Construction duration and the AERIUS calculation: what a 12-month window can and cannot carry

Work in progress (web edition of 30 October 2026). This is a working draft under review, published here so that it can be read and checked. It may change, and its findings are not final. It is version 0.4 of StikstofNu research paper E. The English text is the only edition; a Dutch edition follows (Nederlandse versie volgt). Where the draft says it is not published, that describes it before this web edition.

What differs from the draft. References to files of the research materials are shortened to the file or folder name; those materials are not published. Nothing else is changed except what the change list of this edition names. Corrections and questions: info@alexandermooij.com.

What the calculator does with time, what the Dutch documents say, what the model structure suggests about the gap, and what has and has not been measured. Working draft v0.4, 26 October 2026.

Independence. This is independent desk research by StikstofNu. It is not affiliated with AERIUS, RIVM, the Ministry, a province or a consultant, and none of them reviewed it. Nothing here replaces an official AERIUS Calculator result or legal advice. The paper makes no policy point: it asks only what holds up under scrutiny. Evidence labels: [V] verified by opening the source; [RUN] computed on the box from committed files; [INF] inference from code, text or model structure; [U] unverified.

The answer first

1. The claim holds for the calculation, and it is narrower than "there is no time input" (V, RUN). In AERIUS Calculator 2026.0.0 (IMAER 6.0.2) no input places the emission in a period or a season of the year. Time-related inputs exist, and none of them is a window inside the year: operating hours per year (and an average residence time per visit for ship moorings); the unit of a count (TimeUnitType: hour, day, month, year), which is multiplied up to a year; the number of days a manure store is used per year; one standard emission-variation code for OPS (some codes carry a fixed seasonal or weather-dependent pattern, none lets the user choose a period); the calculation year, which selects emission factors and, for roads within 5 km, background and chemistry, but not the meteorology; and one date field (established, the date an animal housing was established or replaced), which we found no calculation code reading. An older optional field, temporaryPeriod (whole years), exists in IMAER 1.1.0 to 5.1.4, is absent from 6.0.0 onward, and is read by no calculation code we found. Within one 12-month window, two activities with the same annual emission and source characteristics give the same result however long the work runs.
2. The model structure suggests how the gap may point for maximum-type statistics, and only weakly (INF, labelled as inference). OPS is linear in emission for fixed weather (a premise we take from how the model is built; the consistency check in answer 3 does not test it hour by hour). AERIUS reports the climatological expectation of the annual deposition. A short episode meets one realisation of the weather, so at each receptor the realised deposition scatters around the model value, and the scatter is expected to grow as the duration falls (RIVM states this, Erbrink's model computation gives it for one configuration, and we did not measure it). At a single receptor, for a randomly timed episode, the model value is the expected value. The bias appears in statistics that select: the highest value over the receptors is, in expectation, at least as high in the realised field as in the model field, and strictly higher only where the weather can change which receptor is the highest (if the same receptor is always the highest, the two are equal); where it exists, we expect it to grow with the scatter and so with shorter work, but we have not shown that. Where the work falls in the year adds a signed bias that the calculation cannot see. This is an inequality for the expected maximum. It is not a statement about how often a verdict changes: that step is not proved here. We did not measure the size of any of this (Section 6).
3. What the repository tests show (RUN). They show little about duration. D01 and D02 are the same request with different names, so equal results are guaranteed if the service is deterministic: the test shows determinism and shows no gap. V4 (200 kW, 56 operating hours) confirms the emission arithmetic and the offline replica, and is duty-cycle arithmetic, not a duration measurement. One structural fact was measured: comparing V1 and V4 (same site, same source characteristics, NOx emission 4.74 times larger), NOx deposition at all 55,977 receptors scales by 4.7354 to 4.7421, where the emission ratio is 4.7385. The paper therefore holds a code-reading result, a rate ratio, a linearity consistency check and an inference from model structure. It holds no measurement of the size of the gap.
4. Time-varying profiles exist in the file format, but they do not express a work window (V). IMAER 6.0.2 has hourly and monthly profiles. The Dutch (OPS) route reads only a standard diurnal code from a list of ten and drops a custom profile. For standard mobile machinery, Calculator sets the diurnal profile itself from the machine code and overwrites a user value. The hourly and monthly profiles are consumed on the ADMS path.
5. The documents put the 12 months on the user and are silent on the consequence (V). The Instructie (Section 2.3) asks for the 12 consecutive months with the highest deposition. The Handboek says that Calculator always uses multi-year meteorology. Neither says in so many words that duration inside the window has no effect. RIVM (KN-2025-0070) says that uncertainty grows continuously as the duration falls, that there is no breakpoint, and that what to do about it is largely a policy matter. It also says that uncertainty does not mean that a short emission has no effect or can be left out of consideration, and that for some short projects a reliable calculation is possible if enough data are available (Section 4).
6. The one exact number is a rate ratio (RUN, arithmetic). In V4 the work is 56 of 8,760 hours of a year. The model spreads 12.32 kg over the year (0.00039 g/s). During the work the rate is 0.0611 g/s, 156 times higher. That is arithmetic about emission rates and not a measured error in deposition.
7. The first-hand Erbrink text is narrower than it is usually quoted (V). The four-month figure comes from a review text of 20 February 2025 by Hans Erbrink: one low source (5 m), continuous emission, one central site, four receptors at 5 km, STACKS, monthly mean concentrations over 2005 to 2014, and a factor 2 criterion that RIVM calls arbitrary. It is an indication, for one configuration, of how the scatter in point 2 can grow with shorter averaging, not a measured error of AERIUS and not a deposition result. The TNO report of 26 April 2022 that is cited with it contains no duration analysis (Section 4).
8. What is new here, and what is not (Section 8). That duration inside 12 months has no effect follows from the premises that the Handboek and the Instructie state, and we assume practitioners know it. New here: the code-level inventory of the time-related inputs we found, the check that the one project duration field of older schema versions (temporaryPeriod) is not read by calculation code at the pinned commits, the linearity consistency check, the argument from model structure (a weak inequality for maximum-type statistics) with its limits, and the reading of Erbrink against what the text covers.
9. What this paper does not support. It does not support "AERIUS is wrong for short projects". It does not support "AERIUS is fine for short projects" either. It supports a narrower statement: the number AERIUS returns is independent of duration inside 12 months, so any consideration of duration happens outside the calculation, in the choice of the annual total and in the ecological assessment. The next step that would give a measurement is a time-resolved OPS comparison (Section 10); none was run.

