Socioeconomic impacts of land restoration in agriculture: A systematic review

Abstract

Land restoration programmes are an increasingly popular tool to reduce degradation and increase livelihoods. This study aims to summarize the available evidence on the socioeconomic impacts of interventions aiming to promote land restoration or prevent land degradation, including agroforestry, conservation agriculture, integrated soil fertility management, and soil and water conservation. Using a systematic approach to selecting and assessing the quality of relevant studies, this review reveals a paucity of evidence. We identify twenty-nine causal studies, of which just six assess socioeconomic outcomes, only three studies were conducted outside of Africa, and none assessed agroforestry. The twenty-nine studies provide insights into thirty-five intervention–outcome combinations, of which 71 per cent reveal a positive impact on the outcome studies. We discuss the implications for policy and research.

1. Introduction

Land degradation is a significant obstacle to agricultural productivity and can lead to local declines in agricultural production (UNCCD 2017; FAO 2020; Van der Esch et al. 2022). Reversing land degradation and restoring land and soils are essential for a sustainable food system. This includes feeding growing populations while conserving biodiversity and limiting agriculture's impact on climate change (Tilman et al. 2011; Alexandratos and Bruinsma 2012). Land degradation—the loss of soils, nutrients, and water holding capacity—is typically caused by soil nutrient mining, erosion, and loss of vegetative cover. It affects anywhere between 15 and 75 per cent of the global landmass. Recent studies show that around 25 per cent of the global land mass is subjected to human-induced land degradation (see Van der Esch et al. 2022 for an overview). The economic loss resulting from land degradation is estimated to amount to at least US|${\$}$|15 billion, or about 1 per cent of annual global GDP (Nkonya et al. 2016b). Land restoration encompasses the improvement of natural ecosystems and the rehabilitation and sustainable management of lands under human use. Applying land restoration measures in agriculture is therefore synonymous to preventing further land degradation. Restoration measures are thus often similar to sustainable land management measures.

The need to combat land degradation and to invest in land restoration is quickly rising on policy agendas. The UN Decade on Ecosystem Restoration (2021–30) aims to mainstream land restoration in policies and investments. Existing commitments to restore or rehabilitate land by countries under the Rio Conventions and the Bonn Challenge add up to between 0.8 billion and 1 billion hectares, with almost half of these commitments made by sub-Saharan countries (Sewell et al. 2020). Several initiatives aim to upscale land restoration, such as the Bonn Challenge, AFR100, and Initiative 20 × 20. In the EU, a proposal is expected by the European Commission with legally binding targets on nature restoration. The total investment that would be required to implement the current restoration pledges and commitments by countries, falling within the scope of the restoration decade (Sewell et al. 2020), is estimated between US|${\$}$|300 billion and US|${\$}$|1,700 billion (Van der Esch et al. 2022).

Despite these ambitious plans, the potentially high investments required for land restoration warrant a closer inspection of their expected benefits. Many of these relate to water management, biodiversity conservation, and carbon sequestration (Navarro et al. 2017; Barbier and Hochard 2018; IPBES 2018; Pretty 2018; Pretty et al. 2018; Roe et al. 2019). In addition, there are potential projected private benefits to local stakeholders—enhanced resilience and productivity of agriculture, livestock, and forestry—a core motivation for increased investment in restoration measures. Whether, when, and where these benefits will materialize unfortunately remains an open question. At the onset of the decade of restoration (UNEP and FAO 2021), increasing knowledge about potential impacts in this area is imperative. This systematic review aims to fill this void.

We follow established protocols and review the literature to assess the impacts of land restoration practices on farm households. The purpose of our systematic review is to aggregate the available research and assess the causal impact across the available evidence for these types of interventions (see Waddington et al. 2014). We focus on four of the most common interventions (Pandit et al. 2018): (1) soil and water conservation (SWC), (2) integrated soil fertility management (ISFM), (3) conservation agriculture (CA), and (4) agroforestry (AF).

