(C) PLOS One
This story was originally published by PLOS One and is unaltered.
 . . . . . . . . . .



Biome-aligned temperature zones for interpretable climate classification via average monthly temperatures [1]

['Kai Xu', 'Courant Institute Of Mathematical Sciences', 'New York University', 'New York', 'United States Of America', 'William Lauenroth', 'School Of The Environment', 'Yale University', 'New Haven', 'Connecticut']

Date: 2025-07 

Abstract We introduce temperature zones based on average monthly temperatures that closely align with biome boundaries, intended for use in climate classification. This new system retains the simplicity and interpretability of existing classifications, such as those of Köppen-Geiger and Trewartha, while providing an improved fit to biome boundaries. Unlike previous classifications, we developed our system by analyzing contemporary climate and land cover datasets. Our classification focuses on temperature, but it is designed so that future analyses could combine our results with precipitation criteria from new or existing systems to offer a more comprehensive climate classification framework.

Citation: Xu K, Lauenroth W (2025) Biome-aligned temperature zones for interpretable climate classification via average monthly temperatures. PLOS Clim 4(6): e0000579. https://doi.org/10.1371/journal.pclm.0000579 Editor: Jingyu Wang, Nanyang Technological University, SINGAPORE Received: January 21, 2025; Accepted: May 12, 2025; Published: June 12, 2025 Copyright: © 2025 Xu and Lauenroth. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability: The data underlying the findings of this study are publicly available from the Earth Engine Data Catalog, accessible at https://developers.google.com/earth-engine/datasets, and the WorldClim official website, accessible at https://www.worldclim.org/data/worldclim21.html. Funding: Funding for this initiative was provided by the Yale School of the Environment as general faculty support, without allocation of project-specific resources. Competing interests: The authors have declared that no competing interests exist.

Introduction An effective climate classification can serve many scientific and educational purposes. Rather than relying on arbitrary criteria, many existing systems were designed to align with biome and ecosystem boundaries [1–5]. Accurately distinguishing biome boundaries is crucial because biomes represent fundamental ecological units, reflecting the interplay of climate and life. Understanding these boundaries is essential for studying global biodiversity patterns, ecosystem functioning, and predicting the impacts of climate change on ecological systems [6]. A common priority of climate classifications is simplicity; classifications that are easily understood may be preferable to more complex ones with only a marginally better fit to biome boundaries. Many existing climate classifications are based on two criteria, temperature and precipitation, which are used either alone or combined to classify a location. Arguably, the simplest such classification is a chart that plots biome types along axes of mean annual temperature and mean annual precipitation [1]. Although easy to understand, this classification lacks detail because it fails to consider seasonal variation in temperature and precipitation, which significantly affect vegetation. Similarly, the Thornthwaite and Holdridge life zone systems also aimed to use climate to describe dominant life forms but involve more complex calculations of potential evapotranspiration [2, 3], limiting their accessibility compared to simpler classifications. Two of the most popular climate classifications are the Köppen-Geiger and Trewartha classifications. The Köppen-Geiger classification accounts for seasonality of temperature and precipitation [4]. However, it has its flaws; for instance, its definition of subtropical climates is overly broad, covering regions with both deciduous forests adapted to cold winters and evergreen forests adapted to year-round warmth. Another widely used climate classification, the Trewartha model, was designed to address some shortcomings of the Köppen-Geiger system, especially in providing more granularity in mid-latitudes [5]. Nevertheless, it too has notable inaccuracies, such as its placement of an oceanic climate between the subtropical and continental climates in the eastern U.S. and China, despite significant continentality. Inspired by the contributions and limitations of prior climate classifications, we developed temperature zones that are highly correlated with biome locations, using a data-driven methodology guided by geospatial datasets of climate and land cover. Our temperature zones are based purely on average monthly temperatures—a decision that balances simplicity and detail. While more complex climate variables could be considered, monthly averages are readily available, easily interpretable, and capture the essential seasonal patterns relevant to broad-scale climate classifications [6]. Furthermore, the effectiveness of monthly average temperatures is supported by their established use in widely recognized climate classification systems like those of Köppen-Geiger and Trewartha, demonstrating their utility in capturing ecologically meaningful climate distinctions. Indeed, our analysis confirmed that these simple temperature metrics effectively establish critical biome thresholds. For example, criteria such as the mean temperature of the warmest or coldest months, and the number of months exceeding specific temperature values, successfully delineated major biome transitions from polar regions to the tropics—including those for ice cap, tundra, various forest types (boreal, hemiboreal, temperate, subtropical)—and also helped refine distinctions between continental and oceanic climates in the midlatitudes. The resulting temperature zones demonstrate a strong correlation with biome distributions and offer improved alignment with observed ecological boundaries over previous classifications.

