Ecology

Study area and soil propertiesA field experiment was established in May 2012 at Shenyang Agro-Ecological Station (41°31′N, 123°22′E) of the Institute of Applied Ecology, Chinese Academy of Sciences, Northeast China. This region has a warm-temperate continental monsoon climate. The mean annual air temperature and annual precipitation are 7.5 °C and 680 mm, respectively. The soil is classified as Luvisol (FAO classification). The soil properties of the topsoil layer (0–20 cm) at the start of the experiment are as follows: SOC = 9.0 g kg−1, available NH4+–N = 1.18 mg kg−1; available NO3−–N = 9.04 mg kg−1; Olsen-P = 38.50 mg kg−1, available K = 97.90 mg kg−1, bulk density = 1.25 g cm−3, and pH = 5.8. The determination method of soil was shown in “Soil analysis” section.Field experimentThree treatments were established in this experiment: (1) mineral fertilizers (NPK); (2) pig manure incorporation at a local conventional AM application rate of 15 Mg ha−1 yr−1 (NPKM, 126 kg N ha−1 on dry weight); and (3) NPKM plus DMPP (3,4-Dimethylpyrazole phosphate) incorporation at a rate of 0.5% of applied urea (2.39 kg ha−1, 220 kg N/the N content of urea (0.46) × 0.5%) (NPKI + M). The treatments were applied following a randomized design across three replicate field plots (4 m × 5 m). Plots of different treatments remained unchanged in the same locations for 4 years. Each year, the composted pig manure (213 g C kg−1 and 22 g N kg−1 based on dry weight on average, characteristics of pig manure was listed in Table S1) was broadcasted evenly onto the plots a few days before maize planting, and ploughed to a depth of 20 cm by machine (TG4, Huaxing, China). For the respective treatments, urea (220 kg N ha−1 yr−1), calcium superphosphate (110 kg P2O5 ha−1 yr−1), and potassium chloride (110 kg K2O ha−1 yr−1) were applied on the same day as maize (Zea mays L.) was planted. The urea and inhibitor were fully mixed before application.Maize (cultivar was Fuyou #9) was planted on 3rd May 2012, 3rd May 2013, 6th May 2014, and 10th May 2015, at a spacing of 37 cm and 60 cm between rows. No irrigation was applied throughout the experimental period. Maize was harvested on 13th September 2012, 29th September 2013, 29th September 2014, and 29th September 2015, respectively. At harvest, maize yield and aboveground biomass yield were measured by harvesting all plants (20 m2) in each plot. The straw and grain were removed after each harvest and the soil with about 5 cm maize stem was ploughed to a depth of approximately 20 cm in April each year.Each cropping cycle, therefore, consisted of periods of maize (from May to September) and fallow (from October to April) of the following year.The precipitation and air temperature data were acquired from the meteorological station of the Shenyang Agro-Ecological Station. The precipitation during the 2012/2013, 2013/2014, 2014/2015, and 2015/2016 periods were 911.9 mm, 621.7 mm, 485.7 mm, and 585.3 mm, respectively (Fig. 1). 72.3%, 75.5%, 66.5%, and 73.0% of these annual precipitations occurred during maize-growing period, respectively. The mean annual air temperatures in these years were 7.7 °C (− 21.2 to 27.5 °C), 8.1 °C (− 22.7 to 28.3 °C), 9.5 °C (− 21.7 to 28.2 °C) and 9.3 °C (− 17.1 to 27.0 °C), respectively. The soil temperature at a depth of 5 cm varied between − 14 and 35 °C during the four-year period (Fig. 2b). The change trend of soil surface temperature was the same as that of soil temperature at 5 cm depth (Fig. 2a). The mean soil WFPS (0–15 cm) varied between 15 and 73% (Fig. 2c).Figure 1Precipitation and daily mean air temperature during four annual cycles from May 2012 to April 2016 in the experimental field.Full size imageFigure 2Seasonal variations in soil temperature (at soil surface and 5 cm soil depth) and WFPS% at 0–15 cm depth from May 2012 to April 2016.Full size imageGas sampling and analysisThe gas was sampled between 3rd May 2012 and 14th April 2016 using a static closed chamber system as described by Dong et al.16. Briefly, a stainless-steel chamber base (56 cm length × 28 cm width) was inserted into the soil of each plot to a depth of approximately 10 cm, with its long edge perpendicular to the rows of maize. The top chamber (56 cm length × 28 cm width × 20 cm height) was also made of stainless steel. Gas samples were obtained using a syringe 0, 20, and 40 min after the chambers had been closed between 9:00 am and 11:00 am on each sampling day. Gas samples were collected every 2‒6 days and every 7‒15 days during the growing seasons and non-growing seasons, respectively. The first gas sampling time was on day 1, day 3, day 1, and day 3 after maize planting each year. The N2O concentrations in gas samples were quantified using a gas chromatograph (Agilent 7890A, Shanghai, China) with an electron capture detector.Soil analysisThe soil temperature and volumetric water content (SVWC) were measured at depth of 0–15 cm using a bent stem thermometer and a time-domain reflectometry (Zhongtian Devices Co. Ltd, China), respectively. SVWC was converted to soil water-filled pore space (WFPS) using the following equation:$${text{WFPS}} = {text{SVWC}}/(1{-}{text{BD}}/{text{particle}},{text{density}}),$$
(1)
where BD is soil bulk density (g cm−3). Particle density was assumed to be 2.65 g cm−3.Soil samples from the 0–20 cm layer were collected in each plot in April 2012 (before sowing) and October 2015 (maize harvest) using a 5 cm diameter stainless steel soil sampler. The five soil samples collected from different locations in each plot were mixed thoroughly. Visible roots were removed by hand and the samples were air-dried and sieved using a 0.15 mm sieve. SOC was then quantified using an elemental analyzer (Vario EL III, Elementar, Germany). Soil available NH4+–N and NO3−–N were extracted with 2 M KCl and measured colorimetrically using a continuous flow injection analyzer (Futura, Alliance, France)17. Soil Olsen-P was extracted with NaHCO3 and colorimetrically measured using a spectrophotometer (Lambda 2, PerkinElmer, USA). Soil available K was extracted by 1 M CH3COONH4 and analyzed with a flame photometer (FP640, Jingmi, China). Soil pH was determined with deionized water (1:2.5) and analyzed using a pH meter (PHS-3C, LeiCi, China) with a glass electrode.DNA extraction and real-time quantitative PCRThe soil samples for measuring the abundance of nitrification and denitrification functional genes were collected on May 20, 2015. Soil DNA was extracted with the soil DNA extracted kits (EZNA soil DNA Kit; Omega Bio-Tek Inc., U.S.A.). The copy numbers of nitrification and denitrification functional genes were determined by q-PCR with the Roche LightCyler® 96 (Roche, Switzerland). Additional details about the primers and amplification procedure can be found in Dong et al.16.Data analysisThe N2O flux (μg N2O–N m−2 h−1) is calculated based on the increase of N2O concentration per unit chamber area for a specific time interval18 as follows:$${text{F}} = 273/left( {273 + {text{T}}} right) times {text{M}}/22.4 times {text{H}} times {text{dc}}/{text{dt}} times 1000$$
(2)
