From Jan. 1. 2015, Metla is part of the Natural Resources Institute Finland.
Metla-etusivuTutkimusPalvelutJulkaisutMetinfo-metsätietopalvelutTutkimusmetsätTietoa MetlastaStrategiaYhteystiedot
Metla » Research » Climforisk » Methodology » Action 2 » 2. GPP, NPP, NEE and LAI maps for Finland


Sitemap | Search

2. Estimating GPP, NPP and NEE with process-based summary model and generalizing the estimates to regional level with k-NN method

The stand-level forest growth and carbon balance is estimated for the NFI sample plots using a summary model based on carbon production and respiration in different components of the trees (see Härkönen et al. 2010 for detailed description of the model). Soil processes are estimated using the Yasso07 soil carbon model ( based on litterfall data derived from the biomass estimates (Liski et al. 2006). The required input data contains daily weather data and stand characteristics from National Forest Inventory (NFI) (see Fig. 1). The NFI data has been complemented for all the tally trees based on sample tree measurements using a multivariate linear mixed-effects model with species-specific parameters designed for the multi-response NFI data (Eerikäinen 2009). The outputs of the model are gross primary production (GPP), net primary production (NPP) and net ecosystem exchange (NEE) of carbon.

Fig. 1. Model framework.

Estimation of gross primary production (GPP) and net primary production (NPP)

The annual forest growth (kg C ha-1 yr-1) can be expressed as:

NPP = GPP - RM - RG , (1)

where RM is the maintenance respiration and RG is the growth respiration of the trees. NPP can be also expressed as NPP = rNPP GPP, where rNPP is the NPP:GPP ratio depending on the respective rates of maintenance and growth respiration of the stand. Annual biomass production Gt (kg DW ha-1 yr-1) (DW=dry weight) is proportional to NPP as:

Gt=cC-1 NPP, (2)

where cC is the carbon content of biomass dry weight (cC ~ 0.5). GPP depends on environmental driving variables and forest stand data as following:

GPP = fAPAR P0,(3)

where fAPAR is the (effective annual) mean fraction of photosynthetically active radiation (PAR) absorbed by the canopy, and P0 (kg C ha-1 year-1) is annual canopy photosynthesis in a (hypothetical) canopy that absorbs all PAR radiation. This means, that the fAPAR represents the effect of the forest structure to the growth, while the P0 describes the climatic effects. The fAPAR can be estimated using Lambert-Beer formula based on effective extinction coefficient keff, as introduced by Duursma & Mäkelä (2007), and leaf area index (LAI). Effective extinction coefficient can be calculated based on homogenous extinction coefficient KH, crown surface area SA (m2), mean leaf area per tree LA (m2). Leaf area index is derived from the leaf biomass WF (kg DW ha-1) and the assumed specific leaf area (SLA, m2 (kg DW)-1) of the tree species (Luoma 1997). P0 is estimated based on LUE model (Monteith 1977, Mäkelä et al. 2008). Biomasses for different tree components Wi (WF=foliage, WB=branches, WS=stem, WCR=coarse roots and WFR=fine roots) can estimated from stand mean characteristics based on pipe-theory based equations for Scots pine (Mäkelä and Vanninen 2001, Vanninen and Mäkelä 2005), for Norways spruce (Kantola and Mäkelä 2006) and for birches (Ilomäki et al. 2003) or empirical biomass models (Repola et al. 2007). The approach is explained in detail in study by Härkönen et al. 2010.

Estimation of net ecosystem exchange (NEE)

NEE can be derived from the NPP and heterotrophic respiration from the soil, RH, as:

NEE = - (NPP - RH). (4)

Soil processes are estimated using the Yasso07 soil carbon model ( based on litterfall data derived from the biomass estimates (Liski et al. 2006). In Yasso07, total litterfall is divided into non-woody (fine root and foliage) and woody litter (branches, stem, coarse roots). These litter inputs are further divided into four compound groups: 1) compounds soluble in a non-polar solvent, ethanol or dichloromethane (E), 2) compounds soluble in water (W), 3) compounds hydrolysable in acid (A) and 4) compounds neither soluble nor hydrolyzable at all (N) (Tuomi et al. 2008, 2009). Each group has different decomposition rates, which depend on temperature, temperature amplitude between the annual minimum and maximum of mean monthly temperatures and precipitation. Decomposition results in mass loss from the system and inside the system, as well as formation of more recalcitrant humus (H). Annual carbon flux out of the soil, ΔCS (g C m-2 year-1), can be expressed as:

ΔCS=CS,0-CS,1, (5)

where CS,0 (g C m-2 year-1) is soil carbon in the beginning of the simulation year and CS,1 (g C m-2 year-1) is soil carbon at the beginning of the next simulation year . Net ecosystem exchange, NEE (g C m-2 year-1), can thus be expressed also based on the net primary production, PN, total litterfall, LTOT, and the annual soil carbon change, ΔCS , as:

NEE = - ( NPP- LTOT + ΔCS), (6)

negative NEE denoting that the forest is a carbon sink, and the positive that it is a carbon source. Annual litterfall LLS was estimated based on turnover rates defined in the study by Liski et al. (2006) using the above-mentioned biomass estimates.

k-NN imputation

The nearest neighbor method (k-NN) can be used with satellite images either for producing missing input data for the areas to be simulated or for generalizing the already simulated results to the surrounding pixels (see Tomppo 1990, 2006). Here k-NN is used for generalizing the plot-wise simulated results of annual carbon balance to the larger areas based on Landsat 5 TM satellite images, in order to produce raster maps of the carbon fluxes in Finland (Fig. 2). The GPP, NPP and NEE estimates are first produced for the NFI sample plots. The obtained results are then generalized for all the forested areas around the selected sample plots using k-NN imputation based on Landsat 5 TM satellite images. The independent variables can include all the Landsat bands (1-5, 7), or their subset, for example only channels 2-4 (green, red and near-infrared). Further, using several images from the same growing season as well as altitude information from digital elevation model (DEM) as independent variables can be investigated. The nearest neighbors are defined here using the Euclidian distance d as a measure, the estimated Y value being a distance-weighted mean of the nearest neighbors' Y values, with weight 1/(1+d). The k-NN imputations will be done using yaImpute package in the R Statistics (Crookston and Finley 2008). The results of k-NN imputations are evaluated by using leave-one-out cross-validation.

Fig. 2. Example of raster map.


Updated: 10.12.2012 /KBym  |  Photo: Erkki Oksanen, Metla, unless otherwise stated  |  Copyright Metla  |  Feedback