Carbon sequestration potential of street tree plantings in Helsinki

. Cities have become increasingly interested in reducing their greenhouse gas emissions, and increasing carbon sequestration and storage in urban vegetation and soil as part of their climate mitigation actions. However, most of our knowledge on biogenic carbon cycle is based on data and models from forested ecosystems even though urban nature and microclimate are very different to those in natural or forested ecosystems. There is a need for modelling tools that can correctly consider temporal 5 variations of urban carbon cycle and take the urban specific conditions into account. The main aims of this study are to examine the carbon sequestration potential of two commonly used street tree species ( Tilia x vulgaris and Alnus glutinosa ) and their soils by taking into account the complexity of urban conditions, and evaluate urban land surface model SUEWS and soil carbon model Yasso15 in simulating carbon sequestration of these street tree plantings at different temporal scales (diurnal, monthly and annual). SUEWS provides the urban microclimate, and photosynthesis and respiration of street trees whereas the soil 10 carbon storage is estimated with Yasso. Both models were run for 2002–2016 and within this period the model performances were evaluated against transpiration estimated from sap flow, soil carbon content and soil moisture measurements from two street tree sites located in Helsinki, Finland. The models were able to capture the variability in urban carbon cycle due to changes in environmental conditions and tree species. SUEWS simulated the stomatal control and transpiration well (RMSE<0.31 mm h − 1 ) and was able to produce correct 15 soil moisture in the street soil (nRMSE<0.23). Overall, the results indicate the importance of in urban carbon sequestration estimations.


Introduction
The ongoing climate warming is caused by anthropogenic emissions of greenhouse gases (GHGs). A large proportion of these 25 emissions, especially carbon dioxide (CO 2 ), originate from urban areas (Marcotullio et al., 2013). In order to fight against the climate crisis, significant amount of cities have declared themselves to be carbon neutral in the future decades. Carbon neutrality in a city scale means that either GHG emissions and sinks are in balance or alternatively, part of the emissions are compensated elsewhere. Urban green areas have been found to sequester up to 14 % (Vaccari et al., 2013;Hardiman et al., 2017) of cities' GHG emissions. However, urban nature is highly diverse which brings a lot of uncertainty to the estimates. 30 In order for cities to reliably quantify their own carbon sinks to urban vegetation and soil, more information of the biogenic carbon cycle in urban areas is required.

Ecophysiological measurements
The portable gas exchange sensor (CIRAS-2, PP Systems, UK) was used to determine leaf-level responses of transpiration and CO 2 exchange to environmental drivers (light, CO 2 ). A total of 22-25 leaf samples located at different positions in the crown in six to seven trees of each studied species were measured during five field campaigns in 2007(Riikonen et al., 2011. The campaign measurements were normally carried out between 8 am and 4 pm. The measured light and CO 2 responses of leaflevel CO 2 exchange were scaled to stand level ::::::::: stand-level using the forest stand gas exchange model SPP (Mäkelä et al., 2006) and meteorological measurements from Kumpula (See Sect. 2.3). The optimal stomatal control model (Hari et al., 1986)

125
To get estimation for whole-tree transpiration, sap flow sf m (l m −2 h −1 or mm h −1 ) was measured with Granier type heat dissipation sensor pair (Hölttä et al., 2015) from three Tilia and three Alnus trees (Riikonen et al., 2016 (Riikonen et al., 2017). The soil samples were collected in autumn from each soil type from depths varying between 30 to 90 cm. The soil carbon stock estimates were derived from the soil samples, where LOI was determined for each soil type. The proportion of carbon in the LOI was assumed 0.56 (Hoogsteen et al., 2015).

Meteorological measurements
Meteorological variables used to force the models with hourly resolution for years 2002-2016 were primarily from the nearby In order to create continuous meteorological forcing files for the modelled years, missing data from Kumpula were gap filled with observations from a station at Helsinki-Vantaa airport hosted by Finnish Meteorological Institute located 10 km northwest from Viikki. More detailed information of the gap filling procedure is given in Appendix A.

