A restatement of the natural science evidence concerning catchment-based ‘natural’ flood management in the UK

Flooding is a very costly natural hazard in the UK and is expected to increase further under future climate change scenarios. Flood defences are commonly deployed to protect communities and property from flooding, but in recent years flood management policy has looked towards solutions that seek to mitigate flood risk at flood-prone sites through targeted interventions throughout the catchment, sometimes using techniques which involve working with natural processes. This paper describes a project to provide a succinct summary of the natural science evidence base concerning the effectiveness of catchment-based ‘natural’ flood management in the UK. The evidence summary is designed to be read by an informed but not technically specialist audience. Each evidence statement is placed into one of four categories describing the nature of the underlying information. The evidence summary forms the appendix to this paper and an annotated bibliography is provided in the electronic supplementary material.


Introduction
Flooding is among the most damaging natural hazards globally, with inundation leading to disastrous consequences including the loss of lives and destruction of property. Flooding may be fluvial, pluvial, coastal or groundwater related, or caused by a combination of these processes. Here, we focus on fluvial (river) floods, which occur when the amount of water in a river exceeds the channel's capacity. They are caused primarily by the downstream flow of run-off generated by heavy rainfall on wet ground.
Flooding is a natural process, but floodplains are also ideal for agriculture and urban development close to water resources and navigation. Consequently, development in floodplains has increased the exposure of people, property and infrastructure to floods. In many cases it is not practical, cost effective or politically feasible to relocate communities, property and economic activities away from areas prone to flooding, so measures are put in place to manage flood risk by reducing the probability of inundation and/or the negative consequences when a flood does occur.
In this restatement, we concentrate on the scientific evidence concerning the effectiveness of human interventions in river catchments that are intended to reduce fluvial flood hazard. 1 This hazard is typically associated with high river flows. The hazard is characterized by the depth of water at locations where it may cause harm, and also by the velocity of that water, the rate of rise of water levels, duration of inundation and water quality. Interventions in river channels and floodplains that have been widely used to manage flood risk include the building of flood detention reservoirs and flood defences, channel straightening and dredging.
Recent years have seen increasing interest in management interventions that seek to modify land-use and land management, river channels, floodplains and reservoirs (where present), in order to reduce the frequency and severity of flooding, which we refer to here as 'Catchment-Based Flood Management' (CBFM). One subset of CBFM is 'Natural Flood Management' (NFM), which seeks to restore or enhance catchment processes that have been affected by human intervention. These activities aim to reduce flood hazard, while also sustaining or enhancing other potentially significant co-benefits including enhanced ecosystem services (aquatic, riparian and terrestrial) such as greater biodiversity, improved soil and water quality, carbon sequestration, reduced soil erosion, greater agricultural productivity and improved public health and well-being.
While it is recognized that implementation of CBFM or NFM can produce multiple co-benefits, it is not easy to establish the precise nature and extent of those benefits. Often a complex set of trade-offs exists between costs and benefits that accrue to different stakeholder groups within and outside the catchment. Also, while the benefits are well understood in principle, uncertainty around the quantitative predictions of the potential for CBFM/NFM interventions to reduce local and downstream flood hazards remains high, especially in large catchments and for major floods. Differences between river catchments make it difficult to transfer empirical evidence from one location to another. The relative importance of the multiple factors that influence flooding varies spatially and with time, which means that even if an intervention may be beneficial locally, a positive impact on flooding downstream cannot be guaranteed for all possible events in all locations.
The aim of this restatement is to review the scientific evidence for the impacts of CBFM and NFM strategies on downstream flood hazard in the UK. Here, we focus on the natural science evidence base; the social sciences and economics also provide important evidence for policymaking but this is not considered here. The objective is to review processes that impact flood frequency 2 and hazard potential, principally with respect to flood volumes and flood levels but also velocity, duration and water depth. These include modifications to land cover and land management to retain water on and within the land before it flows into rivers, and modifications to and protection of channels and rivers to slow the flow of water and reduce water levels in floodplains downstream where there is a flood hazard (table 1).

