DATimeS: A machine learning time series GUI toolbox for gap-filling and vegetation phenology trends detection

Greening of Svalbard

Svalbard, located between 76°30'N and 80°50'N, is among the regions in the world with the most rapid temperature increase. We processed a cloud-free time-series of MODIS-NDVI for Svalbard. The dataset is interpolated to daily data during the 2000-2022 period with 232 m pixel resolution. The onset of growth, with a clear phenological definition, has been mapped each year. Then the integrated NDVI from the onset (O) of growth each year to the time of average (2000-2022) peak (P) of growth (OP NDVI) have been calculated. OP NDVI has previously shown high correlation with field-based tundra productivity. Daily mean temperature data from 11 meteorological stations are compared with the NDVI data. The OP NDVI values show very high and significant correlation with growing degree days computed from onset to time of peak of growth for all the meteorological stations used. On average for the entire Svalbard, the year 2016 first had the highest greening (OP NDVI values) recorded since the year 2000, then the greening in 2018 surpassed 2016, then 2020 surpassed 2018, and finally 2022 was the year with the overall highest greening by far for the whole 2000-2022 period. This shows a rapid recent greening of Svalbard very strongly linked to temperature increase, although there are regional differences: the eastern parts of Svalbard show the largest variability between years, most likely due to variability in the timing of sea-ice break-up in adjacent areas. Finally, we find that areas dominated by manured moss-tundra in the polar desert zone require new methodologies, as moss does not share the seasonal NDVI dynamics of tundra communities.

Assessment of Phenological Dynamics of Different Vegetation Types and Their Environmental Drivers with Near-Surface Remote Sensing: A Case Study on the Loess Plateau of China

Plant phenology is an important indicator of the impact of climate change on ecosystems. We have continuously monitored vegetation phenology using near-surface remote sensing, i.e., the PhenoCam in a gully region of the Loess Plateau of China from March 2020 to November 2022. In each image, three regions of interest (ROIs) were selected to represent different types of vegetation (scrub, arbor, and grassland), and five vegetation indexes were calculated within each ROI. The results showed that the green chromatic coordinate (GCC), excess green index (ExG), and vegetation contrast index (VCI) all well-captured seasonal changes in vegetation greenness. The PhenoCam captured seasonal trajectories of different vegetation that reflect differences in vegetation growth. Such differences may be influenced by external abiotic environmental factors. We analyzed the nonlinear response of the GCC series to environmental variables with the generalized additive model (GAM). Our results suggested that soil temperature was an important driver affecting plant phenology in the Loess gully region, especially the scrub showed a significant nonlinear response to soil temperature change. Since in situ phenology monitoring experiments of the small-scale on the Loess Plateau are still relatively rare, our work provides a reference for further understanding of vegetation phenological variations and ecosystem functions on the Loess Plateau.

Changes in Onset of Vegetation Growth on Svalbard, 2000-2020

The global temperature is increasing, and this is affecting the vegetation phenology in many parts of the world. The most prominent changes occur at northern latitudes such as our study area, which is Svalbard, located between 76°30'N and 80°50'N. A cloud-free time series of MODIS-NDVI data was processed. The dataset was interpolated to daily data during the 2000-2020 period with a 231.65 m pixel resolution. The onset of vegetation growth was mapped with a NDVI threshold method which corresponds well with a recent Sentinel-2 NDVI-based mapping of the onset of vegetation growth, which was in turn validated by a network of in-situ phenological data from time lapse cameras. The results show that the years 2000 and 2008 were extreme in terms of the late onset of vegetation growth. The year 2020 had the earliest onset of vegetation growth on Svalbard during the 21-year study. Each year since 2013 had an earlier or equally early timing in terms of the onset of the growth season compared with the 2000-2020 average. A linear trend of 0.57 days per year resulted in an earlier onset of growth of 12 days on average for the entire archipelago of Svalbard in 2020 compared to 2000.