SUEWS
The Surface Urban Energy and Water balance Scheme (SUEWS) was originally developed to simulate the urban surface energy and water balance at a local or neighborhood scale (Järvi et al., 2011;Ward et al., 2016). The model includes several submodels for net all-wave radiation (Offerle et al., 2003), storage Sun et al., 2017) and anthropogenic heat fluxes, snow and irrigation (Järvi et al., 2014) to take urban features in the balances appropriately into account. Recently, 160 the surface-atmosphere exchange of anthropogenic and biogenic CO 2 have been included to the model providing integrated information of the energy, water and CO 2 cycles in urban areas, including the impact of increased air temperatures on the water and CO 2 cycles (Järvi et al., 2019). This study used the most recent SUEWS version available V2020a. The model is forced with commonly measured meteorological variables, such as, wind speed, wind direction, air temperature, :: air : pressure, precipitation and shortwave radiation. Specific site information are also needed in the model simulations, such as, surface cover 165 fractions, and tree and building heights.

Biogenic CO 2 flux
The biogenic CO 2 flux components include the carbon uptake by photosynthesis (F GP P ) and carbon emissions by vegetation respiration (F R ). Soil respiration can be included if integrated vegetation and soil parameters are used in the model runs. An empirical canopy-level photosynthesis model (Järvi et al., 2019) was used for the connection of transpiration to photosynthesis 170 via stomatal conductance, and its dependency on local environmental conditions. F GP P (µmol m −2 s −1 ) for deciduous trees is calculated from where the potential photosynthesis (F GP P,max,decid ) is scaled with leaf area index (LAI decid , m 2 m −2 ), surface cover fraction (f r decid ), and by the environmental response functions f (T air ), f (∆q), f (∆θ), and f (K ↓ ) :::::: g(T air ), ::::::: g(∆q), :::::: g(∆θ), ::: and :::::: g(K ↓ ) 175 on air temperature, specific humidity deficit, soil moisture deficit, and shortwave radiation, respectively. The functions have forms (Ward et al., 2016) f g : f g : where and 185 f g Parameter G 2 −G 6 describe the responses of photosynthesis and stomatal conductance on each environmental variable. K ↓,max (W m −2 ) is the maximum observed shortwave radiation, T L and T H ( • C) are the lower and upper limits for temperature to determine when photosynthesis and transpiration switch off, and ∆θ W P (mm) is the wilting point ::::: deficit. The variables ∆q (g kg −1 ), K ↓ (W m −2 ) and T air ( • C) are given to the model as an input at the modelling height, typically well-above the urban 190 surface, but SUEWS has an option to model local values of ∆q and T air at 2-m height (Sun and Grimmond, 2019) allowing to take into account the impact of local climate conditions on the spatial variability of F GP P . ∆θ (mm) is simulated within SUEWS (Järvi et al., 2017).
In SUEWS, F R increases exponentially with measured input or modelled local air temperature. Air temperature is used instead of soil temperature due to its common availability. F R (µmol m −2 s −1 ) is simulated with empirical constants a and b

195
following The lower limit of F R (0.6 µmolm −2 s −1 ) takes into account carbon emissions in winter that can not be achieved with the simple exponential model (Järvi et al., 2019). In this study, F R included only aboveground respiration as the soil respiration is ::: was : determined with Yasso (see Sect. 2.5). In order to correctly simulate the carbon sequestration and respiration of street

Evapotranspiration
The latent heat flux (Q E , W m −2 ) including both evaporation and transpiration is calculated with the modified Penman-Monteith equation for urban areas  205 where Q * (W m −2 ) is the net all-wave radiation, Q F (W m −2 ) the anthropogenic heat flux, ∆Q S (W m −2 ) the net storage heat flux, ρ (kg m −3 ) the density of air, c p (J kg −1 K −1 ) the specific heat capacity of air at constant pressure, VPD (Pa) the vapour pressure deficit, s (Pa • C −1 ) the slope of the saturation vapour pressure curve, γ (Pa • C −1 ) the psychrometric constant, r av (s m −1 ) the aerodynamic resistance for water vapour and r s (s m −1 ) the surface resistance. The surface resistance or its inverse 210 surface conductance g s (m s −1 ) depends on the same environmental factors as photosynthesis (Ward et al., 2016) g s = 1 r s = g max,decid LAI decid LAI max,decid f r decid G 1 f g : (T air )f g : (∆q)f g : (∆θ)f g : where the maximum conductance g max,decid is scaled with maximum leaf area index (LAI max,decid ), f r decid and the environmental response functions. G 1 (mm s −1 ) is a constant obtained from latent heat (Q E ) and sensible heat (Q H , W m −2 ) observations and it connects stomatal conductance to canopy conductance.

