Appendix B. Planting Data Preparation

We received spatial data from the delivery partners, normally in the form of planting boundary polygons. We also received accompanying return forms containing official site-level information such as declared area, number of trees planted, planting type, etc. These data exhibited strong variations in terms of format, digitising standards and completeness, making the generation of the masterfile for analysis a complex task. We document here the key methods and assumptions we used to achieve a complete and representative view of the programme.

Data issues

Point to polygon

Spatial data was not always available. When only a grid reference was provided (2.9 % of total planting), we converted this to a spatial point and buffered this point by the required radius to match the declared planting area. The buffered polygons were only allowed to “grow” on land suitable for planting (i.e. we masked incompatible land uses such as buildings, roads and water). The polygons thus generated do not represent the exact location and configuration of the planting. However, this method allows the sites to be located on realistic land use, achieve their declared area, and be in the vicinity of their true location, so that spatial intersections with datasets used for modelling could be trusted.

Large areas of sparse planting

For some schemes (LATF and UTCF), polygons were not always representative of planting given the nature of the schemes (individual trees or clusters of trees). Some of the polygons were very extensive in size, covering the footprint of whole streets or neighbourhoods, but representing sometimes only tens of trees planted. Retaining these large polygons and encoding them as TOW would have greatly overestimated benefits from these schemes, as the urban baseline would seem completely replaced by trees post-planting. We therefore masked these polygons to represent “plantable” areas (as above), then generated a number of random points to match the declared number of trees within this footprint. These were then buffered by the radius required to achieve the total declared area for the scheme. While not representing the real location of the trees, we are confident that this is more appropriate to calculate uplift for the various metrics and avoids overestimation of benefits. This represented less than 1% of total planting.

Ways of capturing sparse planting

We observed that for planting types like street trees or silvopasture, there may be multiple ways of digitally mapping a scheme. For example, wood pasture can either be mapped as distinct patches of grassland and small clusters of trees, or as one large boundary described as wood pasture. These standards varied across partners and do not always match the level of detail of our baseline mapping, and therefore we attach less confidence to the estimation of benefits from TOW.

Mismatch between spatial and form data

We used Forestry Commission’s official statistics (see next section) as a benchmark to assess completeness of our gathered data. Most partners provided their FC return forms alongside the spatial data. However, in many cases, we found it difficult or impossible to recombine the two, due the absence of a common unique identifier, and had to manually reconcile hundreds of sites. The one key recommendation we would give to all delivery partners is to develop a robust system for assigning unique ID codes to sites. This ID should be short, follow a consistent naming pattern (e.g. Organisation_Year_Number), and appear in all records related to that site (spreadsheets, spatial data). This allows automated data joins and facilitates queries without ambiguity.

A project on data standards is currently being carried out within Defra, and observations from this project have fed into it. We recommended to gather lessons learned by the planting partners who have already developed robust data management pipelines and digitising standards for different woodland types (e.g. England’s Community Forests). These lessons should be shared widely to encourage uptake of good-practice, consistent management and reporting standards.

Imprecision in GIS drawings

Some sites were digitised quite loosely, or encroached on neighbouring features that the actual planting would not reach (e.g. roads). The greatest issue we had is when sites are designed to infill or expand an existing area of trees. Because the habitat baseline is so detailed with the inclusion of the TOW layer (see Appendix A), sometimes NCF schemes appeared to “erase” existing trees. In such situations, especially when the target habitat was not very dense (e.g. wood pasture), our calculation sometimes returned “negative” benefits. In these cases, we capped the benefit values at 0, assuming that no trees were removed to design a scheme. For woodland creation scheme, the presence of these baseline trees is an asset to the modelling, as it avoids over-estimating uplift, when pre-planting capacity was likely relatively good to begin with.

Reconciliation with official statistics

The majority of discrepancies (Table 2.1) could be explained by discussions with partners and Forestry Commission, such as:

  • Not all planting was digitised or had a grid reference in early years of the programme (not included in our masterfile);
  • Not all data from the latest planting season was ready in time for inclusion in the masterfile;
  • Official statistics include some amount of non-NCF planting - our dataset considers NCF planting only;
  • Some schemes were not completed at time of reporting official statistics, but were confirmed and provided to us at a later date;
  • Uncertainty as to how agroforestry schemes feed into official statistics;
  • GIS boundaries are often digitised loosely around the planting, so calculated area is generally higher;
  • Variation in digitising standards for sparse planting: some will capture trees outside woodland as a large field of wood pasture, with calculated area far greater than declared area. The indicators account for this by setting a different habitat type.

Recommendations

Given that the key challenge for this project was to reconcile planting data submitted by delivery partners with centrally held information (such as FC Official Statistics), we recommend that Forestry Commission and Defra work more closely with delivery partners so that spatial data reconciles to the official statistics submissions. Trees outside woodland seem especially inconsistent in their digitising and reporting, and partners feel they “slip through the net” and are not accounted for properly.

A project on data standards is currently being carried out within Defra, and observations from this project have fed into it. We would advise looking up to planting partners who have already developed digitising standards (e.g. England’s Community Forests) and encourage uptake of the key principles from all partners. This could be done through information sessions and technical guides.

Using consistent, unique identifiers

Our most important recommendation would be to ensure that all sites are given unique identifiers that follow a predictable pattern recognisable within the organisation. For instance, some Community Forests use alpha-numeric codes made of 3 letters for the local authority and 3 digits for the site number, e.g. CHE_074. This is robust and much less prone to errors and typos than informal codes or free text. We had several hundreds of mismatches which we had to reconcile manually, e.g. matching a “Simpson Road Gardens” polygon to a “Simpson Rd” spreadsheet entry.

Not all delivery partners have dedicated GIS officers who have the expertise required to validate spatial data against collated statistics. In large organisations where such support exists, we recommend that one person is in charge of reviewing submissions and statistics from all woodland creation officers to ensure that the return form going out to Forestry Commission is internally reconciled. Where the support is not available, it could perhaps be sourced from another delivery partner (short contract), or the validation could be conducted on the receiving end (FC).

Another issue we had was around licencing restrictions on some of the spatial data. Current agreements sometimes restrict the use and sharing of the data, which creates a risk that some benefits may not be recognised. In the end, these issues were resolved with special licencing agreements, but caused severe delays. To future-proof the use of spatial data in similar evaluations, we would recommend that data sharing agreements with landowners have special provisions facilitating the sharing of spatial information (with appropriate restrictions around personally identifiable information such as address, business name or landowner name, etc, but not on the boundaries themselves).