What evidence exists for the efectiveness of on-farm conservation land management strategies for preserving ecosystem services in developing countries? A systematic map

Background: An extensive body of evidence in the ﬁeld of agro-ecology claims to show the positive eﬀects that maintenance of ecosystem services can have on meeting future food demand by making farms more sustainable, productive and resilient, which then contributes to improved nutrition and livelihoods of farmers. However, inconsist-ent eﬀects have commonly been reported, while empirical evidence to support assumed improvements is largely lacking. Overall, a coherent synthesis and review of the evidence of these claims is largely absent from the literature. Methods: Systematic searches of peer-reviewed research were conducted in bibliographic databases of Web of Science, SCOPUS, AGRICOLA, AGRIS databases and CAB abstracts, and grey literature from Google Scholar, and 32 subject-speciﬁc websites. Searches identiﬁed 21,147 articles. After screening, 746 studies were included in the ﬁnal map. Results: Of the 19 conservation land management practices considered, soil fertilisation (24 %), tillage (23 %), agroforestry (9 %), and water conservation (7 %) were most commonly studied. Ecosystem services most commonly studied were supporting (55 %) and regulating (33 %), particularly carbon sequestration/storage, nutrient cycling and soil/ water regulation/supply. Key data gaps identiﬁed included the absence of long-term records (with datasets spanning >20 years), studies located in North and Central Africa, research that focuses on smallholder landscapes, and studies that span diﬀerent scales (regional and landscape levels). Conclusions: The study employs systematic mapping combined with an online interactive platform that geographically maps results, which allows users to interrogate diﬀerent aspects of the evidence through a deﬁned database ﬁeld structure. While studies are not directly comparable, the database of 746 studies brings together a previously fragmented and multidisciplinary literature base, and collectively provides evidence concerning a wide range of conservation land management practices impacting key ecosystem services. The systematic map is easily updatable, and may be extended for additional coding, analysed to assess the quality of studies, or used to inform future systematic reviews.


