Your website and French law

Must my website publish terms of sale?

It depends on what the site does and whom it sells to.

  • Selling to consumers: before the order, the trader must give the seller's identity, the essential characteristics, the price including all taxes, the payment and delivery arrangements, the fourteen-day right of withdrawal for distance selling, the legal guarantees and the contact details of a consumer mediator. On a shop site, the terms are read and accepted before payment.
  • Selling to businesses: drawing up terms of sale is not compulsory, but a business that has them must provide them to any business buyer who asks. They include the payment terms, the basis on which the price is set, late-payment penalties and the fixed €40 charge for recovery costs.
  • Brochure site, with no online sales: no terms of sale to publish. Legal notices remain mandatory.

Source: Entreprendre.Service-Public, "Conditions générales de vente (CGV)" — https://entreprendre.service-public.gouv.fr/vosdroits/R43253

What is the RGAA?

The Référentiel général d'amélioration de l'accessibilité is the French method for checking that a website or an application can be used by people with disabilities. It implements the European standard EN 301 549 and the international WCAG rules. The version in force is RGAA 4.1; the administration has announced a version 5.

Source: https://accessibilite.numerique.gouv.fr

Is my website required to comply with the RGAA?

Those required to apply it are: State services, local authorities, public bodies, organisations entrusted with a public service mission, and private companies whose turnover in France exceeds 250 million euros. They publish an accessibility statement, the notice "Accessibilité : totalement, partiellement or non conforme" on the home page, and a multi-year accessibility plan.

Since 28 June 2025, a European directive also requires certain services to be accessible whatever the size of the company: online commerce, banking, electronic communications, passenger transport, e-books. Microenterprises providing services are exempt.

A small business outside these cases has no legal obligation. An accessible site remains a better-built site, serving more visitors.

Source: Article 47 of the French act of 11 February 2005; Directive (EU) 2019/882 — https://accessibilite.numerique.gouv.fr

What does naturellement vôtre do for accessibility?

The agency carries out accessibility assessments against the RGAA: what complies, what does not, and in what order to fix it. To find out whether your site is concerned or compliant, the simplest route is a half-hour conversation.

Avoid, reduce, offset (the French ERC sequence)

What is the ERC sequence?

A project, plan or programme must first avoid harm to the environment; failing that, reduce it; and only as a last resort offset the harm that could be neither avoided nor sufficiently reduced. ERC stands for éviter, réduire, compenser. The order is a hierarchy: avoidance is the only measure that guarantees there is no harm.

The principle is set out in Article L110-1 of the French Environmental Code, which aims at no net loss of biodiversity, or even a gain. It dates from the act of 10 July 1976 and was strengthened by the act of 8 August 2016. It covers every component of the environment, not biodiversity alone.

Source: French ministry in charge of ecology — https://www.ecologie.gouv.fr/politiques-publiques/eviter-reduire-compenser-impacts-lenvironnement

In which procedures does the ERC sequence apply?

In assessment and authorisation procedures: environmental impact assessment, environmental authorisation, declaration or authorisation under water legislation, Natura 2000 appropriate assessment, derogation from species protection, environmental assessment of planning documents.

Whether a specific project is subject to them depends on the project, the site and the administration: this page cannot say so in your place.

Source: French ministry in charge of ecology — https://www.ecologie.gouv.fr/politiques-publiques/eviter-reduire-compenser-impacts-lenvironnement

What do avoid, reduce and offset mean in practice?
  • Avoid: change the project so that an impact is removed entirely — change its location, route, footprint, timing or technique.
  • Reduce: limit the duration, intensity or extent of the impacts that remain.
  • Offset: provide a counterpart for significant residual impacts.

Source: French national guidelines on the ERC sequence — https://www.ecologie.gouv.fr/politiques-publiques/eviter-reduire-compenser-impacts-lenvironnement

What does the law require of an offset measure?
  • an obligation of result;
  • measures that are effective for the whole duration of the harm;
  • they do not replace avoidance or reduction;
  • they are implemented as a priority on the damaged site or close to it;
  • if the harm can be neither avoided, nor reduced, nor offset satisfactorily, the project is not authorised as it stands.

Official doctrine adds ecological equivalence, additionality, feasibility and permanence. The developer may carry out the offset itself or entrust it to an operator; it remains responsible for it.