where F (μg N2O–N m−2 h−1) is the N2O flux, T (◦C) is the air temperature in the chamber, M (g N2O–N mol−1) is the molecular weight of N2O–N, 22.4 (L mol−1) is the molecular volume of the gas at 101.325 kPa and 273 K, H (m) is the chamber height, dc/dt (ppb h−1) is the rate of change in the N2O concentration in the chamber.Cumulative N2O emissions were calculated as follows:$${text{Cumulative}},{text{emission}} = mathop sum limits_{{{text{i}} = 1}}^{{text{n}}} frac{{({text{F}}_{{text{i}}} + {text{F}}_{i + 1} )}}{2} times ({text{t}}_{{{text{i}} + 1}} – {text{t}}_{{text{i}}} ) times 24$$
(3)
where F is the N2O emission flux (μg N2O–N m−2 h−1), i is the ith measurement, (ti+1 − ti) is the number of days between two adjacent measurements, and n is the total number of the measurements. Annual N2O emissions were calculated between the fertilization dates of each successive year.The SOC stock (Mg ha−1) in the topsoil was calculated as:$${text{C}}_{{{text{stock}}}} = {text{SOC}} times {text{BD}} times {text{D}} times 10,$$
(4)
where BD is soil bulk density (g cm−3), D is the depth of the topsoil (0.2 m).The topsoil SOC sequestration rate (SOCSR) (Mg ha−1 yr−1) was estimated using the following equation:$${text{SOCSR}} = left( {{text{C}}_{{{text{stock2015}}}} – {text{C}}_{{{text{stock2012}}}} } right) times {text{t}}^{ – 1} ,$$
(5)
where Cstock2015 and Cstock2012 are the SOC stocks in 2015 and 2012, respectively, and t is the duration of the experiment (years).Statistical analyses were performed using SPSS 13.0 (SPSS, Chicago, USA). The differences in cumulative N2O emissions and maize yields within a year, and other factors among treatments were assessed using one-way Analysis of Variance (ANOVA) with least significant difference post-hoc tests and a 95% confidence limit. The effects of different treatments, years, and their interactions on N2O emission, maize yield and aboveground biomass were examined using one-way repeated measures ANOVA. Pearson correlation analysis was used to analyze the relationships between cumulative N2O emissions and precipitation (N = 12 (three data each year, four years)), as well as N2O flux and soil available nitrogen content.
Statements of research involving plantsIt is stated that the current research on the plants comply with the relevant institutional, national, and international guidelines and legislation. It is also stated that the appropriate permissions have been taken wherever necessary, for collection of plant or seed specimens. It is also stated that the authors comply with the ‘IUCN Policy Statement on Research Involving Species at Risk of Extinction’ and the ‘Convention on the Trade in Endangered Species of Wild Fauna and Flora’. More

Data sources and pre-processingEach of the predictor variables used in our analysis (Table 1), as well as the dependent variable (fire hotspots) underwent pre-processing to transform the data into a format suitable to be passed to our CNN model for prediction. Here we briefly outline these processes and describe the method of generating a training and validation data set for model development. For further details about each predictor variable pre-processing, see Horton et al. (2021).Table 1 Model input data sources, citation, original resolution, and date ranges.Full size tableFire hotspotsWe used both Moderate Resolution Imaging Spectroradiometer (MODIS) and Visible Infrared Imaging Radiometer Suite (VIIRS) fire hotspot data as the dependent variable for use in our model development. As fire hotspots do not give precise locations, but rather indicate that a fire hotspot occurred within a grid cell of the size of the dataset (MODIS 1 km, VIIRS 375 m), we represented each fire hotspot as a 500 m buffered area around the centre point of each grid square identified. We used all fire hotspot occurrences with a confidence rating >50%.LandcoverWe use a collection of historic land cover maps generated by the Ministry of Forestry Indonesia from 1996 to 2016 at 2–3 year intervals38. Before use, we re-designated the land cover map classifications to reduce the number from 25 to just 8 (supplementary Table S2), which are ‘Primary and secondary dry forest’, ‘Swamp forest, ‘Swamp scrubland’, ‘Scrubland, Transition, and bare land’, ‘Riceland’, ‘Plantation’, ‘Settlements’, ‘water, and Cloud’.In addition to these 8 land cover classifications, we also derived a forest clearance index, which identifies areas cleared of forest and assigns an index value that is large negative (−10) immediately after clearing and degrades back towards 0 as time since clearing increases yearly. Areas that are re-forested are assigned large positive values (10) that degrade towards 0 yearly as time since afforestation increase25.Vegetation indicesAll vegetation indices were taken as pre-fire season 3-month averages from May to July. In addition to the original MODIS ET, PET, NDVI, and EVI products, we also included ‘normalised’ variables, whereby each vegetation index was expressed as the ratio of the same index taken at a reference site. The reference site was an area of dense primary forest outside of the EMRP area.Proximity to anthropogenic factorsThe distance to roads and settlement rasters were derived from OpenStreetMap data as the Euclidean distance to nearest feature in 250 m resolution. The same was done for all water bodies, which were then classified by hand into either canals or rivers. These features are taken as those shown in 2015 for all years, and therefore may misrepresent earlier years. However, the majority of canal development in the region took place between 1996 and 1998 and so should not differ dramatically from this date onwards.Oceanic Niño Index (ONI)We use a single value for the entire study area taken as the three-month average for the early fire season each year (July–September).Number of cloud daysUsing the state_1km band in the daily MODIS terra product (MOD09GA version 6), which classifies each pixel as either ‘no cloud’, ‘cloud’, ‘mixed’, or ‘unknown’, we counted the number of ‘cloud’ or ‘mixed’ designations for each pixel for the pre-fire season period May–July.Cross year normalisationAll predictor variables are normalised to be represented between 0 and 1 as the range between the minimum and maximum values for each variable that occur across all years, such that:$${V}_{{{{{{rm{norm}}}}}}}=frac{V-{V}_{{min }}}{{V}_{{max }}-{V}_{{min }}}$$where ({V}_{{{{{{rm{norm}}}}}}}) is the normalised version of the predictor variable (V), ({V}_{{max }}) is the maximum value within the training dataset across all years (2002–2019), and ({V}_{{min }}) is the minimum value within the training dataset across all years.Training and validation dataset assemblyOnce pre-processed, all predictor variable rasters were resampled to the same dimensions (with a resolution of 0.002 degrees in the WGS84 co-ordinate system) and stacked yearly, so that each year (2002–2019) comprised of a 31 feature maps input as a raster stack, with each feature map representing a different predictor variable. Each yearly stack was then split into tiles matching the input dimensions of the CNN model. Our final model was built to take an input size of 32 × 32 pixels (raster cells). Therefore, each yearly raster stack was split into many 32 × 32 × 31 raster stack tiles that span the defined study area. These were then converted to 3D arrays holding the values of all