One Ecosystem :
Research Article
|
Corresponding author: David N. Barton (david.barton@nina.no)
Academic editor: Soile Oinonen
Received: 07 Apr 2022 | Accepted: 21 Jan 2023 | Published: 22 Feb 2023
© 2023 David N. Barton
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Citation:
Barton DN (2023) Value ‘generalisation’ in ecosystem accounting - using Bayesian networks to infer the asset value of regulating services for urban trees in Oslo. One Ecosystem 8: e85021. https://doi.org/10.3897/oneeco.8.e85021
|
|
In this paper, we demonstrate value generalisation from a sample of ecosystem assets – municipally managed trees - to all tree assets within an urban ecosystem accounting area. A Bayesian network model is used to machine-learn non-parametric correlation patterns between biophysical site condition variables and output variables of an ecosystem service model – here iTree Eco for modelling the regulating services of urban forests. The paper also demonstrates the use of spatial Bayesian network modelling to quantify the reliability of value generalisation for accounting purposes. Value generalisation entails inferring ecosystem service values for all locations in an ecosystem accounting area, where the accounting practitioner has less information about the asset and its context, than in an available sample of managed sites within the accounting area. The modelling is carried out as a “proof-of-principle” of potential value generalisation and uncertainty analysis methods for ecosystem accounting. It does not cover all regulating ecosystem services of urban forests, nor cultural services. While noting that wide confidence intervals for generalised values pose challenges for using monetary accounts for the accounting purpose of change detection, we find that tree-specific asset valuation is possible in an urban accounting setting. Our findings serve the purpose of raising awareness about asset values of urban green infrastructure, to bring them more on a par with grey infrastructure in urban planning. We also argue that the reliability of the asset value of individual trees is also good enough to be used for non-accounting purposes, such as municipal tree damage assessments.
Ecosystem accounting, SEEA EA, value transfer, urban ecosystem accounts, Bayesian belief network, BBN, uncertainty, monetary valuation, ecosystem services
The SEEA EA (
Asset valuation is a an estabilished aspect of the economic analysis of renewable and non-renewable resources (e.g.
Our study focuses on the value of regulating services from trees – for which the iTree Eco model is a well-established model estimating annual flows of regulating ecosystem services (
EU experimental ecosystem accounts (
Urban ecosystem accounts are recognised as one of four key thematic accounts in the SEEA EA (
A challenge for ecosystem accounting is assessing the uncertainty in value estimates generated from using ecosystem service valuation from a few study sites to policy analysis or accounting at other sites, referred to as 'benefit transfer' (
The approach is built around the iTree Eco for estimating the value of regulating service of individual trees and the Hugin QGIS plugin application to infer asset values to the whole urban planning area, while accounting for spatial variation in the condition of the trees’ location. Fig.
Modelling chain for compiling asset accounts of the regulating ecosystem services from trees in Oslo’s built zone, including carbon sequestration, avoidance of stormwater run-off, removal of air pollution and building energy savings.
The condition of tree assets was determined, based on tree canopy cover and height (Fig.
The tree asset extent-condition account considers tree canopy area and canopy height. Source: adapted from https://nina.earthengine.app/view/urban-nature-atlas and https://transect.org/.
The regulating ecosystem service values of an average tree computed using i-Tree Eco by
A Bayesian network (
Machine-learned Bayesian belief network for generalising i-Tree Eco calculated tree asset values for regulating services from municipal trees to all city trees. The graphical user interface displays the nodes, causal links and the non-parametric discrete distributions of each node. Source: BBN model used for value transfer is available as a Hugin Expert file in Suppl. material
The Bayesian network model is then used to infer tree specific regulating services for all 406000 tree canopy polygons identified by
Fig.
As mentioned, Hugin QGIS plugin applies the model to every accounting unit – in our case a tree canopy polygon. In a general ecosystem accounting context, the inference would be done for each basic spatial units of an ecosystem as pixels or polygons in an accounting area.
The Hugin Expert QGIS plugin outputs different statistical parameters calculated on the non-parametric distribution computed for the spatial accounting unit, such as mean, median, st.dev. and 10-90% confidence limits. Reporting of aggregate asset values for the city serve the purpose of awareness-raising about the natural capital value. We discuss a possible accounting table presentation of monetary asset value of urban trees incorporating uncertainty about asset values due to value generalisation. To demonstrate an application of ecosystem accounting to urban policy and planning, we also assessed the change in tree asset values for a particular regulation plan area experiencing sub-urban infill.
Total asset value of urban trees for awareness raising
First, we report on the aggregate asset values of tree canopy for the city’s built zone. Fig.