Source: French Environmental Code, Articles L163-1 et seq.

What is collective agricultural compensation?

When a project permanently takes a significant area of farmland, the French Rural Code requires a prior study of its effects on the agricultural economy of the area, submitted to the CDPENAF for its opinion. It is triggered when three conditions are met: the project is subject to a systematic impact assessment, the land has an agricultural use, and the area taken exceeds the threshold set in the département.

It is distinct from ecological offsetting: one repairs harm to biodiversity, the other a loss to the agricultural economy.

Source: French Rural Code, Article L112-1-3; decree of 31 August 2016

Why look at ecological stakes as early as possible?

Because a sensitive area spotted at sketch stage can still be avoided. Spotted after the application has been filed, it most often leads to reduction and offsetting, which cost more and take longer. That is when TerrArgos is most useful: it says where to look first.

TerrArgos

What is TerrArgos?

Predictive analysis and decision-support software applied to land and the environment, published by the agency naturellement vôtre. It combines statistical modelling with a multi-criteria reading of the terrain (topography, sun exposure, land cover) to rank areas of ecological concern before the first field visit.

Every deliverable states its sources, the version of the processing and its parameters: a study rerun on the same data returns the same result.

Who is TerrArgos for?

Environmental consultancies, developers and local authorities. TerrArgos does not replace consultancies: it gives naturalists what they need to organise their surveys. It is used in particular to prepare:

  • an environmental authorisation or an impact assessment;
  • a protected-species derogation;
  • a Natura 2000 appropriate assessment or a water-legislation file;
  • a building or development permit, a land-clearing application;
  • the environmental assessment of a planning document (PLU, PLUi, SCOT);
  • a prior agricultural study and its review by the CDPENAF;
  • the choice and management plan of an offset site.
What is the difference between solutions A, B and C?

They deal with one species and rise in level of proof.

  • A — the maps alone. Relief, probability of presence, ecological suitability, convergence. No commentary or recommendation. €300 excl. VAT, 2 working days.
  • B — the commented report. Two models combined (Random Forest and MaxEnt), validated on a test set, with variable importance, the ROC curve and a commentary to guide fieldwork. €900 excl. VAT, 3 working days.
  • C — the full audit. Complete traceability, total reproducibility (a third party who reruns the computation lands on the same pixels) and statistical validation of each layer delivered. Designed for files that must stand up to challenge. €2,000 excl. VAT, 6 working days.
What if there is no record of the species on my site?

The model is then built at the scale of the biogeographical region, where the species is documented, and projected at high resolution onto your study area. It shows the areas that most resemble those where the species is known. It is a survey priority, not a certainty of presence.

What does TerrArgos offer for ecological offsetting?

Solution D — ACMEÉ compares a reference area (the site to be offset) with candidate parcels on two criteria: ecological similarity and potential for long-term protection. It can also search the land register for parcels within the documented dispersal radius of the species, and estimate an offset area. What is computed is kept apart from what remains to be investigated: land control, full cost, acceptability. €1,700 excl. VAT, 5 working days.

And for collective agricultural compensation?

Solution E — CCA extends the same method to the agricultural side: locating the farm parcels concerned, estimating the economic loss by sector, and clarifying the scope of the scheme. The economic estimate is an order of magnitude to prepare the discussion with the farming profession and the CDPENAF, not an opposable valuation. €1,700 excl. VAT, 5 working days.

Does TerrArgos deal with invasive species?

Yes, with solution F. It does not ask where the species is, but where there is still time to act: eradicate the isolated patch, contain the front, do not wear yourself out on the established core. It includes an island component. It is priced on quotation, according to the study area and the number of species.

Can several species be combined on the same site?

Yes, with solution G — multi-species report. It remodels nothing: it rereads studies already produced and counts, pixel by pixel, how many species are ranked as high priority there. It is a map of field effort: it says where a day of surveying yields the most. It requires the species to have been modelled over the same study area. It is priced on quotation, according to the study area and the number of species.

In what form are the results delivered?
  • a single, laid-out PDF report;
  • every map and chart as a high-resolution image, usable outside the report;
  • depending on the solution, GIS layers as GeoPackage or GeoTIFF, in Lambert-93, readable in QGIS.