predictor variables for each raster stack tile.The same process was repeated for the yearly fire hotspot rasters used as the dependent variable in building our model. Each year was split into 32 × 32 × 1 tiles across the study area, and then converted to 3D arrays, each of which pairs with one predictor variable array.The 3D predictor variable arrays (dimensions: 32 × 32 × 31) were then stacked into one large 4D array containing all these individual tiles across all years (dimensions: W × 32 × 32 × 31, where W is a large value). The same was done with the 3D dependent variable arrays (dimension: 32 × 32 × 1), preserving the order so that each element in this large 4D array (dimensions: W × 32 × 32 × 1) matches with its counterpart in the predictor variable array.The order of this large 4D training data array was then randomised along the first dimension to avoid bias in passing to the CNN training algorithm, but the randomised re-ordering was repeated with the dependent variable array so as to preserve the elementwise pairing for cross-validation.Model development and applicationFire prediction requires the combination of spatial and temporal indicators to generate a probabilistic output for each location within a given study area. There is a need to preserve a certain level of proximity information, as the location of variables in relation to one another may have a substantial impact on the results. For example, a patch of secondary forest that is immediately adjacent to an area recently deforested may have a significantly higher probability of fire occurrence than an area surrounded entirely by primary forest.CNNs retain spatial features by employing a moving window of reference, known as a kernel, over the input image that captures these proximity relationships within the model structure. For this reason, CNNs are often used for image classification problems, and is an ideal model configuration for the problem of fire prediction across an area. Therefore, we have developed a CNN binary classification model using the Keras API package39 that builds on the TensorFlow machine learning platform40.Model structureCNN models typically apply a combination of kernel layers and dense layers that perform a series of transformations on the multi-channel input to either reduce it down to a single value, or to output an image the same width and height as the input with a single channel. These classification models can either assign a single value (binary classifier), or return one of many possible classifications.Kernels act on a subsection of the input stack (31 feature maps), assigning weights according to each cell’s position within the subsection to transform and combine the values into a new format to pass forward. As the kernel is applied to all subsections of the input stack, it transforms them to the new format, and builds a reconstituted image with dimensions that usually differ from the input. A dense layer will do the same operation, but acting only on a single grid cell of the input stack, acting at the same location upon all input feature maps within the stack at a time—using all values at that location (i.e., the 1 × 1 subsection) and transforming them according to assigned weights to pass forward a new set of channels to a single grid cell on the output stack. Each layer, either kernel or dense, may expand or contract the number of channels it passes forward. A kernel layer may also change the width and height dimensions of the subsection it passes forwards.We require an output that corresponds to a map of fire-occurrences; therefore our model needs to perform a series of transforms that preserve the width and height of the input, but reduce it to a single channel. The single channel in the output then represents the probability of each cell being classified as fire or not-fire (0–1).Our CNN model is comprised of 5 kernel layers (K1–K5 in Fig. 5), each acts on a 3 × 3 subsection and preserves width and height, passing forwards a transformed 3 × 3 section. Kernel K1 takes an input of 31 channels (predictor variables) but passes forward 128 channels to form the transformation T1 (Fig. 6). Kernels K2–K4 take inputs of 128 channels and pass forward 128 channels (T2–T4). Kernel K5 takes an input of 128 channels but passes forward 1 channel—the output. After each kernel applies its weights, there is an activation function applied before the values are passed on, which modify the answer to fit the necessary criteria to be a valid input to the next process. Kernels K1–K4 have a rectified linear (relu) activation function, which returns the input value if positive, and 0 if negative. Kernel K5 has a sigmoid activation function, that transforms the input values to between 0 and 1 such that negative values are transformed to 0.5.Fig. 6: Model structural diagram.Model structural diagram showing the input, 3 × 3 kernel layers (K1–K5), each transformation passed forwards (T1–T4) and the output, with all dimensions labelled.Full size imageModel training and validationWe used a stochastic gradient descent optimising function called Adam41 combined with a binary cross-entropy loss function to train the model against our fire-hotspot dataset iterated over 20 epochs. We split the data 70/30, using 70% as training data and 30% as validation data, recording accuracy, precision, and recall as the performance metrics, as well as the loss function itself.After model training, we applied the model to each yearly raster stack and compared the output against the fire-hotspot data for further model validation. Before validating the model outputs, we applied a simple 3 × 3 moving average window as a smoothing function to reduce the edge effects of tiling that are a by-product of having to split the study area into smaller tiles (32 × 32) for passing to the model. For this yearly validation, we again used the metrics accuracy, precision, and recall, such that:$${{{{{rm{Accuracy}}}}}}=100({{{{{rm{TP}}}}}}+{{{{{rm{TN}}}}}})/({{{{{rm{TP}}}}}}+{{{{{rm{TN}}}}}}+{{{{{rm{FP}}}}}}+{{{{{rm{FN}}}}}})$$$${{{{{rm{Precision}}}}}}=100({{{{{rm{TP}}}}}})/({{{{{rm{TP}}}}}}+{{{{{rm{FP}}}}}})$$$${{{{{rm{Recall}}}}}}=100({{{{{rm{TP}}}}}})/({{{{{rm{TP}}}}}}+{{{{{rm{FN}}}}}})$$where TP is true positive, TN is true negative, FP is false positive, and FN is false negative. These comparisons were made on a raster cell to raster cell basis after designating a 500 m buffer around each fire hotspot observation (MODIS and VIIRS data) and converting the buffers to a raster image of the same resolution and extent as the model prediction.ScenariosAfter validating the model performance, we built future scenarios to investigate the impact on fire occurrence of managing key anthropogenic features of the landscape: canals and land cover (Table 2).Table 2 Future scenario types and descriptions.Full size tableStudies have shown that unmanaged areas of heavily degraded or cleared swamp-forest are most susceptible to fires16,17,25,26,33,42. Therefore, we have built scenarios that investigate the possible impact of managing these areas by altering the model inputs to re-assign the land-cover designations ‘Swamp shrubland’ and ‘Scrubland’, as well as other land designation alterations. The first such