Background
Over the last 80 years, the industrial agricultural revolution has made signiicant advances in food production. For example, in developing regions as a whole, the share of undernourished people in the total population decreased from 24 % in 1990-92 to 13 % by 2012 [1]. However, this transition also created a technology-reliant global food system and associated long-term vulnerabilities, such as dependence on crop monocultures, chemical fertilisers, pesticides, petroleum and antibiotic feed supplements [2]. he agricultural system also causes a complex interaction of environmental deterioration [2] by accounting for: 70 % of water extraction worldwide [2], consumption of one-third of all available energy [3], 75 % of all deforestation [3], 19-29 % of global greenhouse gas (GHG) emissions [4], the largest contribution of non-CO 2 GHG emissions [5], and the leading cause of biodiversity loss [2]. Moreover, these trends are increasing as agriculture intensiies and expands. For example, between 1961 and 2005 agricultural production doubled in Sub-Saharan Africa [5], and was one of the main drivers of degradation in 65 % of natural ecosystems globally [6]. In the years leading up to 2040, an estimated increase of 50-70 % in food production and availability is needed to keep pace with the demands of a global population expected to reach 8-10 billion and a concurrent decline in available arable land [6,7]. As the population grows, per capita supply of farmland is increasingly in competition with accelerated urbanisation [8]. Income growth has also led to expansion of the global demand for meat, which has tripled in the last 50 years [9] so that livestock now consume 40 % of the global feedstock. he threat of climate disruption poses further biophysical barriers to enhancing yields, particularly in rain-fed regions. It is therefore increasingly evident that industrial agriculture is not sustainable at its current level of expansion [2,3,6,10]. his view has been supported by a recent study by Ehrlich and Ehrlich [2], who describe these interconnected challenges as having the potential to lead to a global collapse in the food system, comprising of production, processing, packaging, storage, distribution, retail, consumption and disposal. Other international bodies, including the Global Alliance for Climate Smart Agriculture, the UK Government Foresight Committee [3], and the Oxford Martin Commission [10] have argued that one of the greatest challenges for the 21st century is inding the balance between providing enough food for a growing population and maintaining healthy ecosystems, while meeting the livelihood and nutritional requirements of the most vulnerable in a changing climate.
To address the challenge of identifying how to close the 'yield gap' (i.e. raise yields in less productive systems than those typical of industrial agriculture [2,7]), many studies are now examining the impact of ecosystem service conservation strategies on agricultural productivity [11], and associated with this, the implementation of a variety of alternative practices to conventional or intensive agriculture. he Millennium Ecosystem Assessment (MA) [6] and other more recent frameworks [12,13] have asked the question: what agricultural practices are efective, practical, afordable, and scalable? For the purpose of this study, 'conservation land management practices' are those practices that are adopted with the aim of preserving or enhancing ecosystem services without compromising farm production, and may be adopted before, during, or after cultivation [14]. hese conservation land management strategies may be active, such as surface crop residue management, or passive, such as the preservation of native vegetative patches in ields [15]. Practices may incorporate principles of multifunctional agriculture (e.g. simultaneously producing food and non-food commodities, maintaining wild crop varieties, traditional landraces and local culture [16]), sustainable intensiication (i.e. relieving pressure on land expansion and limiting forest encroachment [5]), and conservation agriculture (i.e. including practices of no-tillage, permanent soil cover using crop residues or cover crops, and crop rotation [17]), amongst others. As conservation land management practices often require minimal capital investment and inputs, provide multiple beneits (e.g. food, fodder, and enhanced soil quality), and show signiicant efects over short periods [14], they have opportunities for enhancing smallholder production.
Despite a growing interest in this area, there still appears to be a lack of a coherent evidence base showing how efectively these management strategies preserve or enhance ecosystem services overall. his may in part be because identifying accurate quantitative measurements, interpreting interrelations, and synthesising how the output can translate into practical management techniques is exceedingly complex for four main reasons. First, changes in conservation land management may afect various ecosystem services diferently, and management requires making judgments about trade-ofs depending on the service, temporal horizon, spatial scale or geography [18]. Second, there is an apparent deiciency of indicators or proxies of ecosystem processes. hose that are available often have incomplete datasets or the beneits and/or disservices of particular practices are inconsistently reported. For example, some studies report that long-term no-till can improve soil fertility, recovery and decrease erosion, but conlicting reports state that no-till also leads to soil compaction, can limit water iniltration and hinders seed germination [19,20]. hird, the issue of scale brings another level of complexity, because some beneits are often only measurable at larger scales. For example, the impacts of conservation land management practiced on a local level may only be visible over a large area or over a long period of time, whereas they may only have been measured at the farm scale, and the reported results are therefore incomplete. Fourth, much of the evidence is spread across diferent disciplinary 'silos' , with very limited synthesis. Some studies may also overstate the beneits of land management strategies.
When evidence is so extensive and disparate, a irst step in an informational synthesis is a systematic map: a rigorous methodological tool for data extraction of peerreviewed and grey literature [21][22][23]. Systematic maps can be used to describe the nature, volume and characteristics of research in a broad topic area, and may be used to identify trends in the literature and knowledge gaps for future analysis. Systematic maps follow the same rigorous systematic processes as systematic reviews, being transparent and repeatable to search for and collate evidence. However, critical appraisal of the quality of the evidence is often either absent or limited in depth, and results from studies are not extracted or synthesised [24,25].
Previous attempts to synthesise the body of research that examines on-farm conservation management practices have focused on particular regions (particularly Africa [26,27]), a limited set of practices [17], or the evaluation of management outcomes purely in terms of crop responses [26]. Our systematic map builds on this research both geographically, being the irst efort to synthesise the evidence in developing countries broadly and in terms of the range of management strategies and ecosystem services studied. his study is among the irst (see also [28,29]) to present the results of a synthesis of environmental evidence using an online interactive geographic map.
he aim of this systematic map is to review the state of evidence that reports on the efectiveness of on-farm conservation land management for protecting or enhancing ecosystem services. he objectives of the systematic map are: 1. To collate studies reporting evidence on the efectiveness of on-farm conservation land management practices on ecosystem service provision in agricultural landscapes in low-income and developing countries. 2. To map regions where on-farm assessments of conservation land management in low-income and developing countries have been undertaken. 3. To make information easily accessible by producing an online interactive map, searchable by topic.
We aim to provide a detailed summary of diferent strategies proposed and tested, for diferent crops in diferent regions, and over diferent timeframes. Moreover, we identify the pathways by which practices are assumed to inluence ecosystem service provision by reporting on measurable indicators assessed in studies. Developing countries are the focus of the review for three key reasons. Firstly, developing regions are where much of the world's agricultural expansion is occurring [3], yet 80 % of arable land is already used [30] and croplands yield well below their potential [31]. Secondly, in some cases, developing regions may also depend on ecosystem services rather than technological inputs to support agriculture, due to lower inancial, technical and credit-borrowing capacity [1]. hirdly, according to the FAO, of the 795 million people classiied as undernourished worldwide, 780 million are in developing regions [1]. We anticipate four key end-users for the information that results from this systematic map: land owners and managers; local decision makers and programme managers; national and international policy makers; and researchers. he information highlights important research directions which can help to develop monitoring baselines, diagnose environmental problems, identify systems close to environmental thresholds, and thus evaluate the beneits and trade ofs of strategies to implement at the farm level.

Methods
he systematic map followed a published a priori protocol [32]. Deinitions of terms were developed collaboratively during the work (Additional ile 1). hese were used principally for searches and screening, to improve rigour and overcome possible ambiguity between reviewers. Full details are provided in Additional ile 2 of the search terms, the number of records generated for speciic searches, and the name, location, date of searches in bibliographic databases, online searches, key international peer-reviewed journals, and specialist organisations and online databases.

Sources of publications
An extensive targeted search of peer-reviewed and grey literature was conducted, including key bibliographic databases, key international journals, specialist organisations, and online databases.
Bibliographic databases he following bibliographic databases were searched: Grey literature for specialist searching hirty-two specialist organisations and online libraries were searched for organisational reports, conference papers or proceedings, policy briefs, station and annual reports (Table 1). Given the limited search capability of databases, a hierarchical approach [33] to searching was used, converting the original string to key words (e.g. in-ield assessment, ecosystem services) and topics (e.g. agriculture) (Additional ile 2). Where the facilities were available, language limits to English were set. Where no search bar existed, websites were also hand-searched.
Estimating the comprehensiveness of the search he search string was the last of 27 iterations tested in Web of Science. A test library of 30 references (see protocol [32]) conirmed the search strings captured relevant literature, balancing speciicity and sensitivity [33].

Study inclusion and exclusion criterion
he following summarised criteria were established through two stakeholder workshops in February 2014, Cape Town, South Africa and June 2014, Oxford, UK.

Population
We examined studies on farms located in 74 low/middle income and developing countries. hese countries were selected from three databases of globally-recognised organisations working to improve ecosystem services, reduce food insecurity and support economic development (i.e. [34][35][36]). Relevant farming systems were based on an extended list of the FAO major commodities list [37], including terrestrial food, cash, oil, and agroforestry crops. As determined through the peer-review process when developing the protocol [32], marine biomes or mangroves with maricultural or aquacultural activities were excluded (see Liquete et al. [38] for a review on marine ecosystem services), and livestock farming and pastures (including land covered with grass or other plants suitable for grazing) were excluded from this study.