1. The claim, and what it covers

The claim as put to this paper: construction duration cannot enter the AERIUS calculation within a 12-month window.

We read it as four statements, and test each:

Scope. We mean the formal OwN2000 route of AERIUS Calculator and Connect 2026.0.0, with OPS for the sources in the test (SRM2 for roads within 5 km; RIVM states that OPS and SRM2 treat short emissions alike, Section 4). We do not claim anything about other models in the AERIUS system (the ADMS path), about earlier versions beyond what is stated, or about the legal basis, if any, of the 12-month rule. The Omgevingsregeling requires the use of AERIUS Calculator for the calculation (art. 4.15, second paragraph [V]); the 12-month rule itself is stated in the Instructie and on the AERIUS product page, which are guidance, and we did not find it in art. 4.15 or 6.15 (Section 9).

"Duration" has three meanings, and only one of them enters:

• (a) How long the work runs ("two weeks", "four months"): does not enter.
• (b) When it runs ("in March"): does not enter the calculation. The meteorology does not depend on the date of the work. In the physical world it can matter, and Section 6 gives an inference about direction (weak, and only for maximum-type statistics).
• (c) How much is emitted in total in the year: enters, because the user states hours per year or movements per year. A user can reduce the total, but the calculator has no notion of the period.

2. What the calculator does with time

Code references are to the AERIUS monorepo at fb8812f ("Release 2026.0.0"), IMAER-java at ecdc81b and the IMAER schema at f2a401b. The full path, line and the text found at each line are in analysis/construction_duration/e_code_refs.json, produced by a script that fails if a cited text is not at the cited line [V, RUN]. Short forms: OJ/ is source/commons-aerius/aerius-commons-ops/src/main/java/nl/overheid/aerius/ops/; IJ/ is source/ in IMAER-java.

Table 1. Where time could enter, and what the code does.

Five observations follow from Table 1.

1. "No input places the work in a period" is true, and "no input is about time" is not [V, INF]. The calculation year, the count unit, the days of use, the residence time, the hours and the standard emission-variation code are time-related. They scale or select inputs, or (for some codes) add a fixed seasonal pattern; none of them places the work inside the year. The one date field, established, records when an animal housing was built and is not read by calculation code that we found. The only field that ever recorded a duration, temporaryPeriod, was in whole years, was optional, and is gone from the current schema.
2. Emission is per year and constant over the year [V, INF]. Nothing in the chain between the GML file and the OPS emission file knows a start or an end. The OPS input carries one standard emission-variation code. Some codes add a fixed seasonal or weather-dependent pattern (space heating, animal housing, manure), but none lets the user place the emission in a chosen period, and for standard machinery the code is not the user's choice. A two-week campaign cannot be expressed at all.
3. The profile that would be needed exists in the schema but is not read for OPS [V]. CustomTimeVaryingProfile is documented in IMAER.xsd 6.0.2 (lines 1334 to 1335) as "Consists of X values, where all values must add up to X * 100", for example 24 hourly values that add up to 2400. A monthly profile would come closest to a work window. It exists for the ADMS source type only. For an OPS source, the GML reader keeps the standard code and drops a custom reference.
4. The meteorology is not chosen by the date of the work [V]. The Handboek states that the calculation year has no effect on the meteorological conditions in an OPS calculation. The Connect option meteoYear picks a dataset. Single-year OPS meteo sets exist in the OPS package (below); whether the production database offers them to Connect users was not checked [U].
5. For work that spans more than 12 months, time enters only through the user's choice of situations [V, INF]. The calculator takes the maximum over the situations the user defines. It does not look for the worst rolling 12 months, and it does not add situations. Two six-month phases entered as two TEMPORARY situations return the larger of the two per receptor, which is the right answer only if the phases do not share a 12-month window. Joining the two documented facts [INF]: the Instructie asks for the 12 consecutive months with the highest deposition; the calculator has no search for them. So for work longer than 12 months the user has to build each TEMPORARY situation from the emissions of one 12-month window, choose by hand which windows to enter, and let the job type MAX_TEMPORARY_EFFECT take, per receptor and substance, the maximum over them (less reference and netting situations, where present). The documents do not describe this procedure step by step (Section 4).