Narratives that land restoration is a ‘triple win’ strategy leading to inclusive green growth feature commonly in high-level policy documents (e.g. World Bank 2012). There is, indeed, substantial evidence from agronomic (often researcher-managed) trials that land restoration interventions improve crop yields and reduce soil degradation (Brouder and Gomez-Macpherson 2014; Corbeels et al. 2014; Droppelmann et al. 2017). However, there is less evidence available on whether these interventions also improve farming outcomes under farmer-managed conditions. Even more so, whether they also improve aspects of economic wellbeing such as household income and food security is poorly understood (Barbier and Hochard 2018; Prince et al. 2018). In fact, an aggregate positive impact is by no means guaranteed. First, the impact of some of the interventions on crop yields is strongly heterogeneous, as documented, for instance, in the case of CA (Brouder and Gomez-Macpherson 2014). Second, many studies take only a partial view on income derived from a specific crop or land restoration method. Studies rarely control for possible reallocations of productive inputs, such as labour, across household activities and the aggregate effect on income. Such reallocations may explain why adopting a promising sustainable agricultural technology does not always enhance income, even when crop yields increase (e.g. Takahashi and Barrett 2014).

Because of this potential discrepancy between agronomic trials and the on-farm reality, in our review we move beyond agronomic trials and focus solely on studies that investigate interventions under farmer-managed conditions. We further make a distinction between farming outcomes, such as crop yield and crop income, and aggregate socioeconomic outcomes, such as household income and food security.

Using forty-nine search terms (see Table 1), we systematically search twelve international databases. We initially obtain 3,786 studies and then use pre-defined criteria to screen out studies with a high risk of bias (following Waddington et al. 2014). The screening retains rigorous studies that make the strongest case that observed changes in outcome indicators are caused by land restoration methods, and not simply correlated with them (when, for instance, wealthier households are more likely to adopt conservation and restoration methods).

Despite long-term promotion of land restoration methods, we identify just twenty-nine relevant studies that are considered rigorous (based on a risk-of-bias assessment assessing the risk of selection bias, spillovers, and reporting bias). Together, these studies assess impact for thirty-five combinations of an intervention with an outcome indicator. Only six document impact on socioeconomic indicators such as income, poverty, and food security. Of these thirty-five combinations, 71 per cent report a positive impact of the interventions studies. For the remaining studies, impact is negative or not significantly different from zero. These findings resonate with similar systematic reviews in the agricultural domain,¹ which suggest that private benefits from land restoration cannot be assumed as given, and that considerable variation exists across the type of restoration method and localities.

In the next section, we discuss the methods underlying this systematic review. Section 3 presents a quantitative descriptive analysis of the characteristics of the studies followed by an in-depth qualitative assessment that we discuss in more detail in Section 4. Section 5 concludes.

2. Methods

We follow the methodologies put forward by the International Initiative to Impact Evaluation (3ie) and Campbell Collaboration (e.g. Waddington et al. 2012) in this review. Below, we describe the strategy used to identify, evaluate, and analyse relevant studies on land restoration, including the database search, screening process, and risk-of-bias assessment. Our final list contains twenty-nine low and medium risk studies (see Fig. 1 for an overview of the steps and papers retained at each step).

2.1 Database search

The database search was conducted from October to December 2018. The search is based on a list of four sets of keywords associated with land restoration: (1) the processes that underlie land degradation, (2) the general aspects of sustainable land management, (3) specific land restoration approaches, and (4) modes of technological transfer and dissemination. In order to increase the odds of finding all relevant studies in the databases screened, we include several keywords (see Table 1) capturing distinct elements of each land restoration approach relating to SWC structures, ISFM practices, CA practices, and AF practices.² For detailed insights on the agronomic principles underpinning these practices, and how these improve productivity, see e.g. Vanlauwe et al. (2010), Farooq and Siddique (2014), Adimassu et al. (2017), Nair et al. (2021), and references therein.

The keywords included in Table 1 form the basis for our search strategy. Using several databases, we search through all titles and abstracts for studies containing at least one keyword associated with the first three categories on land degradation and sustainable land management practices, and at least one keyword associated with the mode of technological transfer. This set-up ensures that we primarily identify studies that focus on farmer-managed adoption (instead of researcher-managed trials) shaped by a clearly defined mode of extension. Moreover, it keeps the search strategy manageable as only searching on categories 1–3 would yield hundreds of thousands of publications. The full search strings used for each database are provided in the online Supplementary Section (Table A1).

The choice of which databases to include was motivated by several pragmatic considerations: the database must be open access or available to one of the co-authors via an existing institutional subscription; it allows for advanced Boolean search using sufficiently long strings; and is able to export citation lists in bulk. Of the twenty-six databases originally considered, twelve met these criteria: AgEcon, AGRICOLA, Agris, ArticleFirst, CAB Abstracts, ECO, EconLit, GreenFile, OpenGrey, Scopus, SocIndex, and Web of Science.³ Given the large number of databases included in the search, it is unlikely that we systematically missed out on key studies. In addition, we asked key experts to screen the final list of retained studies. The search yielded a total of 3,786 publications after removal of duplicates.

