Search

The article presents results of using remote sensing images and machine learning to map and assess land potential based on time-series of potential Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) composites. Land potential here refers to the potential vegetation productivity in the hypothetical absence of short–term anthropogenic influence, such as intensive agriculture and urbanization. Knowledge on this ecological land potential could support the assessment of levels of land degradation as well as restoration potentials. Monthly aggregated FAPAR time-series of three percentiles (0.05, 0.50 and 0.95 probability) at 250 m spatial resolution were derived from the 8-day GLASS FAPAR V6 product for 2000–2021 and used to determine long-term trends in FAPAR, as well as to model potential FAPAR in the absence of human pressure. CCa 3 million training points sampled from 12,500 locations across the globe were overlaid with 68 bio-physical variables representing climate, terrain, landform, and vegetation cover, as well as several variables representing human pressure including: population count, cropland intensity, nightlights and a human footprint index. The training points were used in an ensemble machine learning model that stacks three base learners (extremely randomized trees, gradient descended trees and artificial neural network) using a linear regressor as meta-learner. The potential FAPAR was then projected by removing the impact of urbanization and intensive agriculture in the covariate layers. The results of strict cross-validation show that the global distribution of FAPAR can be explained with an R2 of 0.89, with the most important covariates being growing season length, forest cover indicator and annual precipitation. From this model, a global map of potential monthly FAPAR for the recent year (2021) was produced, and used to predict gaps in actual vs. potential FAPAR. The produced global maps of actual vs. potential FAPAR and long-term trends

Satellite images can be used to derive time series of vegetation indices, such as normalized difference vegetation index (NDVI) or enhanced vegetation index (EVI), at global scale. Unfortunately, recording artifacts, clouds, and other atmospheric contaminants impacts a significant portion of the produced images, requiring the usage of ad-hoc techniques to reconstruct the time series in the affected regions. In literature, several methods have been proposed to fill the gaps present in the images, and some works also presented performance comparisons between them (Roerink et al., 2000; Moreno-Martínez et al., 2020; Siabi et al., 2022). Because of the lack of a ground truth for the reconstructed images, the performance evaluation requires the creation of datasets where artificial gaps are introduced in a reference image, such that metrics like the root mean square error (RMSE) can be computed comparing the reconstructed images with the reference one. Different approaches have been used to create the reference images and the artificial gaps, but in most cases, the artificial gaps are introduced using arbitrary patterns and/or the reference image is produced artificially and not using real satellite images (e.g. Kandasamy et al., 2013; Liu et al., 2017; Julien & Sobrino, 2018). In addition, to the best of our knowledge, few of them are openly available and directly accessible allowing for fully reproducible research. We provide here a benchmark dataset for time series reconstruction method based on the harmonized Landsat Sentinel-2 (HLS) collection where the artificial gaps are introduced with a realistic spatio-temporal distribution. In particular, we selected six tiles that we considered representative for most of the main climate classes (e.g. equatorial, arid, warm temperature, boreal and polar), as depicted in the preview. Specifically, following the relative tiling system shown above, we downloaded the Red, NIR and F-mask bands, Satellite images can be used to derive time series of vegetation indices, such as normalized difference vegetation index (NDVI) or enhanced vegetation index (EVI), at global scale. Unfortunately, recording artifacts, clouds, and other atmospheric contaminants impacts a significant portion of the produced images, requiring the usage of ad-hoc techniques to reconstruct the time series in the affected regions. In literature, several methods have been proposed to fill the gaps present in the images, and some works also presented performance comparisons between them (Roerink et al., 2000; Moreno-Martínez et al., 2020; Siabi et al., 2022). Because of the lack of a ground truth for the reconstructed images, the performance evaluation requires the creation of datasets where artificial gaps are introduced in a reference image, such that metrics like the root mean square error (RMSE) can be computed comparing the reconstructed images with the reference one. Different approaches have been used to create the reference images and the artificial gaps, but in most cases, the artificial gaps are introduced using arbitrary patterns and/or the reference image is produced artificially and not using real satellite images (e.g. Kandasamy et al., 2013; Liu et al., 2017; Julien & Sobrino, 2018). In addition, to the best of our knowledge, few of them are openly available and directly accessible allowing for fully reproducible research. We provide here a benchmark dataset for time series reconstruction method based on the harmonized Landsat Sentinel-2 (HLS) collection where the artificial gaps are introduced with a realistic spatio-temporal distribution. In particular, we selected six tiles that we considered representative for most of the main climate classes (e.g. equatorial, arid, warm temperature, boreal and polar), as depicted in the preview. Specifically, following the relative tiling system shown above, we downloaded the Red, NIR and F-mask bands

