More Area, Less Coffee Per Hectare? Reconciling Expansion and Productivity in Nepal Coffee Sector, 1994–2025

1. Introduction

Arabica coffee (Coffea arabica L.) is among the few perennial export crops that can be grown profitably on small mid-hill holdings in Nepal. Introduced to Gulmi district in 1938, it remained a curiosity crop for decades before commercial plantings began in the late 1970s and early 1980s, followed by a first processing mill in 1983, the National Tea and Coffee Development Board Act in 1993, a first recorded green-bean export in 1994 and a national Coffee Policy in 2004 (KC et al., n.d.). Successive plans and strategies have since framed coffee as a high-value, organic, export-oriented commodity (Government of Nepal & International Trade Centre [GoN & ITC], 2017; National Tea and Coffee Development Board [NTCDB], 2018). Official documents place the area biophysically suitable for coffee at roughly 1.19 million ha (GoN & ITC, 2017), against a planted area of a few thousand hectares.

Two literatures bear on this opportunity but have rarely been connected. The first is climate-suitability modelling: Ranjitkar et al. (2016) estimate that suitable coffee land covers about 7% of Nepal and warn that, by 2050, it may shrink by about 73% in its current zone while shifting upward by roughly 150 m. This echoes global projections of large losses of Arabica-suitable land (Bunn et al., 2015; Davis et al., 2012). The second is the production-statistics and value-chain literature, which describes smallholder plantings of 100–150 plants, a persistent pest problem and rising but unmet demand (KC et al., n.d.; Tuladhar & Khanal, 2020, as cited in Lotz et al., 2026). What is missing is a systematic, quantitative reading of the official production record itself: how fast did Nepal’s coffee sector actually grow, where, and with what productivity?

This matters because policy targets, donor programmes and investment cases (including seedling subsidies and area-based export targets) are anchored in these statistics. If reported area and reported output diverge, or if production is not measured independently of area, then planting-based indicators may overstate progress, and strategies built on them may misdirect resources. A recent plot-level study in four mid-hill districts reports mean yields of 364 kg ha⁻¹ and a clear shade optimum near 40% canopy cover (Lotz et al., 2026), more than three times the national figure implied by the statistics, which sharpens the question of what the national numbers actually measure.

We therefore ask four questions. (RQ1) How have area, production and yield evolved nationally since 1994/95? (RQ2) How are area and output distributed across provinces and districts, and how did district yields change between 2013/14 and 2024/25? (RQ3) Can the “expansion without yield” pattern be explained by the immaturity of new plantings, or does it point to other mechanisms, including statistical artefacts? (RQ4) What do the data imply for the land area, and the productivity, needed to meet claimed domestic demand? We contribute (i) the first harmonised national–district dataset on Nepali coffee that spans three decades, with a documented data-quality audit; (ii) a falsification-style test of competing explanations for the yield decline; (iii) benchmarking against FAO and plot-level yields; and (iv) a measurement agenda for policy.

2. Background and Conceptual Framework

2.1. Institutional History and Policy Emphasis

Table 1 summarises the institutional milestones documented in the sector literature. Coffee promotion intensified after 2002–2004, when the Coffee Policy and donor supported programmes (notably HELVETAS led coffee promotion from 2003) encouraged farmer groups, wet processing and organic certification (KC et al., n.d.). A 2018–2022 coffee development strategy sought to double output from 434 t by expanding area across 41 districts identified as suitable and by managing pests and diseases (The Himalayan Times, n.d.). A national sector export strategy (GoN & ITC, 2017) pursued quality and certification. All of these instruments emphasise expansion of planted area, and area is the headline indicator in the official statistics.

2.2. Climate Suitability and Agronomy

Ranjitkar et al. (2016) model coffee suitability at roughly 600–1350 m elevation across the mid-hills, estimate that about 7% of Nepal (11% of the agricultural landscape) is currently suitable, and project a loss of 72.6% (±4.4%) of the current coffee-suitable area by 2050, partly offset by 11.9% (±2.3%) of newly suitable land at higher elevations. They also find that coffee banana intercropping is feasible in about 37% of currently suitable areas. Plot-level evidence from four mid-hill districts (n = 27 plots, 525–1275 m) shows yields of 36–1081 kg ha⁻¹ (mean 364), a quadratic response to canopy cover peaking near 40% and a trade off between shade-tree carbon storage and yield (Lotz et al., 2026). The coffee white stem borer (Xylotrechus quadripes) is reported as the most harmful pest in Nepal (Pandey et al., 2022, as cited in Lotz et al., 2026), and mid-hill agriculture is affected by outmigration driven land abandonment (Bhawana & Race, 2020; Paudel et al., 2014, both as cited in Lotz et al., 2026).