2.2. Screening and quality assurance

2.2.1. Screening on methods

Moving from the long list of potentially relevant publications, we further refine our search to select studies that use rigorous impact evaluation methods, including randomized controlled trials (RCTs) (or experiments), regression discontinuity (RD) designs, difference-in-differences (DiD), instrumental variable (IV), and propensity score matching (PSM).

We use an automated screening script in Python (see Supplementary Section A3 for the full script) to select studies that mention one of these methods. We first download all full-text PDFs of the studies. This step reduced the number of publications to 2,708, as the remainder either could not be accessed or found online or were not written in English. We then use the resulting PDFs as input for the automated screening process, involving four steps:

1. Extract text from PDF files and set aside PDFs that cannot be extracted.⁴

2. Write text (.txt) files and set aside the studies for which text files are empty or too small (due to failed extraction).⁵

3. Parse through text files and evaluate whether any of the methods are mentioned, using regular expressions.⁶

4. Select papers that mention any of the methods.

This automated process leaves us with 462 papers that mention any of the methods. In addition, 259 studies that are not extracted are manually screened on mentioning of the methods. Of those manually screened, none of the studies are relevant for this review.

2.2.2. Screening on relevance of intervention, outcomes, and methodology

We then conduct a manual screening on the relevance of the 462 papers in terms of the interventions and outcomes studied. In addition, we confirm whether the papers actually apply any of the impact evaluation methods. For each paper, this screening is carried out by two authors independently, using four criteria:

1. The study should focus on a relevant intervention (SWC structures, ISFM, CA, or AF).

2. The study should have a relevant outcome (farm production, farm productivity, farm income, household income, food security, or poverty).

3. The study should use one of the pre-specified evaluation methods (RCT, RD, DiD, IV, and PSM).

4. The study should not be an agronomic field trial.

We also exclude studies that assess the impact of input subsidies, which was reviewed in Hemming et al. (2018). We do, however, retain studies that consider packaged interventions, possibly including input subsidies, that also include any of the four land degradation measures. After this step, we have a list of forty-six studies.

2.2.3. Expert opinion and snowballing exercise

To be certain that we do not miss any important studies, the final list of papers is sent to several experts in the field, yielding two additional studies. Finally, we conduct a snowballing exercise by scanning the reference lists of selected studies so far, leading to fifty-two potentially relevant studies that had not been identified yet. These are subsequently screened, as described in the previous section. Those selected for further review are assessed in terms of their risk of bias, as described below. This snowball exercise yields eighteen additional studies, bringing the total to sixty-four studies.

2.2.4. Risk of bias

We analyse the selected papers by synthesizing key characteristics of the studies (region, intervention type, and outcomes), and provide more in-depth qualitative discussion of the results, and interpretation of the results for each land restoration approach (following the approach of Higgins et al. 2018). As a final step in this analysis, for each study we assess the risk of bias, to assess internal validity of the evidence presented. We follow the tool proposed by Waddington et al. (2014), which was developed for assessing the risk of bias of a paper that applies an impact evaluation method (i.e. the methods we purposefully selected on). With the tool, each paper is assessed on four aspects: (1) selection bias, (2) spillovers, (3) reporting bias, and (4) other sources of bias. For each of these categories, we assess whether the paper meets a list of criteria that is specific to the method used in the paper (see Table A4 in the online Supplementary Section for the list of criteria). Based on this assessment, we simply count for each category whether the criteria were met. If, for one or more of the four categories, the paper does not meet the criteria, it is assessed as having a high risk of bias. If, for two or more of the four categories, it is unclear whether the criteria are met (e.g. because it is not reported in the paper), the paper is assessed as having a medium risk of bias. If otherwise, the paper is assessed as having a low risk of bias. Each study is evaluated by one of the authors and checked by a second author. The study is evaluated by a third author if any disagreement occurs. Detailed outcomes of this evaluation are presented in Table A5 in the online Supplementary Section and summarized in Fig. 2.