Material and methods
The restatement is intended to provide a succinct summary of the natural science evidence relevant to policy-making in the UK as of June 2016. The restatement offers a consensus judgement on the strength of the different evidence components using the abbreviated codes established in previous Oxford Martin School restatements: [D ata ] a strong evidence base involving experimental studies or field data collection, with appropriate detailed statistical or other quantitative analysis; [E xp_op ] a consensus of expert opinion extrapolating results from relevant studies and wellestablished principles; [S upp_ev ] some supporting evidence but further work would improve the evidence base substantially; and [P rojns ] projections made using well-established models that are based on the available physical principles and/or robust empirical evidence gathered in a wide range of settings.
The categories employed are based on those used in previous restatements [2,3], which were themselves developed from the medical and climate change literature. The statements are qualitative in nature and are not intended to form a ranking. We note that, in many cases, evidence is context-or scale-specific. Moreover, interventions that may be effective in one location and at one scale may have a different effect in another setting. Where further gradation is necessary to reflect the quality of evidence, this is done in the accompanying text.
We note in particular the wide range of models used in hydrological science. Some models are based on well-established physical principles such as conservation of mass, energy and momentum, which are fundamental properties of physical systems but which nonetheless require generalizations about parameter values or model equations in order to be applied. Other models represent generalizations from necessarily limited sets of observations whose conclusions cannot be expected to hold in settings different from those in which they were generated.

Results
The summary of the natural science evidence base relevant to catchment-based 'natural' flood management in the UK is given in the appendix, with an annotated bibliography provided as the electronic supplementary material.

