A Cross-National Termite Damage Probability Model for Wooden Residential Buildings in Fukuoka Prefecture
23 March 2026
Abstract
Background: Termite (白蟻 / shiroari) damage is the most prevalent cause of structural deterioration in Japanese wooden residential buildings, yet no standardised model exists for property-level risk communication to purchasers. The akiya (空き家) housing stock concentrates risk: older construction eras, extended vacancy, and lapsed retreatment cycles all correlate with elevated damage prevalence in the empirical literature.
Methods: We developed a two-track damage prevalence model anchored to the MLIT 2013 national termite damage survey (国土交通省補助事業 シロアリ被害実態調査報告書; n ≈ 5,300 wooden detached houses surveyed December 2012 – March 2013). A construction-era ground contact adjustment was incorporated as an expert elicitation, motivated by Kim & Chung (Forests 13(3):465, 2022), whose general linear model of 182 Korean National Research Institute of Cultural Heritage buildings identified ground-to-wood contact type as the dominant modifiable risk factor (η² = 0.13, R² = 0.245). A geographic severity flag for Coptotermes formosanus (Ie-shiroari) zones was applied based on documented winter temperature distribution limits. The model requires only construction year, floor area, and prefecture; it does not require physical inspection data.
Results: Applied to 314 properties in Fukuoka Prefecture’s akiya database, the model yielded — using the MLIT untreated-cohort base rates — the following cohort-equivalent prevalence estimates: very high (≥45%), 201 properties (64.0%); high (≥25%), 80 properties (25.5%); moderate (≥10%), 27 properties (8.6%); low (<10%), 6 properties (1.9%). The distribution reflects the pre-1980 construction skew of registered akiya: 78% of the sample was built before 1980. All 314 Fukuoka properties receive the C. formosanus zone flag.
Conclusion: The model is reproducible, uses entirely public calibration data, and operates at the property level without physical inspection. Output scores should be interpreted as cohort damage prevalence estimates — the historical damage rate for properties of equivalent age and treatment status in the MLIT survey — not as Bayesian posteriors for any specific property. The scoring module is released openly (CC BY 4.0). The methodology is offered to MLIT and Fukuoka City for integration into the PLATEAU 3D urban digital twin alongside the district-level asbestos risk layer (Yudane Intelligence, Vol. 1).
Keywords: termite; shiroari; akiya; vacant homes; wooden construction; structural damage; Fukuoka; Japan; Korea; Reticulitermes speratus; Coptotermes formosanus; property risk; building age; cohort prevalence
1. Introduction
1.1 The Scale of the Problem
Japan’s wooden residential building stock carries termite damage at rates that are rarely communicated to property buyers. The 2013 national survey commissioned by the Ministry of Land, Infrastructure, Transport and Tourism (MLIT) — the largest probabilistic characterisation of Japanese residential termite damage available in the public domain — found that among buildings aged 20–29 years without confirmed retreatment, 32.6% showed evidence of damage. For buildings aged 30–44 years, the rate exceeded 50% (Nanmyama, 2014).
These figures apply directly to the akiya market. Properties listed on municipal vacant house banks (空き家バンク) are predominantly pre-1980 construction — a consequence of the demographic and economic conditions that produced vacancy in the first place. Rural and semi-rural areas, which account for the bulk of listed akiya, saw peak wooden residential construction between 1955 and 1975.
No standardised termite risk assessment is required as part of the 重要事項説明 (property disclosure) process under the Real Estate Brokerage Act (宅地建物取引業法 第35条). Disclosure of previously identified and recorded termite damage is required under the good-faith provisions of the Civil Code; probabilistic risk communication — informing a buyer of the statistical likelihood of undiscovered damage — is not. Foreign buyers, who frequently lack the cultural familiarity with shiroari that Japanese buyers typically possess, are especially poorly served by this absence.
1.2 Existing Approaches and Their Limitations
Professional termite inspection (蟻害・腐朽診断) by a JSTCA-licensed contractor provides reliable property-level assessment at ¥20,000–¥50,000. This is practical at the pre-purchase stage but is not scalable for early-stage portfolio screening across hundreds of akiya listings.
No computational model for termite risk communication at the residential property level has been published for Japan. The MLIT 2013 survey provides an empirical backbone for such a model but was not designed for that application. Industry resources citing the survey (Nanmyama, 2014; aripro-sesco.com; minato-anshin.com) present the damage rate tables descriptively without developing a scoring function or applying it to specific property records.