Probability and uncertainty maps showing the potential current and future natural vegetation on a global scale under three different climate change scenarios (RCP 2.6, RCP 4.5 and RCP 8.5) predicted using ensemble machine learning. Current (2022 - 2023) conditions are calculated on historical long term averages (1979 - 2013), while future projections cover two different epochs: 2040 - 2060 and 2061 - 2080. Files are named according to the following naming convention, e.g.: biomes_graminoid.and.forb.tundra.rcp85_p_1km_a_20610101_20801231_go_epsg.4326_v20230410 with the following fields: generic theme: biomes, variable name: graminoid.and.forb.tundra.rcp85, variable type, e.g. probability ("p"), hard class ("c"), model deviation ("md") spatial resolution: 1km, depth reference, e.g. below ("b"), above ("a") ground or at surface ("s"), begin time (YYYYMMDD): 20610101, end time: 20801231, bounding box, e.g. global land without Antarctica ("go"), EPSG code: epsg.4326, version code, e.g. creation date: v20230410. We provide probability and hard class layers using a revised classification system of the BIOME 6000 project explained in the work of Hengl et al. (2018). The 20 classes from this classification system have then been aggregated in 6 biome classes following the IUCN Global Ecosystem Typology classification system. For probability layers, the uncertainty (model deviation: md) is calculated as the standard deviation of the predicted values of the base learners of the ensemble model. The higher the standard deviation the more uncertain the model is regarding the right value to assign to the pixel. For hard class layers the uncertainty is calculated using the margin of victory (Calderón-Loor et al., 2021) defined as the difference between the first and the second highest class probability value in a given pixel. High values would be measures of low uncertainty, while low values would indicate a high uncertainty. It is highly recommended to use the md layers to properly

The global potential distribution of biomes (natural vegetation) was modelled using 8,959 training points from the BIOME 6000 dataset and a stack of 72 environmental covariates representing terrain and the current climatic conditions based on historical long term averages (1979–2013). An ensemble machine learning model based on stacked regularization was used, with multinomial logistic regression as the meta-learner and spatial blocking (100 km) to deal with spatial autocorrelation of the training points. Results of spatial cross-validation for the BIOME 6000 classes show an overall accuracy of 0.67 and R2logloss of 0.61, with “tropical evergreen broadleaf forest” being the class with highest gain in predictive performances (R2logloss = 0.74) and “prostrate dwarf shrub tundra” the class with the lowest (R2logloss = −0.09) compared to the baseline. Temperature-related covariates were the most important predictors, with the mean diurnal range (BIO2) being shared by all the base-learners (i.e.,random forest, gradient boosted trees and generalized linear models). The model was next used to predict the distribution of future biomes for the periods 2040–2060 and 2061–2080 under three climate change scenarios (RCP 2.6, 4.5 and 8.5). Comparisons of predictions for the three epochs (present, 2040–2060 and 2061–2080) show that increasing aridity and higher temperatures will likely result in significant shifts in natural vegetation in the tropical area (shifts from tropical forests to savannas up to 1.7 ×105 km2 by 2080) and around the Arctic Circle (shifts from tundra to boreal forests up to 2.4 ×105 km2 by 2080). Projected global maps at 1 km spatial resolution are provided as probability and hard classes maps for BIOME 6000 classes and as hard classes maps for the IUCN classes (six aggregated classes). Uncertainty maps (prediction error) are also provided and should be used for careful interpretation of the future projections.

High resolution maps resulting from a trend analysis conducted for the period 2000–2020 on the probability of occurrence maps prepared by Bonannella et al. (2022). For this analysis we selected the realized distribution time series layers at 30m spatial resolution for 6 out of 16 species described in the mentioned publication: Bonannella, C., Parente, L., de Bruin, S. and Herold, M. (2023). Multi-decadal trend analysis and forest disturbance assessment of European tree species: concerning signs of a subtle shift, PREPRINT (Version 1) available at Research Square [https://doi.org/10.21203/rs.3.rs-3288937/v1]

A spatiotemporal machine learning framework for automated prediction and analysis of long-term Land Use/Land Cover dynamics is presented. The framework includes: (1) harmonization and preprocessing of spatial and spatiotemporal input datasets (GLAD Landsat, NPP/VIIRS) including five million harmonized LUCAS and CORINE Land Cover-derived training samples, (2) model building based on spatial k-fold cross-validation and hyper-parameter optimization, (3) prediction of the most probable class, class probabilities and model variance of predicted probabilities per pixel, (4) LULC change analysis on time-series of produced maps. The spatiotemporal ensemble model consists of a random forest, gradient boosted tree classifier, and an artificial neural network, with a logistic regressor as meta-learner. The results show that the most important variables for mapping LULC in Europe are: seasonal aggregates of Landsat green and near-infrared bands, multiple Landsat-derived spectral indices, long-term surface water probability, and elevation. Spatial cross-validation of the model indicates consistent performance across multiple years with overall accuracy (a weighted F1-score) of 0.49, 0.63, and 0.83 when predicting 43 (level-3), 14 (level-2), and five classes (level-1). Additional experiments show that spatiotemporal models generalize better to unknown years, outperforming single-year models on known-year classification by 2.7% and unknown-year classification by 3.5%. Results of the accuracy assessment using 48,365 independent test samples shows 87% match with the validation points. Results of time-series analysis (time-series of LULC probabilities and NDVI images) suggest forest loss in large parts of Sweden, the Alps, and Scotland. Positive and negative trends in NDVI in general match the land degradation and land restoration classes, with “urbanization” showing the most negative NDVI trend. An advantage of using spatiotemporal ML is that the fitted model can be used to predict

