In order to narrow down and describe the specific research options to be assessed, we clustered research interventions addressing Fusarium wilt around three general themes: (i) preventing the spread of the disease (especially Foc TR4) to currently unaffected regions/countries through research on (and implementation of) improved exclusion, surveillance, eradication, and containment (ESEC) measures; (ii) research on integrated crop and disease management (ICDM) to recover banana yields in areas affected by (all strains of) Fusarium; and (iii) research focused on developing banana cultivars resistant to Fusarium wilt. There are two fundamentally different approaches to developing resistant varieties: conventional breeding using the genetic diversity of banana or genetic modification of susceptible cultivars of economic importance, with the latter likely being applicable only for a smaller area due to country biosafety regulations.

Based on these considerations, four distinct potential research options to address FWB were selected and quantitatively assessed for this study:

The adoptable, public good innovations resulting from the research in the form of knowledge, practices, and technologies were formulated for each research option and the specific research agenda was detailed (Table 1). This provided a scope of work for each research option, required to budget expected research costs for the cost-benefit analysis. Even though some topics, e.g., epidemiology, pathogenicity, diagnostic protocols, clean seed, and mapping relate to more than one research option (see ESEC and ICDM overlap in Table 1), we costed each option separately so they can later be compared. If investment in both options occurred, there would be substantial synergies that would result in lower research costs.

Each research option has a distinct target domain, since each option focuses on or is applicable to certain cultivar groups and is thus (more) relevant in certain countries. We considered major banana producing countries in Asia and the Pacific, Africa, and Latin America and the Caribbean (LAC). Our focus was on countries with predominantly small-holder producers and a substantial dependency on bananas for livelihoods.

The research on ESEC is applicable to all six cultivar groups in 29 major banana producing countries threatened by Foc TR4 (Table 2). The agenda will contribute to the effectiveness of national plant protection offices, starting with a better understanding and assessment of risks of Foc TR4 introduction and spread (Dita et al., 2013; Biosecurity of Queensland, 2016). Field studies on movement of planting material, banana products, and soil and other practical experience (Pegg et al., 2019) will contribute to ESEC strategies. Basic information about the disease in the plant is central not only to ESEC, but to ICDM (Warman and Aitken, 2018) and should be expanded. Such basic knowledge is also applicable to a spatial model for the analysis of scenarios for different locations of first outbreak and the speed and likelihood of spread depending on the actions taken (see Flores et al., 2017, for an example on citrus greening disease). Such models will also contribute to more effective surveillance routines and routes and to more strategic actions for eradication and containment. Research also includes development of more effective measures of biosecurity both in large plantations and in zones of diversified small farms with bananas (Pérez Vicente, 2015; Kukulies and Veivers, 2017).

The research on ICDM is primarily applicable to commercial production for national and export markets (Table 2). The deployment of cover crops, microbial organisms, systematic crop rotation, and careful rogueing was considered unlikely among small growers who grow bananas for home consumption and local sale. Delivery systems for inputs and seed are also often deficient where small growers predominate. Research to develop such ICDM approaches centers on healthy soils with capacity to suppress disease build-up and crop management to reduce crop susceptibility and to limit inoculum accumulation. Research results on specific bacterial antagonists of the pathogen (Pérez Vicente et al., 2009; Shen et al., 2015; Bubici et al., 2019), crop suppressive effects (Huang et al., 2012), and integrated systems approaches (Haddad et al., 2018; Huang et al., 2019) already provide evidence for the potential of this strategy. Basic knowledge on Foc populations, cultivar susceptibility, and more advanced quantitative diagnostic tools should underlie the applied management approaches. While the ICDM approaches are applicable for Foc races 1 and SR4, the calculation of returns only takes into account the recovery of losses due to Foc TR4 projected based on the risk index or FOC scale.

The development of CBRC is proposed to have the widest applicability after ESEC with relatively easy uptake for home consumption, local and national markets and export, as long as eating and cooking quality and handling traits are acceptable (Table 2). A broad range of varieties for different cultivar groups is proposed. Different improvement approaches are considered (Table 1) based on improved varieties already available through conventional breeding by Embrapa, the Brazilian Agricultural Research Corporation (Silva et al., 2013), clonal selection in Cavendish by the Taiwan Banana Research Institute already cited, and mutation breeding in Australia (Smith et al., 2006). Screening for resistance among existing lines continues to cover more and more Musa diversity (Zuo et al., 2018; Chen et al., 2019; García-Bastidas et al., 2019). Important areas of work include screening for sources of resistance, more effective phenotyping tools, development of pre-breeding lines, marker-assisted crossing, and multi-site evaluation trials.