Intervention
Studies where conservation land management strategies were adopted to support productive agriculture, while simultaneously preserving or enhancing ecosystem services were examined. However, studies assessing cultivar selection, behavioural ecology, purely agronomic or economic questions, and land uses diferentiated only by crop type were excluded. hose analysing land use gradients and studies of non-point source pollution were also excluded, because these were deemed to be outside the scope of site-speciic management strategies.

Comparators
Within studies, comparators were deined as farms or ields without conservation land management strategies, conventional/intensive agriculture or natural sites.

Outcome
We examined the literature to look for evidence of measured changes in ecosystem services, including supporting services (e.g. carbon regulation, pest regulation, nutrient cycling, biodiversity), regulating services (e.g. water/soil regulation and supply, pollination services), provisioning services (e.g. fuel wood, medicinal and aromatic plants), and cultural services (e.g. education, recreational, spiritual, tourism, bequest or aesthetic value). We excluded non-timber forest products (NTFPs) extracted of the farm (from forests) but included NTFPs on farms (e.g. from trees on farm boundaries or on domestic homesteads). Studies measuring outcomes on health or nutrition, including those measuring fungal pathogens and diseases, were excluded from our analyses.

Study design
We included studies that had provided grounded empirical assessments at the ield level, using direct assessment of social or ecological variables. Lab-based and ex-situ methods were excluded, as were conceptual frameworks, methodologies, training manuals, conservation planning tools and regulatory/legislative frameworks. Studies that looked only at economic valuation were also excluded.

Language
Owing to limitations of time and resources, we only included studies published in English. Authors acknowledge that this may introduce a bias against studies conducted in developing countries where the oicial language or operation language of universities and research institutions is not English, particularly studies in Spanish, French and Portuguese. We suggest expanding to other languages of publication for future iterations.

Date
No time limitation to the search was applied. Single and multiple year studies were included, however long-term paleoecological studies were excluded.

Screening
References were iled in EndNoteX7 ™ reference manager [39] as separate libraries according to bibliographic source, pooled together and duplicates were removed. References were exported into a master database in Microsoft Excel for visual checks at each stage of screening, using categories of publication ID, title, year, abstract, country and URL. References were then imported into DataX systematic review evaluation software for abstract screening [40]. Title, abstract, and full text screening took place between 26 November 2014 and 9 February 2015. Five reviewers at the University of Oxford and CIFOR conducted screening at title and abstract stage. hree reviewers conducted full text screening, recording reasons for exclusion. At the title screening, the Randolph's free-marginal kappa on a random subset of 100 studies was 0.82 [41]. Where there was doubt of studies to be included/excluded, exclusion was conservative. Any ambiguities were discussed through online consultation, and references further veriied by secondary reviewers. Fourteen articles were identiied that were either meta-analyses or systematic reviews (not included in the systematic map database, but as Additional ile 3).
he list of studies excluded at full text with reasons for exclusion is shown in Additional ile 4. Hand-searched articles on organisational websites were excluded when studies described: 1. Cultivar selection, economic valuation, or purely yield impacts of management (e.g. studies in the Africa Rice Centre), 2. Concept notes, proposals, project reports and training, methodology manuals (e.g. TEEB, Comprehensive Africa Agriculture Development Programme (CAADP)), or media releases and workshop reports (e.g. Institute of Environment and Agricultural Research), 3. Programme information of funding agencies (e.g. Department for International Development (DFID)), 4. Institutions, governance and policies of poverty and development without including agriculture data (e.g. many studies of the Overseas Development Institute (ODI), and World Resources Institute (WRI)), or 5. Were not ield based or excluded methodologies to directly assess ecosystem services or were not ieldbased (e.g. Agricultural research companies including Monsanto and Syngenta).

Data coding and extraction
Data extraction was conducted by eleven coders, who regularly met with a secondary reviewer to discuss and resolve inconsistencies between the 10th of February 2014 and the 20th of April 2015. Authors were contacted where studies did not include the site coordinates or study location (n = 4). We used Dropbox and Googleforms to share studies and extract the data. A coding tool (Additional ile 5) for data extraction was developed with division into six categories, as illustrated in Fig. 1.

Data mapping
Following the completion of data extraction, results were placed in a searchable database as a relational Microsoft Access database (Additional ile 6). Results were searchable by keyword, and cross-tabulated to identify where no research outputs were available. Qualitative and quantitative synthesis was conducted on trends, interpretation, and gap analysis from May to July 2015. Graphic visualisation for spatial analysis employed ArcGIS10 mapping software [42], using the World Geodetic System 1984 and decimal degree coordinates. Visualisation in the online interactive map was accomplished with a D3 open source JavaScript framework hosted on the Oxford Long-term Ecology Laboratory website. A summary of each study is available by clicking on the study site, which opens a scrollable side table. he table presents the attributes stored for that study, including bibliographic, biophysical, ecosystem service, and management details, plus a URL link to open access articles. A ilter system allows users to choose criteria for making sites visible on the map. When a database ield is selected as a ilter, all the values within that ield appear in a list. Selecting a set of values causes the map to present only the sites that match those values. Filters may be based on categories (e.g. smallholder farms), ranges (e.g. studies published between 1992 and 2000), or key search terms (e.g. maize crops or tillage). For example, users interested in experimental studies assessing the impact of no-till on regulating and supporting services in maize may use the map to identify 113 studies.