Concentrated Stream Data Processing for Vegetation Coverage Monitoring and Recommendation against Rock Desertification

The vegetation covering regions is confined due to deforestation, mining industries, and environmental factors. The intensified deforestation and industrial development processes impact the vegetation coverage and fail to meet the food demands. Therefore, accurate monitoring of such regions aids in preventing adversary processes and their plant extinction. The monitoring process requires accurate data collection and analysis to identify the root cause that can be due to human/climatic/environmental changes. This article introduces a concentrated stream data processing method (CSDPM) assisted by an extreme learning paradigm. The different causes are analyzed using the extracted features in different learning perceptron layers. In this learning, the accumulated data is analyzed for similar features and trained for the consecutive or lagging input data streams. The monitoring process concluded with the learning output by classifying the plant extinction reason. Therefore, the identified reason is addressed through official policies with new recommendations or alternate vegetation improvements. More specifically, the data concentrated towards deforestation are the fundamental data required for feature matching. The features are initially trained from the existing datasets and previously acquired data from the converted landscapes. This proposed method is analyzed using the metrics analysis rate, analysis time, recommendation rate, and complexity.

Linking Land Use and Plant Functional Diversity Patterns in Sabah, Borneo, through Large-Scale Spatially Continuous Sentinel-2 Inference

Global biodiversity losses erode the functioning of our vital ecosystems. Functional diversity is increasingly recognized as a critical link between biodiversity and ecosystem functioning. Satellite earth observation was proposed to address the current absence of information on large-scale continuous patterns of plant functional diversity. This study demonstrates the inference and spatial mapping of functional diversity metrics through satellite remote sensing over a large key biodiversity region (Sabah, Malaysian Borneo, ~53,000 km2) and compares the derived estimates across a land-use gradient as an initial qualitative assessment to test the potential merits of the approach. Functional traits (leaf water content, chlorophyll-a and -b, and leaf area index) were estimated from Sentinel-2 spectral reflectance using a pre-trained neural network on radiative transfer modeling simulations. Multivariate functional diversity metrics were calculated, including functional richness, divergence, and evenness. Spatial patterns of functional diversity were related to land-use data distinguishing intact forest, logged forest, and oil palm plantations. Spatial patterns of satellite remotely sensed functional diversity are significantly related to differences in land use. Intact forests, as well as logged forests, featured consistently higher functional diversity compared to oil palm plantations. Differences were profound for functional divergence, whereas functional richness exhibited relatively large variances within land-use classes. By linking large-scale patterns of functional diversity as derived from satellite remote sensing to land-use information, this study indicated initial responsiveness to broad human disturbance gradients over large geographical and spatially contiguous extents. Despite uncertainties about the accuracy of the spatial patterns, this study provides a coherent early application of satellite-derived functional diversity toward further validation of its responsiveness across ecological gradients.

Multi-Season Phenology Mapping of Nile Delta Croplands Using Time Series of Sentinel-2 and Landsat 8 Green LAI

Space-based cropland phenology monitoring substantially assists agricultural managing practices and plays an important role in crop yield predictions. Multitemporal satellite observations allow analyzing vegetation seasonal dynamics over large areas by using vegetation indices or by deriving biophysical variables. The Nile Delta represents about half of all agricultural lands of Egypt. In this region, intensifying farming systems are predominant and multi-cropping rotations schemes are increasing, requiring a high temporal and spatial resolution monitoring for capturing successive crop growth cycles. This study presents a workflow for cropland phenology characterization and mapping based on time series of green Leaf Area Index (LAI) generated from NASA's Harmonized Landsat 8 (L8) and Sentinel-2 (S2) surface reflectance dataset from 2016 to 2019. LAI time series were processed for each satellite dataset, which were used separately and combined to identify seasonal dynamics for a selection of crop types (wheat, clover, maize and rice). For the combination of L8 with S2 LAI products, we proposed two time series smoothing and fitting methods: (1) the Savitzky-Golay (SG) filter and (2) the Gaussian Processes Regression (GPR) fitting function. Single-sensor and L8-S2 combined LAI time series were used for the calculation of key crop Land Surface Phenology (LSP) metrics (start of season, end of season, length of season), whereby the detection of cropland growing seasons was based on two established threshold methods, i.e., a seasonal or a relative amplitude value. Overall, the developed phenology extraction scheme enabled identifying up to two successive crop cycles within a year, with a superior performance observed for the seasonal than for the relative threshold method, in terms of consistency and cropland season detection capability. Differences between the time series collections were analyzed by comparing the phenology metrics per crop type and year. Results suggest that L8-S2 combined LAI data streams with GPR led to a more precise detection of the start and end of growing seasons for most crop types, reaching an overall detection of 74% over the total planted crops versus 69% with S2 and 63% with L8 alone. Finally, the phenology mapping allowed us to evaluate the spatial and temporal evolution of the croplands over the agroecosystem in the Nile Delta.