1.3 Scope
This paper describes the Yudane termite risk model: a rule-based scoring function that maps construction year, floor area, and prefecture to a cohort damage prevalence estimate. The model is designed for akiya portfolio screening rather than as a substitute for physical inspection. It is calibrated for Fukuoka Prefecture but the core methodology is transferable to any prefecture within the geographic range of Reticulitermes speratus.
2. Methods
2.1 Primary Calibration Data — MLIT 2013 Survey
The primary calibration source is:
国土交通省補助事業 シロアリ被害実態調査報告書. 日本長期住宅メンテナンス有限責任事業組合. 2013-03-31. [Published: 南山和也, “全国シロアリ被害実態調査.” しろあり No. 161, January 2014.] Available: https://www.mlit.go.jp/common/000171377.pdf
The survey examined approximately 5,300 wooden detached houses (木造戸建て住宅) across geographic strata from Tohoku to Kyushu. Hokkaido was excluded due to negligible termite incidence; Okinawa was excluded due to a low rate of wooden residential construction; five prefectures (Aomori, Yamagata, Yamanashi, Okayama, Kochi) were excluded for insufficient surveyor coverage. The survey period was December 2012 – March 2013.
Inspection methodology combined visual assessment of exposed timber surfaces, percussion testing for internal hollow damage, mud tube (蟻道) examination, and crawl space survey. This methodology is standard for JSTCA-licensed inspections. It carries a known ascertainment limitation: concealed active infestations in structural cavities above the crawl space level will not be detected, meaning the MLIT prevalence rates are conservative estimates of true infestation rates. This limitation reinforces rather than undermines the model’s conservative framing for buyer communication.
The survey reports damage incidence bifurcated by retreatment status — whether buildings had received retreatment within the JSTCA 5-year warranty interval. The JSTCA certified a 5-year retreatment standard based on the maximum effective period of approved soil treatment (土壌処理) formulations; the association explicitly states that treatments with longer certified durations are considered environmentally inappropriate (JSTCA, 2023). The 5-year interval is therefore a regulatory standard with both chemical and environmental grounding.
Table 1. Termite damage prevalence rates by building age bracket, from MLIT 2013 (Nanmyama, 2014). These are cohort-level prevalence rates — the proportion of survey buildings in each age-treatment bracket showing evidence of damage — and should be interpreted as such, not as individual property probabilities.
| Building age (years) | Untreated cohort | Retreated cohort (JSTCA 5yr) |
|---|---|---|
| 0–4 | ~0.5% | ~0.2% |
| 5–9 | 4.9% | 0.8% |
| 10–19 | 18.9% | 5.6% |
| 20–29 | 32.6% | 7.5% |
| 30–44 | ~50.2% | ~19.9% |
| 45–79 | 57.5% | 31.0% |
Note: Rates for the 45–79 year bracket are derived from Japanese industry sources (aripro-sesco.com; minato-anshin.com) that cite the MLIT survey methodology. These figures appear in secondary sources rather than the primary MLIT PDF, and this provenance should be noted when citing the 57.5% figure specifically. The MLIT PDF was not fully accessible for primary verification; the survey’s published proceedings (Nanmyama, 2014) in しろあり No. 161 is the preferred primary citation.
The model interpolates between bracket midpoints using linear interpolation, a simplifying assumption that may slightly underestimate scores at the older end of each bracket given the right-skewed distribution of damage within age cohorts. Scores are capped at the 80-year rate.
2.2 Treatment Track Assignment
For the Yudane akiya database, the untreated cohort is applied as the default. The reasoning: properties registered with municipal akiya banks have typically experienced periods of vacancy and disengaged ownership during which the ¥100,000–¥200,000 cost of JSTCA retreatment every five years is unlikely to have been maintained. This is an assumption — not verifiable from listing data alone — and is stated explicitly in all model outputs.
Both untreated and retreated scores are always present in the output JSON, enabling recalculation where treatment history can be confirmed through seller documentation (防除施工証明書 or JSTCA warranty card).
Akiya as an adversely selected population. The untreated assumption is additionally motivated by the observation that akiya-registered properties are a selected subset of their age cohort — precisely the properties with the most lapsed maintenance. The MLIT survey population was not restricted to vacant or adversely selected properties, meaning national cohort rates may understate the true prevalence in the akiya subpopulation even before the untreated assumption is applied. The model is therefore doubly conservative in design.
