Introduction

Indonesia’s potato production has not kept pace with its consumption levels since the last decade, especially for processed potato (FAOSTAT 2023). This growing imbalance is associated, on the one hand, with a rapidly growing population, an overall higher demand for potato and with a change in diets towards more consumption of convenience foods, including potato-based crisps and French fries (Statista Market Insights 2024). On the other hand, Indonesia’s potato production has been stagnating during the last decade: the harvested area has been stable at about 70,000 ha, and the average yield fluctuates around 20 t ha−1 (Fig. 1, FAOSTAT 2023). The shortage of local supply made Indonesia dependent on imports of large amounts of fresh and frozen potato, despite efforts of the Indonesian Government to stimulate local potato production.

Fig. 1
figure 1

Source: FAOSTAT (2023)

Potato yield progress in West Java and Indonesia over the period 2000–2022

Full size image

Potato in Indonesia is potentially a profitable crop (Arsanti and Böhme 2008), and farmers secure yields through minimizing crop failure risks, with high levels of inputs (Kashyap et al. 2023). The major potato production areas are in Java at altitudes beyond 1000 m above sea level, where low temperatures are favourable to grow the crop. The climate at these altitudes is characterised by a dry and rainy season with temperatures ranging between 15 and 35 °C and relative humidity levels ranging from 70 to 100% throughout the year. In West Java, the main potato-planting period is at the start of the rainy season, from September to December (Adiyoga et al. 1999). A second planting period is possible from January to February, so that the typical growing period of 90 to 100 days ends towards the dry season. When irrigation is available, potato can also be planted in the dry season from May to July. Due to climate conditions, late blight is one of the most prevailing diseases, especially in the rainy season (Schepers et al. 2016b). Furthermore, and as in many countries with emerging economies, potato seed is one of the most costly inputs (Adiyoga et al. 1999; Gunadi and Pronk 2022), and the use of farm saved seed is a common practice (Ginandjar et al. 2020). As potato is a rapidly degenerating crop (Haverkort 2018), fresh and high-quality seed is very important for high yields. Indonesia has restricted imports of seed potato from 2006 onwards even though local certified seed production is insufficient to meet market demand (Taylor and Dawson 2021). Additionally, seed potato generations G3 and G4 are no longer certified (Anonymous 2013), which forces seed growers to have a return on investment on two field multiplications (G1 and G2; Dawson et al. 2011) instead of four, increasing the price of certified seed. Potato is also heavily affected by soil-borne diseases (Dawson et al. 2011; Tantowijoyo et al. 2006; Taylor and Dawson 2021), hence requiring well-designed crop rotations where the crop is rotated with other species (Gunadi et al. 1998), a recommendation not followed by many farmers.

The aforementioned constraints explain partly why on-farm potato yields in West Java are likely far below the observed yields in field trials under controlled conditions, which can reach approximately 40 t ha−1 (Gunadi et al. 2003; Kusmana and Aswani 2020). However, there has been no systematic assessment of yield constraints to potato production in West Java. It is not clear what the size of the yield gap is, what its main constraints are or which interventions are needed to increase potato yield and reduce the difference between growing demand and supply. Yield gap analysis is a well-known method to unravel the relative contribution of different growth factors to actual farm yields (Grados et al. 2020; Haverkort et al. 2014; Ravensbergen et al. 2024; Silva et al. 2017b; Svubure et al. 2015; van Ittersum and Rabbinge 1997). The yield gap can be explained by comparing well-defined yield levels, input amounts and management practices required to achieve them, which was the approach followed in this study.

Against the background of stagnating potato yield and the growing demand for potato in Indonesia, the aim of this study was to assess the yield gap for potato crops at two sites of West Java and to identify the most important yield gap explaining factors constraining potato yield in farmers’ fields. This was achieved through a combination of crop growth simulation modelling and the analysis of farm yield and management practices for 40 farms over five consecutive potato growing seasons, starting in the dry season of 2013. The study documents opportunities to raise potato production in Indonesia to meet the increasing demand for potato in the country.

Materials and methods

Farm Management Registrations

Potato farmers were selected from two sites in West Java in the spring of 2013 with support of the local extension services and based on predefined criteria (De Putter et al. 2014; Pronk et al. 2018; van den Brink et al. 2015). The two sites were Cikajang and Pangalengan. The farmer selection criteria at each site included farmers’ literacy, age (between 25 and 50 years old), a minimum of 5 years of experience in growing potato and the ability and commitment to join bi-weekly training sessions during the potato-growing season (Pronk et al. 2018). An additional selection criterion was that 10 farmers in each district cultivated the variety Atlantic and 10 farmers cultivated the variety Granola on one field, however, not consecutive in time. This criterion was only met in the first year as seeds of the variety Atlantic were particularly difficult to acquire, which made farmers cultivate the variety Granola instead (Table 1). Selected farmers had access to irrigation and were instructed to keep a daily logbook to register their management practices during the dry season of 2013 up to the dry season of 2015 over a total of five consecutive growing seasons. A registration of the practices comprised planting and harvesting dates, amounts and timing of applied inputs (seeds, fertilisers, pesticides, irrigation) and seed source and generation. Crop yield was measured in each field by weighing harvested bags at the end of each growing season. The size of potato fields was measured with GPS to determine the cultivated area. One yield observation (41.8 t ha−1) was found to be exceptionally high, in the dry season of 2015 in Cikajang, and one registration missed the amount of seed planted. Both farm registrations were excluded from further analysis. The remaining 178 registrations were included in the subsequent analysis.