Results
Overall descriptive statistics he low chart for selecting articles for the systematic map, with results for each stage, is shown in Fig. 2 (adapted from Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA [35]). A total of 746 were included in the inal systematic map. (A list extracted from the Endnote database of these articles is included in Additional ile 7).
Of the 746 articles evaluating the impact of farming strategies included in the map, 576 articles were experimental in their design with treatment and control groups, 49 articles were quasi-experimental (i.e. without all factors afecting outcomes, groups controlled), and 121 were non-experimental (i.e. without any explicit manipulation of groups [24]).
Overall, the majority of articles were journal articles (97 %, n = 727). he remaining articles were grey literature articles (2 %, n = 15) composed of reports and policy documents from agricultural, developmental and environmental research agencies (e.g. Integrated Water Management Institute (IWMI)), and articles from conference proceedings (1 %, n = 4). Two hundred and fourteen of the 746 articles were open access of the 746 articles. his analysis diferentiated between articles and studies. For example, where information from one study was presented in one or more articles but did not include the same data, studies were considered multiple outputs and included in the systematic map as such. Furthermore, 244 articles included multiple case studies, which were included in the database as separate entries for geographic mapping.

Journal titles
A total of 202 journal titles were represented. Over half of the relevant articles (n = 434 or 58 %) were published in Recognising that scientiic publications generally have been increasing in number exponentially for the last few decades, we compared the cumulative number of articles in the systematic map to a broader search in Scopus database, following Petrokofsky [25]. Scopus was searched on 10 June 2015, with the search phrase 'conservation agriculture' limited to subject areas of Life Sciences, Physical Sciences, and Social Sciences and Humanities and Environmental Science, Agricultural and Biological Life Sciences and Social Sciences. We found a proportionally greater increase in the number of studies in the systematic map (n = 746) compared with other generalist searches (n = 10,961) over the same time period since 1962 (Fig. 5).
In Asia, most of the studies were from Far East Asia (49 %) and South Asia (42 %). In Africa, the proportion of studies was more evenly split between East Africa, West Africa and Southern Africa (34, 28, 27 %). In West Africa, most studies were located in the English speaking country of Ghana (n = 13) rather than the surrounding Francophone countries. Only one study was located in North Africa, while 18 were in Central Africa. In the Americas, 65 % of studies (n = 118) were located in South America (mostly in Brazil, n = 105, followed by Colombia and Peru). 64 were in North/ Central America (mostly in Mexico, n = 54), followed by Nicaragua, Honduras and Haiti. We acknowledge these results may be an artefact of the language selection criteria.  While 74 countries were admissible in the selection criteria for the systematic map, no studies were found in 28 countries: Afghanistan; Algeria; Bhutan; Botswana; Burundi; Cambodia; Central African Republic; Chad; Comoros; Congo; Democratic Republic of Congo; Djibouti; Eritrea; Guinea-Bissau; Kyrgyzstan; Lesotho; Liberia; Mauritania; Mongolia; Namibia; Samoa; São Tomé and Principe; South Sudan; Sierra Leone; Solomon Islands; Somalia; Tajikistan; and Yemen. Beyond the search criteria, various reasons could explain why these countries were not included, such as a lack of research stations, remote access, limited agricultural production, amongst other factors (see discussion).

Spatial unit of analysis
Geographic scales of plot, farm, landscape, district, and region, as well as social scales of individual, household, and village were diferentiated following Randall and James [23]. he scale of 'plot' referred to studies that compared speciic in-ield interventions (e.g. diferent rates of organic farmyard manure fertiliser application or tillage regimes), whereas 'farm' compared results across a number of farms. Many studies that compared across larger areas could also be considered landscape scale studies. In some cases it was not possible to discern aspects of the spatial scale, because studies did not state whether interventions were replicated within single or on multiple farms [23].
Most studies were conducted at the plot scale (42 %, n = 317) or farm scale (31 %, n = 233), followed by regional (10 %, n = 73), landscape (8 %, n = 62), and district scales (6 %, n = 41). Many studies were performed across various sites and various scales; for instance, 221 studies had multiple study sites. Eleven studies involved cross-country analyses, seven compared two countries, two compared three, one compared four, and one compared ive. Two studies were cross-continental, assessing carbon sequestration rates in no-tillage soils in tropical Brazil and Madagascar [43], and agroforestry and fallow systems in Cameroon, Madagascar, Tanzania, Indonesia, and Laos [44]. Ninety-two per cent of studies used geographic, rather than social, scales of organisation. Figure 7 shows the number of studies according to farm typology and irrigation type, adapted from the categorisation of Dixon, Gulliver and Gibbon [45]. Research stations, which include working farms, ( Table 2) comprised almost half of studies (48 %, n = 358). hese studies usually did not state whether they focused on irrigated or rain-fed crops. Smallholder farms comprised a quarter of studies (24 %, n = 176). Twelve per cent (n = 16) of these were rain-fed, only 4 % were irrigated, and the remainder of studies did not state irrigation techniques. Large-scale and commercial farms constituted the smallest proportion of studies (7 %, n = 50). Very few studies combined assessments on both small and large farms (2 %, n = 17). 145 studies did not state the type of farm.