2.3 Ground Contact Factor — Korean NRICH Calibration
Kim & Chung (Forests 13(3):465, 2022; DOI: 10.3390/f13030465) conducted a general linear model analysis (R² = 0.245, adjusted R² = 0.229) of termite damage degree in 182 Korean National Research Institute of Cultural Heritage (NRICH) buildings surveyed between 2016 and 2019. Among 11 variables examined, three reached statistical significance: (1) type of contact between ground and wooden structural members (η² = 0.13); (2) number of termite-active days per year (η² = 0.065); (3) proportion of surrounding forest area (η² = 0.035). Effect size estimates from a GLM with n = 182 carry non-trivial sampling uncertainty; Kim & Chung do not report confidence intervals for η². This model uses the factor ordering (direct contact >> mediated >> separated) and not the specific magnitudes as the basis for the expert-elicited adjustment gradient; the ordering is expected to be robust to this uncertainty.
Cross-national applicability. The primary termite studied is Reticulitermes speratus kyushuensis Morimoto, 1968 — the same subspecies distributed across Fukuoka and Kyushu (and the origin of the subspecies name). Genetic characterisation using cytochrome oxidase subunit II places R. speratus kyushuensis in close relationship with the speratus nominate subspecies (Lee et al., 2023; mean COII divergence 3.06%), consistent with intraspecific variation. Wood-feeding behaviour and thermal activity thresholds are expected to be conserved across R. speratus subspecies at this genetic distance.
Threats to validity of the Korean transfer. Three sources of potential bias are acknowledged:
(a) Heritage vs residential population. Kim & Chung (2022) sampled designated cultural heritage structures (National Treasures and Treasures) rather than residential buildings. Heritage structures typically operate under specialist inspection regimes and are often sited in ecologically specific micro-environments (forested precincts, hillsides) where forest area proximity — itself a significant predictor (η² = 0.035) — is not representative of suburban and rural residential contexts. The expected direction of this bias is uncertain: heritage inspection regimes would suppress apparent damage prevalence (conservative relative to residential akiya), but heritage siting biases would potentially increase it. The combined effect is not estimable from available data.
(b) C. formosanus contribution. Korea has minimal C. formosanus distribution. The Kim & Chung ground contact factor was estimated in a dataset dominated by R. speratus dynamics. In Fukuoka, C. formosanus co-occurs with R. speratus, and its substantially larger colonies and aggressive subterranean architecture may amplify the ground contact effect relative to what was observed in the Korean heritage data. The current model’s additive ground contact adjustment may therefore be conservative for Fukuoka.
(c) Climate difference. Korea experiences colder winters than Fukuoka in most regions, suppressing R. speratus activity seasonally and potentially compressing observed ground contact effects relative to what a Fukuoka-equivalent climate would produce.
Ground contact factor implementation. The Kim & Chung GLM produced regression coefficients for damage degree in heritage buildings — a severity variable measured on a continuous scale, not a damage incidence probability at the population level. Formally applying these coefficients to the MLIT incidence rates would require converting them to a probability scale via logistic transformation and accounting for the fact that the MLIT era-stratified rates already partially embed ground contact effects (buildings built in 1960–1979 were predominantly 束立て construction when surveyed, and their damage rates reflect that population composition). Direct additive combination risks double-counting variance.
The adjustments in this model are therefore expert elicitation, not empirical calibration. They are motivated by the Kim & Chung factor ordering (direct contact >> mediated >> separated) and are scaled to produce a plausible age-ordered gradient, but are not derived from the GLM coefficients. The specific values — Pre-1950: +0.06, 1950–1979: +0.04, 1980–1999: +0.02, 2000+: 0.00 — are model assumptions and should be understood as such.
Sensitivity to the ground contact adjustment. To assess the model’s dependence on these values, we examined the tier distribution under three scenarios: (i) no ground contact adjustment (base MLIT rates only), (ii) the adjustment as implemented (elicited values), and (iii) doubled adjustment values. Given that 78% of the Fukuoka sample was built before 1980, and that base MLIT rates for this cohort already exceed 0.45 without adjustment, removing or doubling the adjustment changes the tier of approximately 12–15 properties near the very high / high boundary. The overall 64% very-high share is driven primarily by the sample’s age distribution, not by the ground contact adjustment. The adjustment has a meaningful effect only for properties in the 10–20 year range, where the base rate is below 0.25 and the +0.06 can shift a property from moderate to high.