Value generalisation from a modelled sample of trees managed by the municipality (left panel) to tree assets both on public and private land in the Oslo accounting area (right panel) using the spatially distributed Bayesian network in Figure 3. Map data: http://urban.nina.no/maps/396/view.
Fig.
Asset value of regulating services generalised to all city trees in Oslo’s built zone (2014).
Applying the asset account to policy assessment
Ecosystem accounts can help inform planners on the impact of recent urbanisation on trees. We use the change in the tree asset account to demonstrate assessment of urban tree conservation within an urban regulation planning area, particularly in suburban infill of gardens. Oslo's Small House Regulation Plan (SHRP) covers an area of about 3000 ha and 28,000 properties, composed mainly of detached housing with gardens. The SHRP requires, inter alia, a minimum 65% of a property's area to be free from terrain modification and permits for felling large trees with Circumference at Breast Height > 90 cm. Fig.
Extent-condition account for tree assets in Oslo’s Small House Regulation Plan area (2011 to 2017). Source: derived from
This estimated change is less than the Lidar classification and tree segmentation model estimation error (
The loss of larger trees and increase in smaller trees and in the regulation plan area raises an urban planning question relevant for ecosystem accounting. Is the change in height distribution (condition) of urban trees significant for the supply of ecosystem services and more specifically regulating services? As regulating services of trees are primarily determined by a tree’s leaf area index – proxied by crown volume as a function of crown area and height – it is expected that the large increase in small tree planting may have compensated for the loss of tall trees – at the aggregate level of the regulation plan.
Starting from the physical extent-condition account (Fig.
Monetary Asset Account for Urban Trees in the Small House Regulation Plan area in Oslo. Source: derived from
The asset accounts reflect the height class distribution of gains and losses observed in the extent-condition account. Overall, the loss of ecosystem service asset value in the smaller number of larger trees is compensated by the large increase in the canopy area of small trees. The net loss in asset value of about NOK 21 million (approximately -2% over the period) is not significant relative to the combined uncertainty in the tree segmentation modelling (
The Lidar-based mapping of tree canopy underpinning the asset accounts allows for identification of the spatial distribution of the gains and losses for the services. While there is no net aggregate loss in regulating services, Fig.
In this section, we first discuss some sources of uncertainty in the asset accounts for urban trees. In light of this uncertainty, we then discuss policy applications of the tree asset accounts.
Temporal changes in ecosystem service supply due to future changes in urban ecosystem condition. Regulating ecosystem service supply is expected to change over time with the value of stormwater regulation services increasing as climate change brings more frequent and more intense rainfall episodes. Recent work has suggested that the full cost savings to stormwater treatment considering also future climate change projections may be roughly 5 times higher than considered by
Sensitivity of asset value to Tree management assumptions. Ecosystem asset values depend on assumptions about sustainability of management practices and asset life (NCAVES and MAIA 2022). In the present study, individual tree asset values were calculated, based on the discounting of future flows of benefits from trees in their present sizes, using estimates of expected lifetimes for tree species under urban conditions in Oslo (
Aggregation errors. A common practice in unit-value transfers is to transfer a summary statistic of central tendency, such as the mean or median (
Generalisation/value transfer errors. Other uncertainties are present in the asset value generalisation using the BN QGIS model. The asset value all trees was inferred using only observed canopy area and tree height, proxies for the condition indicators which predict regulating services more directly (tree species, total leaf area, stem diameter). The Bayesian network emulates all the variance incorporated in the i-Tree Eco model due to variance of tree characteristics in the input data, as well as variance in the model simulation. Inspecting Fig.
Diagnostic of asset value per canopy trees using the Bayesian Network emulation of the iTree Eco model. In the node “Asset value per canopy area”, we selected the lowest asset value category (0-50 NOK/m2). The network nodes show the changes in likelihoods of input variables relative to the likelihood distribution for all trees. This uses the inductive reasoning feature of BNs made possible by Bayes theorem.
We use the Bayesian network to diagnose the characteristics of the lowest m2 asset value class relative to the whole sample used to estimate the iTree Eco model of municipal trees. The lowest asset value per canopy area class (0-50 NOK/m2, upper right hand distribution) corresponds to the largest crown area, diameter and height classes of trees (distributions in the centre of the figure above) and the lower air pollution zone. The Tilia tree genus is less likely to have low asset values relative to other tree species because they tend to have smaller crown area than other species in Oslo. Different tree species have different leaf area indices, meaning differences in effectiveness per unit canopy area. Lidar data on all tree canopies in the city do not observe tree species, thus introducing uncertainty relative to a traditional iTree Eco modelling approach which is based on ground-based sampling (
Spatially correlated urban condition variables. We generated our prediction model on roughly 16,000 municipally managed trees, of a total of roughly 30,000, as compared to 406,000 identified canopies in the built zone as a whole. Asset value per m2 of canopy are lowest for large canopies; the largest canopies are located in the lowest air pollution zones (Fig.