Ecosystem service
he systematic map diferentiated between broad categorisations of ecosystem services and 16 subtypes, based on the Millennium Ecosystem Assessment categorisation [6] (see protocol [32] for rationale). Categories are nonexclusive, such as soil regulation and nutrient cycling. Supporting services were measured in 55 % (n = 591) of studies, regulating services 33 % (n = 353), provisioning 9 % (n = 101), and cultural services 3 % (n = 32) (nonexclusive categories). Biodiversity was studied more than pest regulation and pollination services. Spiritual and symbolic value often appeared in conjunction with studies of medicinal and aromatic plants, while only ive studies measured tourism as a cultural service (Table 3). One-third of studies (36 %, n = 266) measured ecosystem services in combination with one another, most frequently combining supporting and regulating services (85 %, n = 22). Most studies (92 %, n = 689) measured one or two ecosystem services, while studies of three or more ecosystem services only constituted 8 % (n = 57). he mean number of ecosystem services measured per study was 1.45 ± 0.85. Figure 8 indicates the number of ecosystem services measured, plotted against the duration of studies.

Table 2 Number of studies on in-ield assessments of ecosystem services conducted on farms in research stations, of the ten most frequently cited countries
The

Conservation land management interventions
In total, 19 categories of conservation land management were measured (Table 4). Top interventions included use of organic fertiliser (24 %, n = 358), tillage (23 %, n = 350), agroforestry involving multipurpose and multistorey cropping and home gardening (9 %, n = 129), and seven methods of water conservation (7 %, n = 98). Other practices included: crop rotation; fallowing; cover cropping with legumes or shaded patches; mosaic management including the maintenance of wild vegetative patches on farms; set aside areas and bufer strips; restoration including re/aforestation; ive methods of weed management including mechanical, disc harrowing, herbicide, cover cropping and mulching; integrated pest management; pollination management; two types of mulching using crop residues and plastic ilm; erosion control using terracing and slope re-vegetation; and intercropping with legumes and cereals. he remaining studies involved passive management, rather than active, such as maintaining biodiversity on farms, ire management, and retaining termite mounds within farms to promote nutrient cycling.
he link between management strategy and objective or outcome was not necessarily uniform nor simple. For example, crop residues could be used for mulch, organic fertiliser and weed management. In addition, a number of management practices were frequently adopted in conjunction with one another (e.g. combining leguminous cover crops with minimal tillage). Within studies, conservation land management strategies were compared to other practices, for example organic and inorganic fertiliser, organic and inorganic mulching, integrated pest management and insecticides, mechanical weed control and herbicides, or mosaic and monocropping. Figure 9 plots the ten most frequent land management interventions measured against the year of publication.

Crop types Crops and regions
A total of 72 types of crops were studied (summarised in Table 5). In line with a hypothesised trend stated in the protocol, most studies assessed annual crops (62 %, n = 462) instead of perennial crops (25 %, n = 190). Staples were frequently studied: maize (18 %, n = 285), wheat (12 %, n = 183), and rice (10 %, n = 152). Other major crops studied included tree crops, soybeans, coffee, beans, sorghum, cotton and inger/pearl millet. Despite the extent of global cultivation, our review identiied a limited number of studies in palm oil, tobacco and rubber; although this result may be due to countries excluded, such as Indonesia (see [51]). When compared to the global area of cultivated crops, the proportion of studies looking at rice most similarly relected the proportion of land cultivated globally (Table 6). Cereal grains constituted 45 % (n = 692) of studies, predominantly in Far East Asia (n = 221), South Asia (n = 167) and South America (n = 101). he majority of studies on tree crops (21 %, n = 52) and legumes (24 %, n = 46) were in South Asia. Fruit crops (n = 31) and tubers (n = 16) were most frequently measured in East Africa. Assessments of vegetables were less common, and predominantly in South Asia (n = 14) and East Africa (n = 18). Soybean was most frequently studied in South America (44 %, n = 35), likely because Brazil is the world's second top producer (65.9 m metric tonnes/ annum), following the US [54]. Cofee was most frequently measured in East Africa (28 %, n = 20), including top producing countries of Ethiopia and Uganda. No studies were identiied assessing Adansonia digitata (baobab), Moringa oleifera L. (moringa) and Vitellaria paradoxa (shea tree), even though these are increasingly

Crops and interventions
Most studies assessing the impact of tillage and crop rotation were in wheat and maize cropping systems. Agroforestry was the most prevalent in tree crop and cofee production systems, and water conservation practices were most frequently measured in rice cropping systems. Biodiversity was most frequently studied in sites with cofee, other tree crops, or maize. Crop rotation most often arose in studies of maize, followed by wheat and soybeans (Table 7). Additional ile 9 tabulates all the conservation land management and interventions and ecosystem service measured.

Indicators to measure on-farm conservation interventions
One hundred and seventeen indicators were measured (Table 8). Indicators were identiied using the description given in studies. Some indicators overlap (e.g. yield could be categorised under economic indicators, above ground biomass could also refer to yield), or be closely associated (e.g. Soil Organic Carbon and Total Carbon). Unsurprisingly, most commonly measured indicators are strongly associated with crop productivity, (i.e. Soil Organic Carbon (SOC), macronutrients, yield, bulk density and pH). The overall total (n = 1504) is more than 746, because some studies assessed more than one conservation land management intervention hh household Out of 746 studies, 73 % (n = 544) measured chemical indicators-mostly SOC, followed by N, P, K, pH, Total C, CO 2 , Mg, N 2 O and NO 3 . Physical indicators were measured in 53 % (n = 398) of studies -mostly bulk density, soil texture/particle size, water holding capacity/soil moisture retention, soil type, temperature, soil porosity, runof and soil loss (for leaching), water use eiciency, water iniltration rates and altitude. Biological indicators were measured in 73 % of studies (n = 546)-mostly yield, community diversity/richness and abundance, followed by microbial biomass content, below-ground biomass, stem density/diameter, litter, fruit set, survival rate of trees and crop height. Social indicators were measured in 13 % of studies (n = 96), primarily perceptions of pest incidence, followed by fuel wood supply and usage, medicinal and aromatic plants, water quality and soil regulation. Other variables described the farming context, such as household size, village size, months of food security, land tenure and presence of farmers associations to access information and funds. Twelve per cent (n = 93) measured economic indicators, mostly income, farm inputs, labour days, and farm size and livestock ownership. Indicators of willingness-to-pay or opportunity cost were not included, given that purely economic valuation studies were excluded from the review.