This article describes a data-driven framework based on spatiotemporal machine learning to produce distribution maps for 16 tree species (Abies alba Mill., Castaneasativa Mill., Corylus avellana L., Fagus sylvatica L., Olea europaea L., Picea abies L. H.Karst., Pinus halepensis Mill., Pinus nigra J. F. Arnold, Pinus pinea L., Pinus sylvestrisL., Prunus avium L., Quercus cerris L., Quercus ilex L., Quercus robur L., Quercuss uber L. and Salix caprea L.) at high spatial resolution (30 m). Tree occurrence datafor a total of three million of points was used to train different algorithms: random forest, gradient-boosted trees, generalized linear models, k-nearest neighbors, CART and an artificial neural network. A stack of 305 coarse and high resolution covariates representing spectral reflectance, different biophysical conditions and biotic competition was used as predictors for realized distributions, while potential distribution was modelled with environmental predictors only. Logloss and computing time were used to select the three best algorithms to tune and train an ensemble model based on stacking with a logistic regressor as a meta-learner.An ensemble model was trained for each species: probability and model uncertainty maps of realized distribution were produced for each species using a time window of4 years for a total of six distribution maps per species, while for potential distributions only one map per species was produced. Results of spatial cross validation show that the ensemble model consistently outperformed or performed as good as the best individual model in both potential and realized distribution tasks,with potential distribution models achieving higher predictive performances(TSS = 0.898, R 2logloss = 0.857) than realized distribution ones on average(TSS = 0.874, R 2logloss = 0.839). Ensemble models for Q. suber achieved the best performances in both potential (TSS = 0.968, R2logloss = 0.952) and realized(TSS = 0.959, R 2logloss = 0.949) distribut

Probability and uncertainty maps showing the potential and realized distribution for the common hazel (Corylus avellana, L.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.

Probability and uncertainty maps showing the potential and realized distribution for the silver fir (Abies alba, Mill.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.

Probability and uncertainty maps showing the potential and realized distribution for the Turkey oak (Quercus cerris, L.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.

Probability and uncertainty maps showing the potential and realized distribution for the sweet chestnut (Castanea sativa, Mill.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.

Probability and uncertainty maps showing the potential and realized distribution for the goat willow (Salix caprea, L.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.

The dataset is a collection of presence and absence points for forest tree species for Europe. Each unique combination of longitude, latitude and year was considered as an independent sample. Presence data was obtained from the harmonized tree species occurrence dataset by Heising and Hengl (2020) and absence data from the LUCAS (in-situ source) dataset. A set of 50 different forest tree species was selected from the harmonized tree species dataset and data lacking a temporal observation was overlaid with yearly forest masks derived from land cover maps produced by Parente et al. (2021). We overlaid the points with the probability maps for the classes: 311: Broad-leaved forest, 312: Coniferous forest, 313: Mixed forest, 323: Sclerophyllous forest, 324: Transitional woodland-shrub, 333: Sparsely vegetated area.

Probability and uncertainty maps showing the potential and realized distribution for the holm oak (Quercus ilex L.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.

Probability and uncertainty maps showing the potential and realized distribution for the sweet cherry (Prunus avium, L.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.

Probability and uncertainty maps showing the potential and realized distribution for the Norway spruce (Picea abies, L. H. Karst.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.

Probability and uncertainty maps showing the potential and realized distribution for the cork oak (Quercus suber, L.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.

Probability and uncertainty maps showing the potential and realized distribution for the Scots pine (Pinus sylvestris, L.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.

Probability and uncertainty maps showing the potential and realized distribution for the stone pine (Pinus pinea, L.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.

Probability and uncertainty maps showing the potential and realized distribution for the pedunculate oak (Quercus robur, L.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.

Probability and uncertainty maps showing the potential and realized distribution for the European beech (Fagus sylvatica, Mill.) for Europe from the dataset prepared by Bonannella et al. (2022) and predicted using Ensemble Machine Learning (EML). Potential distribution map cover the period 2018 - 2020; realized distribution cover the period 2000 - 2020.