restoration scenario investigates the impact of reforesting these areas by re-assigning the designations to ‘Swamp forest’. The second such scenario investigates the impact of converting these unmanaged areas to plantations by re-assigning the designations to ‘Plantation’. We also built two further land cover scenarios to investigate the impact of continued deforestation in the region by re-assigning the ‘Swamp forest’ designation to ‘Swamp shrubland’ and ‘Plantation’.We then built a scenario to investigate the impact of canal blocking on fire occurrence, modifying the proximity to canals model input by reducing the number of canals included in our proximity analysis to just two major canals, one that runs north-south, and one that runs west-east (Fig. 1). These canals could not practically be blocked due to their size and importance as navigation conduits.The final scenario simulates the combined impact of both re-foresting unmanaged degraded and cleared forest areas and the blocking of canals simultaneously.To evaluate the impact of each scenario on fire occurrences, we calculated the ratio of model predictions >0.5 probability (i.e., that a fire would occur in that raster cell) for each year for each scenario against the same year for the baseline scenario.Model use as a predictive toolTo evaluate the model’s potential to predict future fire distribution across the wider ex-Mega Rice Project area, we trained a second version of the model following the same methodology outlined above, but included only data from 2002 to 2018 in the training and test data passed to the model fitting algorithm. We then applied the model to the predictor variables corresponding to 2019 and compared model outputs to the observations of fire-occurrences by again looking at the metrics accuracy, precision, and recall. We also present a visual comparison of the outputs from the full model (2019 included in training data), the predictive model (2019 not included), and the observation data (MODIS and VIIRS hotspots). More

In line with the results of present study, we suggest that the exploitation of forage resources by agro-pastoralists occurs in a non-random manner. The use of forage resources is guided by a series of functional characters related to palatability and nutritional value, which determine preferential use due to the better quality of resource. At the same time, we understand that forage uses are complex and multifactorial in nature, and regulated in a substantial way by seasonality and ecological factors (Fig. 5), such as the availability of plant resources and local diversity.Figure 5Diagrammatic representation for the effects of vegetation cover and seasonality on forage resource selection in Dry Forests. Image created with Microsoft Office 2019 PowerPoint (www.office.com).Full size imageThe differences of plant species cited between areas reveal the positive effect of vegetation cover on the use and knowledge of plants by agro-pastoralists. Our findings reveal that the greater number of plant species mentioned by agro-pastoralists in Area II is directly associated with greater availability of resources in this area, as long as we consider vegetation cover as availability of resources, which allows different species to be used throughout the year. On the other hand, in regions with low vegetation cover (Area I), the low availability of resources limits the use and knowledge of plants by residents, which can lead to greater pressure on a small set of available species. Such findings reinforce the importance of vegetation cover for ecosystem provision of goods and services to human populations that depend directly or indirectly on these services.The most represented families found in the present study have also been reported in several other ethnobotanical studies6,16,17,29, with emphasis on Fabaceae and Poaceae, which are recognized for their high forage potential, which derives, above all, from high palatability and nutritional value30. Simultaneously, citations mostly for native species reflect the importance and potential of Caatinga resources as important components of the ruminant diet11, both for the woody and herbaceous strata, corroborating the estimate in the literature that 70% of vegetation has potential use as forage31.The characteristic seasonality of vegetation, on the other hand, represents a limiting factor for forage productivity, culminating in high fluctuations in quality and availability, as well as changes in the dominance of different strata and composition of forage species throughout the seasons11,32. The seasonal distribution of species explains the similarity of seasons between areas, with a higher similarity percentage for the dry seasons, since there is less availability of resources to be exploited compared to the rainy season. In this context, the potentially used species are commonly accessible woody species in both areas. However, during the rainy season, the high availability of herbaceous plants regulates different uses (Fig. 4), but even so, they also exhibit relatively similar patterns, mainly due to the woody component that denotes the common demand by ruminants at the beginning of this season.The effect of climatic variables on vegetation use patterns was documented by16,17, both of which showed greater richness in the use of herbaceous forage during the rainy season, a finding that reflects the seasonal distribution—restriction to that season—and decrease in the qualitative character of annual species33. At the same time, it also reflects the greater number of unique species for the rainy season. However, when compared to woody strata, significant differences in terms of richness are not found because although the diversity of herbaceous species in the Caatinga is greater24, it is much less known than that of the tree-shrub stratum11.Agro-pastoralists even characterize animal preferences for herbaceous stratum, but as its diversity is immense and ephemeral, they claim to have limited ability to identify the species. The high abundance of resources in the rainy season also reduces the concern with forage use, which implies less attention to the species that are consumed. In contrast, woody species, due to multiple uses and greater availability over time, tend to be better known10,34, with a different effect in the dry season making the optimal foraging pattern in this period inherent to the knowledge of agro-pastoralists35.In addition, according to the ecological appearance hypothesis, there is a general tendency for less apparent species to be neglected by populations36. Some studies have corroborated the hypothesis within the context of forage use, with woody species being cited more and having more uses6,15. In addition, people tend to focus on resources whose supply is given continuously10, which may explain why woody species are well represented in both seasons.Security in the provisioning of ecosystem services is an essential component for local populations, and thus woody species are highly valued because they reflect predictability of use15,35. This can be a particularly influential criterion because perennial or late leaf deciduous species, such as Cynophalla flexuosa and Myracrodruon urundeva, had