Taxonomic indicators
Of the 203 studies measuring types of taxonomic groups [following Randall and James (2012) [23]; (Fig. 12)], plants were most commonly studied (32 %, n = 66), followed by invertebrates (non-pollinating) (29 %, n = 59) and pollinators (22 %, n = 44). Very few studies measured birds and mammals, while two studies did not mention which taxonomic indicators were used. Studies measuring mammals included one study of large terrestrial ungulates, carnivores, large rodents, armadillos and primates in cacao agro-forests in southern Bahia, Brazil [55], a study assessing shrews in a sugarcane plantation in the Lowveld of Swaziland [56], and a study on large mammals including bufalo, eland, and hartebeest in Zambia [57]. Various studies also measured changes in dynamics of functional groups in bird and insect communities, including decomposers (e.g. termites, earthworms), phytovores (e.g. weevils), carnivores and frugivores (e.g. birds) [58], tunneller and roller species [59], and pollinators, including stingless bees, solitary wasps, birds, bats and bumblebees [60]. Manipulative or experimental designs were the most common design for all animal taxa (67 %, n = 135) except for birds, where 75 % (n = 6) of studies were nonexperimental. Studies of plants had a comparatively larger proportion of non-experimental designs compared to other taxonomic groups (39 %, n = 23), particularly study systems with agroforestry, reforestation, home gardens or smallholder agricultural mosaics.
Cultural services Most studies of cultural services considered the utilisation and maintenance of wild and cultivated plants on farms that provided biological materials used for medicinal, ritual, edible, ceremonial, timber, ornamental and other purposes [63]. Studies described how the management of biodiversity supports beliefs and cultural continuity [64], and fosters communities' social fabric through the sharing of resources [65]. Other studies measured the aesthetic value of biodiversity on farms, such as birds in isolated trees or open areas outside the forest [66]. Only 32 studies measured these indicators of cultural services (Fig. 14).
Pollination services Studies measured the efects proximity of ields to semi-native habitats have on insect community diversity and abundance [15,70,71], the impact of plant types [72] or landscape efects on pollinator populations and activity [73], or the placement of beehives or trap nests in ields [74]. A total of 48 studies looked at pollination services (Fig. 15).
Carbon regulation Carbon storage and sequestration, included in 282 studies, often measured SOC content [75], and carbon pools in soil, soil litter or in biomass [76]. Carbon regulation services were commonly associated with practices of organic fertiliser application, crop rotation, reducing or eliminating soil tillage [75,76]. Interestingly, only one study assessing carbon regulation measured perceptions of soil organic carbon (Fig. 16).

Outcomes on ecosystem service and yield
he reported overall outcome (or efect) of management strategies ecosystem service provision and on yield was examined using the authors' own conclusions of the studies, following Milder et al. (2014) [77]. In addition to ecosystem service provision, yield was recorded given the association with income, livelihoods and food security, and thus the high value land managers may place on maximizing when choosing land management strategies to implement. Recording both factors is useful for future analysis in assessing beneits and trade ofs of conservation land management strategies. Independent analysis of the studies results is outside the scope of a systematic map, but documenting the claims made regarding yield and ecosystem service outcomes would constitute a useful area for a future systematic review incorporating statistical analysis. Of the studies that explicitly stated the outcomes of interventions: • on ecosystem services and yield (n = 181) (often referred as 'win-win' [78]), 57 % reported dual beneits of ecosystem service and yield improvements; • on ecosystem services (n = 671), 63 % reported an improvement, 28 % found mixed outcomes, 5 % reported no change, and 4 % reported a decline in ecosystem services as a result of the intervention (Table 9); and • on yield (n = 337), 66 % reported an improvement, 22 % reported a mixed outcome, 7 % found no impact, and 5 % reported a decline.
Mixed outcomes were reported in 39 studies (12 %). Very few studies identiied trade-ofs between conservation land management practices. However, ive studies (2 %) found an improvement in ecosystem service but decline in yield and three studies (1 %) showed an Table 7 Summary of crops vs. conservation land management interventions studied Table 7 illustrates tillage and organic fertiliser has been most commonly studied in maize (n = 171, n = 107), while water conservation has been mostly commonly studied in rice (n = 36) and wheat (n = 35)

Crop
Organic    Fig. 13 Water regulation indicators. Eight-five indicators were used to in studies measuring water regulation. Most common indicators included yield, water holding capacity/soil moisture retention, potassium, pH, particle size, water use efficiency, runoff, water infiltration rate, temperature, income and carbon dioxide. Perception of water quality was measured in six studies. Electric conductivity (liquids), cation exchange capability, dissolved organic carbon, and turbidity/water stable aggregates were measured in four studies. Metrics of evapotranspiration, groundwater depth, and erosion control were measured in three studies, and surface flow and base saturation in two studies   Fig. 14 Cultural service indicators. Forty-three indicators were used in studies measuring cultural services. Social indicators (e.g. spiritual and symbolic value) were the best represented cultural service (n = 10 articles), followed by educational (n = 8), aesthetic/bequest (n = 8) [67], recreational (n = 6), and tourism/ornamental value (n = 5). Methods applied included: a preference-based analysis procedure [67]; ethnography [64]; transect walks; semi-structured interviews [68]; stakeholder workshops [67]; focus groups; and life-history interviews [69]. In particular, religious value was commonly inferred by identifying sacred sites, locations where ritual practice was carried out, and the number of ritual artefacts used for festivals and weddings [67]. Other studies used the proxy of farmers' donations of cereal grains to religious institutions [67]     improvement in yield but decline in ecosystem service provision. Although useful for broad interpretation, in the irst instance, further research into speciic pathways of change in ecosystem service provision is necessary.