We exclude twenty-two publications because they are not relevant in terms of interventions, outcomes, and/or methods. Of the remaining forty-two publications, twelve publications are assessed having a low risk of bias, seventeen publications as having a medium risk of bias, and thirteen publications as having a high risk of bias. For the subsequent review (Section 3), only the twenty-nine studies with a low or medium risk of bias are considered. Studies with a high risk of bias often lack sufficient information on methodological choices in addition to incomplete result tables. All excluded studies use an IV or matching method. For IV studies, the strength of the instrument is often not discussed, nor is a first-stage regression always reported. In the matching studies, matching is often carried out on endline characteristics that could have been affected by participation in the programme. None of the selected studies used RD methods.

Based on our experiences of applying the risk-of-bias tool (Waddington et al. 2014), several issues are worth mentioning. Firstly, one of the criteria relates to the way studies deal with spillovers. This is irrelevant for most studies assessed (especially IV studies), we therefore sometimes score a study as having low risk of bias, even though spillovers are not explicitly discussed. Secondly, the risk-of-bias assessment tool does not penalize multiple hypothesis testing. However, many studies do present a large number of regressions and models. Relatedly, we suspect that there is a large chance of a publication bias, as many of the papers that we do find show significant results. It is plausible, similar to other systematic reviews, that studies showing significant results are more likely to be published than studies that find null effects (Franco et al. 2014; Brodeur et al. 2020). Since we do not conduct a statistical meta-analysis (see Section 4), we also do not formally test for publication bias.

3. Results

3.1. General overview

As part of the quantitative description of our analysis, we summarize the number of publications for each land restoration approach for each studied country (Fig. 3). Close to all studies are conducted in Africa, with a regional focus on East Africa (Ethiopia, Kenya, Malawi, Tanzania, and Zambia) and West Africa (Ghana and Nigeria). Research from other contexts is limited to three studies: a study on CA done in Syria, a study on ISFM in India, and a study looking at a number of countries in sub-Saharan Africa. There are no studies conducted in Latin America.

There are also no studies with low and medium risk of bias on AF.⁷ This is surprising given the widespread expectation that AF is crucial for attaining some of the UN Sustainable Development Goals (Waldron et al. 2017; van Noordwijk et al. 2018). A recent systematic review on AF interventions by Castle et al. (2021) paints a similar picture, concluding that AF has the potential to improve agricultural yields, but evidence on socioeconomic outcomes is extremely limited. Our review and the one by Castle et al. (2021) both illustrate an obvious lack of evidence on the impact of AF interventions, despite significant investment in AF by donors, governments, and NGOs.

Next, we construct an evidence gap map (Fig. 4), a visual tool that allows for inspecting combinations of interventions and outcomes that are relatively well studied, versus combinations that are not (i.e. the evidence gaps). To construct this map, we use all intervention–outcome combinations of the included studies, yielding a total of thirty-five combinations (some studies assess impact of multiple interventions or outcomes). The details of the combinations are provided in Table A6 in the supplementary material. When studies assess impact of multiple interventions, or on multiple outcomes, this is evidenced by multiple entries in the respective columns in Table A6. We also include information from the risk-of-bias assessment (low/medium risk) in this evidence gap map. Nearly all studies consider changes in farming outcomes due to SWC, ISFM, and to some extent CA. As mentioned, there are no studies among the twenty-nine that consider impacts of AF. ISFM is the land restoration intervention for which most studies met our criteria. For both SWC and CA we select fewer studies. As for the outcomes, we classify the outcomes into two different types: (1) farming outcomes (all outcomes that measure only partial output of a household, e.g. crop yields or value of production) and (2) socioeconomic outcomes (e.g. household income, food security, or poverty). Most selected studies are limited to farming outcomes, and just six studies also report on socioeconomic indicators.

3.2. Assessment per land restoration type

In this section, we present the results of a qualitative review of the included studies. Figure 5 synthesizes the results by type of intervention and impact (positive, negative, or no effect). Three of the six publications that study the impact of land restoration interventions on a socioeconomic outcome (e.g. household income or poverty) find a positive effect and two-thirds of the studies find a positive effect on farming outcomes (e.g. crop yields or value of production). Below, we describe the main insights for the three types of interventions (SWC, ISFM, and CA).