Fitting environmental response functions
To get correct response of street trees to environmental factors in SUEWS, the environmental response functions (f (T air ), f (∆q), f (∆θ), and f (K ↓ ) :::::: g(T air ), ::::::: g(∆q), :::::: g(∆θ), :::: and ::::: g(K ↓ )) in Eqs. (1) and (9), were separately fitted for Tilia and Alnus trees using a non-linear least-square method. In previous study at the Tilia site, similar fittings were made but only to fit F GP P,max and f (∆q) :::::: g(∆q) assuming the other function forms from park located in England (Järvi et al., 2019). To get more 220 precise parameters to describe the behaviour of street trees, all the response functions were fitted against observations to get parameters G 2 − G 6 and F GP P,max .
The previously calculated stand-level photosynthesis estimates for 2016 were used in the fitting as dependent variable while for independent variables observed T air , ∆q, and K ↓ from Kumpula and SWC from the study sites were used. Fitting was made when K ↓ > 10 W m −2 and ∆q > 1 g kg −1 as otherwise the stomatal conductance may deviate from the fits seen in  (Bosveld and Bouten, 2001). This resulted all together 2492 data points. In the fitting, a bootstrapping method was used by randomly selecting 100 times 7/8th of the available observations with the final parameters calculated as medians with uncertainty from the fittings. Table 2 gives the fitted parameter values needed in Eqs.
(2)-(6). In the calculation of f (∆θ) ::::: g(∆θ) : wilting point (WP) is needed to calculate the limit ∆θ W P . A site specific estimate for ∆θ W P was calculated with soil information from Riikonen et al. (2011).  The respiration parameters a and b in Eq. (7) are ::: were : obtained by fitting canopy-level respiration estimates from the street trees for year 2016 against air temperature measurements from Kumpula. The estimations represent respiration from leaves and branches. To estimate whole tree respiration, one third of the canopy respiration was added to the values before the fittings to represent respiration from the trunk. Using bootstrapping method described above, for Tilia site, parameter values a = 0.78 ± 0.002 and b = 0.08 ± 0.0001, and for Alnus site a = 1.11 ± 0.003 and b = 0.08 ± 0.0001 are obtained.  (Table 1) for both sites were obtained from an airborne laser scanning data with a resolution of 1 m (StromJan, 2020). The modelling areas had buildings, paved surfaces, bare soil, grass and deciduous trees. As SUEWS gives integrated evapotranspiration, photosynthesis and respiration for the whole simulation domain, grass surfaces present in the areas were set to impervious surfaces. This had a minor impact on modelled local air temperature (on average 0.16 • C warmer in summer) and humidity, and furthermore on tree functioning, but this is ::: was : seen more suitable approach when model outputs are :::: were 250 compared with tree observations.
The trees at both sites were planted in 2002 and as SUEWS does not currently include tree growth, information of the development of the trees during the modelled period are :::: were : obtained from the local measurements. Tree height and maximum LAI are :::: were given to SUEWS as model input for each year whereas the seasonal development of LAI is ::: was : based on growing degree days within the model. The tree heights were measured from 2002 until 2011 (Riikonen et al., 2016) and as the tree 255 growths follows ::::: follow : exponential curves, the same exponential growth was assumed for rest of the years. The maximum LAI for both Tilia and Alnus trees was set to 4.8 m 2 m −2 as obtained for Tilia cordata in Breuer et al. (2003) and Alnus glutinosa in Eschenbach and Kappen (1996), respectively. The observations as such were not used for the maximum LAI as they present values for individual trees and not for neighborhood (stand) level as expected by SUEWS.
The vegetation type specific maximum stomatal conductance values (g max,decid ) needed in the model input are significantly 260 different between the two tree species. Alnus glutinosa have larger water use than Tilia x vulgaris. Similarly to maximum LAI values, g max,decid = 8.7 mm s −1 were chosen for Alnus site based on a study made in Germany (Eschenbach and Kappen, 1999), and g max,decid = 3.1 mm s −1 was chosen for Tilia site based on Breuer et al. (2003).
The modelled soil depth under the street trees was 1 m and soil water storage capacity 0.141 m was calculated from laboratory measurements. The amount of water in the top 1 m soil was not sufficient to maintain the high transpiration raters :::: rates of Alnus 265 trees. This can be due to many different reasons, for example, that street trees may not receive enough drainage from paved areas in the model, or tree roots may reach deeper than 1 m, from where they may receive more water if they reach groundwater, which SUEWS can not take into account yet. In order to estimate tree transpiration correctly in Alnus site, a modified simulation with additional water input (0.06 mm h −1 ) to represent the groundwater intake was made. The limit was chosen :: by ::::::::: sensitivity ::::: testing : such that the soil does not dry and limit the modelled transpiration. The run without water input is hereafter called base 270 run and the modified run the final run (See Sect. 3.1.2).