The meteorological period, as far as the file name shows. The Handboek (Section 4.7) says that OPS uses measurement data from 2005 to 2014, while the code selects the file m014123c. AERIUS builds the OPS file name from the start and end year of the meteo dataset, each modulo 100 (ControlFileWriter.toOpsMeteoFile: m0%02d1%02dc for several years, a0%02d1%02dc for one year), so m014123c is the 2014 to 2023 statistic and m005114c the 2005 to 2014 one [V for the code, INF for reading the name]. The OPS package on the box holds both and 20 single-year sets (2005 to 2024), 22 sets and 154 files in all, each listed with its sha256 in e_meteo_files.json [V, RUN]. The TNO report of 26 April 2022 says that using other ten-year meteo files changes deposition "met zo'n 2 tot 3% ofwel ongeveer 10 mol/ha/jaar" at 500 mol/ha/yr (pdf p. 27) [V]. We did not run that comparison ourselves. If the file name is read as AERIUS builds it, the Handboek text appears outdated; the choice of decade is a small effect (2 to 3 % in deposition, TNO). The scatter discussed in Section 6 is a different quantity and we do not compare their sizes.

3. What the repository tests show

Table 2. The two Connect runs that bear on duration. Both used the production Connect service, AERIUS 2026.0.0 (2026.0.0_20260929_655bf03b53), calculation year 2026, situation type PROPOSED. No new job was run for this paper. The site is a rural location in the Netherlands.

Neither run measures the effect of duration, and neither measures a gap. The D01 and D02 comparison is a determinism check; the batch plan itself says that a difference would point to non-determinism or to a release change between the two jobs. The strong evidence for the claim is the code in Table 1.

Emission scaling in the annual statistic (RUN, consistency check). V1 (100 kW, 100 hours, NOx 2.6 kg per year, same site) and V4 (NOx 12.32 kg per year) used the same source characteristics (the offline replica used one set for both machine classes and reproduced both results exactly). If deposition is linear in emission, the NOx deposition at each receptor in V4 divided by that in V1 equals 12.32 / 2.6 = 4.7385. Over all 55,977 receptors the ratio lies between 4.7354 and 4.7421 (median 4.7386), a deviation of at most 0.08 % from the emission ratio, which is the size of the rounding of the four significant digits printed in the result files. This is a consistency check on the annual statistic of two source types with equal characteristics at one site, on the production service. It does not test that OPS is linear in the emission hour by hour for changing weather, which is premise 1 of Section 6 and stays an inference [INF], and a model that multiplies the emission by a fixed transfer would pass it. It is not a test of anything that depends on time.

The row order of the two D result files differs for 138 rows. The earlier comparison, keyed by receptor id, found no difference, and this paper's script repeats that keyed comparison [RUN]. A comparison by row position would have shown differences that are not differences.

4. What the Dutch documents say

Every string in double quotation marks in this paper, in any section, was extracted from the paper text and matched against the source documents by a script (e_quotes.json, key paper_quotations): 85 strings, of which 70 were found in a source document (8 of them single words, matched by presence only) and 15 are listed in the script as the paper's own wording, not quotations (example phrases, the claim under test, labels). The script also matches a fixed list of 62 quotations (key quotes). The check catches a quotation whose words were changed, added or removed. It does not check: a page number given in the text against the recorded page; the attribution of a quotation to a document or author; that a single quoted word is the word of the cited source (presence only); a quotation whose quotation marks were removed (it is then not extracted); or that a quotation is used in the sense of its source. check_quotes.py, part of check_all.py, verifies that the strings in the paper equal the recorded strings and that the recorded counts equal those printed here; it repeats the matching against the sources only where they are available (AERIUS_E_SOURCES, AERIUS_UPSTREAM). Sources not committed (size, copyright) are identified by sha256 in that file. Two sources have no usable text layer or lose ligatures (the TNO report is a scan; the Erbrink review loses "ti"); those quotations were matched against OCR text, which can cause a miss but not a false match.

The 12-month rule. The Instructie gegevensinvoer AERIUS Calculator 2026 (BIJ12, version 1, 24 September 2026), Section 2.3, pdf p. 11: "Uitgangspunt is dat de depositiebijdrage van een project inzichtelijk wordt gemaakt in mol per hectare per jaar en dat daarvoor de aaneengesloten 12 maanden worden gemodelleerd waarvoor de depositie het hoogst is. Deze 12 maanden hoeven niet in hetzelfde kalenderjaar te liggen." The first of the two sentences is also on the AERIUS product page "Vergunningverlening" under "12 maanden"; the second is not [V, fetched 25 October 2026 and 26 October 2026]. For large projects with several realisation years, the Instructie adds that it may be necessary to determine the deposition for several (calculation) years. The Instructie is marked "interbestuurlijk niet afgestemd" (not agreed between government layers), which limits its standing [V, pdf p. 2].

Temporary situations. The Handboek (version 2026 v1; footer date 29 September 2026), Section 7.1.2, pdf p. 83: a temporary situation is meant for "een situatie die - in de tijd - plaatsvindt tussen Referentie en Beoogd, die tijdelijk van aard is". Temporary situations cannot be part of a project calculation; a maximum temporary effect can be calculated for them (same page). Section 7.4.2 (pdf p. 106): for that job type the calculation determines per receptor "de hoogste van alle tijdelijke bijdragen". The Instructie describes the same in Section 2.4.1 (pdf p. 12). So the documents describe a maximum over user-defined situations, which matches the SQL, and say nothing about how to cut a long period into situations beyond the 12-month rule.