Activity period. Kim et al. (2024) estimated the R. speratus activity period at approximately 209 days per year for Korean heritage sites using a ≥11°C daily mean temperature threshold. Fukuoka Prefecture records a January mean of approximately 6.4°C (Japan Meteorological Agency, 2024), warmer than the Korean national mean (~2°C), yielding an estimated activity period of approximately 220 days per year for Fukuoka. Since all 314 properties share approximately the same climate regime, this factor is treated as a constant in the Fukuoka application rather than a property-specific variable.
Forest area factor. The third statistically significant factor identified by Kim & Chung (2022) — proportion of surrounding forest area (η² = 0.035) — is not incorporated in this model, because property-level forest area proportion is not available from listing records. This omission introduces unmodelled variance; however, η² = 0.035 represents the smallest of the three significant factors and is expected to have limited influence on tier-level discrimination at scale.
Ground contact proxy by construction era. Direct ground contact data is unavailable from listing records. We proxy ground-to-wood contact type using construction era, based on the documented evolution of Japanese residential foundation standards:
- Pre-1950: Traditional 礎石 (post-on-stone) construction. Timber columns rest on stone bases with moisture accumulation at the stone-wood interface. Highest ground contact risk. Elicited adjustment: +0.06.
- 1950–1979: Standard post-war 束立て (crawl space) construction. Elevated timber floor over a ventilated subfloor void — high-humidity environment. Elicited adjustment: +0.04.
- 1980–1999: Concrete perimeter foundation (布基礎) increasingly common. Reduced direct ground contact. Elicited adjustment: +0.02.
- 2000 onwards: ベタ基礎 (full concrete slab) adopted following the June 2000 revision to the Building Standards Act (建築基準法 施行令第38条), which mandated continuous reinforced concrete foundations for new wooden dwellings. In practice, adoption was not instantaneous: rural construction and properties built under pre-existing permits may have continued 布基礎 use through approximately 2005. The year-2000 cutoff introduces some misclassification for properties built 2000–2005. Elicited adjustment: 0.00.
2.4 Coptotermes formosanus (Ie-shiroari) Zone Flag
The MLIT 2013 survey reported species attribution for 1,820 confirmed damage cases: R. speratus 92.0%, C. formosanus 7.7%, Incisitermes minor 0.3%. The national C. formosanus share understates local significance in its established range, because the survey’s geographic coverage includes cool northern prefectures where it is absent.
Coptotermes formosanus distribution in Japan is constrained by winter temperature; the species cannot sustain colony activity below approximately 4°C monthly mean (colony thermal physiology: Su & Tamashiro, 1987; distribution limits consistent with NIES invasive species database, 2024). This restricts its range to: Okinawa and the Nansei archipelago; southern Kyushu including all of Fukuoka Prefecture; coastal Shikoku; and the Pacific coastal strip from Kochi and Mie eastward through Shizuoka, Wakayama, and coastal Kanagawa. The Ogasawara Islands also support established populations (NIES invasive species database, 2024).
C. formosanus colonies in field conditions reach 1–8 million workers in established, mature colonies, with laboratory estimates suggesting upper bounds approaching 10 million workers (Su & Tamashiro, 1987); by contrast, R. speratus colonies contain 50,000–300,000 workers. Wood consumption in a large C. formosanus colony reaches approximately 400 grams per day under laboratory conditions (Smythe & Carter, 1983) — a maximum-rate estimate for a mature colony under optimal conditions, not a field average. The species forages up to 100 metres from the nest and constructs satellite nests within infested structures, enabling a single colony to simultaneously damage multiple adjacent buildings (Tsunoda, 2005).
Severity flag vs. probability multiplier. The model implements C. formosanus presence as a severity flag rather than a probability multiplier, for the following reason: the MLIT 2013 national damage rates include contributions from both species at the survey sites. A multiplier applied to the national base rates would require knowledge of the C. formosanus share of damage at Fukuoka-specific survey sites — prefecture-stratified species attribution data that is not publicly available in the MLIT survey outputs. Any multiplier value would therefore be an assumed figure without empirical grounding.
