Philippa Kaur

A hippo swims in Colombia’s Magdalena River, near where Pablo Escobar’s compound was located.Credit: Fernando Vergara/AP/Shutterstock

Colombian environment minister Susana Muhamad has triggered fear among researchers that she will protect, rather than reduce, a growing population of invasive hippos that threaten the country’s natural ecosystems and biodiversity. Although she did not directly mention the hippos — a contentious issue in Colombia — Muhamad said during a speech in late January that her ministry would create policies that prioritize animal well-being, including the creation of a new division of animal protection.
Landmark Colombian bird study repeated to right colonial-era wrongs
The hippos escaped from drug-cartel leader Pablo Escobar’s estate after he died in 1993. Left alone, the male and three females that Escobar had illegally imported from a US zoo established themselves in Colombia’s Magdalena River and some small lakes nearby — part of the country’s main watershed. After years of breeding, the ‘cocaine hippos’ have multiplied to about 150 individuals, scientists estimate.Given that the hippos (Hippopotamus amphibius) — considered the largest invasive animal in the world — have no natural predators in Colombia and have been mating at a steady rate, their population could reach 1,500 in 16 years, according to a modelling study published in 20211. “I do not understand what the government is waiting for to act,” says Nataly Castelblanco Martínez, a Colombian conservation biologist at the Autonomous University of Quintana Roo in Chetumal, Mexico, and co-author of the study. “If we don’t do anything, 20 years from now the problem will have no solution.”Researchers have called for a strict management plan that would eventually reduce the wild population to zero, through a combination of culling some animals and capturing others, then relocating them to facilities such as zoos. But the subject of what to do with the hippos has polarized the country, with some enamoured by the animals’ charisma and value as a tourist attraction and others concerned about the threat they pose to the environment and local fishing communities.‘A bit surreal’Several studies and observations suggest how destructive it could be to allow the Colombian hippo population to explode. A 2019 paper2, for example, showed that, compared with lakes without hippos, those where the animals have taken up residence contain more nutrients and organic matter that favour the growth of cyanobacteria — aquatic microbes associated with toxic algal blooms. These blooms can reduce water quality and cause mass fish deaths, affecting local fishing communities.

A sign near Doradal, Colombia, warns passersby of the danger of invasive hippos.Credit: Juancho Torres/Anadolu Agency via Getty

Other scientists have predicted that the hippos could displace endangered species that are native to the Magdalena River, such as the Antillean manatee (Trichechus manatus manatus), by outcompeting them for food and space. They caution that traffic accidents and attacks on people caused by the hippos will become more common. And they warn that wildlife traffickers are already taking advantage of the situation by illegally selling baby hippos — a trend that could intensify.“It’s a bit surreal,” says Jorge Moreno Bernal, a vertebrate palaeontologist at the University of the North in Barranquilla, Colombia. “This is just a taste of what may come.”When Colombian authorities first recognized the speed at which the hippo population was growing, during the 2000s, they acted to reduce their numbers. But in 2009, when photos appeared online after soldiers gunned down Pepe, Escobar’s fugitive male hippo, the outcry from animal-rights activists and others plunged the environment ministry into an “institutional paralysis”, says Sebastián Restrepo Calle, an ecologist at Javeriana University in Bogotá.Researchers say that the hippos don’t belong in Colombia — they are native to sub-Saharan Africa. Simulations run by Castelblanco Martínez and her colleagues suggest that to reduce the population to zero by 2033, about 30 hippos would need to be removed from the wild population per year1. No other course of action, including sterilization or castration, would eradicate them, according to the modelling of various management scenarios, says Castelblanco Martínez.The cost of inactionThe worry now is that, instead of basing decisions on evidence and expertise in conservation, the government is listening to popular opinion, says Restrepo Calle. Neither Muhamad nor representatives of the environment ministry replied to Nature’s requests for comment.
Ancient stone tools suggest early humans dined on hippo
“Why prioritize one species over our own ecosystems?” — especially a species that isn’t native, asks Alejandra Echeverri, a Colombian conservation scientist at Stanford University in California. Along with her colleagues, Echeverri published a study last month showing that Colombia has few policies governing invasive species compared with its overall number of biodiversity policies3.Animals-rights advocates, meanwhile, argue that they aren’t ignoring environmental concerns. Luis Domingo Gómez Maldonado, an animal-rights activist and specialist in animal law at Saint Thomas University in Bogotá, says “It’s not about saving the hippos on a whim,” but rather about solving the issue while also giving the hippos justice. “My indisputable position is: let’s save as many individuals as possible, let’s do it ethically.”Researchers, too, say they have the animals’ best interests at heart. “Even if [advocates] don’t see it, we care about the hippos,” Castelblanco Martínez says. “The more time that passes, the more hippos will either have to be culled, castrated or captured.”The question is whether environmental authorities will act swiftly to draft and enforce a management plan that is both ethical and effective. Should they sit on the issue for too long, Castelblanco Martínez warns, rural communities that are most affected by the hippos might take matters into their own hands.If the government doesn’t cull them, she says, people will use shotguns to do it. More

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.This is a summary of: Zhu, L. et al. Comparable biophysical and biogeochemical feedbacks on warming from tropical moist forest degradation. Nat. Geosci. https://doi.org/10.1038/s41561-023-01137-y (2023). More

