Land use change - a headache for greenhouse gas emission estimates
Every year, countries have to report their greenhouse gas emissions under the UN's Framework Convention on Climate Change. This is done via a set of sectoral inventories using internationally agreed methodologies, and it is the "Land use, land use change and forestry" (LULUCF) sector that often causes the most headaches.
Land-use change is difficult to track, as detailed on-the-ground surveys are too expensive to undertake every year, even for a relatively small sample of sites. For the UK, a detailed land cover and land use mapping is only performed every 7-10 years. The emission inventory for the period 1990 - present is recalculated every year, however, and the final emissions estimates from these calculations are very sensitive to data availability. So when new latest survey data become available, they can significantly alter the previous year’s emissions estimates for the period, which were based on data extrapolated from the preceding survey. This is understandably frustrating for scientists and policymakers.
In MELODIES, we are responding to this challenge by deriving a new annual land cover change dataset for the UK, using openly-available Earth Observation (EO) data at 500m resolution. The dataset builds on the land cover product developed earlier in the project and the UK’s existing land cover mapping framework, and is a collaboration between EO experts at the University of Reading, and the inventory and land cover mapping experts at the NERC Centre for Ecology and Hydrology.
We assess the likelihood of a change having occurred at a particular location by considering how plausible all potential changes are. For example, we know that a pixel is very unlikely to go from urban to mixed forest to grassland in consecutive years, whereas changes from grassland to arable and back again (crop/grass rotations) are much more likely. This "prior knowledge" can be drawn from many sources, such as the Countryside Survey, settlement data (to give information on urban areas), forestry surveys, expert judgment and so on. A significant advantage of the approach taken in our MELODIES processing chain is that this prior information can be changed, updated or supplemented at any time to give a better estimate of past change. This sort of flexibility and scope for tailoring has been lacking in previous EO-based data products.
Land cover map developed in MELODIES. See previous blog.
At an early meeting with potential users (mainly from relevant UK government departments such as Defra and DECC, but also organisations involved in the compilation of the UK emissions inventories), we discussed how they would like to interactively explore the new MELODIES datasets. In particular, users were interested in being able to translate between one land cover classification and another; for instance, from the 23 classes of the UK detailed land use survey and the six classes for LULUCF reporting. What difference would it make to the UK-wide statistics to map pixels identified as "mixed grassland/arable" all to arable, or all to grassland, for instance? The capability to do this remapping on-the-fly has now been implemented in the MELODIES portal, and was demonstrated at a workshop in Reading in February.
In our follow-up workshop, the response to the datasets and portal has been positive, but cautious. Statements like "It looks a massive improvement over what we've got" (Defra); "I can see lots of benefits to moving to this approach" (DECC) were always followed by a "but" that emphasised the challenges of getting any promising new data source approved for use in the complex reporting process.
It is critical that any new dataset or method is compliant with the reporting guidelines published by the Intergovernmental Panel on Climate Change, which are used by the UNFCCC. The current guidelines are positive about the use of remotely-sensed data in the inventory, saying: "The strengths of remote sensing come from its ability to provide spatially-explicit information and repeated coverage, including the possibility of covering large and/or remote areas that are difficult to access otherwise. Archives of past remote sensing data also span several decades and can therefore be used to reconstruct past time-series of land cover and land use. The challenge of remote sensing is related to the problem of interpretation: the images need to be translated into meaningful information on land cover and land use."
The MELODIES method is a novel way of interpreting the observations that are not currently used in the inventory compilation process, and the important thing (from the point of view of the UN) is that the emissions estimates from different countries can be compared. Use of a new methodology by one country sets a precedent for others, and there is inevitably a lengthy process involved before a new approach is accepted and can become mainstream.
If we are to convince the UK emissions inventory experts and policymakers that our new MELODIES data are fit to be included in the inventory compilation process, we first need to provide appropriate evidence to an expert review panel. The data would then need to be trialled to evaluate the impact on the inventory. If successful, the data could then be used "for real" as part of an improved inventory methodology, which would be reviewed by international inventory experts. There would then be a real opportunity to use the MELODIES approach in other interested countries: one of the significant advantages of an open, EO-based method, and the flexible design of the MELODIES processing chain!