Drawing a forest on a map is much easier than growing one.
And yet Replantio starts with exactly that gesture. Open a satellite view, draw a plot, and the browser turns that polygon into a series of questions: what climate prevails here, what is the soil pH, which species are considered native, which could tolerate these temperatures and rainfall, how quickly might they grow, how much carbon could a stand store?12
The interface gives you an exhilarating sense of control. A real piece of land, often difficult to read as anything other than hectares and coordinates, becomes something you can manipulate.
That is also exactly where you need to slow down.
Replantio is an extremely recent open-source project by Guilherme David. Its own README describes it as a screening tool: it should reduce a huge list of species to a defensible shortlist, not replace a site visit or the work of a forester.2
That sentence is probably the most important part of the software.
Because between “this species seems compatible with the plot’s climate” and “plant this species here,” there is still an entire forest of details.
A polygon becomes a stack of public data
Replantio has no application backend. Most of the work happens in the browser. When an area is drawn, the program calculates its area, then uses its centroid as a reference point to retrieve several layers of data.2
For climate data, the code currently online queries the Open-Meteo archive across ten complete years, from 2015 to 2024. It retrieves daily average and minimum temperatures, precipitation, solar radiation, relative humidity and cloud cover, then aggregates those series into monthly normals on the client side.26
For soil data, a SoilGrids query provides surface pH, among other things. Slope and aspect come from a small sample around the plot in a Copernicus digital elevation model. The country is then used to filter native species, while GBIF can provide an additional signal when a species has recorded occurrences nearby.278
The setup is interesting in its own right. A static application manages to feel like a fairly rich geospatial tool by assembling open services and databases instead of rebuilding an entire GIS system on the server.
The project also does not completely hide where its numbers come from. The results panel names ERA5, SoilGrids and GBIF. The repository documents the transformations. You can open scoring.js and see the function that decides an species is worth 0.63 rather than 0.41.2
That inspectability changes a lot. It also makes it possible to see where the simplifications begin.
The score is not a botanical oracle
Replantio’s core is adapted from EcoCrop, the database and model developed by the FAO to connect cultivated species with their environmental requirements.3
For each species, EcoCrop stores envelopes: acceptable absolute temperatures, an optimal range, precipitation, pH, sometimes cycle length and other traits. Replantio turns these bounds into trapezoidal functions. In the optimal zone, the factor is 1. As the site approaches the absolute limit, the score falls toward 0.2
The combination follows a limiting-factor logic. A species may love the temperature and hate the local rainfall: the best factor does not automatically compensate for a bad one. That is a good way to resist the temptation of a large average score that would make a fundamental incompatibility disappear.
The project then adds several adaptations for perennial plants. Temperature is evaluated over a growing-season window rather than condemning a temperate tree because its winter is cold. A safeguard also checks the annual regime. Frost is tested against the available thresholds, and Replantio even adds a photoperiod penalty calculated from latitude.2
So this is not a form that compares three numbers inside an if.
But a more sophisticated calculation does not make its inputs more precise than they are.
The FAO itself presents EcoCrop as a way to cover a very large number of species using relatively simple environmental requirements. Its documentation also highlights the downside of that breadth: for many species, the available requirements remain sketchy because the literature is limited.34
A very clean formula applied to approximate knowledge remains a very clean approximation.
The centroid is convenient. The land has not read the code
The most useful detail for understanding Replantio may be geometric.
You can draw a large, irregular plot. Its area is calculated correctly from the polygon. But the main climate and soil data are evaluated around its centroid.2
For a small, uniform site, that shortcut may be perfectly reasonable. For a valley climbing up a slope, a plot cut by a river or several hundred hectares crossing multiple soil types, the central point becomes a much harsher abstraction.
The repository acknowledges this under the term “scale honesty.” The ERA5 climate data used operates at a resolution of several kilometers. SoilGrids works at a scale of hundreds of meters. The elevation model is finer but still far from a field survey. Replantio displays radiation, humidity, cloud cover, slope and aspect, but does not feed all of them into the score because EcoCrop’s envelopes do not offer the traits needed to use them properly.2
That is a sound decision.
It would have been very easy to multiply the variables and give the result a more scientific appearance. The software sometimes does the opposite: “I have this data” does not mean “I know how to use it to predict whether this species will succeed.”
The real microclimate remains elsewhere. A pocket of cold air at the bottom of a slope, persistent shade, coastal fog, moisture retained by existing vegetation or exposure to wind can radically change a site without being captured by the centroid’s climate pixel.
A screen can display six decimal places of latitude. The plant, unfortunately, still has to live in the mud.
“Native here” is already a geographic simplification
Replantio tries to make restoration more responsible by favoring native species by default when the information is available.2
Its source is Kew’s World Checklist of Vascular Plants. WCVP provides an expert-reviewed taxonomy and geographic distributions structured by TDWG botanical regions.5
That is much better than manually adding a native: true field.
But mapping those regions to a “native here” button creates another simplification. The repository explains it plainly: an Amazonian species may be classified as native to Brazil, then appear as native on a plot in the south of the country even though its actual natural range is far away. Some TDWG regions cover several countries. A few microstates may inherit the label of a much larger region.2
In other words, “native to this country” is not the same as “ecologically local to this square kilometer.”
For restoration, that difference is enormous. Seed genetic origin, the plant communities already present, connectivity with neighboring fragments and the role of a species in that particular ecosystem are not decorative details.
Replantio can make a first cut through the list. It cannot decide on its own what restoring this place means.