2.3. Conceptual Framework and Hypotheses

Figure 1 organises candidate explanations for a divergence between planted area and output. Policy emphasis on area generates rapid planting. Five mechanisms can then depress reported yield per plantation hectare:

H1, a growing share of immature, non-bearing plants;

H2, statistical artefacts (area derived production estimates, duplicated or stale district entries);

H3, climate stress;

H4, pest and management gaps; and

H5, abandonment of plots that remain in the area register.

With official statistics alone we cannot test H3–H5 directly. We can, however, derive observable implications of H1 and H2 and check whether H3–H5 are at least consistent with the spatial pattern. H1 predicts that yield declines should be larger where area expanded faster. H2 predicts a near unit elasticity of production with respect to contemporaneous area and a stationary yield, with rounded district yields.

3. Data and Methods

3.1. Data Sources

We used only documents whose figures could be read directly in the source and traced to an official or peer-reviewed origin (Table 2). The national series for 1994/95–2013/14 (area, green bean production, district profile for 2013/14 and trade) is reproduced from National Tea and Coffee Development Board and Ministry of Agricultural Development tabulations in KC et al. (n.d.). Later national observations come from GoN and ITC (2017; 2014/15), The Himalayan Times (n.d.; 2015/16), NTCDB (2018; 2017/18), the Ministry of Agriculture and Livestock Development as reported by Lotz et al. (2026; 2022/23) and the Ministry of Agriculture, Forests and Environment (MoAFE, 2026; 2024/25). District level 2024/25 data are from Table 2.8 of MoAFE (2026). Comparator yields are FAO values distributed by Our World in Data (FAO, 2026). Nepal’s fiscal year runs from mid-July to mid-July; we label years by fiscal year and plot them at the fiscal-year end.

3.2. Harmonisation

Area is plantation area (all planted coffee, including non-bearing plants), production is green-bean tonnes, and yield is production divided by plantation area. District names were harmonised (Sindhupalchowk = Sindhupalchok; Kabhrepalanchok = Kavre; Nawalpur = Nawalparasi East). Because the 2013/14 table lists 24 named districts and an aggregate of 15 others, panel comparisons use the 24 named districts (45 districts report coffee in 2024/25). Production in 2013/14 is reported in tonnes with one decimal, so district yields are subject to rounding error.

3.3. Analytical Methods

(A1) Growth was summarised by compound annual growth rates (CAGR). (A2) To test H2 we regressed ln(production) on ln(area) for 2000/01–2013/14 (the period of consistent, single-source reporting), tested a linear time trend in yield, and repeated the elasticity regression with area lagged 0–4 years to examine the maturity lag implied by H1. (A3) Spatial concentration was measured by shares, the Herfindahl Hirschman index (HHI, 0–10,000) and Gini coefficients. (A4) For the 24 matched districts we compared yields in 2013/14 and 2024/25 with Wilcoxon signed rank tests, computed area-weighted yields, and related the area multiplier (2024/25 area ÷ 2013/14 area) to the yield change by Spearman rank correlation (H1 predicts a negative association). (A5) Sensitivity analyses excluded the Lamjung and Syangja entries, which report identical values in 2024/25. (A6) Benchmarking compared Nepal’s reported yield with FAO yields and plot-level yields. (A7) Area-requirement scenarios divide claimed demand by alternative yields; they are arithmetic illustrations, not forecasts. Analyses were run in Python (pandas, SciPy); code and the harmonised tables will be deposited with the final submission.

3.4. Data Quality Audit

Table 3 lists the inconsistencies found. None invalidates the series, but each constrains inference and motivates the sensitivity checks above. Most importantly, the statistics report plantation area, so national yields are not like-for-like with FAO yields, which are based on harvested area.

4. Results

4.1. National Trajectory (RQ1)

Reported plantation area rose from 136 ha in 1994/95 to 1911 ha in 2013/14 and 5501 ha in 2024/25 (Figure 2, Table 4), a roughly 40-fold increase. Area grew at 14.9% yr⁻¹ over 1994/95–2013/14 and 10.1% yr⁻¹ over 2013/14–2024/25. Production grew faster in the first period (20.2% yr⁻¹, from 13.0 t to 429.4 t) but only 3.7% yr⁻¹ in the second (to 642.1 t). Yield on plantation area rose from 95 to a plateau of about 230 kg ha⁻¹ (mean 230.1, SD 28.6 over 2000/01–2013/14) and then declined: 195 (2014/15), 166 (2015/16), 194 (2017/18), 108 (2022/23) and 117 kg ha⁻¹ (2024/25).

