Separating the components of ecosystem-scale carbon exchange is crucial in order to develop better models and future predictions of the terrestrial carbon cycle. However, there are several uncertainties and unknowns related to current photosynthesis estimates. In this study, we evaluate four different methods for estimating photosynthesis at a boreal forest at the ecosystem scale, of which two are based on carbon dioxide (

For the measurement period 2013–2017, the artificial neural network method gave a GPP estimate very close to that of traditional flux partitioning at all timescales. On average, the COS-based methods gave higher GPP estimates than the

Photosynthetic carbon uptake (or gross primary production, GPP) is a key component of the global carbon cycle, with the terrestrial ecosystems removing approximately 30 % of annual anthropogenic carbon dioxide (

Eddy covariance (EC) is widely used to measure the biosphere–atmosphere exchange of

One way to address these uncertainties in flux partitioning is to use machine learning methods, such as artificial neural networks, to separate NEE into respiration and GPP

Yet another approach to addressing uncertainties in GPP estimates is to use proxies for photosynthetic

In this study, we compare the annual, seasonal, daily, and sub-daily variation of (i) a traditional GPP estimate (GPP

Measurements were conducted at the Hyytiälä forest Station for Measuring Ecosystem Atmosphere Relations (SMEAR) II measurement site (

EC measurements were made on a 23 m high tower. The set-up consisted of a Gill HS (Gill Instruments Ltd., England, UK) sonic anemometer measuring horizontal and vertical wind velocities and sonic temperature, as well as a quantum cascade laser (QCL; Aerodyne Research Inc., Billerica, MA, USA) for measuring COS,

Environmental measurements used in the study include air temperature (

This section describes each of the four methods for estimating GPP. Daily average GPP was only calculated if more than 50 % of the measured 30 min flux data were available for each day, and monthly averages were calculated from the daily means. In

NEE was partitioned into respiration (

When NEE measurements were not available, the GPP model followed the formula

Parameters

GPP

Each subnetwork relies on specific predictors. Distinguishing features of this model are that (a) GPP and

Based on previous soil chamber measurements at Hyytiälä forest it is known that the soil COS flux was

LRU was calculated as a function of PAR (

Finally, we estimated GPP from Eq. (

Explanations, literature values, and sources of the parameters used in the LRU

LRU

In addition, LRU

March 2013 was colder than other years (average

Median diurnal variation of

Midday GPP

Median diurnal variation of GPP partitioned using a combined nighttime–daytime method (GPP

Diurnal variation of the difference of GPP

Scatter plots of GPP

Relative

GPP

Based on the CAP stomatal optimization model, LRU

LRU

LRU

We calculated the cumulative GPP estimates over May–July, 13 weeks around the peak growing season for each year (Table 2). Cumulative GPP

It has been suggested that, due to the Kok effect, leaf respiration is inhibited under radiation

Distribution (bars) and probability density functions (lines) of daily average

Cumulative GPP (

All four GPP estimates responded similarly to environmental forcing (PAR,

Responses of the different GPP estimates (GPP

In spring, increasing air temperature increased all GPP estimates similarly until

Because ANN fitting is purely based on the provided examples, GPP

GPP

GPP

One source of uncertainty in GPP estimates based on LRU

Daily GPP

The LRU

Although COS flux measurements are noisier, more expensive, and more difficult than those of

The establishment of large long-term ecosystem research infrastructures (e.g. ICOS, NEON, TERN; see

The general expression for the leaf relative uptake ratio (LRU) derived from the diffusion laws for COS and

If it is assumed that boundary layer and mesophyll conductances are infinite, Eq. (

We used Eq. (

The CAP solution for optimal stomatal conductance

Nevertheless, to assess the performance of LRU

In the case that mesophyll conductance is not assumed to be infinite (but boundary layer conductance is infinite), Eq. (

If we further assume that the ratios of stomatal to mesophyll conductances are the same for

In this case, since Eq. (

We then find the CAP solution for

As for LRU

Modelled against measured NEE using

Scatter plots of LRU

Scatter plots of GPP

ANOVA test results for 30 min GPP data. Gray bars indicate no difference to the reference (blue), and red bars indicate statistical difference to the reference. The results show that only GPP

ANOVA test results for daily GPP data. Gray bars indicate no difference to the reference (blue), and red bars indicate statistical difference to the reference. The results show that both GPP

ANOVA test results for monthly GPP data. Gray bars indicate no difference to the reference (blue), and red bars indicate statistical difference to the reference. The results show that all GPPs are statistically the same at monthly scale.

LRU derived from chamber measurements (gray) and modelled LRU

Net ecosystem exchange (NEE) against photosynthetically active radiation (PAR) close to the compensation point during May, June, July, and August. Data are binned to different air temperature classes:

The flux data and all GPP estimates used in this study are available from

KMK, IM, and TV designed the study. KMK, PK, and LMJK performed the measurements and flux processing. RD, AM, and KMK developed the new LRU formulation. GT provided the GPP estimate by artificial neural networks. DP gave insight into all GPP method uncertainties and study design and commented on the manuscript. All authors contributed by commenting on the study design, results, and the manuscript. KMK wrote the manuscript with contributions from all co-authors.

The contact author has declared that none of the authors has any competing interests.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Special thanks to Helmi Keskinen, Sirpa Rantanen, Janne Levula, and other Hyytiälä technical staff for all their support with the measurements.

This research has been supported by the Academy of Finland (grant nos. 118780, 312571, 282842, 337549, and 342930), the H2020 European Research Council (grant nos. 755617 and 742798), the H2020 Environment (grant no. 820852), and ICOS Finland (grant no. 3119871). Specifically, Gianluca Tramontana was supported by the European Research Council (ERC) under the ERC-2017-STG SENTIFLEX project (grant no. 755617), Linda M. J. Kooijmans was supported by the ERC advanced funding scheme (AdG 2016, grant number 742798, project abbreviation COS-OCS), and Dario Papale was supported by the E-SHAPE H2020 project (grant no. 820852) and the ICOS-ETC. Kukka-Maaria Kohonen was supported by the Vilho, Yrjö, and Kalle Väisälä foundation. Aleksanteri Mauranen was supported by the Doctoral Programme in Atmospheric Sciences of the University of Helsinki.Open-access funding was provided by the Helsinki University Library.

This paper was edited by Nicolas Brüggemann and reviewed by three anonymous referees.