EthicsNo ethics approval was required for this study. Our study was conducted in a state-owned industrial oil palm plantation where we established a cooperation with the estate owner to access the site and collect data. No endangered or protected species were sampled. Research permits were obtained from the Ministry of Research, Technology and Higher Education, and sample collection and sample export permits were obtained from the Ministry of Environment and Forestry of the Republic of Indonesia.Study area and experimental designOur study was conducted in a state-owned industrial oil palm plantation (PTPN VI) located in Jambi, Indonesia (1.719° S, 103.398° E, 73 m above sea level). Initial planting of oil palms within the 2,025 ha plantation area started in 1998 and ended in 2002; planting density was 142 palms ha−1, spaced 8 m apart in each row and between rows, and palms were ≥16 years old during our study period of 2016–2020. The study sites have a mean annual temperature of 27.0 ± 0.2 °C and a mean annual precipitation of 2,103 ± 445 mm (2008–2017, Sultan Thaha Airport, Jambi). The management practices in large-scale oil palm plantations typically result in three contrasting management zones: (1) a 2 m radius around the base of the palm that was weeded (four times a year) and raked before fertilizer application, hereafter called the ‘palm circle’; (2) an area occurring every second inter-row, where pruned senesced palm fronds were piled up, hereafter called ‘frond piles’; and (3) the remaining area of the plantation where less weeding (two times a year) and no fertilizer were applied, hereafter called ‘inter-rows’.Within this oil palm plantation, we established a management experiment in November 2016 with full factorial treatments of two fertilization rates × two weeding practices: conventional fertilization rates at PTPN VI and other large-scale plantations (260 kg N–50 kg P–220 kg K ha−1 yr−1), reduced fertilization rates based on quantified nutrient export by harvest (136 kg N–17 kg P ha−1 yr−1–187 kg K ha−1 yr−1), herbicide and mechanical weeding15. The reduced fertilization treatment was based on quantified nutrient export from fruit harvest, calculated by multiplying the nutrient content of fruit bunches with the long-term yield data of the plantation. Fertilizers were applied yearly in April and October following weeding and raking of the palm circle. The common practice at PTPN VI and other large-scale plantations on acidic Acrisol soils is to apply lime and micronutrients, and these were unchanged in our management experiment. Before each N–P–K fertilizer application, dolomite and micronutrients were applied to the palm circle in all treatment plots using the common rates)52: 426 kg ha−1 yr−1 dolomite and 142 kg Micro-Mag ha−1 yr−1 (containing 0.5% B2O3, 0.5% CuO, 0.25% Fe2O3, 0.15% ZnO, 0.1% MnO and 18% MgO). Herbicide treatment was carried out using glyphosate in the palm circle (1.50 l ha−1 yr−1, split into four applications per year) and in the inter-rows (0.75 l ha−1 yr−1, split into two applications per year). Mechanical weeding was done using a brush cutter in the same management zones and at the same frequency as the herbicide treatment.The 22 factorial design resulted in four treatment combinations: conventional fertilization with herbicide treatment, reduced fertilization with herbicide treatment, conventional fertilization with mechanical weeding and reduced fertilization with mechanical weeding. The four treatments were randomly assigned on 50 m × 50 m plots replicated in four blocks, totalling 16 plots. The effective measurement area was the inner 30 m × 30 m area within each replicate plot to avoid any possible edge effects. For indicators (below) that were measured within subplots, these subplots were distributed randomly within the inner 30 m × 30 m of a plot. All replicate plots were located on flat terrain and on an Acrisol soil with a sandy clay loam texture.Ecosystem functions and multifunctionalityOur study included multiple indicators for each of the eight ecosystem functions23, described in details below (Supplementary Tables 1 and 2). All the parameters were expressed at the plot level by taking the means of the subplots (that is, biological parameters) or the area-weighted average of the three management zones per plot (that is, soil parameters). (1) Greenhouse gas (GHG) regulation was indicated by NEP, soil organic C (SOC) and soil GHG fluxes. (2) Erosion prevention was signified by the understory vegetation cover during the four-year measurements. (3) Organic matter decomposition was indicated by leaf litter decomposition and soil animal decomposer activity. (4) Soil fertility was signified by gross N mineralization rate, effective cation exchange capacity (ECEC), base saturation and microbial biomass N. (5) Pollination potential was designated by pan-trapped arthropod abundance and nectar-feeding bird activity. As such, it does not quantify the pollination potential for oil palm, which is mainly pollinated by a single weevil species, but rather as a proxy for a general pollination potential for other co-occurring plants. (6) Water filtration (the capacity to provide clean water) was indicated by leaching losses of the major elements. (7) Plant refugium (the capacity to provide a suitable habitat for plants) as signified by the percentage ground cover of invasive plants to the total ground cover of understory vegetation during the four-year measurements. (8) Biological control (the regulation of herbivores via predation) was indicated by insectivorous bird and bat activities and the soil arthropod predator activity.All the ecosystem functions were merged into a multifunctionality index using the established average and threshold approaches12. For average multifunctionality, we first averaged the z-standardized values (Statistics) of indicators for each ecosystem function and calculated the mean of the eight ecosystem functions for each plot. For threshold multifunctionality, this was calculated from the number of functions that exceeds a set threshold, which is a percentage of the maximum performance level of each function12; we investigated the range of thresholds from 10% to 90% to have a complete overview. The maximum performance was taken as the average of the three highest values for each indicator per ecosystem function across all plots to reduce effect of potential outliers. For each plot, we counted the number of indicators that exceeded a given threshold for each function and divided by the number of indicators for each function12.Indicators of GHG regulationWe calculated annual NEP for each plot as: net ecosystem C exchange – harvested fruit biomass C (ref. 16), whereby net ecosystem C exchange = Cout (or heterotrophic respiration) – Cin (or net primary productivity)53. The net primary productivity of oil palms in each plot was the sum of aboveground biomass production (aboveground biomass C + frond litter biomass C input + fruit biomass C) and belowground biomass production. Aboveground biomass production was estimated using allometric equations developed for oil palm plantations in Indonesia54, using the height of palms measured yearly from 2019 to 2020. Annual frond litter biomass input was calculated from the number and dry mass of fronds pruned during harvesting events of an entire year in each plot and was averaged for 2019 and 2020. Aboveground biomass production was converted to C based on C concentrations in wood and leaf litter55. Annual fruit biomass C production (which is also the harvest export) was calculated from the average annual yield in 2019 and 2020 and the measured C concentrations of fruit bunches. Belowground root biomass and litter C production were taken from previous work in our study area55, and it was assumed constant for each plot. Heterotrophic respiration was estimated for each plot as: annual soil CO2 C emission (below) × 0.7 (based on 30% root respiration contribution to soil respiration from a tropical forest in Sulawesi, Indonesia56) + annual frond litter biomass C input × 0.8 (~80% of frond litter is decomposed within a year in this oil palm plantation8). SOC was measured in March 2018 from composite samples collected from two subplots in each of the three management zones per plot down to 50 cm depth. Soil samples were air dried, finely ground and analysed for SOC using a CN analyser (Vario EL Cube, Elementar Analysis Systems). SOC stocks were calculated using the measured bulk density in each management zone, and values for each plot were the area-weighted average of the three management zones (18% for palm circle, 15% for frond piles and 67% for inter-rows)15,22.From July 2019 to June 2020, we conducted monthly measurements of soil CO2, CH4 and N2O fluxes using vented, static chambers permanently installed in the three management zones within two subplots per plot11,57. Annual soil CO2, CH4 and N2O fluxes were trapezoidal interpolations between measurement periods for the whole year, and values for each plot were the area-weighted average of the three management zones (above).Indicators of erosion preventionDiversity and abundance of vascular plants were assessed once a year from 2016 to 2020 before weeding in September–November. In five subplots per plot, we recorded the occurrence of all vascular plant species and estimated the percent cover of the understory vegetation. The percentage cover and plant species richness of each measurement year were expressed in ratio to that of 2016 to account for initial differences among the plots before the start of the experiment. For example, percentage cover in 2017 was:$$mathrm{Cover}_{2017} = frac{{left( {mathrm{Cover}_{2017} – mathrm{Cover}_{2016}} right)}}{{mathrm{Cover}_{2016}}}$$The values from five subplots were averaged to represent each plot.Indicators of organic matter decompositionLeaf litter decomposition was determined using litter bags (20 cm × 20 cm with 4 mm mesh size) containing 10 g of dry oil palm leaf litter8. Three litter bags per plot were placed on the edge of the frond piles in December 2016. After eight months of incubation in the field, we calculated leaf litter decomposition as the difference between initial litter dry mass and litter dry mass following incubation. Soil animal decomposer activity is described below (Soil arthropods).Indicators of soil fertilityAll these indicators were measured in February–March 2018 in the three management zones within two subplots per plot22. Gross N mineralization rate in the soil was measured in the top 5 cm depth on intact soil cores incubated in situ using the 15N pool dilution technique58. ECEC and base saturation were measured in the top 5 cm depth as this is the depth that reacts fast to changes in management22. The exchangeable cation concentrations (Ca, Mg, K, Na, Al, Fe, Mn) were determined by percolating the soil with 1 mol l−1 of unbuffered NH4Cl, followed by analysis of the percolates using an inductively coupled plasma-atomic emission spectrometer (ICP-AES; iCAP 6300 Duo view ICP Spectrometer, Thermo Fisher Scientific). Base saturation was calculated as the percentage exchangeable bases (Mg, Ca, K and Na) on ECEC. Microbial biomass N was measured from fresh soil samples using the fumigation-extraction method59. The values for each plot were the mean of the two subplots that were the area-weighted average of the three management zones (above)15,22.Indicators of general pollination potentialFluorescent yellow pan traps were used to sample aboveground arthropods (to determine pollinator communities60) in November 2016, September 2017 and June 2018. The traps were attached to a platform at the height of the surrounding vegetation within a 2 × 3 grid centred in the inter-rows of each plot in six clusters of three traps, totalling 18 traps per plot. Traps were exposed in the field for 48 h. We stored all trapped arthropods in 70% ethanol and later counted and identified to order and species level. The abundance of trapped arthropods in 2017 and 2018 were calculated as the ratio to the abundance in 2016 to account for initial differences among the plots before the start of the experiment. The activity of nectar-feeding birds is described below (Birds and bats).Indicators of water filtrationElement