An NDVI Retrieval Method Based on a Double-Attention Recurrent Neural Network for Cloudy Regions

NDVI is an important parameter for environmental assessment and precision agriculture that well-describes the status of vegetation. Nevertheless, the clouds in optical images often result in the absence of NDVI information at key growth stages. The integration of SAR and optical image features will likely address this issue. Although the mapping of different data sources is complex, the prosperity of deep learning technology provides an alternative approach. In this study, the double-attention RNN architecture based on the recurrent neural network (RNN) and attention mechanism is proposed to retrieve NDVI data of cloudy regions. Overall, the NDVI is retrieved by the proposed model from two aspects: the temporal domain and the pixel neighbor domain. The performance of the double-attention RNN is validated through different cloud coverage conditions, input ablation, and comparative experiments with various methods. The results conclude that a high retrieval accuracy is guaranteed by the proposed model, even under high cloud coverage conditions (R2 = 0.856, RMSE = 0.124). Using SAR images independently results in poor NDVI retrieval results (R2 = 0.728, RMSE = 0.141) with considerable artifacts, which need to be addressed with auxiliary data, such as IDM features. Temporal and pixel neighbor features play an important role in improving the accuracy of NDVI retrieval (R2 = 0.894, RMSE = 0.096). For the missing values of NDVI data caused by cloud coverage, the double-attention RNN proposed in this study provides a potential solution for information recovery.

Quantifying Fundamental Vegetation Traits over Europe Using the Sentinel-3 OLCI Catalogue in Google Earth Engine

Thanks to the emergence of cloud-computing platforms and the ability of machine learning methods to solve prediction problems efficiently, this work presents a workflow to automate spatiotemporal mapping of essential vegetation traits from Sentinel-3 (S3) imagery. The traits included leaf chlorophyll content (LCC), leaf area index (LAI), fraction of absorbed photosynthetically active radiation (FAPAR), and fractional vegetation cover (FVC), being fundamental for assessing photosynthetic activity on Earth. The workflow involved Gaussian process regression (GPR) algorithms trained on top-of-atmosphere (TOA) radiance simulations generated by the coupled canopy radiative transfer model (RTM) SCOPE and the atmospheric RTM 6SV. The retrieval models, named to S3-TOA-GPR-1.0, were directly implemented in Google Earth Engine (GEE) to enable the quantification of the traits from TOA data as acquired from the S3 Ocean and Land Colour Instrument (OLCI) sensor.Following good to high theoretical validation results with normalized root mean square error (NRMSE) ranging from 5% (FAPAR) to 19% (LAI), a three fold evaluation approach over diverse sites and land cover types was pursued: (1) temporal comparison against LAI and FAPAR products obtained from Moderate Resolution Imaging Spectroradiometer (MODIS) for the time window 2016-2020, (2) spatial difference mapping with Copernicus Global Land Service (CGLS) estimates, and (3) direct validation using interpolated in situ data from the VALERI network. For all three approaches, promising results were achieved. Selected sites demonstrated coherent seasonal patterns compared to LAI and FAPAR MODIS products, with differences between spatially averaged temporal patterns of only 6.59%. In respect of the spatial mapping comparison, estimates provided by the S3-TOA-GPR-1.0 models indicated highest consistency with FVC and FAPAR CGLS products. Moreover, the direct validation of our S3-TOA-GPR-1.0 models against VALERI estimates indicated with regard to jurisdictional claims in good retrieval performance for LAI, FAPAR and FVC. We conclude that our retrieval workflow of spatiotemporal S3 TOA data processing into GEE opens the path towards global monitoring of fundamental vegetation traits, accessible to the whole research community.