Yasso
Yasso15 (Viskari et al., 2020) is the most recent version of the soil carbon decomposition model Yasso (Tuomi et al., 2009;Liski et al., 2005), where the rate of decomposition depends on climatic conditions and chemical composition of the soil organic matter. The model can be run as an annual or monthly basis. The annual precipitation, air temperature and air temperature 275 amplitude or monthly precipitation and monthly average air temperatures are needed as model drivers. The model simulates the change in carbon stock based on the balance between the decomposition of soil organic matter and possible litter input.

310
The annual estimates were assumed to evenly distribute over the months. The AWEN shares in the root litter were estimated to be as in Akujärvi et al. (2014) (Table 3) and carbon content in the fine root litter 50 %. The run without roots is hereafter called base run and the model run with roots the final run (See Sect. 3.2).

Model evaluation and statistics
The modelled soil moisture from SUEWS was evaluated against observations to examine the simulation of water balance 315 in the model. Additionally, the performance of the surface conductance and photosynthesis models were evaluated against transpiration estimations from sap flow and leaf gas exchange measurements. The evaluation years were 2008-2011 when  Aleksi ::::::: Lehtonen, ::::::: personal :::::::::::: communication most of the measurements were available. Only months from June to August were included in the evaluation, however, in 2008 measurements were available only for July and August.
In order to compare the modelled and observed soil moisture, the modelled soil moisture deficits (∆θ) were changed to soil 320 water contents (SWC). The observed SWC is an average from depths 10 and 30 cm whereas the modelled SWC represents the average from the whole modelling area, excluding soil beneath buildings. The modelled soil depth depends on the surface type varying between 23 cm for paved areas and 1 m for the street trees. Thus, for the comparisons, both observed and modelled SWC have been normalized between 0 (dry soils) and 1 (wet soils) for each year.
In SUEWS the evapotranspiration for the whole simulation area is estimated from the modified Penman-Monteith model Simulated CO 2 uptake by photosynthesis and emissions by respiration were evaluated against leaf-level measurements that were scaled to canopy level for year 2016. These measurements were used for the stomatal conductance model parameter fittings in SUEWS and thus are not an independent data set. However, the comparison was made to show that SUEWS indeed reproduces similar responses to environmental conditions as the estimations from leaf-level measurements.

335
Yasso model simulations were compared with the carbon pool estimates driven from LOI based soil carbon contents. ::: The :::::::: proportion ::: of :::::: carbon :: in ::: the ::: LOI :::: was ::::::: assumed ::: as :::: 0.56 ::::::::::::::::::: (Hoogsteen et al., 2015) : . However, the first measurements point in 2002 is ::: was not used in the model evaluation as it was given to the model. SUEWS can consider increase in tree height and increase of the canopy horizontally through surface cover fractions, but it cannot currently take densification of the canopy into account. This however must be taken into account when calculating the 340 long-term carbon sequestration of street tree plantings. In the calculation of carbon sequestration of the street tree plantings for 2003-2016, the modelled tree gas exchanges were thus scaled with measured leaf area (LA) to obtain the densification of the canopy. The canopy was allowed to grow (densify) between 2002 and 2008 after which the growth was assumed to cease due to regular pruning of the trees. The calculations for annual carbon sequestration and respiration were done based on how much space was allocated to one street tree. The soil respiration was scaled to 25 m 2 area typical for street trees and the trees were 345 scaled to 9.5 and 4.7 m 2 for Tilia and Alnus site, respectively, based on estimations of canopy area from Riikonen et al. (2016).
The soil respiration estimation was an average from the three soil types.