leaching losses were determined from analyses of soil-pore water sampled monthly at 1.5 m depth using suction cup lysimeters (P80 ceramic, maximum pore size 1 μm; CeramTec) over the course of one year (2017–2018)15. Lysimeters were installed in the three management zones within two subplots per plot. Dissolved N was analysed using continuous flow injection colorimetry (SEAL Analytical AA3, SEAL Analytical), whereas these other elements were determined using ICP-AES. The values for each plot were the mean of the two subplots that were the area-weighted average of the three management zones15,22.Indicators of plant refugiumIn five subplots per plot, the percentage cover and species richness of invasive understory plant species were assessed once a year from 2016 to 2020 before weeding in September–November. We defined invasive species as those plants non-native to Sumatra61 and among the ten dominant species (excluding oil palm) in the plantation for each year. The percentage cover of invasive understory plant species of each measurement year was expressed in a ratio to that of 2016 to account for initial differences among the plots before the start of the experiment. The values for each plot were represented by the average of five subplots.Indicators of biological controlThe activities of insectivorous birds and bats are described below (Birds and bats). In five subplots per plot, soil invertebrates were collected (Soil arthropods), counted, identified to taxonomic order level and subsequently classified according to their trophic groups that include predators60. The values from five subplots were average to represent each plot.BiodiversityBiodiversity was measured by the taxonomic richness of seven multitrophic groups, described in details below (Supplementary Tables 1 and 2).Understory plant species richnessThe method is described above (Indicators of erosion prevention), using the number of species as an indicator (Supplementary Table 2).Soil microorganism richnessThis was determined in May 2017 by co-extracting RNA and DNA from three soil cores (5 cm diameter, 7 cm depth) in five subplots per plot62. While DNA extraction describes the entire microbial community, RNA represents the active community. The v3–v4 region of the 16S rRNA gene was amplified and sequenced with a MiSeq sequencer (Illumina). Taxonomic classification was done by mapping curated sequences against the SILVA small subunit (SSU) 138 non-redundant (NR) database63 with the Basic Local Alignment Search Tool (BLASTN)64.Soil arthropod order richnessFor determination of soil arthropods, we collected soil samples (16 cm × 16 cm, 5 cm depth) in five subplots per plot in October–November 2017. We extracted the animals from the soil using a heat-gradient extractor65, collected them in dimethyleneglycol-water solution (1:1) and stored in 80% ethanol. The extracted animals were counted and identified to taxonomic order level61. They were also assigned to the trophic groups decomposers, herbivores and predators based on the predominant food resources recorded in previous reviews and a local study66,67. Orders with diverse feeding habits were divided into several feeding groups, for example, Coleoptera were divided into mostly predatory families (Staphylinidae, Carabidae), herbivorous families (for example, Curculionidae) and decomposer families (for example, Tenebrionidae). The total number of individuals per taxonomic group in each subplot was multiplied by the group-specific metabolic rate, which were summed to calculate soil animal decomposer activity. The values from five subplots were average to represent each plot.Aboveground arthropod order and insect family richnessIn addition to the fluorescent yellow pan traps described above (Indicators of general pollination potential), sweep net and Malaise trap samplings were conducted in June 2018, which targeted the general flying and understory dwelling arthropod communities. Sweep net sampling was conducted within the understory vegetation along two 10 m long transects per plot, with ten sweeping strokes performed per transect. In each plot, we installed a single Malaise trap between two randomly chosen palms and exposed it for 24 h. Arthropods were counted, identified to taxonomic order level and the insects to taxonomic family level and values from the three methods were summed to represent each plot.Birds and batsBirds and bats passing at each replicate plot were sampled in September 2017 using SM2Bat + sound recorders (Wildlife Acoustics) with two microphones (SMX-II and SMX-US) placed at a height of 1.5 m in the middle of each plot68. We assigned the bird vocalization to species with Xeno-Canto69 and the Macaulay library70. Insectivorous bat species richness was computed by dividing them into morphospecies based on the characteristics of their call (call frequency, duration, shape). In addition, we gathered information on proportional diet preferences of the bird species using the EltonTrait database71. We defined birds feeding on invertebrates (potential biocontrol agents) as the species with a diet of at least 80% invertebrates and feeding on nectar (potential pollinators), if the diet included at least 20% of nectar.Economic indicatorsWe used six indicators linked to the level and stability of yield and profit: yield, lower fifth quantile of the yield per palm per plot, shortfall probability, management costs, profit and relative gross margin. We assessed fruit yield by weighing the harvested fruit bunches from each palm within the inner 30 m × 30 m area of each plot. The harvest followed the schedule and standard practices of the plantation company: each palm was harvested approximately every ten days and the lower fronds were pruned. For each plot, we calculated the average fruit yield per palm and scaled up to a hectare, considering the planting density of 142 palms per ha. Because the palms in each plot have different fruiting cycles and were harvested continuously, the calculation of an annual yield may lead to misleading differences between treatments. Therefore, we calculated the cumulative yield from the beginning of the experiment to four years (2017–2020), which should account for the inter- and intra-annual variations in fruit production of the palms in the plots and thus allowing for comparison among treatments. As effects of management practices on yield may be delayed46, we also calculated the cumulative yield during two consecutive years (2017–2018 and 2019–2020) and checked for treatment effects on yield and profit indicators separately for these two periods.We computed risk indicators on the cumulative yield and on the yield between the two periods. We used the lowest fifth quantile of the yield per palm per plot (left side of the distribution) to indicate the production of the palms with lowest performance. Also, we determined the yield shortfall probability (lower partial moment 0th order), defined as the share of palms that fell below a predefined threshold of yield; the thresholds chosen were 630 kg−1 per palm for cumulative yield and 300 kg−1 per palm per year for the two-year yield, which corresponded to 75% of the average yield.Revenues and costs were calculated as cumulative values during four years of the experiment (2017–2020) using the same prices and costs for all the years. This was because we were interested in assessing the economic consequences of different management treatments, and they might be difficult to interpret when changes in prices and costs between calendar years are included, which are driven by external market powers rather than the field-management practices. For the same reason, we abstained from discounting profits. Given the usually high discount rates applied to the study area, slight differences in harvesting activities between calendar years or months might lead to high systematic differences between the management treatments, which are associated with the variation in work schedule within the plantation rather than the actual difference among management treatments. Revenues were calculated from the yield and the average price of the fruit bunches in 2016 and 201761. Material costs were the sum of the costs of fertilizers, herbicide and gasoline for the brush cutter. Labour costs were calculated from the minimum wage in Jambi and the time (in labour hours) needed for the harvesting, fertilizing and weeding operations, which were recorded in 2017 for each plot. The weeding labour included the labour for raking the palm circle before fertilization, which was equal in all treatments, and the weeding in the palm circle and inter-rows either with herbicide or brush cutter. In addition, we included the time to remove C. hirta, which must be removed mechanically from all plots once a year, calculated from the average weed-removal time in the palm circle and the percentage cover of C. hirta in each plot for each year. We then calculated the profit as the difference between revenues and the total management costs and the relative gross margin as the gross profit proportion of the revenues.StatisticsTo test for differences among management treatments for each ecosystem function and across indicators of biodiversity, the plot-level value of each indicator was first z standardized (z = (actual value − mean value across plots) / standard deviation)4. This prevents the dominance of one or few indicators over the others, and z standardization allows several distinct indicators to best characterize an ecosystem function or biodiversity4. Standardized values were inverted (multiplied by −1) for indicators of which high values signify undesirable effect (that is, NEP, soil N2O and CH4 fluxes, element leaching losses, invasive plant cover, yield shortfall, management costs) for intuitive interpretations. For a specific ecosystem function (Supplementary Figs. 1 and 2) and across indicators of biodiversity (Fig. 2), linear mixed-effects (LME) models were used to assess differences among management treatments (fertilization, weeding and their interaction) as fixed effects with replicate plots and indicators (Supplementary Tables 2 and 3) as random effects. The significance of the fixed effects was evaluated using ANOVA72. The LME model performance was assessed using diagnostic residual plots73. As indicator variables may systematically differ in their responses to management treatments, we also tested the interaction between indicator and treatment (Table 1). For testing the differences among management treatments across ecosystem functions (that is, multifunctionality; Fig. 1), we used for each replicate plot the average of z-standardized indicators of each ecosystem function and ranges of thresholds (that is, number of functions that exceeds a set percentage of the maximum performance of each function12; Supplementary Fig. 3). The LME models had management treatments (fertilization, weeding and their interaction) as fixed effects and replicate plots and ecosystem functions as random effects; the interaction between ecosystem function and treatment were also tested to assess if there were systematic differences in their responses to management treatments (Table 1). As we expected that the type of weeding will influence ground vegetation, we tested for differences in ground cover of understory vegetation, measured from 2016 to 2020, using LME with management treatments as fixed effect and replicate plots and year as random effects. Differences among management treatments (fertilization, weeding and their interaction) in yield and profit indicators, which were cumulative values over four years (Fig. 3) or for two separate periods (2017–2018 and 2019–2020; Supplementary Fig. 4), were assessed using linear model ANOVA (Table 1). For clear visual comparison among management treatments across ecosystem functions, multitrophic groups for biodiversity, and yield and profit indicators, the fifth and 95th percentiles of their z-standardized values were presented in a petal diagram (Fig. 4 and Supplementary Fig. 5). Data were analysed using R (version 4.0.4), using the R packages ‘nlme’ and ‘influence.ME’73.Reporting summaryFurther information on research design is available in the Nature Portfolio Reporting Summary linked to this article. More