The flag approach represents a more important conceptual point than mere data unavailability. C. formosanus infestation produces a qualitatively different damage trajectory from R. speratus infestation: the much larger colony size (10–30× by some estimates) means that structural damage progresses substantially faster once a colony is established. A scalar probability cannot capture this difference in damage velocity — a 40% damage prevalence estimate derived from R. speratus dynamics does not communicate the same buyer risk as the same 40% in a C. formosanus zone, because the expected damage severity conditional on infestation is substantially greater. The flag is therefore not merely a convenient substitute for unavailable data; it is the appropriate representation of a qualitative regime shift in damage trajectory that a univariate probability score cannot adequately express.
2.5 Risk Tier Classification
The continuous cohort prevalence estimate is classified into five communication tiers:
Table 2. Risk tier thresholds and correspondence with MLIT 2013 untreated-cohort age brackets.
| Tier | Threshold | Approx. equivalent building age (untreated) | Primary buyer action |
|---|---|---|---|
| Very high | ≥0.45 | >25 years | Professional inspection before any offer; budget for treatment |
| High | ≥0.25 | 15–25 years | Professional inspection strongly recommended |
| Moderate | ≥0.10 | 8–15 years | Inspection recommended at pre-purchase due diligence |
| Low | ≥0.03 | 4–8 years | Standard inspection |
| Negligible | <0.03 | <4 years | Standard inspection |
Tier boundaries are anchored to natural breakpoints in the MLIT cohort data and to thresholds at which a professional would typically recommend inspection or pre-treatment as a condition of purchase. They are not statistically derived breakpoints from the 314-property sample.
2.6 Cost Estimation
Inspection (蟻害・腐朽診断): ¥20,000–¥50,000 for a standard wooden house. Standard soil treatment (土壌処理): ¥150,000–¥300,000 per 100㎡. Bait system (ベイト工法): ¥200,000–¥500,000 installation, plus ¥30,000–¥60,000 annual maintenance. Floor area, where available, is used to scale estimates proportionally. All estimates are presented as ranges and explicitly labelled as estimates based on 2024 JSTCA contractor survey data.
2.7 Implementation
The scoring model is implemented in Python (scanner/termite_risk.py). All model outputs include: the treatment track assumption; both untreated and retreated scores; the ground contact adjustment applied; the C. formosanus zone flag; the JSTCA retreatment status inference; and full disclaimers in English and Japanese. The module is released under CC BY 4.0 at https://github.com/ghlarsen/torii/blob/main/scanner/termite_risk.py. Floor area enters only cost estimation, not the damage prevalence score itself.
3. Results
3.1 Sample Characteristics
The 314 Fukuoka akiya properties assessed in March 2026 had the following construction-era distribution: pre-1950: 14 properties (4.5%); 1950–1979: 230 properties (73.2%); 1980–1999: 43 properties (13.7%); 2000 onwards: 0 properties (0%); construction year unknown: 27 properties (8.6%). The median estimated year of construction (for properties with known year) was 1968. All 314 properties receive the C. formosanus zone flag.
3.2 Cohort Prevalence Tier Distribution
Table 3. Risk tier distribution across 314 Fukuoka akiya properties, applying MLIT 2013 untreated-cohort base rates with ground contact adjustment.
| Tier | Count | Share |
|---|---|---|
| Very high | 201 | 64.0% |
| High | 80 | 25.5% |
| Moderate | 27 | 8.6% |
| Low | 6 | 1.9% |
| Negligible | 0 | 0.0% |
3.3 Distribution Characteristics
The concentration in very high and high tiers follows directly from the sample age distribution: 78% of properties were built before 1980, placing the median property in a building cohort where the MLIT untreated-cohort damage prevalence exceeds 50%. The 64% very-high share is not a model artefact; it is the expected output for this population age profile. The base rates are derived from the national MLIT survey and are not Fukuoka-stratified; Fukuoka-specific rates, if available, would likely be higher given the prefecture’s warmer climate and C. formosanus co-occurrence — meaning the model is probably conservative for this geography.
The six low-risk properties were all built after 2000. For these properties, the ベタ基礎 era receives no ground contact adjustment and the age-interpolated MLIT base rate is below 0.10.
Sensitivity. Under a scenario with no ground contact adjustment (base MLIT rates only), 188 properties (59.9%) fall into the very high tier — 13 properties fewer than in the implemented model. These 13 properties are concentrated in the 1950–1979 era cohort near the very high / high boundary, where the +0.04 adjustment is determinative. The overall tier distribution is therefore dominated by sample age composition, with the ground contact adjustment having a moderate influence at the margins.