Relevance of asset accounts for urban policy and planning. There are a number of potential uses of tree asset valuation. Policy uptake depends, inter alia, on sufficient accuracy of the value generalisation for the particular policy purpose (
We finish with some examples of purposes of urban tree asset valuation illustrating a broadly increasing order of accuracy requirements (Fig.
Required accuracy of ecosystem service valuation depends, inter alia, on the policy analysis purpose. Error bars in yellow illustrate increasing requirements for accuracy (precision and reliability) across different policy questions from left to right. Source: adapted from
· Awareness raising. The total asset value of regulating services from urban trees of 5.4 billion NOK is a ‘big number’ in absolute terms, raising awareness about the value of trees as green infrastructure providing municipal utilities, similarly to other municipal utilities infrastructure value.
· Change assessment. The physical extent-condition and monetary asset accounts provide time-series indicators of tree stocks and the value of benefits. The tables provide a summary tracking of the ‘sustainability’ of urban intensification in Oslo over time. Accounting statistics can be extracted for planning areas of particular interest. If uncertainty is accounted for, it is also possible to assess whether observed changes during the accounting period are statistically significant and, if so, whether they are also politically important. The spatial resolution of ecosystem accounts also means that change assessment can reported for different sectors, such as private land and publicly managed municipal land and in relation to spatial distribution of different cultural, social and economic demographics (see, for example,
· Priority-setting. Tree asset valuation highlights the value of trees to developers, urban planners, landowners and the public. This potentially makes the economic case for their preservation and for investment in additional tree planting throughout the city. Nevertheless, the monetary values of green infrastructure, such as trees, cannot compete with urban development values in dense urban areas where trees compete for space with building stock. How do tree asset values compare to other property assets values? The asset value of all property in Norway is regularly computed (
· Instrument design. Where urban density allows for nature-based solutions, the tree asset valuation focused on regulating services only makes the case that new urban developments can compensate for the loss of large established tree canopy with planting new trees, provided that compensation is in terms of canopy volume, rather than ‘a tree-for-tree’. It must be noted, however, that tall trees provide cultural services in terms of green views of sites which cannot be compensated on a m2 canopy-basis by small trees. Tall trees are also older and are likely to provide more habitat niches for a variety of species. However, assessment of ecosystem services of trees has, to date, had limited influence on design of Oslo’s Blue-Green Factor instrument (
Physical extent-condition accounts revealed a relative loss of large trees in Oslo’s Small House Regulation Plan area. It showed to be lacking effectiveness of the current permitting requirements for felling large trees, at least relative to an objective of halting the loss of existing tree canopy in suburban areas. Mapping of canopy loss areas may, in future, contribute to both strengthening of conservation incentives and targeting of tree planting in ‘net canopy loss’ neighbourhoods. Reporting ecosystem accounts in terms of change maps will contribute to policy design that considers ‘environmental justice’ of the spatial distributional impacts on different urban populations (
· Damage compensation. Finally, the highest accuracy may be required of ecosystem asset valuation as a basis for damage compensation, especially if used in legal cases. Tree asset valuation by Cimburova and Barton (2020) has been used to argue for the inclusion of damage compensation for regulating services in a Norwegian Standard for the Valuation of Trees (VAT), extending the VAT system used in Denmark (
In this paper, we have demonstrate an ecosystem asset valuation approach using a combination of existing ground-truthed and high resolution remotely-sensed data appropriate for an urban ecosystem accounting setting. We used urban trees as a ‘proof-of-principle’ in an urban ecosystem setting requiring high accuracy to be relevant for planning and policy. The paper demonstrates how machine-learning and Bayesian inference can be used to generalise ecosystem service asset values to an accounting area. The integrated approach avoids time and costs associated with extensive additional surveying and inventorying of urban trees while keeping track of uncertainty. Considering uncertainty, the paper discussed different policy and planning applications in Oslo. The rapid low-cost estimation of regulating services from urban trees and associated asset-based ecosystem accounting may be relevant for other cities. The case study from Oslo also provides an example of so-called 'value generalization' from a sample of ecosystem assets to all ecosystem assets of the accounting area.
Source: BBN model used for value transfer is available as a Hugin Expert file. Model nodes are documented in supplementary material to Cimburova and Barton 2020, available here https://ars.els-cdn.com/content/image/1-s2.0-S161886672030618X-mmc2.docx