Meteorology and the calculation year. Handboek Section 4.7 (pdf p. 45): "In AERIUS Calculator worden alle berekeningen met OPS standaard uitgevoerd met meerjarige meteorologische gegevens", so that "het effect van specifieke meteorologische omstandigheden in een bepaald jaar niet bepalend worden [kan] voor de conclusie van een vergunningaanvraag". Section 7.1.4 (pdf p. 83): "Het ingevoerde rekenjaar heeft géén effect op de gebruikte meteorologische omstandigheden"; the same section says "Het rekenjaar wordt gebruikt bij het ophalen van emissiefactoren uit de database" and, for roads within 5 km, "Een ander rekenjaar betekent een andere achtergrondwaarde en chemie". The documents therefore give a reason for multi-year meteorology (comparability and robustness against one year's weather). They do not discuss the effect on a short activity.

Emission input. The Instructie lists for mobile machinery "De tijdsduur in bedrijf [uur/jaar]" as an input, for the AUB-method in Section 8.1.1 (pdf p. 48) and for the U-method in Section 8.1.2 (pdf p. 49), which is the method of the test files. "Tijdsduur" here is hours of operation per year. For the U-method the Handboek (Bijlage 35, pdf p. 195) says that the calculation uses two kinds of input, "het aantal draaiuren en het maximale vermogen van het werktuig". Bijlage 19 (pdf p. 168) says that for a standard machine Calculator uses fixed default values per power class for the source characteristics, "etmaalvariatie (TimeVaryingProfile)" included, and that values entered by the user are overwritten.

RIVM on short projects. RIVM Kennisnotitie KN-2025-0070 (A. Bleeker and C.W.M. van der Maas, 1 August 2025; the Minister's letter of 19 September 2025 lists it among two recent RIVM notes, pdf pp. 2 and 4) answers three questions from the Ministry. On the model, it says that in permit work "meestal gewerkt wordt met langjarig gemiddelde meteorologie", which is robust for structural emissions and "leidt bij incidentele emissies tot grotere onzekerheden". It states "Er is geen technische grens of 'breekpunt' aan te wijzen waarop AERIUS wel of niet toepasbaar is op basis van emissieduur alleen", that uncertainty rises gradually as the duration falls, and that whether to ignore the deposition of short projects "is een afweging die voor een belangrijk deel beleidsmatig van aard is" (pdf p. 2). It adds that the uncertainty can be limited somewhat by adding season-specific meteorology to the calculation, and that "hoe specifieker de invoergegevens (zoals exacte locatie, tijdstip en emissiegegevens), hoe nauwkeuriger de berekening". On OPS and SRM2: "Beide modellen maken gebruik van dezelfde uitgangspunten voor meteorologie (langjarig gemiddelde weersgegevens) en hanteren dezelfde tijdsresolutie." RIVM attributes the four-month analysis to Erbrink and Duyzer; the text this paper could read is Erbrink's review (below). In its reflection on that analysis RIVM states that "Ook bij kortdurende emissies kan sprake zijn van relevante depositie. Dat een berekening onzeker is, betekent niet dat de emissie geen effect heeft of buiten beschouwing moet blijven" (pdf p. 1), and, in answer to the first question, that "Voor sommige kortdurende projecten is dus een betrouwbare berekening mogelijk, mits voldoende gegevens beschikbaar zijn" (pdf p. 2). RIVM does not give a number for the error and does not say in which direction a short project is mis-estimated.

Erbrink (first-hand text). The source of the four-month figure is a review text of the expert opinion of Prof. A. Petersen, submitted to the Eerste Kamer on 20 February 2025; the Eerste Kamer lists it as "Hans Erbrink, 8 p." [V, EKMETA and EK in e_quotes.json]. In its closing remarks, under "Kortdurende emissies" (pdf pp. 7 and 8), it says what it is about: "Als korte periode geldt hier een periode die belangrijk korter is dan een jaar", and that the international norm is a factor of 2: "Internationaal is een factor 2 nauwkeurigheid gangbaar om te bepalen of een model-uitkomst nog aanvaardbaar is". The analysis was done "voor een lage bron (5m hoog) met continue emissies ergens in het midden van Nederland met het verspreidingsmodel STACKS", computing "de maandgemiddelde concentraties" north, west, south and east of the source "op 5 km afstand", "over een periode van 10 jaar (2005-2014)". The 120 monthly means per receptor are sorted, the lowest three and highest three dropped to give a 95 % interval, and the same is done for 2, 3, 4, 5 and 6 month averages. The conclusion: from 4-month averages the uncertainty factor is 2 or less, so "deposities die het gevolg zijn van emissies die korter duren dan 4 maanden, niet met voldoende zekerheid aan een bron kunnen worden toegewezen". Two further statements: short emissions are mostly in the daytime, so "een grens van 4 maanden is aan de veilige kant"; and taller sources (25 m, 50 m with 2 MW, 110 m with 20 MW) show a larger factor, after which the author sets as starting point that the conclusion holds "voor alle brontypen (mobiele en stationaire bronnen)".

What this text supports and does not:

• It is a concentration analysis (monthly mean concentrations), although the caption of its figure says "depositieberekening". It is not a deposition calculation and not a calculation with OPS.
• It covers one low continuous source at one central site, four receptors at 5 km, one decade of meteorology. The extension to all source types is stated as a starting point by the author, with a check for three taller sources, not shown in the text.
• The factor 2 is a chosen criterion. RIVM calls it "een arbitrair gekozen criterium en geen algemeen geaccepteerde norm" and notes that "In sommige periodes kan de meteorologie juist goed overeenkomen met het langjarig gemiddelde, waardoor de onzekerheid in die gevallen geringer is dan de analyse suggereert" [V, KN].
• It is a text submitted as a review, not a published measurement; RIVM refers to a recent article by Erbrink and Duyzer in the journal Lucht, which we did not obtain [U].
• It is a model estimate (not a measurement) of how much the scatter described in Section 6 grows as the averaging time falls, for that one configuration. It does not measure the error of AERIUS for any project.