Transpiration
SUEWS was able to simulate the observed diurnal dynamics of tree transpiration at Tilia site (Fig. 4 a). At the same time,

460
In this work, we estimated the CO 2 exchange dynamics in common urban street trees and their growing media using validated models. We found that these ecosystems turned from source to sink of atmospheric carbon on annual level during the first 14 years after soil preparation and plantation of the trees. Cumulatively over the years, these street tree plantings would not become sinks until 30 years after the streets were built :: or :::: even :::: later (Riikonen et al., 2017). Commonly used methods to asses the carbon sequestration of street trees, such as i-Tree, estimate the sink strength with biomass equations and growth rate 465 estimations (Nowak and Crane, 2000). However, these methods are unable to provide high temporal variations. Furthermore, these studies have been focusing mostly on the carbon cycle of trees, leaving the soil carbon out of the estimations. The models used in this study allow to consider temporal variations in urban carbon sequestration and respiration by vegetation and soil, and can take climate and local meteorological conditions into account in their estimations.

Dynamics of tree carbon gas exchange
We found that tree CO 2 exchange varied between days, seasons and years due to changes in environmental factors, tree species and tree size. The diurnal cycle of photosynthesis was mainly driven by the changes in incoming shortwave radiation, limiting the uptake at night time and on cloudy days. Additionally, the decrease in air humidity slightly limited the uptake at day time.
The seasonal variability was driven by variations in incoming shortwave radiation, air temperature and LAI whereas the year-

500
Here, the annual tree respiration varied between 1.2 and 6.8 kg C year −1 per tree and photosynthesis ranged between 2.7 and 13.4 kg C year −1 per tree. In the last simulation year 2016, the net uptakes were 7.0 and 6.2 kg C year −1 per tree for Tilia and Alnus sites, respectively. These estimations are lower than those resulted by other methods used to estimate carbon sequestrated by street trees in Europe. Russo et al. (2014) used models (UFORE and CUFR Tree Carbon Calculator), allometric equations and field data to estimate the average aboveground carbon sequestration of street trees in Bolzano, Italy, ranging from 12.1 to 505 17.4 kg C year −1 per tree. Moreover, in Lisbon, Portugal, the street trees were estimated to sequester 43.1 kg C year −1 per tree (Soares et al., 2011). However, the street trees grew in warmer temperate zone and were probably more mature and therefore, could sequester more carbon than the younger trees examined in this study.
Tree biomass equations have been used to estimate the carbon accumulated to woody biomass, roots and leaves in 2003-2011 for the same street trees as used in this study. Riikonen et al. (2017)  Alnus trees, respectively, was sequestrated during the first 10 years after planting. Correspondingly, 39.4 and 35.9 kg C per tree was estimated to accumulate based on the balance between simulated tree respiration and photosynthesis during the decade.
However, respiration from roots was not taken into account in these simulations, which would decrease the accumulated carbon estimations. Moreover, urban biomass estimations have still uncertainty and Riikonen et al. (2017) noted that the estimation for Tilia trees might be an underestimation. Furthermore, i-Tree model has been used to estimate the carbon sequestration of 515 potential Tilia trees in Helsinki, using weather from Maine, USA (Ariluoma et al., 2021). The sequestration potential in 50 years was at best 1.7 t CO 2 , corresponding on average to 7.6 kg C year −1 per tree. The estimation was possibly overestimating the carbon sequestration potential in Helsinki as the weather from Maine had more precipitation than what has been observed in Helsinki. In addition, how the models handle leaves varies depending on the method, as in these streets, we assume that all the leaves end up out of the simulation area, so their decomposition is not taken into account. Overall, the annual carbon sequestration estimated with i-Tree was close to the estimations for Tilia trees in this study.