Discussion
In this restatement, we have drawn attention to some important evidence gaps. We highlight several immediate priorities: 1. National monitoring networks provide essential data for estimating flood risk and determining the efficacy of interventions. Maintenance and enhancement of monitoring systems should pay particular attention to the accurate measurement of high water levels and out-of-bank flows. Significant uncertainty about the impacts of different types of intervention both when used individually and in combination arises in part because there has not been sufficient research to establish causal links between CBFM and NFM actions and downstream effects. Long-term monitoring is necessary because major floods are rare events; it is also necessary that prospective studies establish good experimental controls and collect accurate baseline data. 2. Recent model studies have begun to reproduce field measurements from relatively small monitored catchments. These models could now be used to simulate the impact of changes in land use and management practices in larger catchments. Model studies of large recently flooded catchments (e.g. Yorkshire Ouse, Eden, Parrett, Thames) could help to establish the scale and spatial location of different types of catchment-based intervention that might be required to have a notable effect on flooding. It is important to investigate whether the models' findings can be extrapolated to regions larger than those for which they have been evaluated, given the constraints posed by their formulation and uncertainties in validation data, and to understand whether the benefits of CBFM/NFM measures are more, less or equally predictable than the benefits of hard engineered assets. 3. The Environment Agency's Catchment Flood Management Plans (CFMPs) assess flood risks across a catchment and can include maintaining or restoring natural processes among the measures that might be taken in the course of flood risk management. Moreover, a large number of catchment-based schemes are currently underway, promoted by Rivers Trusts, Wildlife Trusts and flooded community groups, for example. Many of these local initiatives are neither being planned nor evaluated at larger spatial scales. The lack of monitored baselines and experimental controls creates a risk that the wider and scale-dependent impacts cannot be properly investigated or used to inform decision-making. Research and data available within the water management industry (e.g. water companies, Internal Drainage Boards, land and estate management organizations) would add to the evidence base if it were disseminated more widely. 4. The performance, longevity and operation and maintenance of CBFM/NFM should be systematically compared with traditional engineering solutions. The risks and uncertainties and benefits associated with each approach need to be more fully understood and communicated. The interactions between fluvial floods and other flood types (e.g. pluvial, coastal and groundwater) and sequences of events also warrant further systematic study. The potential for CBFM/NFM interventions in groundwaterdominated and heavily engineered and drained river systems needs further research. The extent to which these interventions add resilience to the impacts of climate change is also worthy of further investigation. 5. A practitioner toolkit would help to share practical experience (while noting contextspecific issues), paying attention to appropriate design criteria. A practitioner toolkit might comprise a set of documents outlining best practices and the situations in which their effectiveness has been demonstrated, drawing on well-studied examples. This could be accompanied by a protocol for coordinated, high quality, monitoring of the catchment, river corridor and hydro-meteorological conditions, drawing on modern sensor, communications and information technologies. 6. There would be benefits from improved communication and collaboration between groups undertaking research in river catchments (e.g. water quality, sediment transport, river restoration, biodiversity, agriculture and forestry), which are all relevant to flood risk management. On the basis of current evidence, the cost-effectiveness of NFM at medium-large scales is likely to rely on understanding interactions between flows, debris and sediment management taking into account the range of ecosystem service benefits that accompany NFM.
Competing interests. We declare we have no competing interests.
Funding. The project was funded by the Oxford Martin School (part of the University of Oxford), and though many groups were consulted, the project was conducted completely independently of any stakeholder.  Flooding causes hundreds of millions of pounds of damage in the UK and climate change is projected to increase the frequency and intensity of heavy rainfall in the future. Flooding may be fluvial, pluvial, 3 coastal or groundwater related, or a combination of these. Here we focus on fluvial floods, which are primarily caused by the downstream flow of run-off generated by heavy rainfall on wet ground. Most investment to reduce flood risk goes into engineered systems like flood defences and channel modification. 'Catchment-Based Flood Management' (CBFM) consists of interventions of any kind that seek to modify land-use, land management, upstream river channels and floodplains, in order to reduce the frequency and severity of flooding. CBFM aims to alter flood risk by making changes within the wider catchment rather than managing flood hazard locally at the point where flooding occurs. One subset of CBFM is 'Natural Flood Management' (NFM 4 ) which seeks to restore catchment and river processes that have been adversely affected by human intervention. CBFM and NFM may help reduce the frequency and severity of flooding as well as delivering other environmental, social and economic benefits. However, because CBFM and NFM interventions often occur alongside other factors that influence flooding, including spatial and temporal variability in rainfall and run-off, assessing the effectiveness of these interventions is challenging. (2) Principles The magnitude of fluvial flooding depends upon (i) the rate of run-off from hillslopes into river channels, (ii) the rate of propagation of the run-off downstream in river channels, and (iii) how run-off contributions from multiple hillslopes and sub-catchments combine via the channel network to generate the downstream flood hydrograph. In small catchments, 5 the peak of the flood hydrograph is dominated by run-off from hillslopes in response to storm rainfall. In larger catchments, the river channel network determines which areas of the catchment contribute to the peak of the flood hydrograph to cause flooding [D ata ]. The impact of NFM/CBFM measures on flooding therefore depends on their location within the catchment, the size of the catchment and the connectivity of the channel network [P rojns ]. Simple extrapolation of small-scale changes to larger catchment areas is therefore not possible, and the effects of NFM/CBFM must be assessed within the context of the whole catchment [E xp_op ]. Relatively few studies adopt such a catchment-scale framework; some of those that do are summarized in §20 [E xp_op ]. (

3) Evidence
Many individual studies have investigated the direct effect on run-off and river flow of variations in natural land cover, human-modified land-use and specific details of land and river channel management practices. Several integrated studies have investigated the potential effect of CBFM and NFM on flooding. Together, these strands of research have generated a large amount of important, policy-relevant information. However, because each flood is a consequence of a unique combination of conditions, evidence needs to be interpreted with care. Here, we summarize the evidence base relevant to policy-making in the areas of CBFM and NFM, in the UK, as of June 2016. We look principally at evidence from the UK, but make reference to studies undertaken overseas where appropriate. We focus mainly on peer-reviewed academic studies, although we have indicated the existence of practitioner-led evidence databases and catalogues where relevant.