Monitoring Cropland Phenology on Google Earth Engine Using Gaussian Process Regression

Monitoring cropland phenology from optical satellite data remains a challenging task due to the influence of clouds and atmospheric artifacts. Therefore, measures need to be taken to overcome these challenges and gain better knowledge of crop dynamics. The arrival of cloud computing platforms such as Google Earth Engine (GEE) has enabled us to propose a Sentinel-2 (S2) phenology end-to-end processing chain. To achieve this, the following pipeline was implemented: (1) the building of hybrid Gaussian Process Regression (GPR) retrieval models of crop traits optimized with active learning, (2) implementation of these models on GEE (3) generation of spatiotemporally continuous maps and time series of these crop traits with the use of gap-filling through GPR fitting, and finally, (4) calculation of land surface phenology (LSP) metrics such as the start of season (SOS) or end of season (EOS). Overall, from good to high performance was achieved, in particular for the estimation of canopy-level traits such as leaf area index (LAI) and canopy chlorophyll content, with normalized root mean square errors (NRMSE) of 9% and 10%, respectively. By means of the GPR gap-filling time series of S2, entire tiles were reconstructed, and resulting maps were demonstrated over an agricultural area in Castile and Leon, Spain, where crop calendar data were available to assess the validity of LSP metrics derived from crop traits. In addition, phenology derived from the normalized difference vegetation index (NDVI) was used as reference. NDVI not only proved to be a robust indicator for the calculation of LSP metrics, but also served to demonstrate the good phenology quality of the quantitative trait products. Thanks to the GEE framework, the proposed workflow can be realized anywhere in the world and for any time window, thus representing a shift in the satellite data processing paradigm. We anticipate that the produced LSP metrics can provide meaningful insights into crop seasonal patterns in a changing environment that demands adaptive agricultural production.

Assessing Non-Photosynthetic Cropland Biomass from Spaceborne Hyperspectral Imagery

Non-photosynthetic vegetation (NPV) biomass has been identified as a priority variable for upcoming spaceborne imaging spectroscopy missions, calling for a quantitative estimation of lignocellulosic plant material as opposed to the sole indication of surface coverage. Therefore, we propose a hybrid model for the retrieval of non-photosynthetic cropland biomass. The workflow included coupling the leaf optical model PROSPECT-PRO with the canopy reflectance model 4SAIL, which allowed us to simulate NPV biomass from carbon-based constituents (CBC) and leaf area index (LAI). PROSAIL-PRO provided a training database for a Gaussian process regression (GPR) algorithm, simulating a wide range of non-photosynthetic vegetation states. Active learning was employed to reduce and optimize the training data set. In addition, we applied spectral dimensionality reduction to condense essential information of non-photosynthetic signals. The resulting NPV-GPR model was successfully validated against soybean field data with normalized root mean square error (nRMSE) of 13.4% and a coefficient of determination (R2) of 0.85. To demonstrate mapping capability, the NPV-GPR model was tested on a PRISMA hyperspectral image acquired over agricultural areas in the North of Munich, Germany. Reliable estimates were mainly achieved over senescent vegetation areas as suggested by model uncertainties. The proposed workflow is the first step towards the quantification of non-photosynthetic cropland biomass as a next-generation product from near-term operational missions, such as CHIME.

Methods for interpolating missing data in aerobiological databases

Missing data is a common problem in scientific research. The availability of extensive environmental time series is usually laborious and difficult, and sometimes unexpected failures are not detected until samples are processed. Consequently, environmental databases frequently have some gaps with missing data in it. Applying an interpolation method before starting the data analysis can be a good solution in order to complete this missing information. Nevertheless, there are several different approaches whose accuracy should be considered and compared. In this study, data from 6 aerobiological sampling stations were used as an example of environmental data series to assess the accuracy of different interpolation methods. For that, observed daily pollen/spore concentration data series were randomly removed, interpolated by using different methods and then, compared with the observed data to measure the errors produced. Different periods, gap sizes, interpolation methods and bioaerosols were considered in order to check their influence in the interpolation accuracy. The moving mean interpolation method obtained the highest success rate as average. By using this method, a success rate of the 70% was obtained when the risk classes used in the alert systems of the pollen information platforms were taken into account. In general, errors were mostly greater when there were high oscillations in the concentrations of biotic particles during consecutive days. That is the reason why the pre-peak and peak periods showed the highest interpolation errors. The errors were also higher when gaps longer than 5 days were considered. So, for completing long periods of missing data, it would be advisable to test other methodological approaches. A new Variation Index based on the behaviour of the pollen/spore season (measurement of the variability of the concentrations every 2 consecutive days) was elaborated, which allows to estimate the potential error before the interpolation is applied.