Table 1 Total number of field management registrations of fields where the variety Atlantic or Granola was cultivated during 2013 to 2015 in the dry and rainy growing seasons
Full size table

To characterise potato production at each site and variety combination, an unbalanced analysis of variance (ANOVA) was conducted (VSN International 2022) to identify differences in potato yield (t ha−1); total N, P and K applied (kg ha−1); chicken manure applied (t ha−1); seed amount (kg ha−1); source (bought or farm saved) and generation planted (G2 to G6); preventive and curative fungicides (kg active ingredient, a.i. ha−1), herbicides (kg a.i. ha−1) and insecticides (kg a.i. ha−1); soil treatment (mainly molluscicides, a.i. ha−1); number of sprays and fertiliser applications and the date of first fungicide spray after planting. The total N, P and K applied were calculated as the sum of these nutrients in the mineral fertilisers applied and chicken manure applied with N, P2O5 and K2O values of 1.0%, 1.5% and 1.0%, respectively.

Yield Levels and Yield Gaps

The yield gap analysis comprised four yield levels with their subsequent yield gaps, namely (1) climatic potential yield (Yp), (2) highest farmers’ yield (YHF), (3) technical efficient yield (YTEx) and (4) actual farmers’ yield (Ya; Silva et al. 2017a). Yp refers to maximum yield possible with a given variety in a well-defined biophysical environment with water and nutrients non-limiting and pests, diseases and weeds effectively controlled (van Ittersum et al. 2013). Yp is most often estimated with crop growth simulation models, as in this study, and serves as the yield benchmark for irrigated crops like potato in West Java. YHF reflects the maximum yield observed under on-farm conditions with existing technologies and can be estimated as the mean yield for the fields above the upper 10th percentile of the actual yield distribution, again for a given biophysical environment and variety cultivated. YTEx indicates the maximum yield that can be achieved for a given amount of input in a well-defined biophysical environment (Silva et al. 2017a). YTEx can be estimated by applying methods of frontier analysis to farmer field data. Lastly, Ya reflects the marketable yield observed on-farm with farmers’ management practices.

Four subsequent yield gaps were estimated, that is, the overall yield gap calculated as the difference between Yp and Ya, which was further decomposed into its technology, resource and efficiency components. The technology yield gap refers to the difference between Yp and YHF, hence reflecting the inability of farm technologies to achieve Yp (see Silva et al. 2022; 2017a). The resource yield gap refers to the difference between YHF and YTEx, hence capturing the yield penalty due to sub-optimal rates of applied inputs. Lastly, the efficiency yield gap refers to the difference between the YTEx and Ya, and thus reflects crop management imperfections related to time, form and/or space of the inputs applied.

Stochastic Frontier Analysis and Estimation of YTEx

Stochastic frontier analysis was used to identify the drivers of potato yield variability and estimate YTEx following Silva et al. (2017a). Given the relatively small sample size, a Cobb–Douglas functional form (i.e. considering only first-order linear terms) was assumed to describe the relationship between potato yield and a vector of biophysical and agronomic variables. The stochastic frontier approach considers two random errors—random noise and technical inefficiency (Battese and Coelli 1995)—which are assumed to be independently distributed from each other. Random noise captures random shocks in the production process and is assumed to follow a normal distribution with mean zero. The efficiency yield gap is the agronomic equivalent of technical inefficiency, which captures crop management imperfections related to the time, space and form of applied inputs and is assumed to follow a half-normal distribution truncated at zero.

The stochastic production frontier was estimated using input–output combinations from farm registrations collected for potato farms in West Java (Table 1). The response variable in the model was the actual yield measured for each farm registration. The vector of independent variables was constructed following the concepts of production ecology (van Ittersum and Rabbinge 1997) to capture the relative contribution of growth-defining, growth-limiting and growth-reducing factors to actual farm yields, as explained below. All continuous variables were mean-scaled and log-transformed prior to model fitting, so that parameter estimates can be interpreted as elasticity.