3.2.1. Soil and water conservation

Ten studies explore the impact of SWC on farming and, to a limited degree, socioeconomic outcomes. Geographically, the studies provide a very narrow snapshot, with three studies done in Ghana, one in Kenya, one in Tanzania, and the remaining five in Ethiopia. The types of interventions vary across studies. Usually, a combination of technologies is studied, including bund and terrace construction. Most of the studies look ex-post at the effect of adoption of a technology, rather than at the impact of an actual SWC intervention.

With respect to changing farming outcomes, four studies find positive effects of SWC, with effect sizes ranging 20–24 per cent on crop yield or value of production (Abdulai and Huffman 2014; Arslan et al. 2017; Schmidt and Tadesse 2017; Kassie et al. 2008). Abdulai and Huffman (2014) also report impact on net rice profit and find a significant increase of 16 per cent. Two studies report positive effects but do not report baseline values or effect sizes (Kato et al. 2011; deGraft-Johnson et al. 2014). Two studies report null effects (Faltermeier and Abdulai 2009; Schmidt and Tadesse 2017). However, Kassie et al. (2008) find that for households adopting terraces, net value of crop income is reduced by 15 per cent. Three studies also look at the heterogeneity of impact as a function of climatic variation and find that SWC methods are particularly useful in regions with low rainfall or high climatic variability (Kassie et al. 2008; Kato et al. 2011; Arslan et al. 2017). Schmidt and Tadesse (2017) estimate the marginal effect of each additional year of SWC adoption in Ethiopia, shedding some light on the long-term impact of SWC technologies. They estimate that SWC structures should be in place at least seven years for the technologies to have a significant impact on value of production. Wainaina et al. (2018) is the only study that looks at impacts beyond the farm and finds no significant effect of a number of technologies (including but not limited to SWC technologies) on per capita household income in Kenya.

In sum, while six out of ten studies report positive impact of SWC on farming outcomes, the impacts vary across contexts. SWC seems to hold most promise in areas with low rainfall and high climatic variability. There is a lack of evidence on socioeconomic impacts due to missing data. This makes it hard to draw firm conclusions, as increased farm outcomes may come with higher input costs, making the net benefits for households uncertain.

3.2.2. Integrated soil fertility management

Seventeen studies investigate the impact of ISFM. The study sites are largely in Eastern Africa (Malawi, Ethiopia, Kenya, Zambia, and Tanzania) and some in Western Africa (Ghana and Nigeria) with only one non-African study, located in India. Fourteen studies explore impacts of fertilizer application, in some cases accompanied by a training intervention. Other types of interventions include intercropping and crop rotations. Nine studies explore a package intervention, including, besides ISFM technologies, other technologies such as improved seeds, pesticides/fungicides, SWC technologies, and CA technologies.

In terms of outcomes, the studies present a narrow snapshot, with only two studies assessing a socioeconomic outcome whereas all others study a farming outcome. In these studies, the impact of ISFM on socioeconomic outcomes is negative or insignificant, whereas farming outcomes are, in eleven out of fifteen studies, affected positively. Ragasa and Mazunda (2018), one of the two studies looking at a socioeconomic outcome, study the impact of fertilizer use and access to extension services on value of production and three different food security indicators, in the context of a subsidized input system in Malawi on maize and legume farmers. The authors look at any type of extension services ranging from advice on fertilizer use, crop-specific trainings, and credit. They do not find significant effects of having access to extension services on production or any food security indicators. For the quantity of (subsidized) fertilizer used, they find no effect on production, and inconsistent, sometimes negative results on food security indicators. Wainaina et al. (2018) (also discussed in the previous section) study the impact of fertilizer, crop residue, and manure application on household (per capita) income. Manure is the only ISFM technology that has a significant positive effect (of 20 per cent) on per capita income. For farming outcomes, eleven studies show positive effects with large differences in effect sizes (from approximately 10 to 100 per cent increase in production or productivity) (Deininger and Olinto 2000; Chakravarty 2009; Zeitlin et al. 2010; Bardhan and Mookherjee 2011; deGraft-Johnson et al. 2014; Liverpool-Tasie et al. 2014; Asfaw et al. 2015; Kassie et al. 2015; Burke et al. 2017; Liverpool-Tasie 2017; Biggeri et al. 2018). Four of the found studies show no effect on a farming outcome (Pender and Gebremedhin 2007; Arslan et al. 2017; Abate et al. 2018; Ragasa and Mazunda 2018). In addition, three studies look at farm profitability of fertilizer use. Burke et al. (2017) find that depending on the specific fertilizer technology, fertilizer use may not be profitable for up to 92 per cent of farmers. Liverpool-Tasie (2017) concludes that fertilizer use is not profitable for 65 per cent of farmers. Both studies note that this lack of profitability can explain low adoption rates. The last of these studies does report profitability (Deininger and Olinto 2000).