Deliverables are placed in an online client area opened with a personal code. Other exports, and a report laid out in your own colours, are possible on quotation.

What are the prices and lead times?
  • A: €300 excl. VAT, 2 working days
  • B: €900 excl. VAT, 3 working days
  • C: €2,000 excl. VAT, 6 working days
  • D: €1,700 excl. VAT, 5 working days
  • E: €1,700 excl. VAT, 5 working days
  • F and G: on quotation

Lead times in working days, from receipt of the complete file. The lead time is confirmed when you order; if it has to move, you hear it beforehand. Prices exclude VAT.

Is there a subscription?

Yes, an annual one, for those who order regularly. It entitles you to a number of deliverables each month:

  • Terrain: 3 A deliverables per month, €2,900 excl. VAT per year
  • Étude: 4 A deliverables and 1 B deliverable per month, €5,900 excl. VAT per year
  • Atelier: 8 A deliverables and 2 B deliverables per month, €11,900 excl. VAT per year

Deliverables not used carry over from one month to the next within the year. Solutions not included can still be ordered individually with a subscriber discount.

How do I place an order?

By email: the boundary of the study area (a GeoPackage is enough), the species or subject concerned, and the solution wanted. We confirm within one working day that the file is complete, or tell you what is missing.

Contact: solution@terrargos.fr · +33 6 03 96 70 88

What happens to the data I send?

They are used to produce your deliverables, and nothing else. They do not enter any shared reference database, are not passed to any third party, and are deleted on written request.

Does TerrArgos replace fieldwork?

No. A probability-of-presence map is a decision aid. It replaces neither field survey, nor the judgement of the ecologist, nor the assessment of the administration. TerrArgos produces no naturalist observations: it says where to look first, and it justifies it.

Glossary — the words used in reports

The words below recur in TerrArgos reports. They do not refer to figures but to things: a boundary, a model, a map, a format.

Study area

The boundary of the area under study, as you send it to us. Everything is computed inside it.

Presence-only, presence-absence

Two families of models. The first knows only the places where the species has been seen and compares them with the whole territory. The second also has places where it was looked for and not found: it is more demanding and more robust.

Background

Points drawn within the study area to represent the conditions available across the territory. They are not recorded absences: the species may be there without having been seen.

MaxEnt

The “maximum entropy” model. It works with presence-only data: it looks for what sets the places where the species is known apart from the rest of the territory.

Random Forest

A large number of decision trees whose answers are combined. It also shows which variables weigh most.

Probability of presence

How closely a place resembles those where the species is known, from 0 to 1. A high value marks an area to survey first, not a certainty.

Ecological suitability

A reading of the terrain that is independent of observations: slope, sun exposure and land cover are scored against the known preferences of the species.

Convergence map

Several independent readings overlaid and rendered in five classes. Where they agree, the priority is at its most reliable; where they diverge, it is for fieldwork to decide.

Threshold

The value from which a place is called “suitable”. Several indicators change with it: a figure given without its threshold cannot be read.

Sensitivity, specificity

Sensitivity (or recall) is the proportion of known presences the model recovers. Specificity is the proportion of points without presence that it rightly rules out. Neither is ever read without the other: declaring the whole study area suitable gives perfect sensitivity and zero specificity.

Variable importance

The ranking of variables by their actual weight in the prediction. A measurement taken from the trained model, not an assumption.

Partial response

The effect of a single variable on the predicted probability, all others being held at their mean value.

Co-occurrence index

For each point of the map, the number of species ranked as high or very high priority. A count, not an average.

Ecological equivalence

The principle that an offset must restore habitats and functions comparable to those that are lost.

Random seed

The number that fixes the random draws of a computation. Recording it lets a third party rerun the study and land on exactly the same result.

Traceability manifest

The register attached to the report: processing steps, sources, versions, checksums. It allows the result to be verified from the source data to the final map.

DTM

Digital terrain model: the description of the relief, at one-metre resolution, from which slope and sun exposure are derived.

GeoPackage, GeoTIFF

Two open formats that QGIS can read. A GeoPackage holds outlines (parcels, zones); a GeoTIFF is an image in which each pixel carries a value.

Lambert-93

The official coordinate system of mainland France (EPSG:2154). All delivered data use it.