Growth-defining factors were captured with five categorical variables, namely year (2013, 2014 and 2015), growing season (dry vs. rainy), variety cultivated (Atlantic vs. Granola), seed source (farm saved vs. bought) and site (Cikajang vs. Pangalengan). One continuous variable was included as well, namely altitude in meters above sea level. Altitude was included because potato, being a temperate crop (Haverkort 2018), grows better at higher altitudes in tropical regions where lower temperatures prevail. Farmers in West Java frequently cultivate potatoes on rented land (Rachmina et al. 2013). When selecting land to rent, they include altitude as a criterion for potato cultivation, not below 1000 m above sea level and preferably above 1200 m above sea level. Seed source was also included in the model given the direct impact on the quality of the potato seed cultivated. Climatic variables other than altitude were not considered in the model because the monitored fields were near each other.

Growth-limiting factors were captured with continuous variables related to fertiliser management, namely nitrogen (N), phosphorus (P), potassium (K) and chicken manure application rates (all expressed in kg ha−1). Analysis of the irrigation events in the farm management registrations showed that all fields were irrigated sufficiently to qualify for potential production conditions, and irrigation data was therefore not included in the analysis. The GPS coordinates of each field were used to retrieve the soil type from SoilGrids 2.0 (Poggio et al. 2021). The dominant soil types were Andosols and Cambisols (WRB 2006), yet this information was not included in the analysis due to small variation in soil types across fields.

Finally, growth-reducing factors were captured with five continuous variables related to pesticide use: preventive and curative late blight fungicides, herbicides, insecticides and soil treatments, all expressed in kg a.i. ha−1. One categorical variable was included as well, namely seed generation (G2–G6). The vegetative propagation of the potato crop promotes seed degeneration, mostly due to a decreased health status of potato seeds (Struik and Wiersema 1999). Farmers also planted seed of generation G7 and locally bought seeds of unknown generation. These two seed quality categories were merged with generation G6, because those older generations have comparable low-yielding capacity due to advanced degeneration.

A second stochastic frontier model was estimated to account for inefficiency effects (see Battese and Coelli 1995). This model considered the same functional form and input variables as the model previously described, with the difference that it also included a second-stage multiple regression to identify the determinants of the efficiency yield gap. Inefficiency effects in the second-stage regression included the number of fertiliser and pesticide applications, the date of the first spray with fungicides to control late blight (in days after planting), the length of the growing season (days) and the practice of haulm destruction (yes or no). Neither of these explanatory variables were log-transformed.

The efficiency yield gap was estimated from the stochastic frontier model without inefficiency effects. YTEx was estimated for each farm registration as the ratio between Ya on the one hand and 1 minus the efficiency yield gap on the other (Silva et al. 2017a). Model parameters of the fitted stochastic frontier models were estimated using maximum likelihood as implemented in the sfa() function of the R package frontier (Coelli and Henningsen 2020). Both models were fitted to the pooled data as a cross-section because the small sample size did not allow for reliable inference of panel data models in relation to technology change and time-(in)variant technical inefficiencies.

Appraisal of YHF and Respective Crop Management

Highest-yielding fields were identified as the fields with Ya above the upper 10th percentile of the Ya distribution for unique combinations of variety × growing season. YHF was then estimated as the average Ya for highest-yielding fields and used to estimate the resource yield gap. Likewise, average-yielding fields were identified as the fields with Ya between 10 and 90th percentiles, and lowest-yielding fields were identified as the fields with Ya below the 10th percentile, also for unique combinations of variety × growing season. Average yields for these two groups were also estimated and defined as YAF and YLF, respectively. Statistical differences between YHF, YAF and YLF across different generations of seed potato planted and growing seasons were tested with an unbalanced ANOVA without interactions.

Mean differences in crop management practices, particularly the application of fertilisers and plant protection agents, for highest-, average- and lowest-yielding fields were also tested with ANOVA including effects of site and variety with the two-way interactions. For this analysis, data were pooled across different years as the sample size for a given year × growing season × variety combination was too small to estimate YHF, and only the two-way interactions were included in subsequent statistical inferences. Yet, we noted little inter-annual variation in the simulated Yp, pointing to stable climatic conditions for potato production over time (Supplementary Fig. S1).

Simulation of Yp with LINTUL-POTATO-DSS

Yp was simulated for each farm registration with the LINTUL-POTATO-DSS crop growth model (Haverkort et al. 2015). Temperature is the main driver of crop development considered in the model. Canopy closure is reached at a temperature sum equal to 650 °Cd, and 100% ground cover pertains until harvest. Tuber yield is estimated from the accumulated dry matter assuming that 75% of the aboveground dry matter is allocated to the tubers. The model was calibrated for potato production systems in West Java with data from four field trials conducted with the varieties Granola and Atlantic in three locations. Two additional field trials in Pangalengan and one field trial in Cikajang were used for model evaluation. Weather data were retrieved from NASA-POWER (NASA 2020) specifically for daily minimum and daily maximum temperature (°C), daily precipitation (mm), daily all-sky-surface shortwave downward irradiance (MJ m−2) and altitude-corrected relative humidity (%).