Empirical Approach for Modelling Tree Phenology in Mixed Forests Using Remote Sensing

Phenological events are good indicators of the effects of climate change, since phenological phases are sensitive to changes in environmental conditions. Although several national phenological networks monitor the phenology of different plant species, direct observations can only be conducted on individual trees, which cannot be easily extended over large and continuous areas. Remote sensing has often been applied to model phenology for large areas, focusing mostly on pure forests in which it is relatively easier to match vegetation indices with ground observations. In mixed forests, phenology modelling from remote sensing is often limited to land surface phenology, which consists of an overall phenology of all tree species present in a pixel. The potential of remote sensing for modelling the phenology of individual tree species in mixed forests remains underexplored. In this study, we applied the seasonal midpoint (SM) method with MODIS GPP to model the start of season (SOS) and the end of season (EOS) of six different tree species in Slovenian mixed forests. First, substitute locations were identified for each combination of observation station and plant species based on similar environmental conditions (aspect, slope, and altitude) and tree species of interest, and used to retrieve the remote sensing information used in the SM method after fitting the best of a Gaussian and two double logistic functions to each year of GPP time series. Then, the best thresholds were identified for SOS and EOS, and the results were validated using cross-validation. The results show clearly that the usual threshold of 0.5 is not best in most cases, especially for estimating the EOS. Despite the difficulty in modelling the phenology of different tree species in a mixed forest using remote sensing, it was possible to estimate SOS and EOS with moderate errors as low as <8 days (Fagus sylvatica and Tilia sp.) and <10 days (Fagus sylvatica and Populus tremula), respectively.

Non-Parametric Statistical Approaches for Leaf Area Index Estimation from Sentinel-2 Data: A Multi-Crop Assessment

The leaf area index (LAI) is a key biophysical variable for agroecosystem monitoring, as well as a relevant state variable in crop modelling. For this reason, temporal and spatial determination of LAI are required to improve the understanding of several land surface processes related to vegetation dynamics and crop growth. Despite the large number of retrieved LAI products and the efforts to develop new and updated algorithms for LAI estimation, the available products are not yet capable of capturing site-specific variability, as requested in many agricultural applications. The objective of this study was to evaluate the potential of non-parametric approaches for multi-temporal LAI retrieval by Sentinel-2 multispectral data, in comparison with a VI-based parametric approach. For this purpose, we built a large database combining a multispectral satellite data set and ground LAI measurements collected over two growing seasons (2018 and 2019), including three crops (i.e., winter wheat, maize, and alfalfa) characterized by different growing cycles and canopy structures, and considering different agronomic conditions (i.e., at three farms in three different sites). The accuracy of parametric and non-parametric methods for LAI estimation was assessed by cross-validation (CV) at both the pixel and field levels over mixed-crop (MC) and crop-specific (CS) data sets. Overall, the non-parametric approach showed a higher accuracy of prediction at pixel level than parametric methods, and it was also observed that Gaussian Process Regression (GPR) did not provide any significant difference (p-value > 0.05) between the predicted values of LAI in the MC and CS data sets, regardless of the crop. Indeed, GPR at the field level showed a cross-validated coefficient of determination (R2CV) higher than 0.80 for all three crops.

Green LAI Mapping and Cloud Gap-Filling Using Gaussian Process Regression in Google Earth Engine