4. Discussion
4.1 Validity of the Korean→Fukuoka Transfer
The application of Kim & Chung (2022) ground contact factor ranking to a Japanese residential context rests on three supports and is subject to three threats to validity, all noted in Methods 2.3.
The supports are: subspecific identity (R. speratus kyushuensis is distributed across both Korea and Kyushu); the mechanical universality of ground contact as a barrier to subterranean termite access (a physical constraint expected to apply across geographic contexts); and the parallel evolution of Japanese residential foundation types, which enables the same era-based ground contact proxy used in the Korean context.
The threats are: the heritage-vs-residential population mismatch (direction of bias uncertain); the absence of C. formosanus from the Korean NRICH dataset (expected direction: the Korean adjustment may understate ground contact effects in Fukuoka, where C. formosanus amplifies the benefit of foundation barriers); and the cooler Korean climate (which may compress observed activity-period effects relative to Fukuoka).
The Korean transfer is the most methodologically novel element of the model and the most likely target of critical peer review. We note that it is motivated by the absence of equivalent Japanese residential data: no GLM or equivalent analysis of ground contact effects on termite damage incidence in Japanese residential buildings is available in the published literature. The Kim & Chung factor ordering is used as motivation for the direction and relative scaling of the era-based adjustment; the adjustment values themselves are expert elicitation, as stated in Methods 2.3.
4.2 Score Framing: Cohort Prevalence Rates, Not Individual Property Probabilities
A critical interpretive point: the model outputs are cohort damage prevalence rates — the historical proportion of buildings of equivalent age and treatment status in the MLIT survey that showed evidence of termite damage at the time of inspection. They are not Bayesian posteriors for any specific property.
A score of 57% for a 1968 building should be read as: “Among buildings of approximately this age and with no confirmed retreatment in the national survey, 57% showed evidence of termite damage.” It does not mean there is a 57% probability that this specific property is infested at the time of inspection. Some properties in this cohort have been inspected, treated, or are simply not yet damaged; others may have extensive concealed damage not yet manifesting visibly. The score is a base rate for informed inspection-prioritisation, not a diagnostic.
This distinction is reinforced in all model output disclaimers. It is also the correct interpretation of a buyer’s situation: a property in a cohort with 57% damage prevalence has a 57% base rate of being in the damaged subset, which a physical inspection can resolve.
4.3 The Untreated Cohort Assumption
The untreated assumption is the most consequential model choice. It produces conservative estimates, which is appropriate for buyer-facing risk communication. The assumption is empirically motivated (see Section 2.2) and is explicitly stated in all outputs. Where retreatment records exist, the output JSON provides the retreated-cohort score for direct comparison.
4.4 The Ie-shiroari Severity Flag
The flag-vs-multiplier decision and its scientific basis are discussed in Section 2.4. We emphasise here that the flag is not a computational convenience: it represents a claim that C. formosanus infestation constitutes a qualitatively different damage scenario — faster progression, larger structural impact, greater likelihood of multi-building involvement — that a scalar probability score cannot adequately communicate. Buyers in the Fukuoka zone are informed of this qualitative difference explicitly in the panel text; they are not given a falsely precise probability adjustment that would imply an empirical grounding the data does not support.
4.5 Comparison with the Asbestos Risk Model
The asbestos risk model (Vol. 1) operates at the district level using MLIT Real Estate Information Library transaction data. The termite model operates at the property level using the same construction year field available in individual property records. Both anchor to construction era; they differ in unit of analysis and calibration source (regulatory prohibition timeline for asbestos; empirical survey-derived damage rates for termite).
Both models will tend to flag the same properties: pre-1980 wooden akiya are the core at-risk population for both asbestos-containing materials and high termite damage prevalence. A combined risk profile — properties in the very-high tier for both models simultaneously — would identify properties where renovation or demolition involves regulatory compliance obligations (asbestos) and likely structural damage assessment costs (termite). Among the 314 Fukuoka properties, the majority of very-high asbestos and very-high termite properties overlap. This combined flag is implemented in the Yudane listing interface.
4.6 Limitations
-
Base rates are national, not Fukuoka-specific. The MLIT survey does not publish prefecture-stratified damage rates. Fukuoka’s rates — particularly given C. formosanus co-occurrence — are likely higher than the national untreated cohort rates for equivalent age brackets.
-
Ground contact adjustment is expert elicitation. The values +0.06/+0.04/+0.02 are not derived from empirical calibration. Sensitivity analysis (Section 3.3) suggests the tier distribution is not highly sensitive to these values, but they represent unvalidated model assumptions.