The model calibration procedure of Wang et al. (2018) was followed. First, model parameters governing crop development were estimated. Second, the leaf area–related parameter ground cover was calibrated, and finally, the maximum average daily temperature at which photosynthesis ceases was re-evaluated (see Supplementary Material B for further details). The evaluation of the model was assessed by comparing simulated and measured tuber yield. Given the high agreement between observed and simulated ground cover and yield, the model was thereafter used to estimate Yp (and the technology yield gap) for each variety (Granola and Atlantic) by growing season (dry and rainy) combination pertaining to the 178 farm registrations. LINTUL-POTATO-DSS was programmed with Microsoft Excel macros, and this version of the model was used for all simulations (Haverkort et al. 2015).

Results

Potato Yields and Yield Gaps in West Java, Indonesia

Ya ranged between 4.3 and 34.1 t ha−1, with an average Ya across farms and growing seasons of 18.6 t ha−1 (Fig. 2). ANOVA indicated two significant interactions for Ya, namely variety × site and growing season × site (Supplementary Table S1). For the variety × site interaction, Ya was lowest for Atlantic in Pangalengan (13.8 t ha−1), and no differences were found for Granola in Pangalengan (19.8 t ha−1) and Cikajang (18.0 t ha−1) nor for Granola in Pangalengan and Atlantic in Cikajang (19.9 t ha−1). Moreover, Ya of Atlantic in Cikajang was higher than Ya of Granola (18.0 t ha−1) in Cikajang. For the growing season × site interaction, Ya was lowest in the dry season in Pangalengan (16.8 t ha−1) and similar to Ya in the rainy season in Cikajang (18.0 t ha−1). Furthermore, Ya in the rainy season in Cikajang was similar to Ya in the dry season in Cikajang (19.2 t ha−1) and in the rainy season of Pangalengan (21.1 t ha−1) where no differences were found for the latter two.

Fig. 2
figure 2

Measured (black) and simulated potential (grey) yield of potato crops for the monitored field-year combinations in West Java, Indonesia. The horizontal dashed line depicts the average actual yield of the farm registrations, and the horizontal solid line depicts the average simulated potential yield for the potato varieties Atlantic and Granola in the dry (a) and rainy (b) growing seasons

Full size image

YTEx ranged between 16.1 and 40.1 t ha−1 with an average across farms and growing seasons of 29.1 t ha−1. ANOVA indicated an interaction for YTEx between variety and site (Supplementary Table S1). YTEx was lowest for Atlantic in Pangalengan (24.7 t ha−1), intermediate for Atlantic (27.9 t ha−1) and Granola in Cikajang (29.1 t ha−1) and highest for Granola in Pangalengan (30.7 t ha−1). YHF ranged between 22.1 and 34.1 t ha−1 and was on average 29.8 t ha−1 across farms and growing seasons. The average YHF was 28.3 t ha−1 for Atlantic and 30.8 t ha−1 for Granola in the dry season and 24.8 t ha−1 for Atlantic and 31.3 t ha−1 for Granola in the rainy season.

Yp ranged between 18.0 and 70.0 t ha−1, with an average across farms and growing seasons of 40.1 t ha−1 (Fig. 2). ANOVA showed the same two significant interactions as for Ya, namely variety × site and growing season × site (Supplementary Table S1). For the variety × site interaction, Yp was lowest for Atlantic in Pangalengan (32.9 t ha−1), intermediate for Atlantic in Cikajang (35.6 t ha−1) and Granola in Cikajang (38.1 t ha−1) and highest for Granola in Pangalengan (45.8 t ha−1). For the growing season × site interaction, Yp was lowest in Cikajang in the rainy season (35.6 t ha−1) followed by Cikajang in the dry season (38.5 t ha−1). Similar Yp were found for Pangalengan in the dry (41.1 t ha−1) and rainy seasons (42.5 t ha−1).

The overall yield gap ranged between 2.2 and 50.0 t ha−1 and was on average 21.5 t ha−1, or 53.5% of Yp across farms and growing seasons. The yield gap was 16.4 t ha−1 for Atlantic and 25.3 t ha−1 for Granola in the dry season and 17.4 t ha−1 for Atlantic and 21.7 t ha−1 for Granola in the rainy season (Fig. 3, Supplementary Table S2). The efficiency yield gap was on average 10.4 t ha−1 or 48.5% of the overall yield gap. The efficiency yield gap was 8.1 and 10.6 t ha−1 for Atlantic in the dry and rainy season and 11.3 and 10.8 t ha−1 for Granola in the dry and rainy season, respectively (Fig. 3; Supplementary Table S2). The resource yield gap was 0.7 t ha−1 or 3.3% of the overall yield gap. For Atlantic and Granola in the dry season, the resource yield gap was 2.1 and 1.6 t ha−1, respectively, and a negligible resource yield gap was found for both varieties in the rainy season (Fig. 3). Finally, the technology yield gap was on average 10.3 t ha−1 and 6.2 t ha−1 for Atlantic and 12.4 t ha−1 for Granola in the dry season and 6.8 t ha−1 for Atlantic and 10.2 t ha−1 for Granola in the rainy season.