Materials and methods Given a region delineated by a temperature classifier (for example, all land regions where at most 2 months exceed a mean temperature of 8°C), this section outlines our approach to quantifying how well such a region corresponds to the actual location of a biome. Temperature classifiers Our study used the following classifiers based upon average monthly temperatures:







We selected these temperature classifiers due to their established use in widely-recognized systems such as Köppen-Geiger and Trewartha, as well as their simplicity, availability, and ease of visual derivation from a climate graph. A key modification in our approach was the inclusion of the annual temperature range ( − ), which proved critical for distinguishing continental and oceanic climates. Furthermore, we observed that a location’s annual average temperature—a metric prevalent in traditional systems—can be approximated as the maximum value of T satisfying . This approximation is easier to derive from a climate graph. Quantifying biome-classifier fit We quantified how well a temperature classifier fits a biome using the following methodology: Let the domain D be a portion of Earth’s surface that includes both the biome of interest, B, and the region that it is to be distinguished from, . Note that D should be defined to exclude surface regions that are irrelevant to the classification task. For example, if we wanted to separate land ice cap from all other land biomes, we would define the domain D to consist only of land and not water surfaces. By applying an inequality threshold to a temperature classifier, for instance, , we divide D into a region C that satisfies the inequality, in this case the region that satisfies , and a region that does not satisfy the inequality, in this case the region that satisfies . Our goal was to choose a threshold and classifier such that C corresponds roughly to B and corresponds roughly to . We formalize this approach as follows: We define the false positive area (FPA) as the surface area of the region in the domain where and C overlap. We define the false negative area (FNA) as the surface area of the region in the domain where B and overlap. If the inequality used for the threshold is < or , as the threshold increases, FPA would increase and FNA would decrease. Meanwhile, if the inequality used for the threshold is > or , as the threshold increases, FPA would decrease and FNA would increase. To ensure that both FPA and FNA are low, we aimed to minimize the maximum of false positive and negative area, or MFPNA (Fig 1). We also restricted our classifier to integer parameters and thresholds to ensure simplicity and ease of recall. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 1. Conceptual diagram of domain D, biome B, and classified region C. A temperature-based classifier defines region C. The goal is for C to accurately represent biome B by minimizing both the false positive area (FPA, where C incorrectly includes areas outside B) and the false negative area (FNA, where C fails to include parts of B). This is achieved by minimizing the maximum of FPA and FNA (MFPNA). https://doi.org/10.1371/journal.pclm.0000579.g001 Datasets and data preprocessing We constructed a global gridded dataset of average monthly air temperatures over land and ocean surfaces by combining WorldClim 2.1 monthly temperature averages [7] with ERA5 monthly 2-meter air temperature averages [8] for the years 1970-2000. The 2-meter air temperature variable is directly provided by the ERA5 reanalysis, and was used as such in our study without derivation from pressure-level data. The resulting dataset incorporates high-resolution WorldClim data over all continental regions and uses bicubic-interpolated, smoothed ERA5 data only for oceanic regions. We sourced ground truth data for biome locations from the MODIS MCD12Q1 v061 land cover map of 2001 [9] and the Copernicus CGLS-LC100 land cover map of 2015 [10]. In regions dominated by buildings and cultivated land, we used a potential natural vegetation (PNV) map based on the BIOMES 6000 dataset [11]. We also employed the GTOPO30 digital elevation model [12] to exclude highland areas in certain analyses. We performed all calculations in Google Earth Engine [13].