SUEWS performance and tree measurements
We found that SUEWS is able to simulate evapotranspiration dynamics correctly even though the study sites greatly differed in soil water availability. It is reported that Alnus glutinosa trees tend to have deep roots that can access groundwater (Claessens et al., 2010) and therefore, trees are not dependent only on precipitation but they can access to deep water sources. Our study 525 supports the phenomenon as the modelled transpiration at the Alnus site notably improved when an external water input was fed into the soil at the same time when without additional water, soil moisture in top layer was simulated well. Therefore, the possible existence of unidentified water pools might complicate further simulations of urban photosynthesis in soils with access to groundwater.
Modeling photosynthesis is a relatively new addition to the SUEWS model (Järvi et al., 2019), combining evapotranspiration 530 and photosynthesis with stomatal opening. Model parameters G 1 − G 6 have been previously fitted against surface conductance values estimated from observed latent and sensible heat fluxes (Järvi et al., 2011;Ward et al., 2016), representing integrated conductance for all surface types. The effect of evaporation is eliminated by doing the parameter fittings only for dry conditions.
These kind of general parameters represent the environmental response functions for all vegetation types compared to the method used in this study, where the parameters represent only street trees. Compared to general parameters derived from 535 eddy covariance measurements from Swindon, England (Ward et al., 2016) (G 2 = 200 W m −2 , G 3 = 0.13, G 4 = 0.7, G 5 = 30 • C, G 6 = 0.05 mm −1 , ∆θ W P = 120 mm), f (∆q) :::::: g(∆q) parameters G 3 and G 4 show significant difference. ∆q seems to be less relevant for street trees, however extreme dry conditions were not reached in the fitting period, which could affect the fitted parameters. The same behaviour was found in Riikonen et al. (2016), where they studied the ∆q relation to sap flow measurements. f (K ↓ ) ::::: g(K ↓ ) is slightly more restricting for street trees than the general parameters. f (T air ) :::::: g(T air ) is the 540 same for the general parameters as for the street trees, because the shape and upper and lower limits are the same. The peak air temperature G 5 does not change as this high temperatures are rarely measured in Helsinki. ∆θ W P is slightly smaller for the Swindon site than what was estimated here. f (∆θ) :::::: g(∆θ) for the general parameters is similar to the Alnus site.
The dependencies of the different trees on K ↓ and T air are very similar whereas clearly different responses on ∆q and ∆θ are seen. ∆q relation to stomatal conductance has already been reported to be smaller for these street trees especially for the 545 Alnus site (Riikonen et al., 2016), whereas, in both sites, the soil moisture is expected to have little effect until significant deficit is reached. Especially on Tilia site, SWC is high and therefore no clear dependence to ∆θ is found. The high soil water availability can also affect the stomatal conductance response to ∆q, as even in dry air conditions, the trees have access to water in soil.
Carbon sequestration and evapotranspiration both depend on the tree leaf stomata control. In this study, leaf-level gas ex-550 change measurements were used to parameterize the stomatal control model in SUEWS, whereas sap flow measurements were used to evaluate the functionality of the model. However, both measuring methods have known uncertainties. The leaf-level photosynthetic responses were not used as such, but were scaled to canopy-level with a forest stand gas exchange model SPP (Mäkelä et al., 2006). The measurements were made manually, so no continuous measurement data were available, but rather continuous photosynthesis data were created separately with SPP. For further research, automatic chambers would be rec-555 ommended to get more realistic environmental response functions. The Granier type heat dissipation method (Granier, 1987;Hölttä et al., 2015) used in this study to measure sap flow and estimate whole-tree transpiration has some uncertainties, caused by method related issues, such as, the sensors respond slowly to the changes in flow rate, and by tree related issues, such as, the water stores in trees itself are utilized (Clearwater et al., 1999;Burgess and Dawson, 2008). These issues in the measurement method lead to time lag between the measured sap flow and the actual tree transpiration, and moreover, with meteorological 560 conditions affecting the transpiration. Riikonen et al. (2016) estimated the time lag for the street trees to range between 30 and 90 min depending on the year. Here, the average of 60 min was used for all cases, which may lead to a slight error. The Tilia trees showed a slight morning maximum in the observations, which might be due to transpiration from internal water reservoirs in the tree trunk. Furthermore, the observed sap flows may not be accurate representation of tree transpiration as the sensor location may not represent the whole tree trunk. However, Riikonen et al. (2016) estimated the possible overestimation to be 565 21 % at the highest. The sap flow values also varied between the measurements years, partly due to meteorological conditions.
In 2010, the sap flow values could be at times twice as high than other years, due to higher air temperature and increased VPD observed that year. However, long-term measurements have some uncertainty, as trees grow, the sensors may be buried more deeply, leading to changes in the flow rates (Moore et al., 2010).