ACMEÉ

Comparative Analysis of Habitats and Ecological Stakes: the TerrArgos method for comparing and searching for offset sites.

CCA, CDPENAF

Collective agricultural compensation; the French departmental commission for the preservation of natural, agricultural and forest land, which gives an opinion on projects that consume such land.

RPG

The French land parcel register: the map of declared agricultural parcels and their crops.

EVEE

Invasive alien plant species (the French acronym for espèces végétales exotiques envahissantes).

Glossary — validation indicators

The indicators below are those that appear in TerrArgos reports. For each one, the answer also says what it does not allow you to conclude: that is what prevents a high figure from being over-interpreted.

AUC

Area Under the Curve — the area beneath the ROC curve

What it measures. The probability that a known presence site receives a higher score than a point drawn at random within the study area. 0.5 is equivalent to chance, 1 to perfect separation. It depends on no decision threshold, which makes it the reference indicator in species distribution modelling.

What it does not tell you. Whether the model is useful in the field: AUC measures an ability to RANK, not to locate. It is insensitive to the rarity of the species, and a high value may reflect nothing more than a geographical contrast between the surveyed area and the rest of the study area.

ROC curve

Receiver Operating Characteristic

What it measures. The rate of presences recovered plotted against the rate of false alarms, across every possible threshold. Each point on the curve corresponds to one threshold; the area beneath the curve is the AUC. The name comes from radar detection, where the curve was used to tune a receiver between signal and noise: it says nothing about ecology, it is an engineering term that passed into common use.

What it does not tell you. Which threshold to adopt: the curve shows them all. Choosing a threshold is a study decision, not a model result.

Confusion matrix

What it measures. The four counts obtained AT A GIVEN THRESHOLD: presences recovered, presences missed, false alarms, and points without presence correctly ruled out. “Without presence” means a recorded absence when the model has any, a background point otherwise. Every indicator that follows derives from them.

What it does not tell you. Nothing, if the threshold is not stated: the same matrix changes completely from one threshold to another.

Precision

What it measures. Among the sites the model declares suitable, the proportion that genuinely corresponds to a known presence. It answers: “if I survey where the model sends me, how often am I right?”

What it does not tell you. How many stations the model let slip. High precision obtained by declaring only a handful of very safe pixels suitable is of little use for directing a survey. Against a background it is also underestimated: a “false alarm” may be a station that nobody has surveyed yet.

Recall, or sensitivity

What it measures. Among the known presences, the proportion the model recovers. It answers: “is the model missing stations that are already documented?”

What it does not tell you. The cost of false alarms. A recall of 1.00 can be obtained by declaring the whole study area suitable: it is never read on its own, always together with precision.

F1

F-measure, or F-score

What it measures. The harmonic mean of precision and recall: a single figure that only becomes good if both are. The “1” means the two carry equal weight — one can in theory be favoured over the other, which is not done here.

What it does not tell you. Which of the two is weak. And since it derives from the confusion matrix, it depends on the chosen threshold and on the imbalance between presences and background.

Accuracy

What it measures. The proportion of correct predictions, presences and background taken together.

What it does not tell you. Very little here, and it is the most misleading indicator of the set. These models set a few hundred or thousand presences against a far more numerous background: a model predicting “unsuitable everywhere” would show excellent accuracy. Never to be read on its own.

TSS

True Skill Statistic — true skill index

What it measures. Sensitivity + specificity − 1. It ranges from −1 to 1: 0 for a model that does no better than chance, 1 for a perfect model, a negative value for a model that is wrong more often than chance. Unlike accuracy, it is not distorted by the ratio between presences and background.

What it does not tell you. It does, however, depend on the chosen threshold, which must be stated alongside it.

Boyce index

named after Mark Boyce, who proposed it in 2002

What it measures. Checks that the suitability classes behave monotonically: presences should become more frequent as suitability rises. Designed for presence-only models, where no true absences exist. It ranges from −1 to 1: close to 1, the ranking is coherent; close to 0, it is no better than chance; negative, it is reversed.

What it does not tell you. The quality of a prediction at any given point: it is a coherence check on the overall ranking, not a local measurement.

These answers are for information; they are not legal advice and do not rule on any particular case. For yours: valerie@naturellementvotre.eco · 06 03 96 70 88