significant amounts of citations and perceptions employing high valuation, as represented by some statements by some interviewees: “É um refrigero na seca” (it is savage in the dry season), “É uma ração boa na seca” (it is a good food in the dry season).In turn, differences in richness of the species cited by the two areas corroborate our first hypothesis that populations inserted in environments with greater vegetation cover tend to cite more species. In line with these findings, considerable floristic dissimilarity was also found between the two areas, given the exclusivity of species. Such dissimilarity may suggest particularities in the vegetation attributes of each area, such as greater floristic diversity7,37,38.Since anthropic processes are irregularly distributed in space, variation in the provisioning of ecosystem services by vegetation also occurs, and influences different collection profiles39. On the other hand, areas with greater species richness have been shown to have greater use patterns6,7. The larger number of species cited as woody and native for Area II is, therefore, associated with greater general richness, as well as herbaceous species present in the rainy season. In contrast, common species are reflected in trends of similar foraging patterns, as well as the presence of common species between areas38. In addition to different levels of disturbance, differences in floristic composition between areas may also be due to edaphic variation40.Our second hypothesis was refuted because the difference in the richness of exotic species between the areas. Plausible explanations for this finding are that, in general, exotic herbaceous species are commonly used for forage in the semi-arid region of Brazil41. Herbaceous species comprise the primary component of the ruminant diet. However, in the midst of their occurrence restricted to the short rainy period, exotic species, mainly of Fabaceae and Poaceae, have been introduced to increase the forage availability, which currently represents an important attribute of forage resources in the Caatinga41,42,43. At the same time, and to also increase the availability of forage resources, the cultivation of species by agro-pastoralists may be common in their properties44, mainly exotics, such as Prosopis juliflora, that have high adaptive potential and governmental incentives45.Regarding use patterns, according to the data presented here it is possible to state that agro-pastoralists ’ experiences with herding activities provide an accumulation of a vast knowledge about forage resources15. This knowledge allows forage resources to be characterized by their potential according to a variety of criteria associated with seasonal variation and qualitative attributes, as commonly found by other studies14,15,16,17,37. Such criteria are often revealed by qualitative approaches that define the valuation perception of resources. Thus, nutritional value and palatability can be implicitly associated with the definitions of “É uma ração boa” (it is a good food), “o bicho gosta muito” (the animals like it very much) and “Rico em proteínas” (rich in protein).It should be added that the establishment of intrinsic relationships with resources allows a particular understanding at a high level of detail15,35, such as changes in palatability throughout development with descriptions including chemical17 and structural changes. Studies confirm that some Caatinga species vary in their chemical composition during leaf maturation, which influences nutritional quality17,46.In addition to revealing the domain of information, this body of knowledge allows maximizing forage use based on nutritional properties weighted by availability14,37. Nunes37 confirmed that the forage species selected by informants and the criteria they adopted coincided with nutritional values measured by the literature, and that, as also found in the present study, younger plants were recognized as highly appreciated by animals. This appreciation is due to the greater palatability of plant organs at this stage47. This is a matter of concern for the sustainability of the Caatinga, since direct or indirect grazing has compromised the regeneration process12 since younger individuals are clearly more sensitive to damage48.Also, considering the potential of Caatinga, we suggest that investment through government actions encourage the cultivation of native species to ensure the production of forage and, consequently, guarantee the sustainability of livestock activity and the ecosystem in question. More

This study proposed a protocol to evaluate the behavior of geese reared outdoors in agroforestry systems. A data collection form (i.e., BOF) was developed and validated both in relation to its reliability and its validity. In this context, moreover, ABMs useful for a welfare assessment protocol could be defined, and changes in the behavior of geese due to daily time and environmental context could be identified.Behavioral observations, based on the capture of the major changes in an animal’s body language17, are used daily in the assessment of animal health and welfare. Body language is a type of dynamic expression of the interactions among conspecifics or between animals and their environment. Behavioral changes can happen quickly or as subtle shifts not easily detectable18. Indeed, especially in the case of direct observation in the field, it becomes difficult to identify each behavioral variation. Furthermore, the on-farm use of the BOF proposed in the present study involved focal subgroup sampling, as ten geese were simultaneously observed, which may increase the difficulties. Indirect observation by videos, which allow the review of a certain action several times and the focal-animal approach, is a useful tool to partially overcome these issues and thus improve the accuracy of observation. The validation process of the BOF adopted in this study, therefore, included the definition of both its interobserver reliability and correlation with indirect observations.In this study, the direct observations in the field were performed by both an expert (i.e., main observer) and an inexperienced trained observer. As expected, the main observer was able to detect a higher frequency of behaviors, especially the rarer ones. For example, the inexperienced observer did not report any examples of allo-grooming, squawking, wagging tail, stretching, or panting behavior. However, the two observers showed excellent interobserver reliability (ICC  > 0.75). Major agreements were found for walking, roosting, and foraging. Accordingly, several studies have shown that observers with little experience can also provide a valuable contribution in observational research19,20. Overall, these results support the reliability of the BOF even if the observer’s experience helps him or her to better grasp rarer behaviors, as these behaviors could play an important role as welfare indicators.In the last two decades, important technological developments have occurred in the livestock sector. The use of sensors, cameras, and other devices can generate objective information about individual behavior, thereby allowing its evaluation in large observation areas and for large groups