(4) Aim
We provide a consensus judgement on the nature of the different evidence components using the abbreviated codes, which are based on those used in previous Oxford Martin School Restatements: [D ata ] a strong evidence base involving experimental studies or field data collection, with appropriate detailed statistical or other quantitative analysis; [E xp_op ] a consensus of expert opinion extrapolating results from relevant studies and wellestablished principles; [S upp_ev ] some supporting evidence but further work would improve the evidence base substantially; and [P rojns ] projections made using well-established models that are based on the available physical principles and/or robust empirical evidence gathered in a wide range of settings.
(b) Meteorological drivers of flooding (5) Meteorological data and trends The UK benefits from a meteorological observation network that is dense by global standards. Annual precipitation totals vary considerably from year to year, but there has been no detectable long-term change in spatially averaged annual precipitation totals since the eighteenth century [D ata ]. Over this time period, the UK has, however, experienced a statistically significant increase in winter precipitation, and a reduction in summer precipitation ( figure 1a)   The blue and red shading shows flood-rich and flood-poor periods respectively Data from ref. [5]; https://crudata.uea.ac.uk/cru/data/lwt/. Projections from the latest global and regional climate models do not suggest a systematic change in annual rainfall totals in the UK between now and 2080 (80% of the simulations show between a 16% reduction and a 14% increase) [P rojns ]. The models suggest some change in the spatial distribution of rainfall, with a projected increase in winter rainfall on the west coast of between +9% and +70%, and reduction in summer precipitation in southern England of −65% to −6% by 2080 [P rojns ]. Higher rainfall maxima are expected, and storms are expected to occur more often, especially in the summer [P rojns ]. Winter upland rainfall totals may also increase [P rojns ]. Under warmer conditions, winter precipitation is more likely to be in the form of rain rather than snow [P rojns ]. (15) Impact of land drainage Drainage to control water levels has complex effects on run-off, which depend on the type of drainage used and the sequence of events.
(a) At the plot scale, drainage reduces peak flows from impermeable (e.g. clay) soils, but increases those from more permeable soils [D ata ]. (b) By drying soils, drainage increases their capacity to store water after a rainfall event, but when the soils become saturated drainage increases flows [D ata ].  . This finding was corroborated by a modelling study that showed that the median reduction in the flood peak associated with an extreme rainfall event produced by a realistic suite of land-use management changes was only 2%, assuming that channel conveyance did not change, but with an uncertainty range of a 1% increase to a 6% decrease [P rojns ]. ii. For a hypothetical extreme storm with rainfall of 140 mm over 2 days (estimated 0.6% Annual Exceedance Probability (AEP) 6 ) the simulated reduction in peak flows was 2-11% and there was no reduction in time-to-peak [P rojns ]. iii. The authors note the high levels of uncertainty associated with their model predictions and leave open the question of whether reductions in flood peaks would be possible at spatial scales larger than 6 km 2 [E xp_op ].
(c) In the River Axe catchment (288.5 km 2 ) in southwest England, an assessment of the observable historical effects of land-use change on basin-scale run-off showed they are limited to high flows that arise from moderate rainfalls (10-30 mm day −1 ) after a period of dry weather [D ata ]. For flows of this magnitude (which usually remain within the river's channel), the results suggest that farming practices that minimize soil degradation and compaction may produce a reduction in river flows in this catchment [S upp_ev ]. The authors note, however, that in nine other catchments no significant changes could be identified, owing to natural variability and data limitations [D ata ].

(21) Summary
There is clear evidence that appropriately chosen land-use and land-cover interventions can reduce local peak water flows after moderate rainfall events [D ata ]. The evidence does not suggest these interventions will have a major effect on nearby downstream flood risk for the most extreme events [S upp_ev ]. The evidence available for the downstream effects of upstream land-use changes at large catchment scales is more limited, but at present it does not suggest that realistic land-use changes will make a major difference to downstream flood risk [E xp_op ]. Moreover, it should be recognized that, although the UK landscape has undergone extensive change due to the multiplicity of intensive farming interventions over many decades, the effects of these interventions on flooding have been difficult to detect. Long-term monitoring is needed to separate the effects of land management from those of climatic variability; without this it is unwise to extrapolate the findings from individual studies to larger scales, or to settings with different soil and vegetation types [E xp_op ].