A GBIF point is not a planting certificate
The software adds another rather clever signal: checking whether GBIF has observations of the species near the area.28
Seeing that a species has actually been observed a few dozen kilometers away is more tangible than a purely theoretical temperature match.
But once again, how the signal is used matters more than its existence.
A GBIF occurrence may document a natural population, a plantation, an introduced species or an observation whose ecological context has nothing to do with the plot being studied. A lack of data may also mean “nobody has properly surveyed this area,” not “the species does not grow here.”
Replantio uses it as an additional layer of evidence, not as an absolute rule. That is the right hierarchy.
A pattern is beginning to emerge: each source transforms uncertainty in a different way.
EcoCrop says, “the general conditions appear compatible.” WCVP says, “this species is considered native to this botanical region.” GBIF says, “someone has recorded this species around here.” SoilGrids says, “this soil pixel probably has these properties.” No single source says, “plant 842 individuals on this slope and come back in twenty years.”
The most appealing part is also the most fragile: watching the future grow
Once a species is selected, Replantio can make it grow on screen.
Height follows a Chapman-Richards curve. Diameter is derived from height using a form ratio. Tropical biomass uses an equation from Chave and colleagues, while temperate classes use Jenkins equations. Carbon is then calculated using IPCC factors. The stand starts at a density of 1111 stems per hectare, assumes 85% survival, adjusts the average tree against the dominant-height curve and caps above-ground biomass at 200 tonnes per hectare.2
The apparent precision suddenly becomes very strong. A tree appears, its crown expands, and tonnes of CO2 equivalent emerge.
You have to look at the comment a few lines higher in the code: the model is class-based, not species-specific.2
The 2039 species present in the version we inspected do not each have a locally measured growth curve. They are grouped into six broad behavioral families: tropical or temperate, then fast, medium or slow growth. The repository explains that these classes are assigned using genus heuristics and can be edited.2
That is enough to create a visual order of magnitude. It is not a silvicultural projection for a real plot.
Carbon makes the distinction even sharper. A biomass equation can be scientifically established while becoming highly uncertain when fed a diameter derived from a generic curve, then multiplied by a fixed density and survival rate.
The final number may be useful for comparing scenarios within the model. It would be much more dangerous to read it as the amount of carbon the land is guaranteed to deliver.
The software has already started simulating the climate of 2045
The online version also includes a glimpse of the future.
It can query Open-Meteo’s climate API for the period from 2040 to 2049 and flag a species whose score would fall below the “suitable” level in that scenario.2
The gesture is relevant. Planting a tree today means choosing an organism that will still need to tolerate the site several decades from now.
But this module currently uses one specific climate model in the code we inspected. A single scenario is not a distribution of uncertainty. For a screening tool, the next logical step would be less about producing an even more precise prediction and more about showing how the shortlist changes across several models and assumptions.
That is an important design distinction.
A bad forecasting interface gives you a number.
A good one shows what remains stable when the assumptions change, and what collapses immediately.
Why this tool can still be useful despite all these criticisms
After listing all these simplifications, you might conclude that the button should be removed and an ecologist called directly.
That would be just as absurd.
A professional does not begin with infinite knowledge of every plant on the planet. They filter, consult databases, compare the climate, rule out obvious incompatibilities, look at what grows nearby, visit the site and then refine.
Replantio automates part of the beginning of that chain.
The version we inspected contains more than two thousand species. Even an imperfect reduction to a few dozen compatible options can save a huge amount of research. The software also makes the reasons behind a poor score visible: temperature, rainfall, pH, frost or photoperiod. That makes it possible to discuss the result instead of receiving an opaque recommendation.2
This is probably where the tool is strongest: as a machine for asking better questions before the site visit.
Why is this species ruled out? Why does another appear robust? Which factor becomes limiting? Is it really native to this subregion? Are there nearby observations? How does compatibility change with the future climate?
The model does not decide. It prepares the conversation.
A map can make uncertainty manipulable
Creative tools have a particular quality when they turn an abstract problem into something you can touch.
CAD software turns dimensions into form. A sequencer turns time into blocks you can move around. Replantio tries to do something similar with land, species and several decades of growth.
The collision is interesting because ecological restoration resists simulation much more than a mechanical part does.
A forest is not simply a sum of trees compatible with an average temperature. It is a story of soil, water, competition, succession, disturbance, provenance, animals, fungi, human practices and years in which the climate refuses to follow the norm.
A good planning tool cannot remove that complexity. It can only choose which part to make visible without lying about the rest.
For now, Replantio makes some fairly intelligent choices: it publishes its model, cites its data layers, displays certain variables without scoring them when it does not know how to connect them properly to species, and states clearly that a forester is still needed.2
The risk comes from the interface itself. The prettier the map, the cleaner the number and the more concrete the growth simulation feels, the easier it becomes to forget that you are looking at a hypothesis.
This problem extends far beyond trees.
We are going to use more and more tools capable of taking public data, a few models and an excellent interface, then turning a complex field into a manipulable recommendation. Agriculture, energy, architecture, building health, local climate: the same mechanism keeps returning.
The question is not only “is the model scientific?”
We also need to ask: at what point does the tool stop being a map and start looking like an order?
Replantio is still young enough for that boundary to move. The project’s best future is probably not one where its score becomes impressive enough to replace expertise. It is one where it becomes transparent enough to make that expertise faster, better documented and easier to discuss.
Drawing the forest can take ten seconds. The software is useful if it mainly helps us understand why the next step still has to happen outside.