of animals and resulting in the better detection of natural animal behavior. Thus, in our study, the data collected by a video recording system (Noldus XT) were used as a gold standard measure to define the criterion validity of the BOF. Our results indicated excellent agreement between direct and indirect observations, supporting the BOF criterion validity. A poor correlation was only found for 2 variables (i.e., squawking and wagging tail), which were more difficult to collect by direct observation. The use of the BOF involved the simultaneous observation of 10 animals, but the geese had a synchronized behavior and moved in groups within the grazing area. This greatly facilitated focal subgroup sampling, allowed all animals to always be under observation, and could explain the high correlation between the two observation methods. However, the comparison between the observations collected in the field by the main observer and those recorded using the computerized system confirmed the greater accuracy of the latter. The analysis of the video in continuous with the use of some tools, such as the zoom or slow-motion functions, and the focal-animal sampling provided an easier identification of some behaviors and, in general, greater accuracy. Due to its nonintrusive approach, video recording has become a common practice for behavior assessment21, but it can be expensive and time-consuming. On the other hand, direct observations made by the BOF were valid and less expensive, suggesting that it could be a feasible tool with which to evaluate the welfare principle of Appropriate behavior. As recommended for welfare assessment protocols22, the BOF ethogram included indicators of both positive and negative states; however, it would be necessary to integrate it with behavioral tests and other ABMs evaluating the human-animal relationship.As mentioned above, there is no standardized geese behavior ethogram. Thus, to verify the content validity of the BOF, its behavior variables were analyzed through a PCA. The 4 extracted PCs could represent the broad behavioral dimensions of geese. In particular, the geese’s activity reported in PC1 was characterized by locomotor, foraging, and exploratory behaviors, with opposite signs with respect to roosting. The positive correlation between explorative and grazing activities and their negative correlation with static behaviors has been widely demonstrated in chickens. Chicken genotypes characterized by low exploratory aptitude exhibited low kinetic behaviors but a high frequency of roost and rest behaviors23. Göransson et al.24 showed that 50% of the observed birds exhibited sitting behavior, whereas less than 10% performed foraging activity.PC2 included all the variables that characterized the geese’s social aspects, including both positive and negative interactions. Usually, greylag geese live in a large flock because the offspring remain with their parents for an entire year. Such groups are characterized by complex relationships based on social interactions25. The formation of a group is characterized by agonistic behaviors such as fighting, pecking, and threatening, as well as submissive behaviors such as avoiding contact, crouching, and escaping26 to establish a hierarchical order. After this phase, a tolerance status develops, and birds maintain their social interactions through the use of body postures and vocalizations. Accordingly, the variables reported in PC2 were related not only to aggressive behaviors but also to geese’s vocalization and posture, which probably helped to maintain flock stability. Therefore, a higher PC2 score could indicate the need to establish and maintain a hierarchical order within the group, resulting in high social interactions.PC3 reported comfort and body care behaviors. The opportunity to spend a lot of time on body care, which should also include access to water for bathing, is of paramount importance with regard to fulfilling the biological requirements of geese27. Thus, a higher loading of this PC means that animals showed a good degree of both welfare and adaptability. In our study, a high frequency of self-cleaning and wing flapping behaviors was recorded, and the geese often took advantage of the water tub. In contrast, a very low frequency of aggression behaviors was observed, suggesting that the groups of geese were quite stable and that the animals felt safe in the environment in which they were rear. These findings confirm that agroforestry has a favorable impact on bird welfare by allowing the display of the full range of behavior, improving the animals’ comfort28.PC4 was mainly represented by the neck forward behavior. This position only occasionally represents an attack behavior and is not utilized during the establishment of hierarchical order but when it is necessary to maintain and reinforce the order inside the group. Furthermore, a goose that assumes this posture often does so while continuing another activity29. The neck forward behavior was positively associated with the stretching behavior. Stretching is usually categorized as a comfort behavior for broilers30, but it could also be used when the animal needs to relax stress-related tension in their muscles31,32 or as an adaptive strategy for dealing with unknown contexts33. Neck forward and stretching were eventually considered social avoidance behaviors, although they could be ambivalent and thus require further study, case-by-case assessment, and perhaps a better description in the ethogram.Finally, some interesting results emerged regarding the comparison of geese’s behavior during the morning and afternoon and between the two different agroforestry systems. In particular, geese showed a higher frequency of active behaviors such as walking, foraging, drinking, neck forward, and feeding during the morning compared to the afternoon. All of these behaviors suggest that geese concentrate their grazing and exploration activities during the morning. When and where to move is crucial for the food search and to avoid both predators and adverse climate conditions34. Cartoni Mancinelli et al.35 included exploratory attitude, walking, and eating grass activities in a multifactorial score as important parameters to consider to evaluate the adaptability of different organically reared chicken genotypes. Thus, exploratory and kinetic behaviors are fundamental, especially in animals reared outdoors. Moreover, the positive correlation between walking and grazing behaviors is widely known36,37. In contrast, during the afternoon, geese showed higher frequencies of static behaviors such as resting, roosting, and self-grooming, suggesting that geese are more dedicated to comfort and body care activities during this time. These trials were performed in the hottest season; thus, the geese’s behavioral differences during the day could also depend on the fact that animals preferred to carry out active behaviors during the cooler hours (morning), while in the hottest hours (afternoon), they engaged in static activities. Active behaviors cause an increase in metabolism and body temperature38, whereas static behavior, such as roosting, is considered adaptative behavior to promote heat dissipation31,39.This could also