RESEARCH BRIEFINGS
01 March 2023

Global spatial and temporal patterns of coastal phytoplankton blooms were characterized using daily satellite imaging between 2003 and 2020. These blooms were identified on the coast of 126 of the 153 ocean-bordering countries examined. The extent and frequency of blooms have increased globally over the past two decades. More

There is a recent change in the modus operandi of Brazilian Amazon deforestation. The proportion of illegal deforestation in public land increased from ~43–44% (2015–2018) to ~49–52% (2019–2021)10. Land grabbers occupy public lands (deforesting or raising cattle) in a high-risk expectation of receiving title to the land and/or trading the land with significant returns (land speculation)6,7. Therefore, we argue that it is crucial to rapidly assign most of the Amazon’s UPFs to land tenure regimes associated with conservation. Land-tenure security will bring greater governance and protection to these areas. Achieving this goal requires a combination of three measures: (1) careful attention to the choice of land tenure categories for UPFs, (2) technological improvements, and (3) law enforcement.Choice of land tenure category for UPFsPublic lands in Brazil include several categories, such as conservation areas (with several subcategories under law number 9985/2000), Indigenous lands, and rural settlements, among others. Therefore, the category choice for each undesignated public land area requires studies to determine those lands’ social, environmental, or productive suitability, taking note of their histories of occupation, cultural importance, and potential uses. The unpopulated forest is a myth. Most of the areas in the Amazon have been occupied by human populations—traditional communities, indigenous villages, uncontacted tribes, “riverside” (ribeirinho) peoples, or small farmers—for generations. Ancestral occupation of land without proof or associated studies, however, does not guarantee land rights. Therefore, to avoid unfair competition for land and unilateral political decisions, the best choice of land category for a given UPF to meet social, ecological and economic demands would benefit from active social participation, multidisciplinary scientific studies, in situ observations, and innovative technologies (e.g., remote sensing, data processing capabilities, machine learning, cloud computing) to provide fast, scalable, and quality information.Final allocation decisions, however, must be preceded by participatory and transparent consultation processes to avoid conflicts and safeguard land rights. The measure of assigning tenure categories to the UPFs has a high level of complexity in itself and may benefit from the support of multi-actors (e.g., governments, academia, civil society, private sector) at multi-levels (e.g., studies, participation processes, decision-making processes) and multi-scales (local, regional and national). Despite the complexity, there are examples in the early 2000s of joint efforts to allocate land (“Terra Legal” Program) and create protected areas on a large scale and in a short period of time in the Brazilian Amazon. We emphasize, however, that the tenure categories selected for the UPFs need to maintain forest cover, remain in the public domain in compliance with national laws, and enhance long-term Amazon conservation, respecting the rights of resident populations.Technological improvements to control land grabbing in UPFLasting conservation of the Amazon rainforest depends on ending land-grabbing and illegal deforestation in public forests (designated or undesignated). However, land grabbers are using a self-declaratory tool to declare illegally invaded public lands as private properties, which demands immediate technological improvements to the system.The Rural Environmental Registry (CAR is the Portuguese acronym) is a mechanism of environmental oversight of private lands under the Brazilian Forest Code (Law 12,651/2012). CARs are registered on a web-based platform (Rural Environmental Registry System – SICAR). By law, landowners must self-declare their property boundaries and land use types (e.g., residential, agricultural, protection) in SICAR, respecting legally required protection of certain forest areas and watercourses. Then, a state environmental agency must validate the information. Unfortunately, the validation process has been extremely slow (e.g., More