For the last decade, Gaussian process regression (GPR) proved to be a competitive machine learning regression algorithm for Earth observation applications, with attractive unique properties such as band relevance ranking and uncertainty estimates. More recently, GPR also proved to be a proficient time series processor to fill up gaps in optical imagery, typically due to cloud cover. This makes GPR perfectly suited for large-scale spatiotemporal processing of satellite imageries into cloud-free products of biophysical variables. With the advent of the Google Earth Engine (GEE) cloud platform, new opportunities emerged to process local-to-planetary scale satellite data using advanced machine learning techniques and convert them into gap-filled vegetation properties products. However, GPR is not yet part of the GEE ecosystem. To circumvent this limitation, this work proposes a general adaptation of GPR formulation to parallel processing framework and its integration into GEE. To demonstrate the functioning and utility of the developed workflow, a GPR model predicting green leaf area index (LAI G ) from Sentinel-2 imagery was imported. Although by running this GPR model into GEE any corner of the world can be mapped into LAI G at a resolution of 20 m, here we show some demonstration cases over western Europe with zoom-ins over Spain. Thanks to the computational power of GEE, the mapping takes place on-the-fly. Additionally, a GPR-based gap filling strategy based on pre-optimized kernel hyperparameters is also put forward for the generation of multi-orbit cloud-free LAI G maps with an unprecedented level of detail, and the extraction of regularly-sampled LAI G time series at a pixel level. The ability to plugin a locally-trained GPR model into the GEE framework and its instant processing opens up a new paradigm of remote sensing image processing.

Gap-Free Monitoring of Annual Mangrove Forest Dynamics in Ca Mau Province, Vietnamese Mekong Delta, Using the Landsat-7-8 Archives and Post-Classification Temporal Optimization

Ecosystem services offered by mangrove forests are facing severe risks, particularly through land use change driven by human development. Remote sensing has become a primary instrument to monitor the land use dynamics surrounding mangrove ecosystems. Where studies formerly relied on bi-temporal assessments of change, the practical limitations concerning data-availability and processing power are slowly disappearing with the onset of high-performance computing (HPC) and cloud-computing services, such as in the Google Earth Engine (GEE). This paper combines the capabilities of GEE, including its entire Landsat-7 and Landsat-8 archives and state-of-the-art classification approaches, with a post-classification temporal analysis to optimize land use classification results into gap-free and consistent information. The results demonstrate its application and value to uncover the spatio-temporal dynamics of mangrove forests and land use changes in Ngoc Hien District, Ca Mau province, Vietnamese Mekong delta. The combination of repeated GEE classification output and post-classification optimization provides valid spatial classification (94–96% accuracy) and temporal interpolation (87–92% accuracy). The findings reveal that the net change of mangroves forests over the 2001–2019 period equals −0.01% annually. The annual gap-free maps enable spatial identification of hotspots of mangrove forest changes, including deforestation and degradation. Post-classification temporal optimization allows for an exploitation of temporal patterns to synthesize and enhance independent classifications towards more robust gap-free spatial maps that are temporally consistent with logical land use transitions. The study contributes to a growing body of work advocating full exploitation of temporal information in optimizing land cover classification and demonstrates its use for mangrove forest monitoring.

Optimizing Gaussian Process Regression for Image Time Series Gap-Filling and Crop Monitoring

Image processing entered the era of artificial intelligence, and machine learning algorithms emerged as attractive alternatives for time series data processing. Satellite image time series processing enables crop phenology monitoring, such as the calculation of start and end of season. Among the promising algorithms, Gaussian process regression (GPR) proved to be a competitive time series gap-filling algorithm with the advantage of, as developed within a Bayesian framework, providing associated uncertainty estimates. Nevertheless, the processing of time series images becomes computationally inefficient in its standard per-pixel usage, mainly for GPR training rather than the fitting step. To mitigate this computational burden, we propose to substitute the per-pixel optimization step with the creation of a cropland-based precalculations for the GPR hyperparameters θ . To demonstrate our approach hardly affects the accuracy in fitting, we used Sentinel-2 LAI time series over an agricultural region in Castile and Leon, North-West Spain. The performance of image reconstructions were compared against the standard per-pixel GPR time series processing. Results showed that accuracies were on the same order (RMSE 0.1767 vs. 0.1564 [m2/m2], 12% RMSE degradation) whereas processing time accelerated about 90 times. We further evaluated the alternative option of using the same hyperparameters for all the pixels within the complete scene. It led to similar overall accuracies over crop areas and computational performance. Crop phenology indicators were also calculated for the three different approaches and compared. Results showed analogous crop temporal patterns, with differences in start and end of growing season of no more than five days. To the benefit of crop monitoring applications, all the gap-filling and phenology indicators retrieval techniques have been implemented into the freely downloadable GUI toolbox DATimeS.