The development of GMRC is projected only for Cavendish with application to countries with limited export, but relatively large national markets. Current status of biosafety regulations and laws on GM crops were not taken into account but could reduce the number of countries included. The use of resistance genes for cis and trans modification and the identification of genes linked to resistance to guide gene editing have already advanced beyond proof of concept (Dale et al., 2017; Maxmen, 2019). Resulting materials are screened for resistance to all Foc races in greenhouse and then field trials with evaluation of postharvest handling and taste qualities are conducted.

Data required for our ex ante assessment were collected from three general sources: statistical databases, other published sources, and expert estimates. For consistency, we followed the same procedure for compiling and cleaning/adjusting data that was used in the previous assessment of RTB banana research options that is described in detail in Pemsl and Staver (2014). To facilitate the disaggregation of production data by cultivar group, we used information from FRuiTRoP (2012) in addition to the FAO⁴ statistics. To avoid bias due to annual abnormalities, we computed a 3-year average for banana yield and price for each country included (using 2010–2012 data for consistency with the previous assessments, older data if necessary). Since FAO does not separate production from large scale, commercial plantations from (semi−) subsistence production under smallholder conditions, some yield figures for cultivar groups other than Cavendish for countries with sizable banana export industry were capped using expert judgment to reflect smallholder conditions. Yield and production data were then used to compute the banana production area for each included country (see Table 3).

To populate the poverty impact model, we relied on data included in the World Development Indicators⁵, namely, the most recent information for each included country on poverty prevalence (total population and poverty rate) and (agricultural) gross domestic product (GDP) by country.

The benefit of the FWB research interventions is the yield loss avoided⁶ by either preventing or delaying the spread of the disease (ESEC) to an area, (partially) recovering yields in areas with infested soils through crop or disease management (ICDM), or replacing susceptible with resistant banana cultivars that are not affected by the disease (CBRC and GMRC). The magnitude of the benefits from each of these options largely depends on the scale and pace of disease spread, i.e., the size of the affected area that constitutes the target domain for the intervention.

In the absence of an established epidemiological model to predict the future spread of Foc TR4, we relied on the risk-index model developed by Scheerer et al. (2018a) to project the future disease spread and thus determine the expected yield losses that could be avoided by investing in the four research options. The risk-index model consists of two steps where for each country a score is assigned based on (i) the likelihood of initial outbreak of Foc TR4 (time lag in years) and (ii) the internal spread rate of the disease (disaggregated per cultivar type) once present in the area. Factors considered when scoring for the time lag for Foc TR4 to reach a country include the importance of mono-cropped Cavendish, global banana traffic to and from a country, quality of borders and internal plant quarantine measures, and land and other links to countries where Foc TR4 is already present. The higher the aggregated score for a country, the shorter the time lag for Foc TR4 to be introduced and established in the country. The internal spread rate is scored based on three factors: the quality of internal quarantine measures, the importance of Cavendish, and the importance of banana for research investment and public policy. The higher the aggregated score for a country, the faster the spread and thus the higher the expected loss of banana production due to Foc TR4 with differentiated losses depending on cultivar types. Loss of production was proposed between 1 and 8% of area planted during the first 5 years after detection depending on cultivar group and the country score for internal spread risk. Assuming an accelerating expansion of the disease-affected area, especially in the early years, in each successive 5-year period after first detection, the spread rate was calculated to increase by 50%. In a second, more conservative scenario, the spread rate only increased by 25% for each successive 5-year interval.

In Table 3, we show the projected banana area lost to FWB using the risk index for each country over time (2014–2039, in 5-year intervals) in percent of the national banana production area lost. The last column gives the absolute area rendered unsuitable for production due to FWB in the absence of interventions thus constituting the target domain for our research interventions. The projected national losses are organized by region to illustrate the projected impact of the disease in different parts of the world over the next 25 years. The large effect in Asia/Pacific is due to the fact that Foc TR4 has already been detected in several countries in this region and internal spread has progressed. Though the current spread of the disease is more limited in Africa, the results indicate the potential for very high negative impact especially on smallholders. In the model results, spread onset was most delayed for LAC. However, the recent detection of Foc TR4 in Colombia is expected to accelerate the spread in LAC and shows how difficult the prediction of initial outbreak is.

The accelerating nature of the spread over time as well as the slowing effect of a more conservative internal spread assumption is visualized in Figure 1.

Following the methodology used in the multi-crop RTB priority assessment, we simulated adoption of the innovations resulting from ICDM, CBRC, and GMRC over time based on the target domain and estimated parameters on adoption lag (time until first adoption of innovation), the adoption pace, and ceiling. The adoption pace is indicated by the time until full adoption. The adoption ceiling is defined as the maximum share of the total production area affected by the disease on which the innovation(s) will be adopted. The definition of adoption is different for the ESEC research option, since the decision to implement is taken on the national (or even regional) level instead of the individual producer level and thus either none or all of the (affected) production area in the country benefits. Table 4 gives an overview of the parameter assumptions used in the assessment.