OverviewThis study establishes a framework for mapping carbon stocks at the level of individual trees at a sub-continental scale in semi-arid sub-Saharan Africa north of the Equator. We used satellite imagery from the early dry season (Extended Data Fig. 1). The deep learning method developed by a previous study1 allowed us to map billions of discrete tree crowns at the 50-cm scale from West Africa to the Red Sea. Then we used allometry to convert tree crown area into tree wood, foliage and root carbon for the 0–1,000 mm year−1 precipitation zone in which our allometry was collected (Extended Data Fig. 2). We introduce a viewer that enables the billions of trees to be viewed at different scales, with information on location, metadata of the Maxar satellite image used, tree crown area and the estimated wood, foliage and root carbon content based on our allometry (Fig. 4). We also make available our output data for the 1,000 mm year−1 precipitation zone southward to 9.5° N latitude with information on location, precipitation, metadata of the Maxar satellite image used, tree crown area, tree wood carbon, tree root carbon and tree leaf carbon.Satellite imageryWe used 326,523 Maxar multispectral images from the QuickBird-2, GeoEye-1, WorldView-2 and WorldView-3 satellites collected from 2002 to 2020 from November to March from 9.5° N to 24° N latitude within Universal Transverse Mercator (UTM) zones 28–37 for Africa (Extended Data Table 1a). These images were obtained by NASA through the NextView License from the National Geospatial-Intelligence Agency. Data were assembled over several years with a focus on later years to achieve a relatively recent and complete wall-to-wall coverage.When using satellite data from different satellites over several years, with varying sun–target–satellite angles, with varying radiometric calibration of satellite spectral bands and different atmospheric compositions through which the surface is imaged, there are two possibilities for using hundreds of thousands of satellite images together quantitatively. One approach, used extensively in NASA’s, NOAA’s and the European Space Agency’s Earth-viewing satellite programmes, is to quantitatively inter-calibrate radiometrically the satellite channels through time; correct these data for time-dependent atmospheric effects such as aerosols, clouds, haze, smoke, dust and other atmospheric constituent effects and then normalize the viewing perspective to the same sun–target–satellite angle38. Another approach is to use the satellite data as collected; assemble training data of trees viewed from different satellites under different sun–target–satellite angles, different times, different atmospheric conditions and use machine learning with high-performance computing to perform the tree mapping at the 50-cm scale. The key to successful machine learning is to account for all the sources of variation within the domain of study in the training data to ensure accurate identification of trees under all circumstances. We included trees viewed substantially off-nadir, trees collected under different aerosol optical thicknesses, trees collected under cirrus cloud conditions, trees viewed in the forward and backward scan directions, trees on sandy soils, trees on clay soils, trees on burn scars, trees in laterite areas and trees in riverine settings. Our training data were collected by one team member and are a carefully selected manual delineation of 89,899 individual trees under a range of atmospheric conditions, viewing perspectives and ecological settings.All multispectral and panchromatic bands associated with our Maxar images were orthorectified to a common mapping basis. We next pan-sharpened all multispectral bands to the 0.5-m scale with the associated panchromatic band. The absolute locational uncertainty of pixels at the 0.5-m scale from orbit is approximately ±11 m, considering the root-mean-square location errors among the QuickBird-2, GeoEye-1, WorldView-2 and WorldView-3 satellites (Extended Data Table 1). We formed the normalized difference vegetation index (NDVI)39 from every image in the traditional way from the pan-sharpened red and near-infrared bands. We also associated the panchromatic band with the NDVI band and ensured that the panchromatic and NDVI bands were highly co-registered. The NDVI was used to distinguish tree crowns from non-vegetated background because the images were taken from a period when only woody plants were photosynthetically active in this area36. Our training data were labelled on images from the early dry season when only trees have green leaves. Because most semi-arid savannah trees continue to photosynthesize in the early dry season after herbaceous vegetation senesces, green leaf tree crowns are easily mapped because of their higher NDVI values than their senescent herbaceous vegetation surroundings. We substantiate this by analysis of 308 individual trees using NDVI time series with 4-m PlanetScope imagery that emphasized the importance of satellite data from the November, December and January early dry-season months (Extended Data Fig. 1).We next formed our data into mosaics by applying a set of decision rules, resulting in a collection of 16 × 16-km tiles within each UTM zone from 9.5° N to 24° N latitude for Africa. The initial round of scoring considered percentage cloud cover, sun elevation angle and sensor off-nadir angle: preference was given to imagery that had lower cloud cover, then higher sun elevation angle and finally view angles closest to nadir. In the second round of scoring, selections were assigned priority to favour early dry-season months and off-nadir view angles: preference was given to imagery from November, December and January with off-nadir angle less than ±15°; second to imagery from November to January with off-nadir angle between ±15° and ±30°; third to imagery from February or March with off-nadir angle less than ±15°; and finally to imagery from February or March with off-nadir angle between ±15° and ±30°. Image mosaics were necessary to eliminate multiple counting of trees. We formed mosaics using 94,502 images for tree segmentation, with 94% of these being from November, December and January. Ninety percent of our selected mosaic imagery was within ±15° of nadir, 87% were acquired between 2010 and 2020 and 94% were from the early dry season (Extended Data Fig. 7). A summary of month, year, solar elevation and off-nadir angle by UTM zone can be found in Supplemental Information Fig. 1.Possible obscuration of the surface by clouds totalled 4.1% of our input mosaic data area and aerosol optical depth >0.6 at 470-nm (ref. 40) areas totalled 3.4% of our input data. However, we mapped 691,477,772 trees in our possible cloud-cover-affected and aerosol-affected areas, indicating that cloud and aerosol effects were lower than these numbers. In addition, 0.9% of our input data did not process. We include a data layer in our viewer for these three conditions.Mapping tree crowns with deep learningWe used convolutional neural network models developed by a previous study1. The models were trained with manually delineated and annotated 89,899 individual trees along a north–south gradient from 0 to 1,000 mm year−1 rainfall1. Only features that showed a distinct crown area and associated shadow were included, which excluded small bushes, grass tussocks, rocks and other features that might have green leaves or cast a shadow from our classification. All training data and model training was done in UTM zones 28 and 29. Because tree floristic diversity in the 0–1,000 mm year−1 zone of our study is highly similar from the Atlantic Ocean to the Red Sea across Africa41,42,43, we added no further training data as our study moved further eastward. We used state-of-the-art deep learning to segment trees crowns at the 50-cm scale1. We used two different models based on a U-Net architecture, one for lower-rainfall desert regions with 150 mm year−1. Details about the network architecture, training process and hyperparameter choices can be found in ref. 1. Previous evaluation showed that early dry-season images performed better than late dry-season images, which was a limitation of our previous study. We reduced this error by using early dry-season images with only 6% of our area being covered by images from February and March. The models were also designed to separate clumped trees by highlighting spaces between different crowns during the learning process, similar to a strategy for separating touching cells in microscopic imagery22.AllometryVery-high-resolution satellite images and deep learning have achieved mapping of individual trees over large areas1. Each tree is georeferenced in the satellite data and defined by crown area. The challenge was to develop allometric equations for foliage, wood and root dry masses or carbon based on crown area regardless of species. This was met by reanalysing existing Sahelian and Sudanian woody plant data from destructive sampling. Overall, the seasonal maximum foliage, wood and root dry masses were measured on 900, 698 and 26 trees or shrubs from 27, 26 and 5 species, respectively, for which crown area was also measured. Several allometric regression models tested for foliage, wood or root masses are power functions and independent of species. All the regression outputs were inter-compared for fit indicators, by systematic estimates of prediction uncertainty and by root-to-wood ratios and foliage-to-wood ratios over the range of crown areas. This resulted in a set of ordinary least squares log–log equations with crown area as the independent variable. The Sahelian and Sudanian allometry equations were also compared with published allometry equations for tropical trees, primarily from more humid tropics, which are generally based on stem diameter, tree height and wood density. Our allometric predictions are within the range of other allometry predictions, reinforcing the confidence in their use beyond the Sahelian and Sudanian domains into sub-humid savannahs for discrete trees19.On the basis of ref. 19, we predicted the wood (w), foliage (f) and root (r) dry mass as functions of the crown area (A) of a single tree as:$$begin{array}{c}{text{mass}}_{{rm{w}}}(A)=3.9448times {A}^{1.1068},({N}_{{rm{w}}}=698)\ {text{mass}}_{{rm{f}}}(A)=0.2693times {A}^{0.9441},({N}_{{rm{f}}}=900)\ {text{mass}}_{{rm{r}}}(A)=0.8339times {A}^{1.1730},({N}_{{rm{r}}}=26)end{array}$$The tree mass components of wood, leaves and roots were combined to predict the total mass(A) in kg of a tree from its crown area A in m2:$$text{mass}left(Aright)={text{mass}}_{{rm{w}}}left(Aright)+{text{mass}}_{{rm{f}}}left(Aright)+{text{mass}}_{{rm{r}}}left(Aright)$$As in ref. 1, a crown area of size A  > 200 m2 was split into ({rm{lfloor }}A/100{rm{rfloor }}) areas of size 100 m2 and one area with the remaining m2 if necessary. We converted dry mass to carbon by multiplying with a factor of 0.47 (ref. 44).Uncertainty analysisWe evaluated the uncertainty of our tree crown area mapping and carbon estimation in two ways. First, we quantified our tree crown mapping omission and commission errors by inspecting randomly selected areas from UTM zones 28–37, validating that our neural network generalized over UTM zones consistently (Extended Data Fig. 8).Second, we quantified the relative error of our tree crown area estimation. We consider the uncertainty Δx of a quantity x and the corresponding relative uncertainty δx defined by the absolute and relative error, respectively45. To assess the relative error in crown area estimation resulting from errors by the neural network, we considered external validation data from ref. 1, which were not used in the model-building process. We considered expert-labelled tree crowns as well as the predicted tree crowns from 78 plots of 256 × 256 pixels. The hand-labelled set contained 5,925 trees and the system delineated 5,915 trees. The total hand-labelled tree crown area was 118,327 m2 and the neural network predicted 121,898 m2. This gave a relative error in crown area mapping of δarea = 3.3%. We matched expert-labelled and predicted tree crowns and computed the root-mean-square error (RMSE) per tree, taking overlapping areas and missed trees into account (see Extended Data Fig. 8). We estimated the allometric uncertainty (δallometric) using the data from ref. 19 (see below). The two relative errors δarea and δallometric were combined to an overall uncertainty estimate for the carbon prediction of ±19.8% (see below).Omission and commission errorsWe evaluated our tree crown mapping accuracy by analysis of 1,028 randomly selected 512 × 256-pixel areas over the 9.5° N to 24° N latitude within UTM zones 28–37. Because the drier 60% of our study area only contains 1% of the 9,947,310,221 trees we mapped in the 0–1,000 mm year−1 rainfall zone, we applied an 80% bias for selecting evaluation areas above the 200 mm year−1 precipitation line46, as >98% of tree identifications were above the 200 mm year−1 precipitation isoline. Identified tree polygons were further categorized into tree crown area classes from 0–15 m2, 15–50 m2, 50–200 m2 and >200 m2, with a total of 50,570 trees evaluated. Although a previous study reported greatest uncertainty in both the smallest and largest area classes1, our more expansive work found the greatest uncertainty in our smallest tree class. We excluded from evaluation any tiles that had annual precipitation46 >1,000 mm year−1 and all areas that were devoid of vegetation, leaving us with 850 areas.Seven members of our team evaluated the accuracy in terms of commission and omission by tree crown area classes for the 850 areas. Input data provided for every area were the NDVI layer, the panchromatic shadow layer and the neural net mapping results in each of the four crown area classes. Ancillary data available to evaluators included the centre coordinates for comparison with Google Earth data, the Funk et al.46 rainfall, the acquisition date of the area evaluated and the viewing perspective.We identified areas wrongly classified as tree crowns (commission errors), missed trees (omission errors) and crown areas corresponding to clumped trees (Extended Data Fig. 8). Clumped trees were most common for >200 m2 tree crown area. They were rare in the 3–15 m2 and 15–50 m2 tree classes, which comprise 88% of our tree crowns. In the 850 patches, the number of trees ranged from one tree to 326 trees, with a total of 50,570 trees evaluated and 3,765 errors identified. Overall, the commission and omission error rates were 4.9% and 2.7%, respectively, a net uncertainty of 2.2%.Allometric uncertainty estimationThe prediction of tree carbon from the crown area for a single tree based on crown area alone is inherently uncertain47,48. As the allometric equations are based on three different datasets, we compute their uncertainties independently, combine them and put them in relation to the total carbon measured in the three datasets.The allometric equations were established using an optimal least-squares fit of an affine linear model predicting the logarithmic carbon from the logarithmic tree crown area19. To estimate the uncertainty of the allometric equations, we repeated the fitting using random subsampling. The datasets were randomly split into training data (80%) for fitting the allometric equations and validation data (20%) for assessing the uncertainty. For example, from the root measurements, (({A}_{1},{y}_{1}),ldots ,({A}_{{N}_{{rm{r}}}},,{y}_{{N}_{{rm{r}}}})), we compute ({mu }_{{rm{r}}}=frac{1}{{N}_{{rm{r}}}}mathop{sum }limits_{i=1}^{{N}_{{rm{r}}}}{y}_{i}) and ({hat{mu }}_{{rm{r}}}=frac{1}{{N}_{{rm{r}}}}mathop{sum }limits_{i=1}^{{N}_{{rm{r}}}}{text{mass}}_{{rm{r}}}({A}_{i})). The corresponding error is ({varDelta }_{{rm{r}}}=|{mu }_{{rm{r}}}-{hat{mu }}_{{rm{r}}}|).Because the total carbon for a tree with a certain crown area is the sum of the three carbon components, we add the absolute uncertainties assuming independence45.$${varDelta }_{{rm{a}}{rm{l}}{rm{l}}{rm{o}}{rm{m}}{rm{e}}{rm{t}}{rm{r}}{rm{i}}{rm{c}}}simeq sqrt{{varDelta }_{{rm{f}}}^{2}+{varDelta }_{{rm{w}}}^{2}+{varDelta }_{{rm{r}}}^{2}}$$and compute the relative uncertainty as ({delta }_{text{allometric}}=frac{{varDelta }_{text{allometric}}}{{mu }_{text{mass}}}), in which the average mass μmass is given by the sum of the averages for wood (μw), leaves (μf) and root (μr). This process was repeated ten times, resulting in a mean relative uncertainty of$${bar{delta }}_{{rm{allometric}}}=19.5 % .$$Total carbon uncertaintyWe combine the uncertainties from the neural net mapping and our allometric equations, which can be viewed as considering (1 + A)·(1 + B) with A and B being random variables with standard deviations δarea and δallometric. Neglecting higher-order and interaction terms, we combine the two sources of uncertainty to (delta simeq sqrt{{delta }_{{rm{area}}}^{2}+{bar{delta }}_{{rm{allometric}}}^{2}}), resulting in an uncertainty in total tree carbon for our study of ±19.8%. See also Extended Data Fig. 9 for the RMSEs of our predicted crown areas calculated on external validation data from ref. 1, binned on the basis of the 50th quantiles of the hand-labelled crown areas and converted also into carbon. Extended Data Fig. 10 is a flow diagram summarizing our methods.Our viewerVisualizing our large tree-mapping dataset in an interactive format was essential for quality-control purposes, exploration of the data and hypothesis creation. Creating a web-based viewer serves the purpose of being the initial point of interaction with our dataset for fellow researchers, local stakeholders or the general public. The visualization of more than 10 billion trees in a web browser required maintaining performance, interactivity and individual metadata for each polygon. Users should be able to zoom in to any area within the dataset to view individual tree polygons and query their statistics while at the same time accurately depicting the overall trends of the dataset at lower zoom levels. The visualization also needed to clearly denote where data were missing or possibly affected by clouds or aerosols. Finally, the extent and origin of the source imagery, its acquisition date and a preview of the imagery needed to be available. To accomplish these goals, a vector-tile-based approach was taken, with the data visualized in a Mapbox GL JS map within a React web application. To create vector tiles covering the entire study area, we developed a data-processing pipeline using high-performance computing resources to transform the data into compatible formats, as well as to package, optimize and combine the vector tiles themselves.We used two tracks to store and visualize the results of this study on the web: vector polygon data and generalized rasters representing tree crown density. At the native spatial resolution of 50 cm, the map shows the full-resolution tree polygon dataset. At lower-spatial-resolution zoom levels, rasterized representations of tree density are shown. Visualizing generalized rasters in place of vector polygons improves performance substantially. As users zoom in to higher spatial resolutions, the raster layer fades away and is replaced by the full-resolution polygon layer. Once zoomed far enough to resolve individual polygons, users can click to select a polygon to show a map overlay containing various properties of the tree, as well as the date on which the source imagery was acquired and a link to preview the source imagery.Rainfall dataWe used the rainfall data of Funk et al. to estimate annual rainfall at 5.6-m grids46. We averaged the available data from 1982 to 2017 and extracted the mean annual rainfall for each mapped tree and bilinearly interpolated it to 100 × 100-m resolution. The rainfall data were also used to classify the study area into mean annual precipitation zones: hyper-arid from 0–150 mm year−1, arid from 150–300 mm year−1, semi-arid from 300–600 mm year−1 and sub-humid from 600–1,000 mm year−1 zones. The rainfall data are found at https://data.chc.ucsb.edu/products/CHIRPS-2.0/africa_monthly/ (ref. 46). More