Fig. 3
figure 3

Efficiency, resource and technology yield gaps for potato in West Java, Indonesia. Data represent absolute (a) or relative (b) averages of the efficiency, resource and technology yield gaps across farm registrations for a given variety × growing season combination

Full size image

Input Use and Field Altitude across Farm Registrations

Potato production in West Java is characterised by high levels of mineral fertiliser inputs with an average application rate of 180 kg N ha−1, 109 kg P ha−1 and 135 kg K ha−1 per growing season. Moreover, chicken manure was also commonly used with an average application rate of 21 t ha−1 per growing season, corresponding to about 214, 140 and 178 kg N, P and K ha−1 (Supplementary Fig. S2), yielding total application rates of 394 kg N ha−1, 249 kg P ha−1 and 313 kg K ha−1. These high application rates of chemical fertilisers and chicken manure confirm findings of Kashyap et al. (2023) and Pronk et al. (2020) for vegetable production, including potato. There were no significant differences in N or manure applied across sites, varieties and growing seasons, whereas P and K application rates were significantly affected by a variety × site interaction (Supplementary Table S3). P and K application rates were higher in Cikajang than in Pangalengan and for Atlantic than for Granola in Cikajang.

High input levels were also observed for plant protection agents both in terms of amounts of active ingredients applied (average of 43 kg a.i. ha−1 per growing season) and number of applications (average of 15 applications per growing season). Herbicides and insecticides applied were significantly different across sites, with both inputs being used more intensively in Pangalengan than in Cikajang (Supplementary Fig. S3b; Table S3). The amount of preventive and curative fungicides applied per growing season was also high, 37.5 and 2.5 kg a.i. ha−1, respectively, with significant mean differences between varieties and sites and varieties and growing seasons. The preventive fungicide use was higher for Atlantic (64 kg a.i. ha−1) than for Granola (23 kg a.i. ha−1) in Cikajang, whereas the opposite was true in Pangalengan, where preventive fungicide use was higher for Granola (35 kg a.i. ha−1) than for Atlantic (32 kg a.i. ha−1). Both varieties received similar rates of preventive fungicides in the rainy season (45 and 34 kg a.i. ha−1 for Atlantic and Granola, respectively), whereas Atlantic received more preventive fungicides (57 kg a.i. ha−1) than Granola (27 kg a.i. ha−1) in the dry season (Supplementary Fig. S3a). Similar results were found for curative fungicide application rates, yet higher rates of curative fungicides were used in Pangalengan (3.5 kg a.i. ha−1) than in Cikajang (1.3 kg a.i. ha−1) and for Atlantic (2.9 kg a.i. ha−1) than for Granola (2.3 kg a.i. ha−1, Supplementary Fig. S3b). Plant protection agents applied as soil treatment varied across varieties and sites. The highest dose of soil treatment was recorded for Atlantic in Cikajang (0.5 kg a.i. ha−1) and the lowest for Granola in Cikajang (0.04 kg a.i. ha−1). No differences between varieties were found in Pangalengan, 0.18 and 0.26 kg a.i. ha−1 for Atlantic and Granola, respectively.

A significant interaction between variety and site was found for seed rate (Supplementary Table S3). Seed rates used for Granola in Cikajang (1620 kg ha−1) were the lowest of all seed rates used, and seed rates used for Atlantic in Cikajang (1960 kg ha−1) were the highest of all seed rates used. No differences in seed rates used were found between the varieties in Pangalengan, 1699 and 1738 kg ha−1 for Atlantic and Granola, respectively. Seed source showed a significant interaction between variety and growing season instead. In the dry season, all Atlantic seeds were bought, whereas in the rainy season, only 67% of the seed used by farmers was bought. No differences in seed source were found for Granola between growing seasons. Approximately 40% of the Granola seeds were bought, and the remaining seeds were farm saved from the previous growing season.

The altitude of the monitored fields showed a three-way interaction between growing season, variety and site (Supplementary Table S3). In Cikajang, no significant differences in altitude were found between varieties and growing seasons. In Pangalengan, however, all monitored fields were planted at higher altitudes compared to Cikajang, except for Atlantic in the dry season, which was planted at comparable altitudes as Atlantic in the rainy season in Cikajang. Furthermore, in Pangalengan, Granola in the dry season and Atlantic in the rainy season were planted at the highest altitudes.

Biophysical and Management Drivers of Potato Yield Variability

The biophysical and crop management drivers of potato yield variability are presented in Table 2. P and insecticide application rates had a positive effect on potato yield, and a 1% increase in the use of these inputs increased potato yield by 0.34% and 0.03%, respectively. No effects were observed for the total amount of N, K and manure applied on potato yield nor for the amount of fungicide applied, probably due to the high rates used for these inputs in the monitored fields.