The largest adoption area reached over the assessment period represents the reach of the intervention and is reported as one impact indicator in the results section. This largest adoption area figure is also the basis for the computation of the number of beneficiaries, an additional impact indicator. The conversion is conducted by division with country specific estimates of the average banana area per household and multiplication with the country specific average household size (RTB, 2011).

To quantify the benefits from adopting the innovations developed under each of the research options at the national level, we used a partial equilibrium ES model estimated over a 25-year period (2014–2039). This quantitative approach of computing the ES resulting from a research-induced supply shift is a standard procedure in the agricultural economics and impact assessment field (see, e.g., Alston et al., 1995) and has been used in many previous studies (e.g., Alene et al., 2009, 2018; Fuglie and Thiele, 2009). We assumed elasticities of supply and demand to be 1 and 0.5, respectively, across all technologies and countries due to lack of other information.

In the subsequent cost-benefit analysis, we discount the benefits computed in the ES model (i.e., apply an interest rate to account for the difference in time when benefits occur) and compare them with the discounted costs associated with each research options. The estimated research and development (R&D) costs for each research option (Table 4) were doubled for the cost-benefit analysis to account for (in-kind) contributions of national partners. Additionally, dissemination costs based on the marginal annual adoption area were included. Dissemination costs were set at US$50 per hectare for improved varieties and US$80 per hectare for knowledge intensive innovations such as crop or pest management practices. For ESEC, additional costs were included for (i) establishing the quarantine system in the amount of US$50 per hectare reflecting the initial capacity strengthening effort and (ii) maintaining the quarantine and surveillance system in the amount of US$5 per hectare and year prior to Foc TR4 arrival and US$10 per hectare and year once Foc TR4 is present.

The key indicators resulting from the cost-benefit analysis are the net present value (NPV) and the internal rate of return (IRR) on the investment.

In addition to computing the standard economic indicators NPV and IRR, we also assessed the potential impact of each of the research options to reduce poverty. To do so, we estimated the marginal impact of an increase in the value of agricultural production on poverty reduction using elasticities of agricultural productivity growth. We applied the regional elasticities proposed by Thirtle et al. (2003) who found that a 1% growth in agricultural productivity reduces the number of rural poor by 0.72% in Africa, 0.48% in Asia, and 0.15% in LAC. Following Alene et al. (2009), we calculated the reduction in the number of poor by considering the estimated economic benefit of the respective research option as an increase in the agricultural production value.

Ex ante assessments by their nature predict uncertain future outcomes of potential investments and results are based on (expert) estimates of the costs and effects. Results have notoriously been too optimistic with regard to the benefits while underestimating (unknown) costs and problems. Alston et al. (1995) proposed sensitivity analysis to remedy this situation. In order to probe the robustness of results and take into account the tendency of technical experts to be overly optimistic with regard to the performance of and demand for their technologies, we embedded different adoption scenarios in the assessment. We also conducted sensitivity testing for additional key variables, focusing on parameters elicited from experts rather than model inherent parameters (such as elasticities and discount rate) or parameters populated based on statistics (e.g., banana production area, yield, or farm-gate prices). In order to keep the number of scenarios manageable, we focused on the three most crucial parameters which at the same time seem most prone to overly optimistic assumptions: (1) adoption area of the new technology, (2) time when adoption starts, and (3) magnitude of the farm-level benefit realized when adopting the technology.

For ICDM, CBRC, and GMRC, we assessed two different adoption scenarios: the first uses the adoption ceiling estimated by experts, while the second one is more conservative and uses a 50% reduced adoption assumption. Subsequently, we tested additional scenarios with delayed adoption and reduced effects for these three research options.

Since adoption is either 0 or 100% at the national level for ESEC, we constructed three increasingly cautious scenarios in order to test sensitivity of the results to changes in key parameters (see Scheerer et al., 2018b). For Scenario 1, we assume that ESEC will lead to a doubling of the arrival time (e.g., the disease will first occur in a country after 10 instead of 5 years) and a 50% reduced increase in spread rate once Foc reaches the country (i.e., 12.5 instead of 25% increase of spread rate every 5 years). Scenario 2 has a less delayed first arrival [Scenario 1 minus 5 years, i.e., 5 years earlier than under Scenario 1 in our example after (5 years × 2) – 5 = 5 years] and the same reduced increase in spread rate of 12.5%. Scenario 3 has the same less delayed arrival from the second scenario and a small reduction in the increase in spread rate (18.75 instead of 25%) once the disease breaks out in a country.