Soil carbon 570
Here, we demonstrated the relative importance of soil carbon in the carbon cycle of street trees. Cities have already used soil carbon models to estimate their soil carbon stocks, but relatively few studies exist about the applicability of these models to urban soils Qian et al., 2003;Trammell et al., 2017). We showed that Yasso soil model is mainly able to simulate the initial decrease in soil carbon pool after planting of trees but there seems to be increasing misfit over the simulation period. The reasons behind remain unsolved in this study but we assume that the differences arises from unknown 575 initial AWENH of the soil substrates, spatially limited sampling of soil carbon pool and possibly overestimated soil moisture on paved systems. Next, we discuss these in detail.
Yasso simulates the decomposition of soil carbon depending on the solubility of the carbon compounds. The used AWENH fractions were based on qualitative description of the soil composed of different organic materials (Riikonen et al., 2017). Their proportions in the mixture, such as the share of peat, were unclear and therefore leading to uncertainty in the initial AWENH.

580
Further, setting of these initial fractions had high impact to the model results. For example, bark was ignored in the soil 2 as we assumed the share of it to be minor but the lack of it in the model runs might explain some of the underestimation in comparison with the measurements. On the other hand, the soil measurements have also a large uncertainty, as they were spatially measured only from two locations even though vertically from multiple depths. The samples were taken app. 2-3 meters from trees whereas we simulated the whole soil volume where the distance especially between the Tilia trees were 585 notably higher. According to the measurements, the soil carbon pool was stable or even increasing 7-15 years after the planting.
Such finding in nature can result only from notable litter input, notable decrease in decomposition of organic matter or most likely combination on those two.
In the simulations, the fine roots had a minor impact in the soil carbon stock as the study trees were still young and thus the root biomass low. As the fine roots were assumed to be evenly spread in the model runs, the simulated fine root litter 590 input and decomposition represent an average from the whole soil volume. In nature, fine roots probably are denser close to the trees i.e. at the sampling locations than further away. Besides, high root mass decreases soil moisture and therefore also the decomposition rate. Higher root litter input and decreased decomposition rate at the sampling locations could cause the observed underestimation in the model simulation in the long run. With current knowledge, quantifying the fine root litter input is difficult as the amount and the turnover rate are still unknown especially in urban areas. The turnover rates have been 595 estimated to vary between one to nine years in forest ecosystems (Matamala et al., 2003) and therefore, future estimations would benefit from studies revealing more accurate root lifetime in urban ecosystems.
The forcing meteorology for Yasso was generated from the 2-m local air temperature simulated by SUEWS to get the local temperatures. Local temperatures vary spatially in urban areas, because build environments tend to warm more, and vegetative environments cool down because of evapotranspiration (Oke, 1982). However, the study sites in Viikki are similar to the 600 measurement site in Kumpula, so the difference between measured air temperature from Kumpula and the modelled local temperatures remained small. In theory, increased soil temperature would lead into increased decomposition of soil organic matter. At the same time, the role of soil moisture is more complex as the decomposition is decreased both in high and low soil moisture conditions (Moyano et al., 2012). Yasso soil carbon model is driven by precipitation but in these kind of paved systems, the soil moisture might be lower than expected as notable part of the water never enters the soil volume. Changing the 605 drivers belowground would probably lead to improved model performance but on the other hand, observations of soil moisture and temperature are rare. Nevertheless, further efforts are needed in studying the role of soil moisture in the decomposition of urban soil carbon pool.
The estimated SOC densities in 2016 ranged from 1.7 to 5.7 kg C m −2 , mostly depending on the soil type. Soils 1 and 2 reached similar SOC in 2016 (4.5-5.7 kg C m −2 ) even though the initial SOC was almost twice as high for soil 2. These street 610 soil estimates are much lower than those previously measured in the parks of city of Helsinki (10.4 kg C m −2 ; Lindén et al. (2020)) and even lower than forest soils in Finland (6.3 kg C m −2 ; Liski et al. (2006)). However, direct comparison between SOC estimations can be challenging due to different soil types, vegetation and age. On the other hand, limited amount of new carbon enters the soil of these streets, which may explain part of the difference. The time of construction or renovation of the park had a major impact on SOC (Scharenbroch et al., 2005;Setälä et al., 2016), as also Lindén et al. (2020) found in the parks 615 of the city of Helsinki where SOC accumulation stabilized after 50 years. The effect of construction of the streets is clearly seen also from the street SOC estimations. The estimations show a decrease of SOC during the study period as the root litter input is not enough to stabilize the decomposition of SOC. Compared to other urban soil studies outside of Finland, the average SOC storage in greenspace was 9.9 kg C m −2 in Leicester, UK (Edmondson et al., 2014), which show similar estimates as parks in Helsinki. However, in warmer climates, the estimated SOC values have been lower. In Singapore, under turfgrass, the The maximum monthly soil respiration estimates varied between 0.08 and 0.26 kg C m −2 month −1 after the high initial carbon loss, which corresponds to 2.5 and 8.1 µmol CO 2 m −2 s −1 . These estimates compare reasonably well to previous research on soil respiration in urban areas. In greater Boston's residential areas (Decina et al., 2016), the soil respiration 625 of urban forests, lawns, and landscaped cover types were 2.6, 4.5, and 6.7 µmol CO 2 m −2 s −1 , respectively. In Singapore, turfgrass soil respiration was measured to be an average 2.4 µmol CO 2 m −2 s −1 and highest mean value of 4.4 µmol CO 2 m −2 s −1 (Velasco et al., 2021). No seasonal trends were observed as the tropical weather is favourable to constant soil respiration.