General discussion
he systematic map illuminates some general trends in the available evidence measuring the efectiveness of onfarm conservation land management for preserving or enhancing ecosystem services. Currently, the types of interventions reported are closely related to contemporary understanding and priorities in agriculture, and in particular conventional agricultural research. he review showed most indicators measured have a strong association with crop productivity, economic considerations being key drivers of farmers' decision-making. Of particular note, a higher number of studies assessed inorganic fertiliser in contrast with organic fertiliser.
While previous reviews suggest that most assessments measured pollination and provisioning services (e.g. [79]), half of the studies in this review measured supporting services (55 %) and one-third measured regulating services (33 %). he majority of these studies have measured the efects of production on carbon sequestration and storage (26 %), nutrient cycling (17 %) and soil regulation (19 %). his emphasis could be attributed to recent concerns about the efects of increasing atmospheric concentration of CO 2 and other GHGs on climate change, and growing interest in how terrestrial sinks can mitigate warming.
While the review covered a wide range of management practices, a large proportion of studies assessed tillage (24 %) and organic fertilisation regimes (24 %). Since 2011 [80,81] there has been a notable increase in interest in three particular soil conservation practices, namely mulching (~4.5-fold), organic fertiliser (~threefold) and alternative tillage regimes (~twofold). his suggests a shift in the research agenda towards agro-ecological practices. However, the practices currently studied are still limited in scope, focusing on the efect of farmyard manure fertilisation (10 %), no tillage (9 %), multipurpose or multistorey cropping (8 %), or dry fallowing (5 %).
he systematic map found an exponential increase since 1992 in publications covering the topic of on-farm conservation land management. Based on analysis of date of publication of these articles, we suggest increases could be associated with (1) the publication of seminal reports (e.g. MA in 2005 [6]), TEEB in 2008 [12]); (2) the launch of the 'climate-smart agriculture' concept of the FAO in 2010 [80,81]; (3) international meetings (e.g. the Rio Summit in 1992 and Rio + 20 in 2012); (4) the formation of international alliances (e.g. Intergovernmental Panel on Biodiversity and Ecosystem Services (IPBES) in 2012); and/or (5) a general increase in inancial investment, greater corporate sponsorship, and a wider variety of inance tools available for ecosystem service valuation [82].
We identiied three geographical gaps in research of on-farm conservation land management. First, there is a dearth of research in Africa (constituting 25 % of studies) and in Central and Latin America (constituting 25 % of studies), dwarfed in comparison to the good representation of studies in Asia, particularly India and China. Second, few studies were conducted at the regional and landscape levels, which may skew results away from ecosystem processes that operate at these scales (e.g. certain hydrological services). hird, more studies were conducted on research stations (76 %) than on working farms; large-scale/commercial farms constituted only 7 % of studies and combined assessments of small-and largescale farms constituted 2 %. hese trends may be associated with a broader movement away from experimental work towards theoretical work [83]. he trend could also be as a result of methodological barriers to working in the ield (e.g. labour and time investments, site access, permissions, language, payments), and other practical reasons that require researchers to draw on existing networks and institutional ailiations, rather than setting up new ield sites. he reported research is therefore biased towards a limited set of known research areas or stations. More work is needed to connect broader scale modelling and ieldbased, ground-truthed data. Centralising open access data can help to ensure the efective recording and use of this valuable information when it is collected.
We further identiied that multiservice, mixed-method, and multidisciplinary studies were conspicuously absent, even though the importance of multidisciplinary research Table 9 Overall outcomes on ecosystem service provision and yield, as reported by authors Table 9 shows the conclusions drawn by study authors, based on their results for yield and ecosystem services outcomes: 181 studies claimed on-farm conservation land management both improved ecosystem services and yield. This inding is a key topic for further exploration in a systematic review that includes statistical analysis (ES ecosystem service provision) is widely acknowledged [84]. Most studies (92 %) measured just one or two ecosystem services. Only one study measured perceptions and management of carbon storage and sequestration even though this was one of the most frequently measured services. Further, only four studies used an experimental design to assess cultural services, while the remaining studies were non-experimental. Cultural services also constituted the smallest proportion of studies (3 %), conirming that we have limited data (not only in agricultural landscapes) of nontangible beneits and non-marketable functions of ecosystem services [67,85]. Interestingly, only ive studies on tourism value in agriculture as were detected in our search, and six in recreation, aesthetic, or educational value. herefore, our results indicate a clear need for a broader suite of indicators to be studied in new research projects so that land management decisions are based on a more realistic array of ecosystem outcomes. he review further showed that most studies took place on a time-scale that is inadequate for understanding whether impacts of management decisions are temporary, or have more permanent consequences for ecosystem services (54 % were <4 years). his presents a particular challenge for ecosystem processes and services that operate on longer time-scales in the order of decades, such as soil carbon storage. herefore, more longterm experiments and monitoring are needed.
Finally, the inding that most studies assessing the efectiveness of on-farm conservation land management are published in journal articles (97 %), only onethird of which are open access, suggests that access to this knowledge may be limited for decision-makers outside of academia. Few studies based on direct local evidence are available in the literature published by relevant organisations and institutions (2 %). his raises a potential disconnect between the types of management practices promoted by organisations likely to have a presence in the ield and the empirical evidence of their eicacy in promoting ecosystem service delivery.