The assessment results show that all research options for FWB generate positive NPVs under all adoption scenarios (Table 5), indicating that investment in all research options is profitable. The magnitude of NPVs, however, varies considerably across options ranging from US$35 million for the most conservative scenario of improved quarantine and surveillance measures to reduce Foc TR4 spread (ESEC) to slightly over US$1 billion for ICDM to facilitate commercial production of partially Foc-resistant cultivars on Foc-infested soils (expert adoption assumption scenario). Since R&D costs, or the level of investment required, vary substantially across research options (US$8.5—47.7 million), the NPVs cannot be used to rank the research options. Instead, the IRRs (the interest rate realized on the invested amount) are a preferred measure for ranking alternative technologies.

All assessed FWB research options yield positive IRRs that are above the standard 10% interest rate (Table 5). Even under the lower adoption scenario the IRRs are positive and mostly well above 10%. Again, there is considerable variation in the return on investment among research options and adoption scenarios. ESEC Scenario 3 yields an estimated 11% return on the investment while the higher adoption scenario for ICDM reaches an estimated 36% IRR. The three ESEC scenarios show the lowest returns on investment, just slightly above the 10% threshold. These lower returns result from additional cost variables we included compared to the other research options. In addition to the R&D costs (matched 1:1 with national partner costs for the assessment) and the dissemination costs included for all research options, we added the costs of establishing quarantine systems reflecting the initial capacity strengthening effort and the costs of maintaining the quarantine and surveillance system. The resulting costs during the first ten years are exceptionally high, thereby lowering the IRR. At the same time, we did not include any additional benefits resulting from reduced or prevented losses due to pests and diseases other than Foc, that would very likely result from strengthening national level of surveillance and quarantine systems. We consider our results for ESEC to be very conservative.

Table 5 displays the estimated area on which the new technologies will be adopted under each of the adoption scenarios. In the case of ESEC, since the adoption decision takes place on the national level, the adoption area represents the area after 25 years for which losses could be avoided (in this case all area affected) due to the execution of the quarantine and surveillance measures at a national level. In comparison, the adoption area for the other three research options is the area on which farmers apply the new technologies and thus individually avoid or reduce losses. The estimated adoption area translates into the number of people that benefit from the new technologies which is highlighted in the second last column of Table 5. These figures are based on the largest adoption area reached over the assessment period.

The estimates show that conventional breeding of FWB-resistant cultivars (CBRC) reaches the largest number of beneficiaries across all research options under both the higher (14 million beneficiaries) and lower adoption scenario (7 million beneficiaries) because of the largest estimated adoption area. The investment in breeding GMRC reaches the lowest number of beneficiaries (2.7 million and 1.4 million beneficiaries for the two adoption scenarios, respectively) due to the assumption that countries with export-oriented banana production would not adopt GM varieties due to political and consumer concerns in importing countries. Similar to the NPV and IRR results, these numbers should be interpreted in combination with the size of the investments required for each research option.

The last column of Table 5 shows the poverty reduction impact of the different research option, measured as the estimated number of persons lifted out of poverty. This indicator is largely driven by the total economic benefits, national poverty rates, and region-specific elasticities of poverty reduction with respect to agricultural productivity growth. With Africa having the highest poverty rates and largest poverty elasticity, the poverty reduction measure thus favors research options that generate a larger part of the benefits in Africa compared to the other regions. This partly explains why the estimated impacts on poverty reduction are highest for ESEC (615,000–807,000 people lifted out of poverty based on the scenario) and CBRC (850,000 and 422,000 for the two adoption scenarios, respectively), whereas the investment in GMRC has the lowest poverty reduction effect. The lower importance of Cavendish cultivars in Africa also contributes to these lower adoption figures. Research options that generate comparable global economic benefits may have different poverty reduction impacts depending on share of benefits generated in countries in Africa.

We estimated the regional distribution of the adoption area (Table 6) and find that most adoption occurs in Asia/Pacific (45–92%) followed by Africa (2–44%). The regional benefits are determined by the extent of spread over the period of the calculation with only small areas affected in Latin America⁷. The benefits are mainly determined by the adoption area, but also other parameters used in the model, such as cost effects, yield levels (which are currently much higher in LAC compared to most countries in Africa), crop prices, and likely success rate.

In addition to the adoption scenarios included in results (Tables 5, 6), we conducted sensitivity analysis for the pace of adoption and size of the effects. We also computed impacts for an even more conservative adoption scenario (25% of expert estimated adoption). The results of these additional scenarios are displayed in Table 5 for ESEC and Table 7 for the other options.

Even under these increasingly, extremely, and conservative scenarios, all but one assessed research options yield positive NPVs and IRRs well above the 10% benchmark level (Tables 5, 7). These findings confirm our conclusion that investments in the assessed FWB research options generate positive economic and poverty reduction effects.