The TNO report. The review's footnote 5 cites the TNO report "Afbakening in de modellering van depositiebijdragen van individuele projectbijdragen (Fase 2) Versie 3. Referentie 100342643." with the note "[Auteurs: J. Duyzer en H. Erbrink]" for the interpretation of "not significant" (the arithmetic lower bound), not for duration. We obtained that report (34 pages, the copy published by the Eerste Kamer; a scan, read through OCR). We found no analysis of emission duration in it: the terms "emissieduur", "kortdurende", "maandgemiddeld", "tijdsduur", "emissieperiode" and "korter dan" do not occur in the OCR text (e_quotes.json, term_counts_in_text; the sha256 of the OCR text of each page and the tesseract version are recorded in the same file, and the OCR text itself is not committed, like the source documents). The duration analysis therefore rests on the review text alone. The report does contain one number that bears on Section 2: the choice of the ten-year meteorological period changes deposition by 2 to 3 %.

The Ministry. The letter of the Minister of 19 September 2025 (DGLGS/100048561, p. 4) says that researchers raised "technisch-modelmatige bezwaren" against calculating the deposition from short projects with OPS, that RIVM confirms other uncertainties apply, and that the Ministry wants to look with the provinces at how to handle this. The passage in which the letter says this gives festivals and walking tours as its examples of short projects ("Het gaat daarbij bijvoorbeeld over festivals of wandeltochten.", p. 4) [V]. It then adds, in full: "Als de rekenkundige ondergrens wordt ingevoerd, dan is dit minder relevant. Naar verwachting zijn er immers vrijwel geen kortdurende projecten met een depositiebijdrage van meer dan 1 mol/ha/jaar. De rekenkundige ondergrens wordt met voorrang opgepakt." The letter does not mention construction work, so we do not use the sentence as a statement about construction durations; the RIVM note itself lists "festivals, vreugdevuren of kleine bouwprojecten" as short projects (p. 1) [V]. The expectation is conditional on a bound being introduced, it is an expectation and not a measured fact (no figure supports it in the letter), and it concerns a bound of 1 mol/ha/yr [U for the expectation]. The condition has not been met in the version tested here. Calculator 2026.0.0 still applies the 0.005 mol/ha/yr bound (Paper D, draft, internal change request 13). The Raad van State, asked for advice on such a bound, wrote on 23 May 2025 that introducing it is, "alles bijeen genomen, kwetsbaar is", and that "Allerminst zeker is of deze dan overeind blijft in het licht van artikel 6, derde lid, van de Habitatrichtlijn" (voorlichting W11.25.00063/IV, pdf p. 2) [V]. The Tweede Kamer voted on 2 July 2026 on the amended motion of Boomsma and others (33576, nr. 511), which asked the government to raise the bound to at least 0.5 mol/ha/yr (1 mol/ha/yr after rounding) at the next AERIUS update in October and in any case before 1 January 2027, and rejected it: "zodat zij is verworpen" (Handeling 2025-2026, nr. 90, item 26) [V]. The letter's sentence concerns events, is conditional on a bound that has not been introduced, and says nothing about construction work.

Practice in rulings (examples, not a survey). We read three decisions that show where duration appears when the calculation does not carry it.

• ECLI:NL:RVS:2022:3093 (Afdeling bestuursrechtspraak, 26 October 2022), consideration 17.2: the nitrogen report concludes that only the construction phase causes an increase on the Veluwe, "maximaal 0,01 mol/ha/jr voor een tijdelijke periode van naar verwachting 1, en maximaal 2 jaar". The duration (in years) is stated in words and used in the ecological conclusion.
• ECLI:NL:RBOBR:2026:6238 (Rechtbank Oost-Brabant, interim relief, 28 August 2026), considerations 8 to 8.2: the applicant points to a calculated construction-phase contribution of "0,02 mol N/ha/jaar" ("in de aanlegfase"); the court holds that the preliminary assessment (voortoets) of "de tijdelijke beperkte toename van stikstofdepositie" meets the requirements.
• ECLI:NL:RVS:2022:3159 (2 November 2022): the Afdeling recites the legislator's view that, because AERIUS had become more accurate, "is het mogelijk om de tijdelijke deposities van activiteiten in de bouwfase op verschillende Natura 2000-gebieden individueel te berekenen en te beoordelen", and concludes that the legislator's standpoint that the partial construction exemption does not conflict with Article 6 of the Habitats Directive "niet juist is". We read the cited passages, not the whole decision.

[INF] In these decisions duration is a fact in the text of the assessment, not a parameter of the calculation. We did not review the statutory framework (Omgevingswet, Omgevingsregeling) beyond one provision: art. 4.15, second paragraph, of the Omgevingsregeling provides that AERIUS Calculator applies to the calculation of nitrogen deposition (text read in Staatscourant 2021, 31421, and in a consolidated copy) [V]. The AERIUS product page names art. 4.15 and 6.15 for the obligation to use AERIUS, and carries a disclaimer that the page has not yet been updated for the Afdeling's decisions of 18 December 2024 [V for the page].