Two features of the 2000/01–2013/14 record are informative about H1 and H2. First, production tracked contemporaneous area almost proportionally: the log–log elasticity was 1.005 (SE 0.076; R² = 0.94), and yield showed no trend (slope −0.66 kg ha⁻¹ yr⁻¹, p = 0.74) even though area grew 4.5-fold. Second, the fit deteriorated monotonically as area was lagged: R² fell from 0.96 (lag 0) to 0.93, 0.88, 0.81 and 0.76 for lags of one to four years (n = 15). If production depended mainly on the age structure of plantings, lagged area should fit at least as well as contemporaneous area. The pattern is instead what one would expect if output were estimated as area multiplied by a norm yield the footprint of H2 although independent measurement with a stable yield cannot be excluded.

4.2. Spatial Structure (RQ2)

Output is concentrated, and not where the area is (Figure 3, Table 5). Kavre (121 t), Nuwakot (90.5 t) and Ilam (83 t) supply 45.9% of national production from 20.9% of area; the five largest producing entries supply 56.1% of output (HHI 880 for production vs. 490 for area; Gini 0.69 vs. 0.57). Bagmati Province holds 28.7% of area but 45.5% of output (185 kg ha⁻¹), whereas Koshi holds 31.5% of area but 20.0% of output (74 kg ha⁻¹), driven by large new or underperforming areas such as Dhankuta (491 ha; 17 kg ha⁻¹), Taplejung (207 ha; 51 kg ha⁻¹) and Panchthar (180 ha; 25 kg ha⁻¹), with Ilam the exception (241 kg ha⁻¹). Districts with yields below 100 kg ha⁻¹ account for 56.5% of national plantation area. Karnali and Sudurpashchim, which have been the focus of westward-expansion arguments, together report 55.9 ha and 2.8 t.

4.3. District Yield Dynamics, 2013/14 to 2024/25 (RQ2–RQ3)

Across the 24 matched districts, plantation area grew from 1858 to 4244 ha (2.3-fold), while the area-weighted yield fell from 231 to 139 kg ha⁻¹ and the median district yield fell from 238 to 111 kg ha⁻¹ (Figure 4, Table 6). Yields declined in 21 of 24 districts (Wilcoxon signed-rank p < 0.001). Excluding Lamjung and Syangja (identical entries), 19 of 22 districts declined (median 241 to 112 kg ha⁻¹; p < 0.001). Only Kavre, Nuwakot and Rasuwa recorded higher yields.

The decline was unrelated to how fast area expanded (Spearman ρ = −0.05, p = 0.81). Yields fell by 131, 175, 104, 81, 346 and 58 kg ha⁻¹ in Lalitpur, Arghakhanchi, Palpa, Baglung, Makwanpur and Myagdi, the six districts whose area grew at most 1.5-fold (mean change −149 kg ha⁻¹), while the fast-expanding districts of Kavre (3.2-fold) and Ilam (6.6-fold) held yields at 244 and 241 kg ha⁻¹. A pure immaturity mechanism (H1) predicts the opposite gradient. H1 may contribute Panchthar, Khotang, Dhading and Gorkha expanded 4–5-fold and recorded large losses but it does not explain the decline in stable districts, where H2 (stale or estimated data), H4 (pests, management) and H5 (abandonment) are more plausible, and H3 (climate) cannot be ruled out.

4.4. Benchmarking (RQ3–RQ4)

Nepal’s reported yield of 117 kg ha⁻¹ equals 17.5% of Ethiopia’s FAO yield in 2024 (667 kg ha⁻¹), 16% of India’s (739), 12% of Colombia’s (1003) and 6% of Brazil’s (1743) (Figure 5). Because Nepal’s denominator is plantation area, the gap overstates the agronomic shortfall. Plot-level data point the same way: Lotz et al. (2026) measured a mean of 364 kg ha⁻¹ (range 36–1081; median plot age about 7 years), 3.1 times the national statistic, though their sample was drawn from cooperative farms and may be biased towards well-managed plots. The gap between 117 and 364 kg ha⁻¹ is therefore a plausible bracket for the combined effect of non-bearing area, abandonment, and measurement error.

4.5. Trade and Value Signals

Export volumes of green beans remained small and volatile (Figure 6): 35.7 t in 2004/05 and 66.5 t in 2013/14, excluding the anomalous 508.6 t recorded in 2008/09. Unit export values, however, rose from NPR 302 kg⁻¹ to NPR 788 kg⁻¹ over the same period (11.2% yr⁻¹, nominal). Imports exceeded exports in value in 11 of 20 years between 1994/95 and 2013/14. Later data show 111.2 t exported and 105.0 t imported in 2015/16 and 84.2 t exported against 163.4 t imported in 2017/18 (as cited in a farm-economics study of Arghakhanchi and Gulmi); we do not use these in the analysis because the primary table was not retrievable. The picture is of a sector whose price signal is strong but whose export base is thin; domestic consumption absorbs most output.

