Comparison of global z\(_\text{root}\) estimate with RPGE data

Distribution of values

The global z\(_\text{root}\) estimate can be compared with the ecosystem-level RPGE data.

The figure below shows the distribution of modelled values (bars, where ‘Balland’ and ‘SaxtonRawls’ refers to the WHC estimate based on pedotransfer functions from Balland et al., 2008, and Saxton & Rawls, 2006, respectively.). The blue line represents the distribution of depth estimates for the extrapolated 95% quantile of root mass (D95_extrapolated) in the RPGE data.

## Parsed with column specification:
## cols(
##   .default = col_double(),
##   ID = col_character(),
##   Schenk_Jackson_2002 = col_character(),
##   Reference = col_character(),
##   location = col_character(),
##   Elevation = col_character(),
##   MaxDiameter = col_character(),
##   Measurement = col_character(),
##   Sampmax = col_character(),
##   Texture = col_character(),
##   Tree_phenology = col_character(),
##   Anthropogenic = col_character(),
##   Wetland = col_character(),
##   Vegetation = col_character(),
##   Rainfall_season = col_character()
## )
## See spec(...) for full column specifications.
## Warning: Removed 31 rows containing non-finite values (stat_bin).
## Warning: Removed 4 rows containing missing values (geom_bar).

This suggests:

  • A good general agreement of the magnitude and the distribution, especially at intermediate values (between about 0.5 m and 2.3 m),
  • The frequency of very shallow rooting (below about 0.5 m) is overestimated by the model, and the frequency of rather deep rooting (values around 2.5 m and above) tends to be underestimated. This may be related to the scale mismatch. The model uses gridcell-average fAPAR to estimate the water demand per unit area, with must be lower than the water demand per unit area of vegetetated land only. This discrepancy is highest in areas with low fAPAR, including arid regions with large dry season water deficits.

Site-by-site evaluation

A site-by-site comparison of modelled vs. observed rooting depth (scatterplot), where modelled is extracted from the global simulation, is done below.

## Loading required package: LSD
## Loading required package: ggthemes
## Loading required package: RColorBrewer

This suggests:

  • A poor model performance.
  • An likely challenge of scale mismatch.

Rooting depth map

The values of the points in the RPGE data are overlaid onto the rooting depth map from the model. The question is: Are there broad patterns that we could focus on?

## Loading required namespace: ncdf4
## Loading required package: rworldmap
## ### Welcome to rworldmap ###
## For a short introduction type :   vignette('rworldmap')
## Loading required package: scales
## 
## Attaching package: 'scales'
## The following object is masked from 'package:readr':
## 
##     col_factor
## Regions defined for each Polygons
## Warning: Removed 637 rows containing missing values (geom_path).
## Warning: Removed 5 rows containing missing values (geom_point).

Let’s discuss what we can get from this …

Addressing the scale mismatch

We may argue that we cannot expect a global model that does not account for local topography to accurately simulate rooting depth measured at the site scale. Can we instead require the model to capture known patterns in the rooting depth across some class of vegetation type, climate, biome, … ? The challenge is to identify such patterns where a priori expect rooting depth variations.

Let’s discuss this further …