Table 2 Parameter estimates of the stochastic frontier models estimated for five consecutive potato growing seasons in West Java, Indonesia. σ2v and σ2u refer to the estimated variance of the two error terms, random noise and technical inefficiency, respectively. γ values closer to 1 indicate the stochastic frontier model is a more appropriate specification for the data than ordinary-least squares. Model 1 assumed a Cobb–Douglas functional form considering first-order linear terms only and no efficiency effects whereas model 2 was specified similarly to model 1 but considering inefficiency effects
Full size table

Seed generation had a negative effect on potato yield, indicating that older seed generations had lower Ya compared to younger generations (Table 2; Supplementary Fig. S4). For instance, potato yield was on average 27% and 33% lower in fields cultivated with seed generations G5 and G6 compared to fields cultivated with seed generation G2, respectively. Fields cultivated with seed generation G3 and G4 also yielded less than fields cultivated with seed generation G2, yet these results were not statistically significant at 5% significance level (Table 2). There were also yield differences between growing seasons, with Ya being ca. 10% higher in the rainy season than in the dry season. No yield differences were observed among varieties, years, and districts (Table 2). But farmers that planted generation G5 or G6 were also able to achieve high yields, as shown in Table 3 for the rainy season. Table 3 also shows that the lowest-yielding fields were not planted with G2 and G3 generations in both growing seasons, and that highest-yielding fields in the dry season were not planted with generations G5 and G6.

Table 3 Potato yield for different seed generation classes and growing seasons across highest- (YHF), average- (YAF), and lowest-yielding fields (YLF). Row means or column means with different letters differ significantly at 5% significance level. No lowest-yielding fields were planted with seed of generations G2 and G3 in the dry and rainy season (†) and no highest-yielding fields were planted with seed of generations G5 and G6 in the dry season
Full size table

The model specification with inefficiency effects also captured the significant effect of P applied, seed generation and growing season on potato yield, but not of insecticide rates (Table 2). Instead, altitude had a significant positive impact on potato yield. These results indicate that adding inefficiency effects had relatively little impact on the biophysical and management drivers of potato yield variability.

Determinants of the Efficiency Yield Gap

The drivers of the efficiency yield gap included the number of growing days, the number of fungicide sprays and the timing of the first fungicide spray (Table 2; Fig. 4), whereas active haulm destruction and the number of fertiliser applications had no significant impact on the efficiency yield gap (Table 2). Growing days had a negative effect on the efficiency yield gap, indicating that fields with more growing days had smaller efficiency yield gaps compared to fields with less growing days (see also Fig. 4a and b). Conversely, the number of fungicide sprays and the timing of the first fungicide spray had a positive effect on the efficiency yield gap, indicating that more fungicide sprays and later first sprays resulted in larger efficiency yield gaps compared to less fungicide sprays and early first sprays, respectively.

Fig. 4
figure 4

Crop management determinants of the efficiency yield gap for Atlantic (n = 57) and Granola (n = 121) potato varieties in West Java, Indonesia: a, b number of growing days, c, d total number of fungicide sprays, and e, f first spray after planting. The efficiency yield gap (t ha−1) is displayed for each farm registration in the dry (n = 103) and rainy (n = 75) season. YHF highest farmers’ yield; YAF average farmers’ yield; YLF lowest farmers’ yield

Full size image

Figure 4 shows a large variability of the efficiency yield gap across farmers’ fields. A growing season by yield class interaction was found for growing days (Supplementary Table S4), indicating that lowest-yielding fields in the dry season had significantly less, and highest-yielding fields significantly more, growing days than average-yielding fields (Fig. 4 a and b). No major differences were found for the number of sprays nor the timing of first spray (Supplementary Table S4) across yield classes, meaning the impact of these factors on the efficiency yield gap depends on other production factors considered in the analysis (Table 2), but Atlantic received the first spray earlier (17 days) after planting than Granola, 21 days after planting. Thus, timely disease control at the onset and during the growing season is important to narrow efficiency yield gaps of potato in West Java.

Determinants of the Resource Yield Gap

P application rate was an important driver of potato yield variability (Table 2), hence explaining part of the resource yield gap. Although large amounts of P were applied in both growing seasons, highest-yielding fields received significantly more P than average-yielding fields (and average- more than lowest-yielding fields; Fig. 5a and b). P application rates from both mineral fertilisers and chicken manure in highest-, average- and lowest-yielding fields with Atlantic were on average 355, 281 and 178 kg ha−1. Fields with Granola received significantly less P, on average 256, 235 and 219 kg ha−1 for highest-, average- and lowest-yielding fields, respectively. Regarding insecticide application rate, fields cultivated in the dry season received 38% more insecticides than fields in the rainy season (Fig. 5c and d). Differences in insecticide application rates between highest-, average- and lowest-yielding fields were observed in both growing seasons (Supplementary Table S5). In the dry season, insecticide rates applied were lower on highest-yielding fields (0.7 kg a.i. ha−1) than on average- (1.9 kg a.i. ha−1) and lowest-yielding fields (2.3 kg a.i. ha−1). The opposite was true in the rainy season where insecticide rates applied were higher on highest-yielding fields (3.0 kg a.i. ha−1) than on average- (1.2 kg a.i. ha−1) and lowest-yield fields (0.3 kg a.i. ha−1).