Discussion and conclusion Our temperature zones were highly correlated with biome locations and represented a clear improvement on past analyses (Fig 9). One of the challenges that this and any analysis of the relationship between climate and biome locations must address is that changes in land cover lag behind changes in climate, but the exact duration of this time lag is uncertain and very likely varies across different biomes [14]. Hence, it is challenging to determine which historical period’s average temperatures, if held in a long-term steady state, would result in the land cover reflected by the datasets in this paper. Our approach used data from 1970-2000. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 9. Proposed temperature zones and corresponding map. Basemap from Natural Earth (public domain). Data: Proposed temperature zones and corresponding map. Basemap from Natural Earth (public domain). Data: https://www.naturalearthdata.com/downloads/10m-cultural-vectors/10m-admin-0-countries/ . Terms: https://www.naturalearthdata.com/about/terms-of-use/ https://doi.org/10.1371/journal.pclm.0000579.g009 Both Köppen-Geiger and Trewartha defined an ice cap climate as one satisfying . We propose adhering to this criterion, despite yielding a slightly lower MFPNA, due to its simplicity and potential to account for time-lag. Both Köppen-Geiger and Trewartha defined a polar climate, corresponding to tundra and ice cap, as one satisfying . We found that this definition is prone to false negatives, and instead recommend the classifier . Even though yielded a slightly lower MFPNA, we preferred a slightly more conservative classifier to account for potential time-lag. Both Köppen-Geiger and Trewartha used the criterion of to separate polar and subpolar biomes from temperate biomes. We found that this threshold yields an acceptable MFPNA, and maintain its usage. However, in temperate zones, the warm-summer continental, hot-summer continental, oceanic, and subtropical temperature classifications of Köppen-Geiger align poorly with temperate biomes. Instead, we propose a hemiboreal or cool temperate zone defined by , corresponding to regions with a mixture of deciduous trees resembling more equatorward forests and conifers resembling the more poleward boreal forest. We also propose a regular temperate zone for regions satisfying but not satisfying subtropical or tropical criteria, corresponding to more equatorward temperate regions with mostly deciduous forests. In both temperate and subpolar regions, we find that the simple classifier is suitable for distinguishing continental from oceanic regions, vastly outperforming the Köppen-Geiger definition. We differ from both Köppen-Geiger and Trewartha in our definition of a subtropical climate. We believe that an intuitive definition of a subtropical climate is one that is, in colloquial terms, “never too cold” and “warm on average.” As such, we define a subtropical climate as one satisfying both and , but not satisfying the criterion for a tropical climate. We define two subtropical subtypes: subtropical warm, where , and subtropical hot, where . These categories are designed so that a temperate continental climate transitions into a subtropical hot climate if the mean temperatures of all months increased by the same amount, while a temperate oceanic climate transitions into a subtropical warm climate under the same transformation. Finally, we find that the tropical threshold of used by both the Köppen-Geiger and Trewartha classifications performs fairly well for distinguishing between non-highland terrestrial regions in the tropics from those in the subtropics, and maintain its usage. Plotting our temperature zones on a map reveals several major global climatic phenomena. In the Atlantic Ocean, the Gulf Stream shifts the subtropical warm climate type into far northerly latitudes. Meanwhile, cold ocean currents off the west coasts of North and South America sometimes shift subtropical climates in coastal regions into strictly tropical latitudes. In the Northern Hemisphere, the continental climates extend farther south on the eastern sides of their respective continents, likely due to decreasing oceanic moderation of temperature. Continental climates are virtually nonexistent in the Southern Hemisphere, likely due to a lack of wide land regions enabling a large annual temperature range. For the same reason, bands of oceanic climates in the Northern Hemisphere feature more latitudinal variation than in the Southern Hemisphere. Future research will likely focus on combining these temperature zones with precipitation zones, allowing further classification into well-known humidity-based subtypes such as arid, semi-arid, humid, Mediterranean, and monsoonal. Another direction of future research is mapping historic and forecasted changes of these temperature zones to visualize the effects of climate change.

[END]
---
[1] Url: https://journals.plos.org/climate/article?id=10.1371/journal.pclm.0000579

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/