explain why higher frequencies of walking and foraging and lower frequencies of static behaviors were found in the orchard system than in the vineyard system. Studies carried out on chickens have reported that, among different pasture enrichments, the presence of trees promotes walking animal activity compared with crop inclusion40,41. The cover provided by trees made the animals feel protected from predators and provided shade during the hottest part of the day40, thereby stimulating the animals to explore all the available space in the pen. Accordingly, geese reared in the apple orchard ingested more grass than those reared in a vineyard36. However, there were no differences between the two systems for social behaviors. Moreover, the highest frequency of roosting and self-cleanliness behaviors was recorded in the vineyard, suggesting that this space offered a comfortable environment and that both systems seem respectful of the biological needs and welfare of the geese.The behavioral assessment protocol proposed in this study involving the BOF ethogram was feasible, low-cost, fast, and responsive both over time and between housing systems. It could thus be used for the assessment of Appropriate behavior in a welfare assessment protocol for geese reared in outdoor or free-range systems, although it lacks indicators of the human-animal relationship, such as avoidance distance or handling tests; such a scoring system should be developed. Regarding the specific behaviors in the two agroforestry systems, it should also be noted that they are difficult to generalize, as the characteristics of the plants, the environment, and management could have influenced these traits. Specifically, the behaviors could have been affected by the temperatures; therefore, further trials at different altitudes, seasons (i.e., autumn and winter), and climate are necessary for external validation. More

In early 2020, my colleagues and I realized that animal-tracking data collected before, during and after the pandemic lockdowns could provide invaluable insights into human–wildlife interactions and conservation benefits on a global scale. We launched a research consortium — the COVID-19 Bio-Logging Initiative — to investigate how animals behaved while much of the world’s human population sheltered at home.But we had no way to establish how many, and which, animals were wearing tags. Miniature tracking devices are routinely attached to a vast range of species — from songbirds to whales — to collect detailed data on their movements, behaviour and physiology. Yet, of the thousands of ‘bio-loggers’ deployed every year, many generate data sets that remain effectively undiscoverable — they are saved on personal hard drives or institutional servers, inaccessible to the wider community. This problem can be solved by setting up a global registry for all tags on wild animals.Although individual tracking studies make important contributions to our understanding of the ecological needs of animal species, pooling data (across taxa, longer time periods or multiple locations) can reveal general patterns, aiding the design of particularly effective conservation strategies. For example, integrating the tracks of 4,060 animals across 17 marine species (including albatrosses, penguins, seals and whales) has helped to identify conservation priority areas in the Southern Ocean (M. A. Hindell et al. Nature 580, 87–92; 2020).In an ideal world, all animal-tracking data would be archived — with either open or restricted access — in public repositories, such as Movebank. Excellent progress has been made towards this goal, but universal uptake is hindered by time constraints, governmental or institutional restrictions and concerns over inappropriate data use.To encourage as many data owners as possible to join the COVID-19 Bio-Logging Initiative, we launched a recruitment campaign through Movebank, social media, mailing lists, newsletters, personal contacts and a published call to action (C. Rutz et al. Nature Ecol. Evol. 4, 1156–1159; 2020). Our consortium has grown to more than 600 international collaborators, accumulating a staggering one billion location records for some 200 animal species. Despite this impressive community response, we know that this is only the tip of the iceberg.The global tag registry that I suggest would contain metadata for tags (including tag type and settings, information on the animal, and date and location of deployment), as well as researchers’ contact details — but not the actual tracking data. This decoupling of information would unlock the field’s full conservation potential in the short term and would build the trust required to allow raw data to be archived routinely in public repositories in the longer term. Over time, the tag registry is likely to evolve naturally into a ‘meta-repository’, linking to raw data sets hosted across a multitude of repositories.The registry would enable researchers to check data availability at the push of a button — for example, for a particular taxonomic group, such as terrestrial carnivores, or a specific region, such as the Pacific Ocean — and to get in touch with the relevant data owners. Registry management must comply with international best practices, so robust processes would need to be set up to vet queries, pass on collaboration proposals to data owners and minimize overlap between studies.For the registry to fulfil its intended purpose, it must be used by the entire animal-tracking community. How can this be achieved? I see an opportunity to integrate tag registration into existing ethical-review processes. Governmental authorities, research institutions, funders, publishers and fieldworkers agree that permits must be in place before animals can be tagged. Building on this international consensus, ethical review boards could make tag registration a condition of study approval.To complement this bottom-up approach, well established initiatives — such as those associated with the United Nations Environment Programme or the International Union for Conservation of Nature — could help to build an international policy mandate and provide independent oversight. The International Bio-Logging Society, which has been working to unite animal-tracking efforts on land and at sea, could provide crucial support.This vision is no doubt ambitious, but it is achievable. Every civil aircraft on the planet must be registered — so I am convinced that, with effective coordination, we can accomplish the same for tagged animals. Furthermore, the basic principle of hosting metadata, but not raw data, is being used productively by other databases, such as AviSample — a registry for biological samples collected from wild birds.Many researchers, myself included, feel a moral obligation to the animals carrying our tags. A global tag registry would help to realize the full conservation potential of all tracking data, minimize duplication of tagging efforts and facilitate sharing of welfare-related expertise. The conservation cost of missing data in large-scale collaborative projects cannot be easily measured, but is probably substantial. We simply cannot afford this, and must ensure that all animal-tracking data are immediately discoverable.