Besides studying (in)organic fertilizer use, only two studies look at other ISFM practices. Kassie et al. (2015) look at crop diversification (intercropping and crop rotation), in addition to minimum tillage, a CA practice. They find positive effects of crop diversification on yield of around 30 per cent increasing to 80 per cent when used in combination with minimum tillage. Arslan et al. (2017) report no effect of intercropping (or organic and inorganic fertilizer use) on maize yield in Tanzania but do find yield gains for ISFM practices when they are used simultaneously, or in combination with SWC or improved seeds.

The only two experimental studies in this review were on ISFM technologies. Chakravarty (2009) studies the impact of a randomized fertilizer intervention on maize productivity for vulnerable farmers (HIV patients). Treated farmers increased yields by 9 per cent on average compared to the control group. In addition, treated farmers increased maize sales by 70 per cent (though this large effect can be explained by low baseline level of sales as most farmers were net consumers). Abate et al. (2018), with an RCT on a package of training, inputs (improved seed on credit, fertilizer, and gypsum), and marketing support on wheat yield in Ethiopia, find no significant impact on wheat yields.

In sum, even though ISFM is supposed to entail a variety of technologies aimed at improving soil fertility, all but two studies focused on fertilizer alone (although often in combination with non-ISFM practices). Most studies find predominantly positive effects on yields or other farm outcomes. We find no robust evidence that ISFM adoption also increases household incomes, and studies that try to explain low fertilizer adoption rates find that for many farmers, fertilizer use is not profitable at the farm level.

3.2.3. Conservation agriculture

Seven studies look at the impact of CA. Five studies were conducted in Eastern Africa (Zambia, Malawi, Kenya, and Ethiopia), one in multiple countries across sub-Saharan Africa, and one in Syria. Again, the evidence provides a geographically narrow picture. Four studies look at CA in combination with other technologies. The CA practice mostly studied is minimum or zero tillage. Just three studies also look at a socioeconomic outcome.

Most studies find positive effects of CA on farming and socioeconomic outcomes. Abdulai (2016) reports positive effects on socioeconomic outcomes measured by a reduction in poverty of 27–69 per cent in Zambia. Tambo and Mockshell (2018) look at the impact of three main CA techniques (minimum soil disturbance, residue retention, and crop rotation) and combinations thereof for a range of sub-Saharan African countries, also using observational data. They find no impact of the three practices separately but do find significant impacts when they are combined. Per capita household income increases by about 30 per cent. Wainaina et al. (2018) (also discussed in the previous section) evaluate, amongst other technologies, zero tillage in their study. Though they do find a significant positive impact on household income, this effect disappears when looking at per capita income. The remaining studies all find positive effects of CA on farming outcomes of 30–80 per cent (Pender and Gebremedhin 2007; Kassie et al. 2015; Abdulai 2016; El-Shater et al. 2016; Abdulai and Abdulai 2017).

Taken together, similar to the other two types of interventions, there is little rigorous evidence out there on how CA affects farm households. Most studies are conducted in African countries and focus on zero/minimum tillage, often in combination with other technologies. In all but one of the found studies, CA is associated with improvements in farming and socioeconomic outcomes.

4. Discussion

The evidence base revealed in this review is small, with considerable diversity in findings including many null results. Such a limited number of high-quality studies are equally observed in other systematic reviews, also within the agricultural domain in developing countries (Waddington et al. 2014; Lawry et al. 2016; Hemming et al. 2018). Within the field of study this review considers, the paucity of studies could also stem from the fact that ‘counterfactual thinking is only a rather recent development in the reservation and conservation sector’ (Mirzabaev and Wuepper 2023). A concurrent explanation is the long time horizon before these practices yield benefit to practicing households. Even for ISFM practices these could be considerable (Takahashi et al. 2020), as well as for SWC (see section 3.2.1) while it arguably takes a couple of seasons for trees and woody shrubs under AF practices to be fully grown. By that stage, development programmes and projects may have moved on, explaining the lack of impact assessments. Finally, the uneven geographical spread, particularly the lack of studies from Latin America, is equally observed elsewhere (Jain et al. 2023) and could result from language differences. Indeed, the fact that non-English studies were excluded is a limitation of this study.