Conclusions
Quantification of the carbon cycle of urban nature is needed in planning of green areas, carbon neutrality assessments, and urban climate studies. In this study, an urban land surface model SUEWS and soil carbon model Yasso were evaluated and used to estimate the carbon sequestration of street trees and soil in Helsinki, Finland. The compensation point when street tree plantings turn from annual source to sink was achieved after 14 years of the planting of the street trees :: but ::: as ::: the :::: setup ::::: does 635 ::: not ::::::: represent ::: the :::: full ::::: variety ::: of ::: soil ::::::: growing :::::::: mediums, ::::::: planting :::::::: densities ::: and ::::: plant ::::: types, :::: these :::::: results :::::: should :: be :::::::: upscaled :::: with :::::: caution. The annual carbon sequestration depended on environmental factors such as air temperature and humidity indicating the need for modelling techniques allowing to take appropriately the local climate conditions into account. Yasso and SUEWS are able to simulate the carbon cycle of street tree plantings as shown against observed soil moisture, sap flow and soil carbon from two street tree sites, but the used substrates vary widely and the indeterminable soil properties cause great uncertainty in 640 estimations of the longevity of soil organic carbon. However, Yasso developed for a non-urban area performs reasonably but further studies especially on root litter input and on the role of soil moisture in the decomposition process would decrease the model uncertainties.
Code and data availability. The data sets are openly available at Havu et al. (2022), including the model runs for SUEWS and Yasso, the fittings of the environmental response functions, the gap filling of the meteorological measurements, and codes to reproduce the figures.
Precipitation was gap filled with multiple measurement devices and locations. The order of measurements used in the gap filling was: hourly PWD (since 2014), hourly SYNOP from Kumpula (since 2006), hourly Ott (since summer 2002), daily SYNOP from Kumpula (since 2006) and daily SYNOP from the airport (since 2002). Daily SYNOP data were divided evenly over the day to get the hourly values.
Temperature, wind speed, wind direction and incoming radiation were measured at tower, rooftop and airport, whereas, 655 relative humidity and air pressure only from rooftop and airport. Primary measurements were either the tower and rooftop measurements which were gap filled using airport measurements using a linear correlation. Rest of the missing hours were gap filled by linear interpolation if less than 5 hours were missing (2 hours for radiation), or with the average of the same hour from the previous day and the following day, if less than day was missing. If more than day was missing, the values were filled by calculating the average for the same hour of three previous days and three following days. forestry, 103, 411-416, 2005.