(d) Channel flow
(22) Geomorphic processes and river channel form Erosion, transport and deposition of sediment can, over time, result in major changes in channel morphology (cross-section and profile) and even channel pattern in some cases [D ata ]. If the amount of sediment flow from a catchment increases, some of it will tend to accumulate downstream in places where the pattern of water flow is insufficient to keep material in suspension. Sedimentation reduces the channel's capacity to convey flow, resulting in higher water levels for a given discharge, and increasing the frequency of flooding [D ata ]. These processes are commonly overlooked in flood risk mapping exercises, but are likely to be important in any river system which receives high rates of sediment delivery and which in the past would have deposited much of its sediment on the floodplain [E xp_op ].
(a) A simulation study of a river in Yorkshire showed that over a 16-month period change in channel configuration due to coarse sediment deposition led to a substantial increase in the area that flooded for the highest flows that were recorded during the study period [P rojns ]. (26) River restoration River restoration seeks to recreate natural channel properties in rivers that have been modifiedoften 'channelized'-in the past. Where it increases the ability of the river to flow onto its floodplain, or creates storage in areas that were once part of the river's floodplain, river restoration can reduce flood risk downstream [E xp_op ]. In many cases, the motive for river restoration is conservation of biodiversity, with the aim that there should be no negative flood impact.
(a) In the New Forest, river restoration has been used to reconnect channelized rivers to their floodplains. For small and medium-sized drainage basins (less than 100 km 2 ) there is evidence that restoration of river channel morphology and floodplain woodland with associated large wood log-jams may reduce flood risk, although only for flows with AEP greater than 50% [D ata ]. (b) Floodplain forest restoration can reduce peak discharge at the catchment outlet by a combination of the processes described in § §27-29. For an event with AEP of 3%, peak discharge was reduced by up to 19% under mature forest. In areas where only 20-35% of the overall catchment area was restored to forest, peak discharge was reduced by 6% [D ata ].

(27) Channel and bank vegetation
Vegetation growing on banks and in the river itself can increase 'channel roughness' which slows water flows and increases sediment deposition. Reduced vegetation in winter reduces roughness and accounts for higher seasonal flows (by up to 50% in a study of the River Stour in Dorset) [D ata ]. The cultivation and maintenance of bankside and river channel vegetation, which is often of value for biodiversity, can induce a small decrease in water flow and hence reduce downstream flood risk in their immediate vicinity in narrow rivers where the width is less than 16 times the depth [D ata ].

(28) Riparian buffer strips
Non-agricultural riparian buffer strips of 10-30 m around channels limit catchment sediment inputs to river channels, which is important in maintaining channel conveyance [D ata ]. Buffer strips also provide co-benefits in the form of reducing movements of agricultural pollutants into watercourses, and shading of river channels from excess heat which benefits aquatic biodiversity   [6]. sediment transport in affecting flood hazard is less well understood than that of hydraulics, but it is known that accelerated sedimentation in rivers can significantly increase downstream flood hazards, and that CBFM and NFM have the potential to reduce catchment sediment yields elevated by intensive farming [E xp_op ].

