Meng²  Ollinger, Scott, V^3,4  Wassen, Martin J.¹  Karssenberg, Derek² 

[1] Univ Utrecht, Fac Geosci, Copernicus Inst Sustainable Dev, Environm Sci, Utrecht, Netherlands

[2] Univ Utrecht, Fac Geosci, Phys Geog, Utrecht, Netherlands

[3] Univ New Hampshire, Earth Syst Res Ctr, Durham, NH 03824 USA

[4] Univ New Hampshire, Dept Nat Resources & Environm, Durham, NH 03824 USA

关键词: Canopy nitrogen;    Foliar nitrogen;    Plant traits;    ICP Forests;    Remote sensing;    MODIS;    MERIS;    Environmental predictors;    Random forests;    Vegetation indices;   

DOI  :  10.1016/j.rse.2020.111933

来源: Elsevier

【 摘 要 】

Canopy nitrogen (N) influences carbon (C) uptake by vegetation through its important role in photosynthetic enzymes. Global Vegetation Models (GVMs) predict C assimilation, but are limited by a lack spatial canopy N input. Mapping canopy N has been done in various ecosystems using remote sensing (RS) products, but has rarely considered environmental variables as additional predictors. Our research objective was to estimate spatial patterns of canopy N in European forests and to investigate the degree to which including environmental variables among the predictors would improve the models compared to using remotely sensed products alone. The environmental variables included were climate, soil properties, altitude, N deposition and land cover, while the remote sensing products were vegetation indices and NIR reflectance from MODIS and MERIS sensors, the MOD13Q1 and MTCI products, respectively. The results showed that canopy N could be estimated both within and among forest types using the random forests technique and calibration data from ICP Forests with good accuracy (r(2) = 0.62, RRMSE = 0.18). The predicted spatial pattern shows higher canopy N in mid-western Europe and relatively lower values in both southern and northern Europe. For all subgroups tested (All plots, Evergreen Needleleaf Forest (ENF) plots and Deciduous Broadleaf Forest (DBF) plots), including environmental variables improved the predictions. Including environmental variables was especially important for the DBF plots, as the prediction model based on remotely sensed data products predicted canopy N with the lowest accuracy.

【 授权许可】

Free