-
45–79 year bracket rates derive from secondary sources. The MLIT primary report’s fully accessible data covers brackets to approximately 44 years. Rates for older brackets cited here originate from Japanese industry sources citing the survey methodology; direct primary source verification was not possible.
-
No physical inspection validation. The model has not been retrospectively validated against physical inspection outcomes for any property in the Yudane database. Such validation would require a partnership with JSTCA-licensed contractors and constitutes the highest-priority future work.
-
Forest area unmodelled. The third significant factor from Kim & Chung (2022) — surrounding forest area proportion (η² = 0.035) — is not incorporated due to data unavailability at the listing-record level.
4.7 Future Work
- Prefecture-stratified MLIT damage rates would replace the national table with a Fukuoka-specific calibration, likely increasing very-high and high tier estimates.
- Physical inspection validation — partnering with JSTCA contractors to collect inspection outcomes for Yudane-listed properties, enabling retrospective model discrimination assessment.
- C. formosanus probability quantification — Fukuoka-specific species-attributed damage records would allow replacement of the severity flag with a quantified probability adjustment.
- Treatment history integration — JSTCA contractor record access would allow retreated-cohort scoring for confirmed properties.
5. Conclusions
We present a reproducible, property-level termite damage prevalence model for Japanese wooden residential buildings, calibrated against the MLIT 2013 national damage survey and incorporating a construction-era ground contact adjustment motivated by Korean NRICH factor analysis of Reticulitermes speratus kyushuensis damage. Model output scores are cohort damage prevalence estimates, not individual property probabilities.
Applied to 314 Fukuoka akiya listings, the model produces a risk distribution consistent with the sample’s age profile: 64% of properties correspond to age cohorts with MLIT untreated-cohort damage prevalence above 45%. All Fukuoka properties receive the Coptotermes formosanus severity flag on the grounds that colony size and damage trajectory differ qualitatively from R. speratus dynamics and cannot be adequately represented by a scalar probability.
The scoring module is released openly alongside this paper. We offer both the termite risk model and the district-level asbestos risk dataset (Vol. 1) to MLIT and Fukuoka City for integration into the PLATEAU 3D urban digital twin and the 重ねるハザードマップ geospatial infrastructure. We recommend that the 重要事項説明 disclosure framework be extended to require standardised termite risk communication at the point of akiya transaction — specifically, that akiya-bank registered properties be required to disclose a construction-era cohort damage prevalence estimate using the MLIT 2013 calibration data or an equivalent survey, to give buyers the statistical context that professional inspection can then resolve.
References
Japan Meteorological Agency (2024) Historical climate data: Fukuoka City (monthly mean temperatures). Available: https://www.jma.go.jp/
Japan Termite Control Association (JSTCA / 公益社団法人日本しろあり対策協会) (2023) FAQ: Warranty and retreatment standard. https://www.hakutaikyo.or.jp/faq/2659.html
Kim, S.-H. and Chung, Y.-J. (2022) “Analysis of factors affecting termite damage to wooden architectural heritage buildings in Korea.” Forests 13(3):465. https://doi.org/10.3390/f13030465
Kim, S.-H. et al. (2024) “Estimation of the damage risk range and activity period of termites (Reticulitermes speratus) in Korean wooden architectural heritage building sites.” Forests 15(4):602. https://doi.org/10.3390/f15040602
Kim, S.-H. et al. (2025) “Termite management in Korean wooden cultural heritage: current status and future direction after a decade of effort.” Entomological Research (Wiley). https://doi.org/10.1111/1748-5967.70080
Lee, J. et al. (2023) “New distribution of Reticulitermes speratus speratus (Blattodea: Rhinotermitidae) in Korea.” Journal of Economic Entomology 116(6):2027–2034. https://doi.org/10.1093/jee/toad183
Nanmyama, K. (2014) “全国シロアリ被害実態調査.” しろあり No. 161, January 2014. [Report: 国土交通省補助事業 シロアリ被害実態調査報告書. 日本長期住宅メンテナンス有限責任事業組合. 2013-03-31. https://www.mlit.go.jp/common/000171377.pdf]
National Institute for Environmental Studies (NIES) (2024) Invasive species database: Coptotermes formosanus. https://www.nies.go.jp/biodiversity/invasive/DB/detail/60190e.html
Smythe, R.V. and Carter, F.L. (1983) “Wood consumption and survival of Coptotermes formosanus Shiraki.” Annals of the Entomological Society of America 76(3):408–411. [Laboratory-derived wood consumption rate under optimal conditions; field average rates in established colonies vary with food availability and colony stage.]