Limitations in searching
We acknowledge that studies published before 1984 are less likely to be available in electronic format and were therefore not a source of literature for the current systematic map. Although the search strategy was widely circulated amongst our stakeholder groups and published in the protocol [32], there may have been terms we may have missed to produce additional records, such as practices (e.g. direct and mulch tilling) or ecosystem services (e.g. soil organic carbon stores in cases where the phrase 'carbon sequestration' is not used in the title, abstract or keywords).
Low representation of particular taxonomic groups or crops may be a result of the exclusion of biomes in the search. Two of the five journals selected during stakeholder workshops for targeted searches (Field Crops Research and Agriculture and Ecosystems and Environment) were the most frequently cited journal titles, which may have resulted in familiarity bias [86]. Websites from regional agricultural research consortiums, rather than national research agencies were selected to reduce bias towards governmental agencies, although might have limited the number of studies from some countries. Bias towards countries most frequently studied is also likely to be reflected in the language and country population densities (e.g. China, Brazil, India and Mexico).

Limitations in interpretation
Limitations in interpretation could have arisen from categorising overlapping practices, aggregating multiple practices, or reporting the number of articles rather than multiple outputs of one study. Further, as this map did not set out to evaluate the quality of evidence, results may illuminate evidence gaps, but we cannot infer the robustness of studies beyond whether they are experimental, quasi-experimental, or correlative. Although the terms experimental, quasi-experimental, or correlative were deined, there is also room for interpretation of what is randomised or manipulated-thereby introducing the possibility of an overrepresentation of experimental studies. Nevertheless, users of the map may read the studies relevant to the ecosystem service, intervention or crop of interest in order to assess their quality using existing criteria [23].

Conclusion
his systematic map provides a robust synthesis of the evidence on the efectiveness of on-farm conservation land management for preserving and enhancing ecosystem service provision in agricultural landscapes in developing countries. he systematic map of 746 studies, in combination with an interactive online platform that geographically maps results (accessible at https://oxlel. zoo.ox.ac.uk/resources/ecosystem-services-onfarmconservation-map), allows users to interrogate diferent aspects of previously fragmented evidence through a deined database ield structure. he database provides evidence concerning a wide range of conservation land management practices-particularly tillage, agroforestry, organic fertilisation and water conservation-which impact key ecosystem services-particularly carbon sequestration, soil/water regulation, nutrient cycling and biodiversity.

Implications for policy and management
Natural landscapes will continue to be converted to agricultural landscapes, and extensive land converted into intensive land, with 7.5 million km 2 expected to be converted by 2050 [87]. In the face of unprecedented agricultural expansion and land use change, in the future there is a high likelihood the management will be geared to favour some services (e.g. provisioning) over others (e.g. supporting). While future work is needed on alternative public and private payment schemes, institutional arrangements, and the important matter of explicit trade-ofs arising from conservation management, results emphasise that more value should be attached by development planners to the importance of minimal input, multifunctional agriculture, sustainable intensiication, and conservation agriculture.

Implications for future agriculture and ecosystem services research
Future research needs include: • long-term experiments (20 years+) that assess temporal stability, and response and recovery from a variety of disturbances, particularly of biodiversity on farms and functional diversity; • studies in 28  • studies of perennial crops, especially cash crops, such as tobacco and rubber.

Implications for extension of the systematic map
he systematic map is easily updatable and may be extended to include additional data analysis. An extension of the map might consider key economic and livelihood metrics, funders of studies (to test whether funding may determine emphases on fertilisation and major commodity crops, for example), and other likely environmental variables that could lead to heterogeneity in determining overall efect of land management (e.g. soil type, mineral texture class, altitude, slope, species mix in cropping systems, previous land uses, and surrounding land uses) [32]. he systematic map could also be expanded to include studies in other major languages, especially French and Spanish.

Implications for future systematic reviews
Future work could usefully look in more detail at elements of this systematic map through a full systematic review. Such systematic reviews could focus on individual crops, or particular regions, indicators, or management strategies. We identiied the following key questions that appear to have suicient primary research to carry out systematic reviews, and have current global relevance to policy and management: • 'What is the ecological impact of key agroforestry crops with therapeutic properties and livestock fodder crops (including Adansonia digitata (baobab) [88], Moringa oleifera L. (moringa) [89] and Vitellaria paradoxa (shea tree) [90]), on improving smallholder farmer livelihood in Sub-Saharan Africa and Asia?' • 'What are the economic, ecological and social costs and beneits for smallholders of shifting from conventional to short maturing seed varieties?' • 'What is the impact of livestock, population density and land holding size on organic manure availability and soil organic carbon?' • 'What is the impact of intercropping leguminous cover crops and traditional staple cereals (e.g. sorghum, maize and cassava) on soil/water regulation, and nutrient cycling?' • 'What is the impact of ire management regimes on carbon sequestration, water resources, air quality, and biodiversity in developing regions?' (While previous reviews have been conducted in speciic countries, such as the US [91], Australia [92], and Ghana [93], we identiied no such regional reviews.) • 'What is the impact of large-scale or commercial production on preserving ecosystem services in palm oil, jatropha, and soya bean cropping systems?'