(e) Flood storage and floodplain conveyance (32) Storage
Water is stored naturally in catchments in forest canopies, wetlands, soils, aquifer rocks, river channels and floodplains. Management actions to increase storage may range from widespread small-scale impoundments (such as blocked ditches and micro-ponds) to large-scale flood detention reservoirs, which are major engineering works. All of these schemes store water upstream and then release it slowly over varying lengths of time depending on capacity and flood conditions. Their effectiveness at reducing flood hazard downstream depends on whether the stored water would have contributed to the flood peak. Small stores may fill up early and have no further effect in a large flood, while controllable larger storage (i.e. with gates or sluices) can be synchronized to maximize the effect on the peak of the forecast flood wave.
(a) In the 5.7 km 2 Belford Burn catchment in north Northumberland, a pond adjacent to the river with 800-1000 m 3  River flows that just exceed the bank-full level are stored in the floodplain, while the channel remains the main mechanism for conveying water downstream. For higher flood flows, velocities on the floodplain approach that in the channel and the floodplain has an active role in conveyance.
Modifications to the floodplain cross-section, typically by encroachment of built-up development protected with flood defences, will modify these floodplain functions-by reducing the natural floodplain storage and reducing the floodplain conveyance during extreme floods. Reducing the floodplain cross-section in this way will increase the water depth in the flood plain for a given flow. On the other hand, removing these obstructions increases the floodplain cross-section and reduces water depths [D ata ]. When flood defences are overtopped or breaches occur, their effect on water levels diminishes, although they may provide additional floodplain roughness and resistance to flow [E xp_op ]. These modifications will not only influence local water levels but will also have downstream impacts, which can be verified with a well-calibrated hydrodynamic model. In a model study of a 5 km reach of the River Cherwell, central England, construction of embankments separating the river from its floodplain (thus reducing the floodplain cross-section) increased peak flood flows downstream by 50-150% and raised water levels by up to 0.5 m [P rojns ]. (f) Co-benefits (36) Co-benefits CBFM and NFM can yield multiple co-benefits, including mitigation of diffuse pollution from agricultural land, reduced water discolouration from peatland-fed watercourses, and mitigation of soil erosion impacts on in-stream and lake ecology. The creation and restoration of terrestrial (riparian, moorland, forest) and aquatic (river, wetland) habitats and associated carbon storage may also be significant. Additional co-benefits may include aquifer recharge and retention of water upstream that can supplement water resources at times of low flow, and protection from the adverse ecological impacts of high water temperatures. These benefits can help to reduce downstream water treatment costs, sustain the productivity of agricultural soils, preserve and enhance ecosystems and biodiversity, enhance recreational value and help build the resilience of ecosystems to other stressors, including climate change. While there are many co-benefits that might arise from CBFM/NFM, to date there have been few studies that have systematically quantified these co-benefits [E xp_op ].  The hazard associated with small floods in small catchments may be significantly reduced by CBFM and NFM although the evidence does not suggest these interventions will have a major effect on the most extreme events. Large fluvial floods are caused primarily by heavy rainfall on wet, frozen or impermeable ground. It is possible that a flood will occur that is so extreme that it will overwhelm any risk management measures or flood defences, natural or otherwise. Land-use and channel form influence the severity of these floods in a fairly subtle way [E xp_op ]. The effectiveness of NFM and CBFM varies with the severity of the event-for example, tree shelterbelts or drain blocking may offer mitigation against small floods, but are likely to be less effective during extremely intense or prolonged high rainfall [E xp_op ]. Actions that provide small-scale local benefits have not been shown to provide significant benefits at the spatial scale of a larger catchment [S upp_ev ]. Although a simple extrapolation would imply that many small interventions (each creating local benefits) should combine to create large benefits at large scale, this is not always the case because (i) local benefits are attenuated downstream by the channel network, and (ii) interactions among local events mean that slowing water flow in one catchment can make a flood worse further downstream when waters from several catchments meet [E xp_op ]. Where multiple interventions have taken place it can be difficult to disentangle the effects of an individual intervention, the effect of which depends upon catchment properties (in particular size, shape, topography, geology, soils, and both hydrological and sediment connectivity) and the extent and location of the intervention within the catchment [E xp_op ]. With the current state of scientific knowledge, it is not possible to state unequivocally whether the lack of demonstrable effect at large scale is because noticeable flood mitigation could not be achieved in a large catchment, or because a sufficiently large-scale set of interventions have not yet been implemented [E xp_op ].

(38) Conclusion II
The larger the catchment and the larger the flood, the smaller is the scope for slowing the flood or storing the floodwater to reduce the flood hazard. We highlight the following main conclusions, which are summarized graphically in figure 3a,b: (i) Interventions that increase the ability of soils to absorb and retain water (through changes to land cover and land management) are at their most effective in smaller floods and at smaller scales. Once soils become saturated the effect is no longer noticeable; (ii) storage (from distributed micro-ponds, through natural floodplains, to large detention basins) can be effective in reducing flood risk, depending on how much storage is provided, where it is located, and how and when it is used; and (iii) increasing the cross-sectional area of floodplains by setting back flood defences that have disconnected areas of the floodplain from the river can reduce peak river flows and flood water levels.