No clear tendencies emerge across the types of restoration practices as impact (changes in outcomes) range from relatively large to very small (or no impact). The twenty-nine included studies report on thirty-five intervention–outcome relations, of which twenty-five (71 per cent) suggest a positive relation and the remainder non-significant or negative effects. Non-significant effects are particularly observed in five studies that investigate impact on household-level indicators. Somewhat more studies report positive impacts when considering on-farm impact.

The initial goal of this review was to conduct a quantitative meta-analysis thereby computing aggregate estimates of impact. But this turned out to be infeasible as either the data provided in many studies, or the research methods followed, are not apt for a formalized meta-analysis. Next, computing standardized impact scores across the great variation of interventions and outcomes studied would obscure more information than it would yield.

Calculating aggregate effects requires a comparison group, which was often not readily available. Very few studies follow an experimental approach (only two) and most studies (twenty-three) use an IV regression or PSM. Thirteen studies are regression-based and include a number of control variables and interactions with the treatment variable, which makes it difficult to compare treatment effects across studies. Secondly, the quality of reporting in many studies is insufficient to extract the necessary data for doing a meta-analysis. Descriptive statistics are often missing, as are necessary P values or t-statistics, in some studies the sample size remains unclear. In the case of IV studies the validity of instruments is often not being discussed, while some PSM studies match on endline data using variables that could logically have been affected by the intervention. For all of these reasons a quantitative meta-analysis was deemed impractical, also preventing further exploration of publication bias, and motivated the qualitative review used instead. At the same time, we urge researchers to consistently publish all such necessary statistics in future reports and articles in order to allow for meta-analyses in the future.

Considering the specific land restoration interventions, the impact of promoting AF practices remains particularly underreported. We retained only one study that investigates the impact of a package intervention with AF as a component. Seven and ten studies were identified on CA and SWC practices, respectively. Fifteen studies on ISFM are included, in absolute terms the greatest number amongst the four practices. However, the impact of components within ISFM (like composting, manuring, and specific crop rotations) remains unclear, as most of these assess packages also including inorganic fertilizer, improved seeds, and pesticides.

As argued, this review purposely moved beyond farm-level indicators (such as crop yields or aggregate farm production) in order to better understand the impact of such changes on household level, such as aggregate income or poverty. Our study thereby partially overlaps with another recent systematic review focusing on sustainable intensification (Jain et al. 2023). While our study focuses on land restoration methods, the study by Jain et al. (2023) encompasses a different range of methods, including improved crop varieties. Our studies overlap with respect to ISFM methods, whereby Jain et al. (2023) generally observe positive effects, partially mirroring our findings on ISFM. Moreover, they equally find only a limited set of high-quality impact assessments and to a subsequently narrow geographical coverage from which to draw conclusion.

The paucity in insights into the impact of land restoration practices at households stands in contrast with studies highlighting positive economic benefits of land restoration due to greater ecosystem services (e.g. Nkonya et al. 2016a). A major source of this discrepancy stems from different scales and the distinction between private and public benefits. However, Nkonya et al. (2016a) focus on quantifying the, mostly public, benefits of land restoration ecosystem services to societies at larger scales, beyond the farm level. Our study, conversely, focuses exclusively on the private benefits at the farm level. Note that even the studies that show null effects in our review could still harbour positive effects beyond the farm level. It is for this reason that Jain et al. (2023) call to include environmental indicators, beyond those measurable at the farm level, in impact assessments more often, to which we concur.

Indeed, this discrepancy also highlights the need to focus more on monitoring, quantifying, and, where possible, monetizing the actual ecosystem services that adopters provide to societies, such as carbon storage or biodiversity conservation. Such a call is, obviously, not a novel insight, as many authors highlight that most of the benefits of land restoration are public in nature to begin with (e.g. Mirzabaev and Wuepper 2023). Proposals to link smallholders to carbon markets or promoting Payment for Environmental Services are further testimony to this point. As our study queries the common narrative that land restoration methods increase household wellbeing, always and everywhere, the need to understand the magnitude of externalities provided to societies becomes even greater. Such insights are greatly needed to inform the design of effective policies to support land restoration in order to make the UN decade on ecosystem restoration a success.