Balashov, S. P. et al. Xanthorhodopsin: a proton pump with a light-harvesting carotenoid antenna. Science 309, 2061–2064 (2005).Article 
ADS 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Imasheva, E. S., Balashov, S. P., Choi, A. R., Jung, K.-H. & Lanyi, J. K. Reconstitution of Gloeobacter violaceus rhodopsin with a light-harvesting carotenoid antenna. Biochemistry 48, 10948–10955 (2009).Article 
CAS 
PubMed 

Google Scholar 
Fuhrman, J. A., Schwalbach, M. S. & Stingl, U. Proteorhodopsins: an array of physiological roles? Nat. Rev. Microbiol. 6, 488–494 (2008).Article 
CAS 
PubMed 

Google Scholar 
Vollmers, J. et al. Poles apart: Arctic and Antarctic Octadecabacter strains share high genome plasticity and a new type of xanthorhodopsin. PLoS ONE 8, e63422 (2013).Article 
ADS 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Bertsova, Y. V., Arutyunyan, A. M. & Bogachev, A. V. Na+-translocating rhodopsin from Dokdonia sp. PRO95 does not contain carotenoid antenna. Biochem. Mosc. 81, 414–419 (2016).Article 
CAS 

Google Scholar 
Misra, R., Eliash, T., Sudo, Y. & Sheves, M. Retinal–salinixanthin interactions in a thermophilic rhodopsin. J. Phys. Chem. B 123, 10–20 (2019).Article 
CAS 
PubMed 

Google Scholar 
Béjà, O. et al. Bacterial rhodopsin: evidence for a new type of phototrophy in the sea. Science 289, 1902–1906 (2000).Article 
ADS 
PubMed 

Google Scholar 
Béjà, O., Spudich, E. N., Spudich, J. L., Leclerc, M. & DeLong, E. F. Proteorhodopsin phototrophy in the ocean. Nature 411, 786–789 (2001).Article 
ADS 
PubMed 

Google Scholar 
Atamna-Ismaeel, N. et al. Widespread distribution of proteorhodopsins in freshwater and brackish ecosystems. ISME J. 2, 656–662 (2008).Article 
CAS 
PubMed 

Google Scholar 
Frigaard, N.-U., Martinez, A., Mincer, T. J. & DeLong, E. F. Proteorhodopsin lateral gene transfer between marine planktonic Bacteria and Archaea. Nature 439, 847–850 (2006).Article 
ADS 
CAS 
PubMed 

Google Scholar 
Finkel, O. M., Béjà, O. & Belkin, S. Global abundance of microbial rhodopsins. ISME J. 7, 448–451 (2013).Article 
CAS 
PubMed 

Google Scholar 
Gómez-Consarnau, L. et al. Microbial rhodopsins are major contributors to the solar energy captured in the sea. Sci. Adv. 5, eaaw8855 (2019).Article 
ADS 
PubMed 
PubMed Central 

Google Scholar 
DeLong, E. F. & Béjà, O. The light-driven proton pump proteorhodopsin enhances bacterial survival during tough times. PLoS Biol. 8, e1000359 (2010).Article 
PubMed 
PubMed Central 

Google Scholar 
Munson-McGee, J. H. et al. Decoupling of respiration rates and abundance in marine prokaryoplankton. Nature 612, 764–770 (2022).Article 
ADS 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Wang, W.-W., Sineshchekov, O. A., Spudich, E. N. & Spudich, J. L. Spectroscopic and photochemical characterization of a deep ocean proteorhodopsin. J. Biol. Chem. 278, 33985–33991 (2003).Article 
CAS 
PubMed 

Google Scholar 
Man, D. Diversification and spectral tuning in marine proteorhodopsins. EMBO J. 22, 1725–1731 (2003).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Lanyi, J. K. & Balashov, S. P. in Halophiles and Hypersaline Environments (eds. Ventosa, A., Oren, A. & Ma, Y.) 319–340 (Springer, 2011).Balashov, S. P. et al. Reconstitution of Gloeobacter rhodopsin with echinenone: role of the 4-keto group. Biochemistry 49, 9792–9799 (2010).Article 
CAS 
PubMed 

Google Scholar 
Kopejtka, K. et al. A bacterium from a mountain lake harvests light using both proton-pumping xanthorhodopsins and bacteriochlorophyll-based photosystems. Proc. Natl Acad. Sci. USA 119, e2211018119 (2022).Article 
CAS 
PubMed 

Google Scholar 
Pushkarev, A. & Béjà, O. Functional metagenomic screen reveals new and diverse microbial rhodopsins. ISME J. 10, 2331–2335 (2016).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Pushkarev, A. et al. A distinct abundant group of microbial rhodopsins discovered using functional metagenomics. Nature 558, 595–599 (2018).Article 
ADS 
CAS 
PubMed 

Google Scholar 
Chazan, A. et al. Diverse heliorhodopsins detected via functional metagenomics in freshwater Actinobacteria, Chloroflexi and Archaea. Environ. Microbiol. 24, 110–121 (2022).Article 
CAS 
PubMed 

Google Scholar 
Inoue, K. et al. A light-driven sodium ion pump in marine bacteria. Nat. Commun. 4, 1678 (2013).Article 
ADS 
PubMed 

Google Scholar 
Bhosale, P. & Bernstein, P. S. Microbial xanthophylls. Appl. Microbiol. Biotechnol. 68, 445–455 (2005).Article 
CAS 
PubMed 

Google Scholar 
Demmig-Adams, B., Polutchko, S. K. & Adams, W. W. Structure–function–environment relationship of the isomers zeaxanthin and lutein. Photochem 2, 308–325 (2022).Article 

Google Scholar 
Barreiro C. & Barredo J. L. Microbial Carotenoids: Methods and Protocols (Humana Press, 2018).Ram, S., Mitra, M., Shah, F., Tirkey, S. R. & Mishra, S. Bacteria as an alternate biofactory for carotenoid production: a review of its applications, opportunities and challenges. J. Funct. Foods 67, 103867 (2020).Article 
CAS 

Google Scholar 
Shibata, M. et al. Oligomeric states of microbial rhodopsins determined by high-speed atomic force microscopy and circular dichroic spectroscopy. Sci. Rep. 8, 8262 (2018).Article 
ADS 
PubMed 
PubMed Central 

Google Scholar 
Luecke, H. et al. Crystallographic structure of xanthorhodopsin, the light-driven proton pump with a dual chromophore. Proc. Natl Acad. Sci. USA 105, 16561–16565 (2008).Article 
ADS 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Chuon, K. et al. Assembly of natively synthesized dual chromophores into functional actinorhodopsin. Front. Microbiol. 12, 652328 (2021).Article 
PubMed 
PubMed Central 

Google Scholar 
Yoshizawa, S., Kawanabe, A., Ito, H., Kandori, H. & Kogure, K. Diversity and functional analysis of proteorhodopsin in marine Flavobacteria. Environ. Microbiol. 14, 1240–1248 (2012).Article 
CAS 
PubMed 

Google Scholar 
Ahmed, F. et al. Profiling of carotenoids and antioxidant capacity of microalgae from subtropical coastal and brackish waters. Food Chem. 165, 300–306 (2014).Article 
CAS 
PubMed 

Google Scholar 
Shihoya, W. et al. Crystal structure of heliorhodopsin. Nature 574, 132–136 (2019).Article 
ADS 
CAS 
PubMed 

Google Scholar 
Kishi, K. E. et al. Structural basis for channel conduction in the pump-like channelrhodopsin ChRmine. Cell 185, 672–689.e23 (2022).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Balashov, S. P., Imasheva, E. S., Wang, J. M. & Lanyi, J. K. Excitation energy-transfer and the relative orientation of retinal and carotenoid in xanthorhodopsin. Biophys. J. 95, 2402–2414 (2008).Article 
ADS 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Lakowicz, J. R. (ed.) in Principles of Fluorescence Spectroscopy 27–61 (Springer, 2006).Dana, J. et al. Testing the fate of nascent holes in CdSe nanocrystals with sub-10 fs pump–probe spectroscopy. Nanoscale 13, 1982–1987 (2021).Article 
CAS 
PubMed 

Google Scholar 
Polívka, T. et al. Femtosecond carotenoid to retinal energy transfer in xanthorhodopsin. Biophys. J. 96, 2268–2277 (2009).Article 
ADS 
PubMed 
PubMed Central 

Google Scholar 
Iyer, E. S. S., Gdor, I., Eliash, T., Sheves, M. & Ruhman, S. Efficient femtosecond energy transfer from carotenoid to retinal in Gloeobacter rhodopsin–salinixanthin complex. J. Phys. Chem. B 119, 2345–2349 (2015).Article 
CAS 
PubMed 

Google Scholar 
Doi, S., Tsukamoto, T., Yoshizawa, S. & Sudo, Y. An inhibitory role of Arg-84 in anion channelrhodopsin-2 expressed in Escherichia coli. Sci. Rep. 7, 41879 (2017).Article 
ADS 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Nagiri, C. et al. Crystal structure of human endothelin ETB receptor in complex with peptide inverse agonist IRL2500. Commun. Biol. 2, 236 (2019).Article 
PubMed 
PubMed Central 

Google Scholar 
Yamashita, K., Hirata, K. & Yamamoto, M. KAMO: towards automated data processing for microcrystals. Acta Crystallogr. D Struct. Biol. 74, 441–449 (2018).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Kabsch, W. XDS. Acta Crystallogr. D Biol. Crystallogr. 66, 125–132 (2010).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
McCoy, A. J. et al. Phaser crystallographic software. J. Appl. Crystallogr. 40, 658–674 (2007).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021).Article 
ADS 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of Coot. Acta Crystallogr. D Biol. Crystallogr. 66, 486–501 (2010).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Afonine, P. V. et al. Towards automated crystallographic structure refinement with phenix.refine.Acta Crystallogr. D Biol. Crystallogr. 68, 352–367 (2012).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Zivanov, J. et al. New tools for automated high-resolution cryo-EM structure determination in RELION-3.eLife 7, e42166 (2018).Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nat. Methods 14, 290–296 (2017).Punjani, A., Zhang, H. & Fleet, D. J. Non-uniform refinement: adaptive regularization improves single-particle cryo-EM reconstruction. Nat. Methods 17, 1214–1221 (2020).Article 
CAS 
PubMed 