5. What the gap is, in numbers

Within one 12-month window, AERIUS models an activity as if its annual emission were released all year. We can state the consequence for the emission rate exactly.

Table 3. Ratio of the emission rate during the work to the rate AERIUS uses, for work that is continuous and then stops.

For V4, the same ratio is 8,760 hours of the year over 56 operating hours: 156.4. The model uses 12.32 kg / 31,536 = 0.00039066 g/s all year. During the 56 hours of operation, the same mass is released at 12,320 g / (56 x 3,600 s) = 0.0611 g/s. The ratio does not depend on how the 56 hours are spread over days; it depends only on the share of the year during which the source runs [RUN, arithmetic].

What this ratio is and is not:

• It is the exact factor between two emission rates. It is not a factor between two deposition results. Deposition accumulates over the period and depends on the weather in that period. We have no time-resolved calculation and so no estimate of the size of the difference.
• It tells us that for short work the model's assumption (the rate is the same all year, the weather is the long-term average) is further from the physical situation than for work that runs all year. Section 6 gives an inference about the direction of that difference.
• The legal metric is itself a yearly mean in mol/ha/yr, and the 12-month rule is a rule about the unit of assessment. So for the number that the rule asks for, the annualisation is the rule, not an error [INF]. The open question is how well the long-term mean represents a short episode, and that is the question RIVM answered qualitatively.

6. What the model structure suggests about the gap (inference)

Status. This section is inference [INF]. It rests on three premises and a piece of statistics, none of it measured for duration. It is offered so that what the model structure does imply is stated, and it says where it stops.

Premises.

1. OPS is linear in the emission for fixed weather [INF]. Doubling the emission doubles the deposition at every receptor. This is taken from how the model is built (the result is the emission times a transfer that depends on weather and source characteristics) and not measured for changing weather. The scaling check in Section 3 is consistent with it for the annual statistic at one site, and would be passed by any model of that form, so it is a consistency check and not a test.
2. AERIUS reports a climatological expectation [V for the documents, INF for the reading]. The meteorology is a multi-year statistic (Section 2), "jaarstatistieken zijn hierbij de standaard" (Handboek Section 4.7), and the annual emission is spread evenly (Table 1). The model value at a receptor is therefore what a source with that annual emission would deposit there on average over the weather of the statistic.
3. A short episode meets one realisation of the weather. The weather during the work is not the long-term average; at a given receptor it is a draw, with the wind in some directions and not others, with some rain and some stability classes.

What follows.

• At one receptor, for an episode that starts at a random time, the model value is the expected value of the realised deposition, and the error is scatter. Linearity (premise 1) makes the realised deposition the emission times the transfer for that window, and the climatological expectation is the mean transfer over all windows. The spread of the transfer is expected to grow as the window shortens (RIVM states it; Erbrink's model computation gives it for one configuration; we did not measure it). We expect it not to vanish at one year, because the weather also differs from year to year and only the multi-year statistic removes that part; we have not checked this here. We have not measured the spread at any window. This is the direction of RIVM's statement that uncertainty grows as the duration falls, and Erbrink's model computation (Section 4) gives that growth for one low source and monthly mean concentrations, with a factor of 2 reached at about four months.
• For a statistic that selects, the expected maximum is at least the maximum of the expectations. The model gives a smooth field, the average over all wind directions. A realised short episode gives a rougher field with the same average emission going somewhere. The highest value over receptors in the realised field is, in expectation, at least as high as the highest value in the model field: the expected maximum of scattered values is at least the maximum of their expected values (Jensen's inequality for the maximum). Equality holds when the same receptor is the highest in every realisation. The difference is therefore strictly positive only where the weather can change which receptor is the highest, that is where several receptors have model values close to each other; for a receptor that is far ahead of all others in every realisation, the argument gives no bias in the maximum. Where a difference exists, we expect it to grow with the scatter and so with shorter work; we have not shown that. The verdict at a bound is a test of whether any hexagon exceeds it (Paper D), so it is a maximum-type statistic. This does not by itself show that a result of "no hexagon above the bound" is less secure for short work than for long work. That would need the probability that the realised peak exceeds the bound, and a higher expected peak does not give it: the probability depends on the whole distribution of the realised peak and on how far the modelled peak lies below the bound. We have neither derived nor measured it, so "less secure" is unproven here. We also did not measure whether the highest hexagon moves for any project. Conversely, a specific receptor whose model value lies above a threshold may receive less than the model says. For the number of receptors above a threshold, the sign depends on how many receptors lie on each side of it and we do not claim one.
• Season adds a signed bias that the calculation cannot see. If the work falls in a period whose typical transfer to a receptor differs from the annual one (for example a prevailing wind that differs by season), the realised value at that receptor is biased up or down for that period. RIVM names season-specific meteorology as a way to limit the uncertainty: "Dit zal de onzekerheid en systematische fout verminderen, maar niet wegnemen." (KN-2025-0070, pdf p. 2) [V, KN]. RIVM therefore itself speaks of a "systematische fout" for short projects calculated with long-term meteorology, and gives no direction for it. The calculator accepts neither a season nor a start date, so this part is invisible in the result.
• Daytime work. Erbrink writes that, because short emissions are mostly in the daytime, "De onzekerheid zal in praktijk dus wellicht nog wat groter zijn dan in de figuur is aangegeven" [V]. We have not checked this; the diurnal profile for standard machinery is fixed by Calculator (Table 1).