Several other issues warrant further investigation. First, some of the studied interventions may mitigate downside risks. This implies these practices could still be beneficial to households, even in the absence of average income effects, when associated with a reduction in the risk of crop failure. This has been documented particularly for SWC. Three studies (Kassie et al. 2008; Kato et al. 2011; Arslan et al. 2017) show that the impact of SWC is especially significant in areas with temperature shocks or lower rainfall. The benefits of reducing downside risks in agricultural production systems are equally discussed elsewhere (e.g. Takahashi et al. 2020; Sarr et al. 2021).

Second, synergies between the restoration techniques require further investigation. Two studies (Kassie et al. 2015; Wainaina et al. 2018) find synergistic effects when minimum tillage and ISFM practices are used simultaneously, but other intervention combinations are possible.

Third, our findings show that little is known on how enhanced on-farm productivity could translate into improved socioeconomic outcomes at the household. The key avenue linking these two is through labour reallocation as also discussed elsewhere (e.g. Takahashi et al. 2020). Restoration practices are typically more labour intensive and the overall income effects could be negligible when factoring in labour being shifted from other on-farm or off-farm activities. This makes insights into differences in initial factor endowments crucial in understanding adoption and impact.

Fourth, in our review, as explained in Section 2, spillover effects are treated as something undesirable as it potentially biases a clean observation of the impact. While this is common practice in similar systematic reviews, this reasoning bypasses the option that the return to these technologies may be inherently shaped by spillover effects or network externalities. For instance, joint learning on the optimal local implementation of an ISFM method may increase the overall returns to the technology as well as the number of farmers adopting it (e.g. Barrett et al. 2022). Next, positive network externalities could be present when the overall returns to using SWC or AF in a community increase with a larger share of households practicing such methods, for instance when water holding capacity due to SWC only materializes when applied on larger acreages. Investigating such effects is out of the scope of this study and warrants future study.

5. Conclusions

Widespread land degradation poses a threat to long-term agricultural productivity and wellbeing. For this reason, projects that aim to promote land restoration are receiving more attention from policymakers, as evidenced for instance by the current UN decade on ecosystem restoration (e.g. UNEP and FAO 2023). Yet, implementing such large-scale programmes requires better a priori insights into the returns of various potential technologies in order to devise the most effective supportive policies. This systematic review was conducted with the aim of contributing to this knowledge gap, specifically by building a greater understanding of the impact of four commonly promoted technologies to restore land within farming communities: SWC, ISFM, CA, and AF. We purposely focus on assessing impact beyond the farm-level indicators (such as crop yields or production) at the household level (aggregate income and poverty effects) to assess the private returns to such methods.

Our study concludes that the number of studies assessing impact of these technologies at the household level remains extremely limited. Many of the retained studies rely on IV estimations, display considerable diversity in impact observed, either null or positive effects, and our findings could still be adversely shaped by publication bias. Moreover, the majority of the included studies focus solely on farm-level outcomes. Together, these points strongly suggest that the four land restoration practices do not raise incomes universally, but in some specific instances only. Much greater effort should thus be placed on rigorous impact assessments of land restoration programmes—shifting focus from estimating their impact on technology adoption to the impact on households as a result thereof, and on disentangling the different impact pathways as well as discerning which work under which conditions.

Acknowledgements

We would like to thank Willem Verhagen (PBL) for help with compiling the list of keywords on land restoration interventions. For their support in gathering the studies under review, we would like to thank Dewy Verhoeven, Frank Koot, and Joanne Rink (WUR). We would like to thank Emily Schmidt (IFPRI) for reviewing our final selection of studies. For creating some of the graphs of this study, we would like to thank Filip de Blois (PBL). We thank two anonymous reviewers for their useful and constructive comments.

Funding

PBL Netherlands Environmental Assessment Agency received funding from the Netherlands Ministry of Infrastructure and Environment, Economic Affairs and Ministry of Foreign Affairs to conduct policy research supporting the activities of the United Nations Convention to Combat Desertification (UNCCD). PBL, in turn, used part of this funding to support the involved researchers at Wageningen University and Research (WUR) and the International Institute of Food Policy Research (IFPRI). This research was jointly designed and conducted by the involved researchers at PBL, WUR and IFPRI.

Conflict of interest

We declare no competing interests. Neither the funders, nor UNCCD, had a role in the study design, data collection, data analysis, data interpretation, or writing.

Data availability

The data supporting this article are accessible in its online supplementary material.

Footnotes

References