Google Scholar 
Rosenthal, P. B. & Henderson, R. Optimal determination of particle orientation, absolute hand, and contrast loss in single-particle electron cryomicroscopy. J. Mol. Biol. 333, 721–745 (2003).Article 
CAS 
PubMed 

Google Scholar 
Emsley, P. & Cowtan, K. Coot: model-building tools for molecular graphics. Acta Crystallogr. D Biol. Crystallogr. 60, 2126–2132 (2004).Article 
PubMed 

Google Scholar 
Adams, P. D. et al. PHENIX: a comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr. D Biol. Crystallogr. 66, 213–221 (2010).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Yamashita, K., Palmer, C. M., Burnley, T. & Murshudov, G. N. Cryo-EM single-particle structure refinement and map calculation using Servalcat. Acta Crystallogr. D Struct. Biol. 77, 1282–1291 (2021).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Edgar, R. C. Search and clustering orders of magnitude faster than BLAST. Bioinformatics 26, 2460–2461 (2010).Article 
CAS 
PubMed 

Google Scholar 
Inoue, K. et al. Exploration of natural red-shifted rhodopsins using a machine learning-based Bayesian experimental design. Commun. Biol. 4, 362 (2021).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Salazar, G. et al. Gene expression changes and community turnover differentially shape the global ocean metatranscriptome. Cell 179, 1068–1083.e21 (2019).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Chen, I.-M. A. et al. The IMG/M data management and analysis system v.6.0: new tools and advanced capabilities. Nucleic Acids Res. 49, D751–D763 (2021).Article 
CAS 
PubMed 

Google Scholar 
Nayfach, S. et al. A genomic catalog of Earth’s microbiomes. Nat. Biotechnol. 39, 499–509 (2021).Article 
CAS 
PubMed 

Google Scholar 
Sunagawa, S. et al. Metagenomic species profiling using universal phylogenetic marker genes. Nat. Methods 10, 1196–1199 (2013).Article 
CAS 
PubMed 

Google Scholar 
Wickham, H. in ggplot2 (eds Gentleman, R., Hornik, K. & Parmigiani, G.) 189–201 (Springer, 2016).Katoh, K., Misawa, K., Kuma, K. & Miyata, T. MAFFT: a novel method for rapid multiple sequence alignment based on fast Fourier transform. Nucleic Acids Res. 30, 3059–3066 (2002).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Capella-Gutiérrez, S., Silla-Martínez, J. M. & Gabaldón, T. trimAl: a tool for automated alignment trimming in large-scale phylogenetic analyses. Bioinformatics 25, 1972–1973 (2009).Article 
PubMed 
PubMed Central 

Google Scholar 
Nguyen, L.-T., Schmidt, H. A., von Haeseler, A. & Minh, B. Q. IQ-TREE: a fast and effective stochastic algorithm for estimating maximum-likelihood phylogenies. Mol. Biol. Evol. 32, 268–274 (2015).Article 
CAS 
PubMed 

Google Scholar 
Hoang, D. T., Chernomor, O., von Haeseler, A., Minh, B. Q. & Vinh, L. S. UFBoot2: improving the ultrafast bootstrap approximation. Mol. Biol. Evol. 35, 518–522 (2018).Article 
CAS 
PubMed 

Google Scholar  More

Lawrence, D. & Vandecar, K. Effects of tropical deforestation on climate and agriculture. Nat. Clim. Change 5, 27–36 (2015).Article 
ADS 

Google Scholar 
Spracklen, D. V., Baker, J. C. A., Garcia-Carreras, L. & Marsham, J. H. The effects of tropical vegetation on rainfall. Annu. Rev. Environ. Resour. 43, 193–218 (2018).Article 

Google Scholar 
Bonan, G. B. Forests and climate change: forcings, feedbacks, and the climate benefits of forests. Science 320, 1444–1449 (2008).Article 
ADS 
CAS 
PubMed 

Google Scholar 
Spracklen, D. V., Arnold, S. R. & Taylor, C. M. Observations of increased tropical rainfall preceded by air passage over forests. Nature 489, 282–285 (2012).Article 
ADS 
CAS 
PubMed 

Google Scholar 
Staal, A. et al. Forest-rainfall cascades buffer against drought across the Amazon. Nat. Clim. Change 8, 539–543 (2018).Article 
ADS 

Google Scholar 
Baker, J. C. A. & Spracklen, D. V. Divergent representation of precipitation recycling in the Amazon and the Congo in CMIP6 models. Geophys. Res. Lett. 49, e2021GL095136 (2022).Article 
ADS 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Guan, K. et al. Photosynthetic seasonality of global tropical forests constrained by hydroclimate. Nat. Geosci. 8, 284–289 (2015).Article 
ADS 
CAS 

Google Scholar 
Staal, A. et al. Hysteresis of tropical forests in the 21st century. Nat. Commun. 11, 4978 (2020).Article 
ADS 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Zemp, D. C. et al. Self-amplified Amazon forest loss due to vegetation-atmosphere feedbacks. Nat. Commun. 8, 14681 (2017).Article 
ADS 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Hansen, M. C. et al. High-resolution global maps of 21st-century forest cover change. Science 342, 850–854 (2013).Article 
ADS 
CAS 
PubMed 

Google Scholar 
Chagnon, F. J. F. & Bras, R. L. Contemporary climate change in the Amazon. Geophys. Res. Lett. 32, L13703 (2005).Article 
ADS 

Google Scholar 
Khanna, J., Medvigy, D., Fueglistaler, S. & Walko, R. Regional dry-season climate changes due to three decades of Amazonian deforestation. Nat. Clim. Change 7, 200–204 (2017).Article 
ADS 

Google Scholar 
Garcia-Carreras, L. & Parker, D. J. How does local tropical deforestation affect rainfall? Geophys. Res. Lett. 38, L19802 (2011).Article 
ADS 

Google Scholar 
Leite-Filho, A. T., Soares-Filho, B. S., Davis, J. L., Abrahão, G. M. & Börner, J. Deforestation reduces rainfall and agricultural revenues in the Brazilian Amazon. Nat. Commun. 12, 2591 (2021).Article 
ADS 
CAS 
PubMed 
PubMed Central 

Google Scholar 
McAlpine, C. A. et al. Forest loss and Borneo’s climate. Environ. Res. Lett. 13, 044009 (2018).Chapman, S. et al. Compounding impact of deforestation on Borneo’s climate during El Niño events. Environ. Res. Lett. 15, 084006 (2020).Spracklen, D. V. & Garcia-Carreras, L. The impact of Amazonian deforestation on Amazon basin rainfall. Geophys. Res. Lett. 42, 9546–9552 (2015).Article 
ADS 

Google Scholar 
Jiang, Y. et al. Modeled response of South American climate to three decades of deforestation. J. Clim. 34, 2189–2203 (2021).Article 
ADS 

Google Scholar 
Harris, I., Osborn, T. J., Jones, P. & Lister, D. Version 4 of the CRU TS monthly high-resolution gridded multivariate climate dataset. Sci. Data 7, 109 (2020).Article 
PubMed 
PubMed Central 

Google Scholar 
Fassoni-Andrade, A. C. et al. Amazon hydrology from space: scientific advances and future challenges. Rev. Geophys. 59, e2020RG000728 (2021).Article 
ADS 

Google Scholar 
Haiden, T., Janousek, M., Vitart, F., Ferranti, L. & Prates, F. Evaluation of ECMWF Forecasts, Including the 2019 Upgrade. ECMWF Technical Memorandum No. 853 (ECMWF, 2019).Esquivel-Muelbert, A. et al. Compositional response of Amazon forests to climate change. Glob. Change Biol. 25, 39–56 (2019).Article 
ADS 

Google Scholar 
Brum, M. et al. ENSO effects on the transpiration of eastern Amazon trees. Philos. Trans. R. Soc. B 373, 20180085 (2018).Article 

Google Scholar 
Bagley, J. E., Desai, A. R., Harding, K. J., Snyder, P. K. & Foley, J. A. Drought and deforestation: has land cover change influenced recent precipitation extremes in the Amazon? J. Clim. 27, 345–361 (2014).Article 
ADS 

Google Scholar 
Wunderling, N. et al. Recurrent droughts increase risk of cascading tipping events by outpacing adaptive capacities in the Amazon rainforest. Proc. Natl Acad. Sci. USA 119, e2120777119 (2022).Article 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Fu, R. & Li, W. The influence of the land surface on the transition from dry to wet season in Amazonia. Theor. Appl. Climatol. 78, 97–110 (2004).Article 
ADS 

Google Scholar 
Leite-Filho, A. T., de Sousa Pontes, V. Y. & Costa, M. H. Effects of deforestation on the onset of the rainy season and the duration of dry spells in southern Amazonia. J. Geophys. Res. Atmos. 124, 5268–5281 (2019).Article 
ADS 

Google Scholar 
Negri, A. J., Adler, R. F., Xu, L. & Surratt, J. The Impact of Amazonian deforestation on dry season rainfall. J. Clim. 17, 1306–1319 (2004).Article 
ADS 

