Section 1 Introduction
This report describes work carried out on the project “Improving Land Use Change Tracking in the UK Greenhouse Gas Inventory” for the Department for Business, Energy & Industrial Strategy (reference TRN 2384/05/2020). The aim of the project was to make improved estimates of land-use change in the UK, using multiple sources of data, using a Bayesian data assimilation approach. Two previous reports describe the background to the project and results from Work Package (WP) A (Levy et al. 2020), using pre-existing data sets, and work developing new data sets based on Earth Observation in WP-B (Rowland et al. 2021). This report describes subsequent work which focussed on assessing the uncertainties in data sets, and incorporating these in the data assimilation procedure. Specifically, we aimed to:
- quantify random uncertainty in the different data sources;
- quantify systematic uncertainty, i.e. biases which may make a data source under- or over-estimate land-use change;
- represent the uncertainty associated with the different sampling frequencies of the data sources (e.g. annual versus decadal surveys);
- handling the uncertainty associated with the different temporal coverage of the data sources (avoiding step changes when data coverage starts or stops);
- incorporate some additional data sets which were not previously available because of data access or processing time constraints;
- improve the representation of the rotational change between crop and grassland.
The above were incorporated in the data assimilation procedure, and results produced for each of the Devolved Administrations (DAs) of the UK. Land-use change on mineral soil and organic soil was estimated separately in each DA.
The remainder of this report describes these tasks and the resulting estimates of land-use change produced after their inclusion.
The first section on [quantifying uncertainty][Quantifying Uncertainty in Land-Use Data Sources] describes how uncertainty is represented, and estimates random and systematic uncertainty by comparison with a reference data set.
The next section describes an alternative method for systematic uncertainty based on the length of the interval between surveys. As a third method, we then look at how errors in map classification propagate into errors in estimates of land-use change.
Next, we describe a method for more accurately representing the frequency of rotational land-use change, using the idea of “life tables” borrowed from polulation modelling.
The previous reports give detailed descriptions of the methods. However, here we reproduce the basic rationale and approach of the project for background.
1.1 Tracking land-use change
The tracking of land use and land-use change is fundamental to producing accurate and consistent greenhouse gas inventories (GHGI) for the Land Use, Land-Use Change and Forestry (LULUCF) sector. This is necessary to meet the international requirements of the Kyoto Protocol to the UN Framework Convention on Climate Change (UNFCCC) and the Paris Agreement and the national requirements of the UK’s Climate Change Act and related legislation within the UK’s Devolved Administrations.
The estimation of land-use change in the current UK GHGI is based on a combination of infrequent CEH Countryside Surveys and afforestation/deforestation statistics from the GB Forestry Commission. It uses Approach 2 (non-spatial land-use change matrices) as described in the KP Guidance. However, this matrix-based approach, and its implementation in the UK, have some important limitations. Firstly, the non-spatial matrix-based approach is insufficient for tracking annual land-use change: the matrices have no time dimension, and are defined independently each year. There is therefore no possibility of representing a sequence of land-use on the same parcel of land (such as afforestation followed by deforestation, or crop-pasture rotations). Secondly, the data used to estimate these matrices in the UK are rather limited. The CEH Countryside Surveys were only carried out approximately once per decade, and whilst the geographical extent was very wide, the actual ground area surveyed was small as a fraction of the total UK area. The afforestation/deforestation statistics from the Forestry Commission have good national coverage (excluding Northern Ireland) but do not contain any information on the spatial location or land use prior to afforestation or following deforestation.
In October 2019, the UNFCCC Expert Review of the UK 1990-2017 GHG inventory raised concerns in relation to the reporting requirements of the second commitment period of the Kyoto protocol. They questioned whether the current approach is appropriate for the identification and tracking of lands where the elected Article 3.4 activities occur (i.e. Cropland Management, Grazing Land Management and Wetland Drainage and Rewetting). They recommended that the UK explore how to make the best possible use of available data and prepare and implement a work-plan to enable the use of these data. The UK has already explored several approaches to land use tracking, including a data assimilation approach to integrate available land-use data into land-use vectors, which was successfully piloted in Scotland (Levy et al. 2018). This project builds on that approach to assess gross land-use change, and land-use history for the whole of the UK from 1990 to 2019.
As well as improving accuracy of the GHGI, a time series of spatially explicit land-use change would enable better tracking of mitigation activities and improve baseline data for scenario modelling. These baseline data are needed for understanding the potential of land-based mitigation and adaptation options. The government’s ambitions for Net Zero by 2050 or sooner means that the LULUCF sector will have an increasingly critical role in the UK’s overall GHG balance. This kind of scenario modelling will become very important to inform the setting of future carbon budgets and monitor progress towards the UK’s legal obligations to GHG emissions reductions. An accurate spatio-temporal land-use change data set would be useful to other stakeholders and UK government departments. For example, from the perspectives of biodiversity conservation, air quality, or ecosystem services, there are clear applications of these data for understanding and tracking the effects of land use.
1.2 Approach
If we had reliable maps of land use each year, we could infer land-use change by difference. However, even with advances in satellite sensors, GIS and spatial data handling, the accuracy of change detection from EO-based products is generally too poor to do this; the different EO products are inconsistent (with each other, and with themselves over time), irregular, and become more infrequent as we go back in time. Change is more reliably detected by repeat gound-based surveys, but these have other short-comings. For example, the annual June Agricultural Census gives a long record of areas in different land uses, but does not provide spatial data, or any information on gross change (i.e. what land uses have changed to which other land uses). The CEH Countryside Survey did provide spatial data with gross change, but without complete coverage, and only at infrequent intervals.
In light of the above, some data assimilation method, which combines the spatio-temporal data with non-spatial repeat survey data, would appear to provide a solution. To this end, we previously developed a methodology using a Bayesian data assimilation approach, and this has been applied successfully to Scotland (Levy et al. 2018). This method allowed for the use of a wider range of data types, including high-resolution spatial data, and combined them in a mathematically coherent way. Importantly, the method produced the appropriate data structure needed for modelling the effects of land-use change on GHG emissions - the set of unique land-use vectors (i.e. unique sequences of land use, or land-use histories) and their associated areas. An important feature is that the uncertainty in land-use change can be easily propagated to provide the uncertainty in GHG emissions, because the procedure explicitly handles the distribution of plausible vectors of land-use change. The approach provides a general framework for combining multiple disparate data sources with a simple model which describes how these data sources inter-relate. This allows us to constrain estimates of gross land-use change with reliable national-scale census data, whilst retaining the detailed spatial information available from several other sources. Here, we apply this methodology to improve and update the tracking of land-use change for the UK. Our aim was to apply a Bayesian approach to make spatially- and temporally-explicit estimates of land-use change in the UK, using multiple sources of data.
All the code is written in R using the “literate programming” paradigm implemented with Rmarkdown, which combines the source code, text/graphical output, documentation, and report text within the same document. This ensures integrity of documentation, code and corresponding outputs. All the Rmarkdown files are held in a GitHub repository, for version control and wider accessibility. The documentation is rendered using bookdown and made publicly available as a web site via GitHub Pages. This documentation describes the data processing workflow so as to make it reproducible.