Where this stops.

• No number follows from the argument. We did not run any time-resolved calculation, so we cannot say how large the scatter or the bias is for a given project, at which duration it matters, or whether it crosses any value that decides a case.
• The legal quantity is a mean over 12 months. If the assessment asks for that mean, the annual model value is the right number by definition (Section 5). The argument concerns how well it represents a short episode, not whether it is correct for the metric.
• The scaling check (Section 3) was done at one site and one source type, for annual values only. The climatological reading of the model is a reading of the documents and of how the statistic is described, not a check of the OPS code.
• The effect of a different ten-year meteorological period is about 2 to 3 % in deposition (TNO report, pdf p. 27; not run by us). Erbrink's factor 2 at four months is for monthly mean concentrations at one configuration. These are different quantities and we do not compare their sizes.

7. What a user can and cannot do inside the calculator

[INF, from Sections 2 to 4]

• Can: state the true annual total (hours, fuel, vehicles), define several situations, run the maximum temporary effect over them, and describe duration in the supporting text. For custom machinery and for other sources on the OPS route, choose one of the ten standard diurnal codes.
• Cannot: state a start date or period, restrict the emission to a season, use a custom profile on an OPS source, choose the diurnal profile of a standard machine, or pick a rolling worst 12-month window automatically.
• Risk of reading too much into it: two activities with equal annual totals but very different durations give the same hexagon values, so a reader who sees the number cannot recover the duration from it. The duration has to be carried in the assessment text.

8. What is new beyond the Handboek, the Instructie and RIVM

Already in those documents: the 12-month rule and the instruction to enter hours per year; that Calculator uses multi-year meteorology and why; that temporary situations exist and that the maximum is taken; that uncertainty grows as the duration falls and that no breakpoint exists (RIVM). That duration inside 12 months has no effect on the result follows from these premises, and we assume that practitioners know it. The paper does not claim that this is a discovery.

What this paper adds:

• A code-level inventory of the time-related inputs we found (Table 1; it is not shown to be complete), including the ones that are easily overlooked (calculation year, count unit, days of use, residence time, the fixed emission-variation code of standard machinery, the one date field), with a script that fails if a cited line does not contain the cited text.
• A check across all 20 IMAER schema versions and the importer: a project duration field (temporaryPeriod) existed until 5.1.4 and is not read by calculation code at the pinned commits (history not examined).
• A scaling consistency check on the production service (Section 3), which bears on the premise of linearity in that argument without testing it.
• The argument from model structure itself, with what it does and does not give (Section 6), against RIVM's qualitative statement and Erbrink's single-configuration result.
• The first-hand reading of Erbrink's text and its scope, and the finding that the TNO report cited with it has no duration analysis (Section 4).
• The Ministry's sentence with its condition, set against the later status of the 1 mol bound.
• The exact rate ratios (Table 3). These are arithmetic and carry no information beyond the share of the year.

What it does not add: any measurement of the size of the gap, any comparison against time-resolved weather, or any statement on what the rules legally require beyond the one provision of the Omgevingsregeling named in Section 4.

9. Limits and what is not established

• We did not run any time-resolved or season-specific calculation. The size of the scatter and of the bias in Section 6, and the duration at which they matter, are unknown to us. RIVM does not give them either, and Erbrink gives a factor for one configuration only.
• Only the formal OwN2000 route was read. The Connect meteoYear option and the ADMS profiles were read at code level only; no run used them. Whether the production database offers the single-year meteo sets to users is unchecked [U]; they exist in the OPS package.
• D01 and D02 test determinism. V4 tests emission arithmetic and the offline replica. V1 and V4 together are a consistency check on scaling with the emission. None is a duration test, and none used a TEMPORARY situation or the maximum temporary effect job type.
• The Erbrink text is a review text for one low source, concentrations and a chosen criterion (Section 4). The article in Lucht was not obtained [U]. The Ministry's expectation about 1 mol/ha/yr is an expectation, conditional on a bound that has not been introduced [U].
• The effect of the choice of meteorological decade (2 to 3 %) is taken from the TNO report [V]; we did not run it.
• Three rulings are examples, not a survey of case law; only the cited passages were read. Of the statutory text only art. 4.15, second paragraph, of the Omgevingsregeling was read (it requires the use of AERIUS Calculator); the 12-month rule was not found in it.
• The AERIUS product page and the Instructie are guidance. The Instructie says it is not agreed between government layers.

10. Open questions that the files cannot answer

These are questions, not protocols. Nothing here is pre-registered.

1. The next step that would measure the gap. For one real site and one short activity, how far is the annual-mean AERIUS deposition from a deposition computed with OPS on season-specific or hourly meteorology? RIVM names seasonal meteorology as a possible mitigation. The offline replica holds OPS and the single-year meteo sets (Section 2), but a seasonal or monthly statistic is not part of the standard OPS distribution, so this would need its own method.
2. Does AERIUS (the helpdesk, or the maintainers) state anywhere that duration inside 12 months has no effect? We found the premise and not the sentence.
3. How do competent authorities in practice cut a project of more than 12 months into temporary situations, and does the maximum over situations understate the deposit where phases overlap in a 12-month window? This is a question about practice that the code cannot answer.
4. Does the Lucht article of Erbrink and Duyzer give deposition results, or only the concentration analysis of the review text?

