Epsilon Archive for Student Projects

Abstract

Mapping of e�ective leaf area index (LAIe) over the Swedish boreal forest test site
Krycklan (64°N19°E) was performed using ground-based �eld estimates of LAIe and
remote sensing data sources. The LAIe data were collected 2017 and 2018 using
the LAI-2200 Plant Canopy Analyzer and its later version LAI-2200C Plant Canopy
Analyzer. The remote sensing data used were airborne laser scanning (ALS) data,
Interferometric Synthetic Aperture Radar (InSAR) data from TanDEM-X, and stereo
matched drone images. The stereo matched drone images only covered a small subset
of the Krycklan catchment, the ICOS grid area. Point cloud metrics were calculated
from the ALS data and the drone data such as height percentiles, intensity percentiles,
point cloud density and cover metrics. Three metrics from the TanDEM-X data were
evaluated as predictors; interferometric phase height, coherence and backscatter.
Estimations were done by �tting regression models of LAIe and the predicting remote
sensing data sources. The best ALS regression model for predicting LAIe used the
canopy density gap metric, giving an R
2
adj=0.93 for catchment level estimations and
R
2
adj=0.97 for the ICOS grid area. The TanDEM-X metric interferometric phase height
was the single best predictor of the three InSAR metrics, predicting LAIe with a
R
2
adj=0.85 at catchment level and R
2
adj=0.93 at the ICOS grid area. The drone data
model included the variables canopy cover gap and the 99th height percentile, which
resulted in a R
2
adj value of 0.95. The models were used to generate wall-to-wall rasters
and evaluated with the leave-one-out cross validation method. It was concluded
that the ALS model was best suited to predict LAIe as it was able to handle varying
forestation, which both the other methods struggled with. When applied over mature
and homogeneous boreal forest all models performed with similar accuracy.