Google Scholar 
Chagnon, F. J. F., Bras, R. L. & Wang, J. Climatic shift in patterns of shallow clouds over the Amazon. Geophys. Res. Lett. 31, L24212 (2004).Article 
ADS 

Google Scholar 
Chambers, J. Q. & Artaxo, P. Biosphere–atmosphere interactions: deforestation size influences rainfall. Nat. Clim. Change 7, 175–176 (2017).Article 
ADS 

Google Scholar 
Baudena, M., Tuinenburg, O. A., Ferdinand, P. A. & Staal, A. Effects of land-use change in the Amazon on precipitation are likely underestimated. Glob. Change Biol. 27, 5580–5587 (2021).Article 
CAS 

Google Scholar 
Duku, C. & Hein, L. The impact of deforestation on rainfall in Africa: a data-driven assessment. Environ. Res. Lett. 16, 064044 (2021).Akkermans, T., Thiery, W. & Van Lipzig, N. P. M. The regional climate impact of a realistic future deforestation scenario in the Congo basin. J. Clim. 27, 2714–2734 (2014).Article 
ADS 

Google Scholar 
Staal, A. et al. Feedback between drought and deforestation in the Amazon. Environ. Res. Lett. 15, 044024 (2020).Xu, X. et al. Deforestation triggering irreversible transition in Amazon hydrological cycle. Environ. Res. Lett. 17, 034037 (2022).Kooperman, G. J. et al. Forest response to rising CO2 drives zonally asymmetric rainfall change over tropical land. Nat. Clim. Change 8, 434–440 (2018).Article 
ADS 

Google Scholar 
Chen, Z. et al. Global land monsoon precipitation changes in CMIP6 projections. Geophys. Res. Lett. 47, e2019GL086902 (2020).Stickler, C. M. et al. Dependence of hydropower energy generation on forests in the Amazon Basin at local and regional scales. Proc. Natl Acad. Sci. USA 110, 9601–9606 (2013).Article 
ADS 
CAS 
PubMed 
PubMed Central 

Google Scholar 
Challinor, A. J. et al. A meta-analysis of crop yield under climate change and adaptation. Nat. Clim. Change 4, 287–291 (2014).Article 
ADS 

Google Scholar 
Strand, J. et al. Spatially explicit valuation of the Brazilian Amazon forest’s ecosystem services. Nat. Sustain. 1, 657–664 (2018).Article 

Google Scholar 
Potapov, P. et al. Global maps of cropland extent and change show accelerated cropland expansion in the twenty-first century. Nat. Food 3, 19–28 (2022).Article 

Google Scholar 
Li, Y. et al. Deforestation-induced climate change reduces carbon storage in remaining tropical forests. Nat. Commun. 13, 1964 (2022).Article 
ADS 
PubMed 
PubMed Central 

Google Scholar 
Aragão, L. E. O. C. et al. Interactions between rainfall, deforestation and fires during recent years in the Brazilian Amazonia. Philos. Trans. R. Soc. B 363, 1779–1785 (2008).Article 

Google Scholar 
Marengo, J. A. et al. Changes in climate and land use over the Amazon region: current and future variability and trends. Front. Earth Sci. https://doi.org/10.3389/feart.2018.00228 (2018).Jiang, Y. et al. Widespread increase of boreal summer dry season length over the Congo rainforest. Nat. Clim. Change https://doi.org/10.1038/s41558-019-0512-y (2019).Van Der Ent, R. J. & Savenije, H. H. G. Length and time scales of atmospheric moisture recycling. Atmos. Chem. Phys. 11, 1853–1863 (2011).Article 
ADS 

Google Scholar 
Sorí, R., Nieto, R., Vicente-Serrano, S. M., Drumond, A. & Gimeno, L. A Lagrangian perspective of the hydrological cycle in the Congo River basin. Earth Syst. Dyn. 8, 653–675 (2017).Article 
ADS 

Google Scholar 
van der Ent, R. J., Savenije, H. H. G., Schaefli, B. & Steele-Dunne, S. C. Origin and fate of atmospheric moisture over continents. Water Resour. Res. 46, W09525 (2010).ADS 

Google Scholar 
Feng, Y. et al. Doubling of annual forest carbon loss over the tropics during the early twenty-first century. Nat. Sustain. 4, 441–451 (2022).
Google Scholar 
Tuinenburg, O. A., Bosmans, J. H. C. & Staal, A. The global potential of forest restoration for drought mitigation. Environ. Res. Lett. 17, 034045 (2022).Met Office. Cartopy: a cartographic python library with a Matplotlib interface 2010–2015. Met Office https://scitools.org.uk/cartopy (2022).Hoyer, S. & Hamman, J. xarray: N-D labeled arrays and datasets in Python. J. Open Res. Softw. https://doi.org/10.5334/jors.148 (2017).Zhuang, J. xESMF. Zenodo https://doi.org/10.5281/zenodo.1134365 (2022).Baker, J. C. A. & Spracklen, D. V. Climate benefits of intact Amazon forests and the biophysical consequences of disturbance. Front. For. Glob. Change https://doi.org/10.3389/ffgc.2019.00047 (2019).Schaaf, C. & Wang, Z. MCD43A3 MODIS/Terra+Aqua BRDF/Albedo Daily L3 Global – 500m V006. NASA EOSDIS Land Processes DAAC https://doi.org/10.5067/modis/mcd43a3.006 (2015).Waskom, M. Seaborn: statistical data visualization. J. Open Source Softw. 6, 3021 (2021).Article 
ADS 

Google Scholar 
Chen, M. et al. Global land use for 2015–2100 at 0.05° resolution under diverse socioeconomic and climate scenarios. Sci. Data 7, 320 (2020).Article 
PubMed 
PubMed Central 

Google Scholar 
Funk, C. et al. The climate hazards infrared precipitation with stations—a new environmental record for monitoring extremes. Sci. Data 2, 150066 (2015).Article 
PubMed 
PubMed Central 

Google Scholar 
Xie, P. et al. NOAA Climate Data Record (CDR) of CPC Morphing technique (CMORPH) high resolution global precipitation estimates, version 1. NOAA National Centers for Environmental Information https://doi.org/10.25921/w9va-q159 (2019).Xie, P. et al. A gauge-based analysis of daily precipitation over East Asia. J. Hydrometeorol. 8, 607–626 (2007).Article 
ADS 

Google Scholar 
Hersbach, H. et al. The ERA5 global reanalysis. Q. J. R. Meteorol. Soc. 146, 1999–2049 (2020).Article 
ADS 

Google Scholar 
Elke, R., Hänsel, S., Finger, P., Schneider, U. & Ziese, M. GPCC Climatology Version 2022 at 0.25°: monthly land-surface precipitation climatology for every month and the total year from rain-gauges built on GTS-based and historical data. GPCC https://doi.org/10.5676/DWD_GPCC/CLIM_M_V2022_025 (2022).Huffman, G. J. A., Behrangi, R. F., Adler, D. T., Bolvin, E. J. & Nelkin, G. G. Introduction to the new version 3 GPCP monthly global precipitation analysis. GPCP https://docserver.gesdisc.eosdis.nasa.gov/public/project/MEaSUREs/GPCP/Release_Notes.GPCPV3.2.pdf (2022).Hou, A. Y. et al. The global precipitation measurement mission. Bull. Am. Meteorol. Soc. 95, 701–722 (2014).Article 
ADS 

Google Scholar 
Kobayashi, S. et al. The JRA-55 reanalysis: general specifications and basic characteristics. J. Meteorol. Soc. Japan 93, 5–48 (2015).Article 
ADS 

Google Scholar 
Gelaro, R. et al. The modern-era retrospective analysis for research and applications, version 2 (MERRA-2). J. Clim. 30, 5419–5454 (2017).Article 
ADS 

Google Scholar 
Chen, M., Xie, P. & Janowiak, J. E. Global land precipitation: a 50-yr monthly analysis based on gauge observations. J. Hydrometeorol. 3, 249–266 (2002).Article 
ADS 

Google Scholar 
Nguyen, P. et al. The CHRS data portal, an easily accessible public repository for PERSIANN global satellite precipitation data. Sci. Data 6, 1180296 (2019).Article 

Google Scholar 
Ashouri, H. et al. PERSIANN-CDR: daily precipitation climate data record from multisatellite observations for hydrological and climate studies. Bull. Am. Meteorol. Soc. 96, 69–83 (2015).Article 
ADS 

Google Scholar 
Nguyen, P. et al. Persiann dynamic infrared–rain rate (PDIR-now): a near-real-time, quasi-global satellite precipitation dataset. J. Hydrometeorol. 21, 2893–2906 (2020).Article 
ADS 
PubMed 
PubMed Central 

Google Scholar 
Sadeghi, M. et al. PERSIANN-CCS-CDR, a 3-hourly 0.04° global precipitation climate data record for heavy precipitation studies. Sci. Data 8, 157 (2021).Article 
PubMed 
PubMed Central 

Google Scholar 
Huffman, G. J. et al. The TRMM Multisatellite Precipitation Analysis (TMPA): quasi-global, multiyear, combined-sensor precipitation estimates at fine scales. J. Hydrometeorol. 8, 38–55 (2007).Article 
ADS 

Google Scholar 
Matsuura, K. & Willmott, C. J. Terrestrial precipitation: 1900-2017 gridded monthly time series. Global Precipitation Archive http://climate.geog.udel.edu/~climate/html_pages/Global2017/README.GlobalTsP2017.html (2018). More