Fig. 5
figure 5

Crop management determinants of the resource yield gap for Atlantic (n = 57) and Granola (n = 121) potato varieties in West Java, Indonesia: a, b total amount of phosphorus applied with mineral fertilisers and chicken manure, c, d insecticides applied, and e, f preventive fungicides applied. The resource yield gap (t ha−1) is displayed for each farm registration in the dry (n = 103) and rainy (n = 75) season, respectively. YHF highest farmers’ yield; YAF average farmers’ yield; YLF lowest farmers’ yield; a.i. active ingredient

Full size image

The lack of significant effects of N and fungicide application rates on potato yield indicates these factors were likely not limiting potato production in the monitored fields and further explains the small resource yield gaps quantified in this study (Fig. 3). Despite differences in N and preventive fungicide application between highest-, average- and lowest-yielding fields (Fig. 5e and f, Supplementary Table S5), we note that lowest-yielding fields received an average of 319 kg N ha−1 and 21 kg a.i. ha−1 as preventive fungicide, rates way above the agronomic recommendations to overcome N limitations and reductions due to leaf diseases. There were also no evident potato yield responses to N and preventive fungicide application rates when data were pooled across variety by growing season combinations (Table 2). These results indicate considerable scope to improve the use efficiency of these inputs, especially through reductions in the amount of N and fungicides applied.

Determinants of the Technology Yield Gap

Since the technology yield gap is defined as the difference between Yp and YHF, its explanatory factors are not directly captured in the farm registration data. Possible determinants of the technology yield gap include soil borne diseases due to narrow crop rotations for potato (Gunadi et al. 1998), degraded seed quality due to, e.g. poor storage- and/or sub-optimal haulm destruction practices, farm saved seed practices (Gunadi et al. 2021), overfertilisation with N fertilisers, inappropriate use of adjuvants (Schepers et al. 2014b) and too large water volumes for fungicide application, as discussed in detail in the next section.

Discussion

A substantial yield gap of about 55% of Yp was quantified for two important potato varieties in West Java (Fig. 3). Agronomic factors explaining the existing yield gap were not related to sub-optimal input amounts, as most fields reported high input levels, often beyond agronomic recommendations. Comparable large yield gaps were found for potato in Chile, on average 60% of Yp (Haverkort et al. 2014); China, 66% of Yp (Wang et al. 2018), Peru, 43% of Yp (Grados et al. 2020) and Zimbabwe, 77% of Yp (Svubure et al. 2015), whereas much smaller yield gaps were found in the Netherlands, between 20 and 31% of Yp (Ravensbergen et al. 2024; Silva et al. 2017b). The studies of Grados et al. (2020) and Svubure et al. (2015) concluded that agronomic factors explaining the existing yield gaps were not related to sub-optimal input amounts, as also found in this study for potato production systems in West Java, Indonesia.

The P application rates and insecticide rates applied were high (Fig. 5) and in case of P application rates, mostly above optimum levels for potato (van den Brink et al. 2016). The historical and currently high application rates of P with mineral fertilisers and manure products on potato crops (Kashyap et al. 2023; Moll 1981) mitigated the soil P-fixating properties of volcanic soils in West Java under vegetable cultivation (Anda et al. 2020; Yatno and Zauyah 2008). The moderate positive effect on yield is possibly through enhanced soil properties such as soil porosity, water holding capacity and wilting point from the manure applications (Setiyo et al. 2016), as the dominant available P for uptake is released from the soil (Syers et al. 2008) rather than from the P fertiliser applied during the growing season. The supposed side effects of high P application rates through enhanced soil properties supports our finding that the resource yield gap quantified in the dry season was small and non-significant.

Input reductions can likely be combined with yield increases for potato in West Java since inputs were mostly not used efficiently in the monitored fields. This translated into an overall efficiency yield gap of 10.4 t ha−1, nearly half of the overall yield gap (Fig. 3). Improved resource use efficiency under current farm practices can best be achieved through a focus on management practices that increase the number of growing days (i.e. duration of the growing season), where fungicide management is an important practice to keep the leaf canopy clean and healthy. The number of growing days contributes to increased potato yield by ensuring a green closed canopy for a longer period, which in turn ensures higher light interception and higher bulking for the crop (Haverkort 2018). Yet, maintaining a green closed canopy longer in the field poses obstacles, particularly regarding late blight control with fungicides.

