Global greening trends have been widely reported based on long‐term remote sensing data of terrestrial ecosystems. Typically, a hypothesis test is performed for each grid cell; this leads to multiple hypothesis testing and false positive trend detection. We reanalyze global greening and account for this issue with a novel statistical method that allows robust inference on greening regions. Based on leaf area index (LAI) data, our methods reduce the detected greening from 35.2% to 15.3% of the terrestrial land surface; this reduction is most notable in nonwoody vegetation. Our results confirm several greening regions (China, India, Europe, Sahel, North America, Brazil, and Siberia), that are also supported by independent data products. We also report evidence for an increasing seasonal amplitude in LAI north of 35°N. Considering the widespread use of spatially replicated trend tests in global change research, we recommend adopting the proposed multiple testing procedure to control false positive outcomes.

Vegetation state is largely driven by climate and the complexity of involved processes leads to non-linear interactions over multiple time-scales. Recently, the role of temporally lagged dependencies, so-called memory effects, has been emphasized and studied using data-driven methods, relying on a vast amount of Earth observation and climate data. However, the employed models are often not able to represent the highly non-linear processes and do not represent time explicitly. Thus, data-driven study of vegetation dynamics demands new approaches that are able to model complex sequences. The success of Recurrent Neural Networks (RNNs) in other disciplines dealing with sequential data, such as Natural Language Processing, suggests adoption of this method for Earth system sciences. Here, we used a Long Short-Term Memory (LSTM) architecture to fit a global model for Normalized Difference Vegetation Index (NDVI), a proxy for vegetation state, by using climate time-series and static variables representing soil properties and land cover as predictor variables. Furthermore, a set of permutation experiments was performed with the objective to identify memory effects and to better understand the scales on which they act under different environmental conditions. This was done by comparing models that have limited access to temporal context, which was achieved through sequence permutation during model training. We performed a cross-validation with spatio-temporal blocking to deal with the auto-correlation present in the data and to increase the generalizability of the findings. With a full temporal model, global NDVI was predicted with R² of 0.943 and RMSE of 0.056. The temporal model explained 14% more variance than the non-memory model on global level. The strongest differences were found in arid and semiarid regions, where the improvement was up to 25%. Our results show that memory effects matter on global scale, with the strongest effects occurring in sub-tropical and transitional water-driven biomes.

The statistical analysis of environmental data from remote sensing and Earth system simulations often entails the analysis of gridded spatio-temporal data, with a hypothesis test being performed for each grid cell. When the whole image or a set of grid cells are analyzed for a global effect, the problem of multiple testing arises. When no global effect is present, we expect $$ \alpha $$α % of all grid cells to be false positives, and spatially autocorrelated data can give rise to clustered spurious rejections that can be misleading in an analysis of spatial patterns. In this work, we review standard solutions for the multiple testing problem and apply them to spatio-temporal environmental data. These solutions are independent of the test statistic, and any test statistic can be used (e.g., tests for trends or change points in time series). Additionally, we introduce permutation methods and show that they have more statistical power. Real-world data are used to provide examples of the analysis, and the performance of each method is assessed in a simulation study. Unlike other simulation studies, our study compares the statistical power of the presented methods in a comprehensive simulation study. In conclusion, we present several statistically rigorous methods for analyzing spatio-temporal environmental data and controlling the false positives. These methods allow the use of any test statistic in a wide range of applications in environmental sciences and remote sensing.

This study analyzed highly correlated, feature-rich datasets from hyperspectral remote sensing data using multiple statistical and machine-learning methods. The effect of filter-based feature selection methods on predictive performance was compared. In addition, the effect of multiple expert-based and data-driven feature sets, derived from the reflectance data, was investigated. Defoliation of trees (%), derived from in situ measurements from fall 2016, was modeled as a function of reflectance. Variable importance was assessed using permutation-based feature importance. Overall, the support vector machine (SVM) outperformed other algorithms, such as random forest (RF), extreme gradient boosting (XGBoost), and lasso (L1) and ridge (L2) regressions by at least three percentage points. The combination of certain feature sets showed small increases in predictive performance, while no substantial differences between individual feature sets were observed. For some combinations of learners and feature sets, filter methods achieved better predictive performances than using no feature selection. Ensemble filters did not have a substantial impact on performance. The most important features were located around the red edge. Additional features in the near-infrared region (800–1000 nm) were also essential to achieve the overall best performances. Filter methods have the potential to be helpful in high-dimensional situations and are able to improve the interpretation of feature effects in fitted models, which is an essential constraint in environmental modeling studies. Nevertheless, more training data and replication in similar benchmarking studies are needed to be able to generalize the results.

The REDD+ framework requires accurate es- timates of deforestation. To derive these estimates, ground measurements are supported by methods based on remote sensing data to automatically detect and delineate deforesta- tions over large areas. In particular in the tropics, optical data is seldom available due to cloud cover. As Synthetic Aperture Radar (SAR) data overcomes this limitation, we performed a separability analysis of two statistical metrics based on the Sentinel-1 SAR signal over forested and deforested areas. We compared the range between the 5th and 95th temporal percentiles (PRange) and the Recurrence Quantification Analysis (RQA) Trend metric. In contrast to the PRange, the RQA Trend considers the temporal order of the SAR data acquisitions. This enables the estimation of the timing of deforestation events. We assessed the impact of polarisation, acquisition orbit, and despeckling on the separability between forested and deforested areas, and between different deforestation timings for two test sites in Mexico. We found that the choice of the orbit impacts the detectability of deforestation. In all cases VH data showed a higher separability between forest and deforestations than VV data. The PRange slightly outperformed RQA Trend in the separation between forest and deforestation. However, the RQA Trend exceeded the PRange in the separation between different deforestation timings. In this study C-Band data was used, although it is commonly not considered as the most suitable SAR frequency for forestry applications Nevertheless, our approach shows, the potential of dense C-Band time series to support the REDD+ framework.