Competing Interests
This article is a contribution of the COVID-19 Bio-Logging Initiative, which is funded in part by the Gordon and Betty Moore Foundation (GBMF9881) and the National Geographic Society (NGS-82515R-20) (both grants to C.R.), and endorsed by the United Nations Decade of Ocean Science for Sustainable Development. More

Rio Conservation and Sustainability Science Centre, Department of Geography and the Environment, Pontifical Catholic University, Rio de Janeiro, BrazilBernardo B. N. Strassburg, Alvaro Iribarrem, Carlos Leandro Cordeiro, Renato Crouzeilles, Catarina Jakovac, André Braga Junqueira, Eduardo Lacerda & Agnieszka E. LatawiecInternational Institute for Sustainability, Rio de Janeiro, BrazilBernardo B. N. Strassburg, Alvaro Iribarrem, Carlos Leandro Cordeiro, Renato Crouzeilles, Catarina Jakovac, André Braga Junqueira, Eduardo Lacerda, Agnieszka E. Latawiec, Robin L. Chazdon & Carlos Alberto de M. ScaramuzzaPrograma de Pós Graduacão em Ecologia, Universidade Federal do Rio de Janeiro, Rio de Janeiro, BrazilBernardo B. N. Strassburg, Renato Crouzeilles & Fabio R. ScaranoBotanical Garden Research Institute of Rio de Janeiro, Rio de Janeiro, BrazilBernardo B. N. StrassburgSchool of Biological Sciences, University of Queensland, St Lucia, Queensland, AustraliaHawthorne L. BeyerAgricultural Science Center, Federal University of Santa Catarina, Florianópolis, BrazilCatarina JakovacInstitut de Ciència i Tecnologia Ambientals, Universitat Autònoma de Barcelona, Barcelona, SpainAndré Braga JunqueiraDepartment of Geography, Fluminense Federal University, Niterói, BrazilEduardo LacerdaDepartment of Production Engineering, Logistics and Applied Computer Science, Faculty of Production and Power Engineering, University of Agriculture in Kraków, Kraków, PolandAgnieszka E. LatawiecSchool of Environmental Sciences, University of East Anglia, Norwich, UKAgnieszka E. LatawiecDepartment of Zoology, University of Cambridge, Cambridge, UKAndrew Balmford, Stuart H. M. Butchart & Paul F. DonaldInternational Union for Conservation of Nature (IUCN), Gland, SwitzerlandThomas M. BrooksWorld Agroforestry Center (ICRAF), University of The Philippines, Los Baños, The PhilippinesThomas M. BrooksInstitute for Marine & Antarctic Studies, University of Tasmania, Hobart, Tasmania, AustraliaThomas M. BrooksBirdLife International, Cambridge, UKStuart H. M. Butchart & Paul F. DonaldDepartment of Ecology and Evolutionary Biology, University of Connecticut, Storrs, CT, USARobin L. ChazdonWorld Resources Institute, Global Restoration Initiative, Washington, DC, USARobin L. ChazdonTropical Forests and People Research Centre, University of the Sunshine Coast, Sippy Downs, Queensland, AustraliaRobin L. ChazdonInstitute of Social Ecology, University of Natural Resources and Life Sciences Vienna, Vienna, AustriaKarl-Heinz Erb & Christoph PlutzarDepartment of Forest Sciences, ‘Luiz de Queiroz’ College of Agriculture, University of São Paulo, Piracicaba, BrazilPedro BrancalionRSPB Centre for Conservation Science, Royal Society for the Protection of Birds, Edinburgh, UKGraeme Buchanan & Paul F. DonaldSecretariat of the Convention on Biological Diversity (SCBD), Montreal, Quebec, CanadaDavid CooperInstituto Multidisciplinario de Biología Vegetal, CONICET and Universidad Nacional de Córdoba, Córdoba, ArgentinaSandra DíazUnited Nations Environment Programme World Conservation Monitoring Centre, Cambridge, UKValerie Kapos & Lera MilesBiodiversity and Natural Resources (BNR) program, International Institute for Applied Systems Analysis (IIASA), Laxenburg, AustriaDavid Leclère, Michael Obersteiner & Piero ViscontiDivision of Conservation Biology, Vegetation Ecology and Landscape Ecology, University of Vienna, Vienna, AustriaChristoph PlutzarB.B.N.S. wrote the first version of the paper. All authors provided input on subsequent versions of the Reply. More

Treasure our treesNearly one-third of tree species are threatened with extinction. This is more than twice the number of threatened mammals, birds, amphibians and reptiles combined.The loss of tree species is often overlooked, as our News Feature reports. In 2021, after a huge tree-hunting exercise called the Global Tree Assessment, plant conservationists announced that they had found 58,497 tree species, of which 17,510 were threatened. Since then, almost 2,800 of those have been labelled critically endangered. Some 142 species are thought to be extinct in the wild.

Killer cancersThis chart shows some of the results from the largest study yet of the link between cancer burden and risk factors. Researchers used extensive data on death and disability from more than 200 countries to estimate that potentially avoidable risk factors were responsible for more than 44% of global cancer deaths in 2019. Of these, tumours of the lung, trachea and bronchus were the leading cause of death.Smoking, alcohol use and a high body-mass index were the risk factors with the biggest contribution to cancer. The findings emphasize familiar health advice not to smoke, drink too much or become overweight.

New breed of vaccinesIt was hoped that a new breed of COVID-19 vaccine — based on Omicron variants of the virus SARS-CoV-2 — would offer substantially greater protection than older vaccines that are based on the strain of the virus that emerged in 2019. But an analysis of data from several studies suggests that updated boosters offer much the same level of protection as does an extra dose of the older vaccines. The study is a preprint that has not yet been peer reviewed.The team’s modelling showed that, in a population where half of people are already protected against a symptomatic SARS-CoV-2 infection through previous vaccination or infection, an updated vaccine booster bumped protection up to 90%, compared with 86% protection provided by an extra dose of the original vaccine. For protection against severe disease, however, the difference was less than 1%. But the relative benefits of variant-based boosters could grow stronger if a new variant appears, as our News story explains. More

RESEARCH BRIEFINGS
07 September 2022

A diverse range of marine predators — including tunas, billfishes and sharks — in the North Pacific Ocean cluster together in clockwise-rotating eddies, seemingly to hunt deep-ocean prey, which are unusually abundant there. This suggests that there is a relationship between the foraging opportunities of predators and the energetics of this marine biome. More