4.6. Land and Yield Requirements for Claimed Demand (RQ4)

Claimed domestic and export demand ranges from about 6000 t (KC et al., n.d.) to 8000 t (NTCDB, 2018). At the 2024/25 reported yield, supplying 6000–8000 t would require 51,400–68,600 ha, which is 4.3–5.8% of the 1.19 million ha described as suitable (GoN & ITC, 2017) but about 9–12.5 times today’s area (Table 7). At the 2013/14 yield (225 kg ha⁻¹) the requirement falls to 26,700–35,600 ha, and at plot-level yields of 364–486 kg ha⁻¹ (Lotz et al., 2026) to 12,300–22,000 ha. Even after the projected 72.6% loss of suitable area by 2050 (Ranjitkar et al., 2016; roughly 0.28 million ha remaining if 7% of Nepal is suitable today), land is not the binding constraint; yield, quality and the reliability of statistics are.

5. Discussion

5.1. What the Pattern Does and Does Not Show

The central empirical result is a divergence: area multiplied about 2.9-fold between 2013/14 and 2024/25 while output rose 1.5-fold. The tests in Section 4 narrow the explanations. H1 (immature plantings) has some support in fast-expanding eastern districts but fails as a general account because yield fell in slowly expanding districts and was unrelated to expansion speed. H2 (estimation artefacts) is supported by near-unit elasticity, a trendless yield, rounded district yields, duplicated entries and non-monotonic national output; however, it is an inference from statistical signatures, not proof of how the figures are compiled. H3–H5 cannot be tested here, but they are consistent with the geography: the largest reported falls are in the traditional western districts (Gulmi, Arghakhanchi, Palpa, Kaski), where the oldest plantings are exposed to pests and abandonment, and in newer frontier districts where site selection may be marginal. The national yield of 117 kg ha⁻¹ should thus be read as an administrative productivity indicator, not as the productivity of bearing coffee.

5.2. Implications for Policy and Investment

Area-based targets reward planting but not production. We suggest four changes. First, replace plantation area with bearing area and verified cherry or parchment deliveries as headline indicators, reported by district and altitude band. Second, establish a bearing-area census and farmer register, supplemented by remote sensing (optical and radar) to detect abandonment. Third, prioritise yield-enabling services white stem borer management, shade-canopy management near the 40% optimum reported by Lotz et al. (2026), pruning and processing quality over additional seedling distribution, particularly in districts with yields below 100 kg ha⁻¹. Fourth, align site selection with suitability mapping (Ranjitkar et al., 2016), which advises upward shifts of about 150 m and discourages expansion into low-suitability districts; the near zero output of the far western districts in Table 5 illustrates the risk of area led expansion.

For investors and parliamentary policymakers, the region’s case rests on price premia rather than volume: unit export values rose 2.6-fold in nominal terms over 2004/05–2013/14, but export tonnage is small and a yield shortfall is the main risk to financial returns. Cost–benefit analyses that assume long-run yields near the 225–486 kg ha⁻¹ range should state which statistic they rely on and test sensitivity to the 117 kg ha⁻¹ official value.

5.3. Limitations

This study is based on secondary data and cannot separate H1–H5 without plot-level information on tree age, bearing status, pest incidence and abandonment. Several national observations are drawn from policy documents and press reports that cite NTCDB or ministry tables; their definitions may differ over time. The 2013/14 district table is available from a secondary compilation (KC et al., n.d.) whose original publication details should be verified in the final manuscript. District yields in 2013/14 are subject to rounding. The comparison with FAO yields mixes plantation and harvested area. Finally, our scenarios use claimed demand figures that are themselves unsourced estimates.

5.4. Research Agenda

Priority work includes (i) a national bearing-area and tree-age census with GPS plot boundaries; (ii) remote-sensing detection of coffee plots and abandonment; (iii) a panel of farm-level yield and cost data stratified by altitude, shade and pest incidence; (iv) attribution of yield change to climate using local weather records and the suitability projections of Ranjitkar et al. (2016); and (v) integration of coffee agroforestry carbon and yield trade-offs (Lotz et al., 2026) into district level planning.

6. Conclusions

Nepal’s official coffee statistics tell a story of rapid expansion and falling productivity: plantation area rose about 40-fold in three decades, whereas reported yield on that area halved in the last decade to 117 kg ha⁻¹. The decline cannot be explained solely by young trees, because it occurs in slowly expanding districts, and the statistical signatures of area-derived production, duplicated entries and rounded yields suggest that measurement quality is itself part of the problem. Output is concentrated in a few districts, and the claimed demand could be met on a small fraction of the suitable land if plot-level yields were realised. For a climate exposed smallholder crop, the policy priority is therefore to measure bearing area and yields credibly and to invest in productivity enhancing services rather than in planted area alone.