Using remotely sensed imagery to estimate ecosystem services on farmland

The world population is projected to reach over 9 billion by 2050. To meet the demand of an increasing population that is also increasing its per capita demand for food the FAO estimates almost a doubling of global food production will be required. This will increase the pressure on land which not only provides food, but also other "ecosystem services" such as the production of fibre, clean water, fertile soils, pollination or natural pest control services as well as cultural services such as maintaining biodiversity or land for recreational use. Increasing food production - whilst not impacting on the other services - is widely recognized as one of the major societal challenges for the 21st century.

Recent research suggests that in intensively farmed areas, a landscape that combines land farmed to maximise yields with land managed for nature could provide both greater yields and more biodiversity (or ecosystem services in general) than were the whole landscape farmed with what is traditionally seen as sustainable farming methods (e.g. organic farming). This is because if yields are high, a smaller land area is needed, allowing some land to be "spared" for supporting ecosystem services. Maintaining high ecosystem service levels (e.g. pollination and natural pest control services) by proper management of non-cropped areas can also produce an increase in yield. The "optimal" mix of farm land and non-farm land to maximise yields and services is specific to each landscape and depends on the achievable yields and the background biodiversity. Given such data it is possible to assess each landscape and advise on the best way to manage land for biodiversity whilst maintaining production systems.
However, whilst yield data is available, as farmers record this, data on the amount and quality of habitat, and its associated biodiversity is currently prohibitively expensive to obtain as they are typically collected by field surveys. Remote sensing can offer a cost effective alternative. Reliably mapping of biodiversity could be possible if remote sensing can 1) distinguish non-cropped land from cropland (the former typically contains a higher biodiversity than neighbouring cropland) 2) map the spatial configuration of non-cropped land (and so provide information on habitat connectivity or hotspots of suitable habitat, etc.), 3) map habitat type and 4) map vegetation structure (habitat type and structure are often associated with certain animal groups). Until now, the major constraint has been the limited ecological detail that habitat types mapped from remotely sensed data contained. Good mapping of vegetation communities (detailed information on plant species composition) has been difficult to achieve. Recently, the applicants completed the first study that classified National Vegetation Classification (NVC) communities at a high resolution (5 m) for a large extent with a high accuracy of 87-92% in an upland area of the UK. If mapping at this level of detail and accuracy can be replicated in lowland agricultural areas, mapping biodiversity for large areas could become possible and so ultimately the modelling of designs that optimise both yield and ecosystem services.
In the first step, we will map the amount and layout of non-cropped features from remotely sensed imagery. This requires mapping at a high resolution to resolve small, but widespread features such as field margins. In the second step, we will develop our existing methodology for the lowland agricultural areas to estimate vegetation composition and structure. In the third step, we will estimate biodiversity and associated ecosystem services using the habitat maps from previous steps and ecological knowledge of processes at multiple spatial scales. Throughout the grant, we will seek to use cost-effective imagery (e.g. aerial is cheaper than LiDAR imagery) and explore the contribution of different imagery (e.g. satellite versus aerial) to achieving the above objectives.

To advise on the optimal way to manage ecosystem services in production landscapes requires information on the amount of, the quality of, and the biodiversity associated with non-cropped land. As biodiversity surveys for large areas are prohibitively expensive, we aim to develop the potential of remote sensing to offer an affordable alternative.

If the amount and spatial configuration of non-cropped land (which often contains higher biodiversity than adjacent crop land) as well as its vegetation composition, structure and diversity can be mapped, the mapping of biodiversity and associated ecosystem services for large areas could become possible. Until now a major constraint of using remotely sensed data was that only limited ecological detail of vegetation composition could be obtained. However, our recent work derived for the first time 24 detailed National Vegetation Classification (NVC) communities at a high accuracy (87 - 92%) and a high resolution for an area of the UK uplands. Our methods coupled available data at local and neighbourhood scales (digital terrain models, soil maps, aerial images) and sophisticated statistical classification models (the machine learning classifier "Random Forest") to produce a 5 m resolution vegetation map of 1630 km2. Such maps require some survey work (to "train" the classification algorithm and validate the model), but this is a relatively small effort relative to the field work required to survey a whole landscape.

This grant aims to 1) adapt the methodology to the lowland agricultural landscape and map vegetation composition; 2) develop cost-effective methodologies to map the amount, 3) spatial configuration, 4) vegetation structure and 5) diversity of non-cropped land at a high resolution (to resolve small, but common features such as field margins). Derived habitat maps will then be used to model biodiversity and associated ecosystem services using ecological knowledge of processes at multiple spatial scales.

Who will benefit from this research?

The beneficiaries of this research are academics and society.
The toolset and resulting maps would benefit society by enabling the design of landscapes that maximise yield and ecosystem services, thereby allowing us to meet the demands of an increasing global population sustainably. This would be of particular interest to societies where strategies to implement desired land use designs are already in place (e.g. in the UK via agri-environment schemes) and to societies in parts of the world where most of the population increase is projected to occur and crop yield can be increased considerably (e.g. Africa). The toolsets would therefore be of interest to agriculturalists concerned with ecosystem service provisioning as well as to policy makers, land managers and farmers concerned with sustainable production (nationally and internationally). The toolset would also benefit land managers outside of intensively farmed agricultural areas, such as conservation managers of species or habitats as the methodology is likely to be easily adaptable to other environments and could for example facilitate designing reserve networks.

The toolset developed in this grant would enable the landscape scale production of detailed habitat maps as well as maps of biodiversity and associated ecosystem services. These maps would be a valuable resource for research in a variety of disciplines, such as agricultural policy, landscape ecology, zoology, conservation or population dynamics.

How will they benefit from this research?

The results from this research will be published in peer-reviewed journals to maximise dissemination among the academic community. They will also be disseminated to the wider public through an established procedure of working with the media (via a PR company: CampusPR). Stakeholders (Defra, Natural England, Syngenta, IFPRI, other academics working on land use issues internationally) will be engaged by setting up an informal stakeholder board through which we will disseminate project results quickly and circulate ideas. In the final year of the project a workshop will be organised in which the results of the project and results and ideas from other teams working on similar issues will be presented. We anticipate seeking follow on funding to develop methods (e.g. software) that allows usage of the developed methodology by a wider community of practitioners in the field.

Using these routes, we will maximise dissemination of the methodology. We will also seek future funding to develop the methodology into wider areas (both geographically) and in terms of what can be classified (e.g. soils). The technology of using sophisticated statistical approaches to utilise information from maps and images is a powerful one and one can imagine a variety of other uses, some of which have commercial applications. As the methodology develops, we will explore the potential for developing a consultancy outfit to commercialise this expertise to different solutions.