Three aspects related to fungicide management require attention in the future: the timing of the first spray, the number of fungicide sprays and the rates of preventive fungicides (Table 2). First, the need for an early first spray relates to an early detection of late blight. Late blight infections may be induced by favourable weather conditions, but also from seed (Cooke et al. 2011) due to late blight development in storage. This can happen when haulm destruction practices are not adequate, which is the case in our sample where only 30% of the farms reported haulm destruction at the end of the growing season. Instead, the common practice to destroy the crop foliage is to stop spraying preventive fungicides, so that late blight infections do so. Such practices may infect tubers and induce early-season late-blight problems (Cooke et al. 2011; Yuen et al. 2008), which may explain our findings that an early fungicide spray can be an effective measure to increase potato yield. Second, our results show that more fungicide sprays are detrimental to potato yield, which contrasts with earlier research (Schepers et al. 2014a, 2015, 2016a). Moreover, a longer growing season with frequently spraying leads to an overall higher number of sprays. We found that lowest-yielding fields received a comparable number of fungicide sprays in a shorter growing season compared to highest-yielding fields, and a shorter growing season was associated with larger efficiency yield gaps, which explains the contrast with earlier findings. Most likely, farmers notice late blight soon after emergence and act on that by increasing the frequency of fungicide sprays, apparently without effective control due to large spraying volumes and additions of adjuvants with surfactant properties to the sprays (Schepers et al. 2014b). Lastly, we found no effect of the total amount of preventive fungicides applied on potato yield (Table 2). This can be explained by the farmer practices of combining different commercial fungicide products, with similar active ingredients (Adiyoga 2009), into one spray (tank mixtures). Such risk aversion practise leads to overapplication of preventive fungicides, which if prevented, could lead to both environmental and financial benefits to farmers.

Active haulm destruction practices may reduce late-blight infections of seed potatoes during storage and early-season infections. Active haulm destruction done mechanically or spraying a desiccant or a herbicide, combined with a preventive fungicide spray, can effectively reduce late-blight infections on harvested tubers (Struik and Wiersema 1999) with subsequently reduced losses during storage and early-season infections.

Older seed generations resulted in lower potato yield than younger ones (Table 2). This effect, however, is not straight-forward to interpret in the context of the technology yield gap as some highest-yielding fields also planted seed generation G5 and G6 (Table 3). Seed is the most costly input in potato farming (Swastika and Gunawan 2021), and farmers’ demand for good quality seed is increasing. The potato seed value chain, however, has difficulties to establish (Taylor and Dawson 2021), and the heterogeneity of seed quality entering the market contributes to unexpected yield losses as found in our study. To improve the seed potato value chain, policy makers are advised to evaluate recent policies, eliminate distortions in the input and output markets and focus on policies to increase the availability of quality potato seed (Saptana et al. 2022). Additionally, farmers save potato seeds from their own yield for the next growing season to reduce costs (Fuglie et al. 2006), and such practice may increase fast degeneration when farmers select the smaller, possible degraded, potatoes as seeds for the next growing season. Well-practiced farm-saved seed selection can reduce degeneration (Gunadi et al. 2017), which explains the high yields of some farmers planting older generations.

Finally, the technology yield gap ranged between 6.2 and 12.4 t ha−1 with an average across farms and growing seasons of 10.3 t ha−1, nearly half of the overall yield gap (Fig. 3). Such large technology yield gap indicates that current farm practices cannot reach agronomic best practices due to multiple factors not captured in this study. Soil-borne diseases, particularly potato cyst nematodes and bacterial wilt, may cause great damage to potato yield. Where potato cyst nematodes are solely associated with the production of potato, bacterial wilt is a general and much more difficult-to-control soil borne disease. Cyst nematode infections are on the rise in some intensive potato-producing regions of Indonesia (Handayani et al. 2020), yet seldom recognised by farmers. Bacterial wilt has a wide range of hosts, including potato (Machmud 1986; Nurbudiati and Wulandari 2020), and control measures include scouting for unaffected fields when renting land for potato planting, disease-free seeds and disease-free irrigation sources (Nurbudiati and Wulandari 2020).

Conclusions

The small resource yield gap quantified in this study for potato crops in West Java, Indonesia, indicates that the high amounts of fertilisers and plant protection agents, most notably fungicides, currently used by farmers are unlikely to constrain potato productivity in their fields. Indeed, the rates reported in the farm registrations for most inputs are above the agronomic recommendations required to reach the potential yield of potato in the studied sites. Narrowing potato yield gaps and breaking the potato yield plateau in West Java would require more efficient use of inputs, particularly the operational management of fungicides and probably changes in tactical farm management to reduce the frequency of potato in the crop rotation. Increasing the availability of high-quality potato seed remains important, despite recent attempts to develop a local healthy seed system, as seed generation was identified as an important yield constraint. The study also found considerable scope to reduce the use of N (and P) inputs and of plant protection agents (particularly fungicides) without compromising potato yield among the monitored fields. We recommend future research to train farmers on late blight management practices and on the importance of agronomically healthy crop rotations for potato cultivation and to test combinations of practices with more efficient use and lower input amounts than currently used by farmers.