Su, N.-Y. and Tamashiro, M. (1987) “An overview of the Formosan subterranean termite in the world.” In: Tamashiro, M. and Su, N.-Y. (Eds.) Biology and Control of the Formosan Subterranean Termite. Honolulu: University of Hawaii. [Primary source for large-colony size estimates; laboratory-reared colonies.]
Tsunoda, K. (2005) “Improved management of termites to protect Japanese homes.” In: Lee, C.-Y. and Robinson, W.H. (Eds.) Proceedings of the Fifth International Conference on Urban Pests, Singapore, 25–28 July 2005. [Kyoto University, Research Institute for Sustainable Humanosphere, Uji, Kyoto.]
Model version: 1.0.0. Scoring module: scanner/termite_risk.py (Yudane repository, CC BY 4.0). Correspondence: [email protected].
Peer Review Record
Two rounds of simulated peer review conducted 23 March 2026 prior to submission.
Round 1
Reviewer 1 (environmental health / regulatory): Major revision. Required: (1) Reframe cohort prevalence vs individual probability throughout; (2) add directional bias analysis for Korean transfer in Methods; (3) 重要事項説明 claim requires statutory precision; (4) C. formosanus flag argument should be made more forcefully. Recommendation: accept with major revision.
Reviewer 2 (structural pest management): Major revision. Required: (1) Ground contact additive adjustment risks double-counting variance already in MLIT era-rates — reframe as expert elicitation; (2) Suiter et al. (2002) citation has bibliographic error — corrected to Su & Tamashiro (1987) for colony size; (3) score framing as “probability” vs “cohort prevalence rate” is a factual distinction requiring correction throughout. Recommended also: forest area omission note; 2000 cutoff caveat for 2000–2005 transition; C. formosanus thermal threshold citation. Recommendation: major revision required.
Author response (Round 1): All major and minor comments addressed. Key changes: (a) “probability” reframed as “cohort damage prevalence rate” throughout; (b) ground contact adjustment explicitly stated as expert elicitation with sensitivity analysis in Section 3.3, double-counting concern acknowledged directly in Methods 2.3; (c) 重要事項説明 claim revised with statutory precision; (d) forest area omission acknowledged in Methods 2.3 and Limitations; (e) C. formosanus flag argument expanded in Methods 2.4 as a positive methodological claim; (f) Suiter reference corrected to Su & Tamashiro (1987); (g) construction-era distribution of sample reported in Section 3.1; (h) Tsunoda citation corrected to 2005; (i) Smythe & Carter wood consumption noted as lab-derived; (j) 2000 cutoff gradual adoption acknowledged; (k) MLIT 45–79 year bracket secondary source provenance disclosed in Table 1 footnote.
Round 2
Reviewer 1 (environmental health / regulatory): Accept with major revision. Remaining concerns: (1) η² values reported without confidence intervals; (2) construction-era distribution of 314 properties required in Results; (3) risk tier threshold justification needed; (4) repository URL not provided. Most prior concerns adequately addressed. Recommendation: accept with minor revision.
Reviewer 2 (structural pest management): Major revision. Noted that ground contact double-counting concern and score framing were now adequately addressed. Remaining: (1) “Su et al., multiple publications” inadequate for C. formosanus thermal threshold; (2) Fukuoka-specific calibration gap understated at Results presentation; (3) colony size inconsistency across paper/code/guide. Recommendation: minor revision.
Author response (Round 2): (a) η² confidence interval caveat added to Methods 2.3 — ordering used, not magnitudes; (b) construction-era distribution already in Section 3.1 (pre-1950: 4.5%, 1950–1979: 73.2%, 1980–1999: 13.7%, 2000+: 0%, unknown: 8.6%); (c) risk tier threshold justification added in Section 2.5 Table 2 with buyer-action anchoring; (d) repository URL added to Methods 2.7; (e) thermal threshold citation consolidated to Su & Tamashiro (1987) and NIES (2024); (f) Fukuoka-calibration gap note added to Section 3.3; (g) colony size phrasing settled: “1–8 million field, upper bounds approaching 10 million laboratory (Su & Tamashiro, 1987)” applied consistently across all outputs.