11. Reproduction

construction_duration/:

• e_duration_check.py reads the committed D01 and D02 files and V4 from commit 4c0a979 (git show), recomputes the keyed comparison and the arithmetic, and writes e_numbers.json.
• e_linearity_check.py compares NOx deposition of V1 and V4 from commit 4c0a979 and writes e_linearity.json.
• e_code_refs.py resolves each code citation at the pinned commits and writes e_code_refs.json (56 references).
• e_schema_scan.py lists time-window elements in every IMAER schema version and the Java files that mention or call temporaryPeriod, and writes e_schema_scan.json.
• e_meteo_files.py lists the OPS meteo sets in the OPS package and writes e_meteo_files.json.
• e_quotes_check.py extracts every string in double quotation marks outside code spans from the paper text and matches it against the source files and the IMAER schema (85 strings), and matches a fixed list of 62 quotations; it writes e_quotes.json.
• e_build_numbers.py registers the figures of this paper in numbers.json.

Run from the repository root: python3 e_duration_check.py . > .../e_numbers.json, and likewise for the linearity check. The V4 files are read from the internal change request 8 commit; if that commit is merged, the same paths exist in the tree. The code scripts need the upstream checkouts, the quotation script the source documents, the upstream checkouts and tesseract, and the meteo script an OPS package.

References

• [C1] AERIUS monorepo, gitlab.com/AERIUS/AERIUS @ fb8812fd74f9b5a548c8772fde3a7e21be238e54 (Release 2026.0.0); IMAER-java @ ecdc81b0f1e25f484ca5cc28f289c669eea78989; IMAER @ f2a401b3eff26d20103fe1b0fc887d805228b808.
• [R1] AERIUS, Handboek werken met AERIUS Calculator 2026 (version 2026 v1, 6 October 2026), Sections 4.7, 7.1.2, 7.1.4, 7.4.2 and Appendices 19 and 35. https://www.aeriusproducten.nl/site/binaries/site-content/collections/documents/2026/10/06/handboek-werken-met-aerius-calculator-2026/handboek-werken-met-calculator-2026.pdf (read 2026-10-25).
• [R2] BIJ12 (Expertiseteam Stikstof en Natura 2000), Instructie gegevensinvoer AERIUS Calculator 2026 (version 1, 24 September 2026), Sections 2.3, 2.4.1 and 8.1.1. https://www.aeriusproducten.nl/site/binaries/site-content/collections/documents/2026/10/06/instructie-gegevensinvoer-aerius-calculator-2026/instructie-gegevensinvoer-aerius-calculator-2026.pdf (read 2026-10-25).
• [R3] AERIUS, Vergunningverlening, section "12 maanden". https://www.aeriusproducten.nl/aerius-en-stikstofbeleid/aspecten-vergunningverlening (read 2026-10-25).
• [R4] RIVM, Bleeker, A. and Van der Maas, C.W.M., Modellering kortdurende projecten, Kennisnotitie KN-2025-0070, 1 August 2025. doi:10.21945/RIVM-KN-2025-0070. https://www.rivm.nl/bibliotheek/rapporten/KN-2025-0070.pdf (read 2026-10-25).
• [R5] Minister of Agriculture, Fisheries, Food Security and Nature, Kamerbrief Doorontwikkeling berekening stikstofdepositie, 19 September 2025, DGLGS/100048561. https://www.rijksoverheid.nl/documenten/kamerstukken/2025/09/19/kamerbrief-doorontwikkeling-berekening-stikstofdepositie (read 2026-10-25).
• [R6] Erbrink, H., review of the expert opinion and position paper of Prof. A. Petersen (8 p.), Eerste Kamer, 20 February 2025. https://www.eerstekamer.nl/overig/20250220/review_gegeven_over_het/document (read 2026-10-26).
• [R7] TNO, Duyzer, J. and Erbrink, H., Afbakening in de modellering van de depositiebijdragen van individuele projectbijdragen (fase 2, versie 3), ref. 100342643, 26 April 2022, 34 p. Copy published by the Eerste Kamer: https://www.eerstekamer.nl/overig/20220727/afbakening_in_de_modellering_van/document (read 2026-10-26).
• [R8] Afdeling advisering van de Raad van State, voorlichting W11.25.00063/IV, 23 May 2025. https://zoek.officielebekendmakingen.nl/blg-1198305.pdf (read 2026-10-26).
• [R9] Tweede Kamer, Handeling 2025-2026, nr. 90, item 26 (vote of 2 July 2026 on motion 33576 nr. 511). https://zoek.officielebekendmakingen.nl/h-tk-20252026-90-26.html (read 2026-10-26).
• [R10] Staatscourant 2021, 31421 (Omgevingsregeling, art. 4.15 and 6.15 as first published). https://zoek.officielebekendmakingen.nl/stcrt-2021-31421.html (read 2026-10-26), and a consolidated copy of art. 4.15, https://concordans.nl/wetten/BWBR0045528/artikel/4.15 (read 2026-10-26).
• [J1] ECLI:NL:RVS:2022:3093; [J2] ECLI:NL:RVS:2022:3159; [J3] ECLI:NL:RBOBR:2026:6238, on rechtspraak.nl (read 2026-10-25).
• [D1] connect-batch-1/ (D01, D02, manifest.json, duurtoets.json); [D2] connect-validation-1/ at commit 4c0a979 (internal change request 8).
• Related papers: P1-aerius-deconstructed.md (Sections 5.8 and 5.10), calculation-chain.md (Section 2.5), legal-and-documentation.md (row b). Paper D (draft, internal change request 13) on the 0.005 mol/ha/yr bound.