Gaussian processes retrieval of crop traits in Google Earth Engine based on Sentinel-2 top-of-atmosphere data

Remote Sensing of Environment, Год журнала: 2022, Номер 273, С. 112958 - 112958

Опубликована: Март 4, 2022

The unprecedented availability of optical satellite data in cloud-based computing platforms, such as Google Earth Engine (GEE), opens new possibilities to develop crop trait retrieval models from the local planetary scale. Hybrid are interest run these platforms they combine advantages physically- based radiative transfer (RTM) with flexibility machine learning regression algorithms. Previous research GEE primarily relied on processing bottom-of-atmosphere (BOA) reflectance data, which requires atmospheric correction. In present study, we implemented hybrid directly into for Sentinel-2 (S2) Level-1C (L1C) top-of-atmosphere (TOA) traits. To achieve this, a training dataset was generated using leaf-canopy RTM PROSAIL combination model 6SV. Gaussian process (GPR) were then established eight essential traits namely leaf chlorophyll content, water dry matter fractional vegetation cover, area index (LAI), and upscaled variables (i.e., canopy content content). An important pre-requisite implementation is that sufficiently light order facilitate efficient fast processing. Successful reduction by 78% achieved active technique Euclidean distance-based diversity (EBD). With EBD-GPR models, highly accurate validation results LAI obtained against situ field study site Munich-North-Isar (MNI), normalized root mean square errors (NRMSE) 6% 13%. Using an independent similar types (Italian Grosseto test site), showed moderate good performances canopy-level variables, NRMSE ranging 14% 50%, but failed leaf-level estimates. Obtained maps over MNI further compared Level 2 Prototype Processor (SL2P) estimates ESA Sentinels' Application Platform (SNAP) Biophysical Processor, proving high consistency both retrievals (

Язык: Английский

---

Comparing and combining data-driven and model-driven approaches to monitor wheat green area index with high spatio-temporal resolution satellites

Remote Sensing of Environment, Год журнала: 2024, Номер 305, С. 114118 - 114118

Опубликована: Март 19, 2024

Monitoring crops with high spatio-temporal resolution satellites provides valuable observations to ensure food security in the global change context. This study focuses on estimating Green Area Index (GAI) monitor wheat a spatial of 3 m and daily satellite from SuperDove constellation. With an easier access large training datasets ground GAI measurements, improvement realism radiative transfer model simulations, choice optimal approach (data-driven or model-driven) constitutes key question when retrieving observations. compares data-driven model-driven estimate satellites. Both approaches are based Gaussian Process Regression (GPR) machine learning techniques. The uses over 300 measurements collected 12 sites China France, each 20 51 contrasting plots. 10,000 simulations top canopy reflectance corresponding values generated by LESS applied 3D scenes built ADEL-Wheat (Architectural Development L-systems) model. Results confirm that reliable consistent Sentinel-2 values. When using GPR reflectance, (R2 = 0.83, RMSE 0.80, Accuracy 0.01 Precision 0.80) generally outperforms 0.88, −0.13 0.87), except for small In-silico experiments show uncertainties ground-measured size diversity limit approach. In contrast, is mostly constrained particularly low Two ensemble solutions weighted average two previous then proposed: solution 0.86, 0.75, A −0.06 P 0.74) where weight assumed independent values, adaptive 0.85, 0.76, −0.08 0.76) depends perform similarly, improving both approaches. Finally, applying plots along growth cycle allows clear differentiation nitrogen modalities cultivar effects. However, minimum plot × (4 4 pixels) recommended minimize co-registration errors increase precision.

Язык: Английский

---

Estimating Crop Biophysical Parameters Using Machine Learning Algorithms and Sentinel-2 Imagery

Remote Sensing, Год журнала: 2021, Номер 13(21), С. 4314 - 4314

Опубликована: Окт. 27, 2021

Global food security is critical to eliminating hunger and malnutrition. In the changing climate, farmers in developing countries must adopt technologies farming practices such as precision agriculture (PA). PA-based approaches enable cope with frequent intensified droughts heatwaves, optimising yields, increasing efficiencies, reducing operational costs. Biophysical parameters Leaf Area Index (LAI), Chlorophyll Content (LCab), Canopy (CCC) are essential for characterising field-level spatial variability thus necessary enabling variable rate application technologies, irrigation, crop monitoring. Moreover, robust machine learning algorithms offer prospects improving estimation of biophysical due their capability deal non-linear data, small samples, noisy variables. This study compared predictive performance sparse Partial Least Squares (sPLS), Random Forest (RF), Gradient Boosting Machines (GBM) estimating LAI, LCab, CCC Sentinel-2 imagery Bothaville, South Africa identified, using importance measures, most influential bands parameters. The results showed that RF was superior all three parameters, followed by GBM which better LAI CCC, but not where sPLS relatively better. Since could be achieved RF, it can considered a good contender operationalisation. Overall, findings this significant future product development reduce reliance on many specific facilitating rapid extraction actionable information support PA monitoring activities.

Язык: Английский

---

Crop specific inversion of PROSAIL to retrieve green area index (GAI) from several decametric satellites using a Bayesian framework

Remote Sensing of Environment, Год журнала: 2022, Номер 278, С. 113085 - 113085

Опубликована: Июнь 2, 2022

Язык: Английский

---

Validation of Simplified Level 2 Prototype Processor Sentinel-2 fraction of canopy cover, fraction of absorbed photosynthetically active radiation and leaf area index products over North American forests

Remote Sensing of Environment, Год журнала: 2023, Номер 293, С. 113600 - 113600

Опубликована: Апрель 27, 2023

Canopy biophysical variables such as the fraction of canopy cover (fCOVER), absorbed photosynthetically active radiation (fAPAR), and leaf area index (LAI) are widely used for ecosystem modelling monitoring. The Sentinel-2 mission was designed systematic global mapping these at 20 m resolution using imagery from MultiSpectral Instrument. Simplified Level 2 Prototype Processor (SL2P) is available a baseline solution. Previous validation over limited sites indicates that SL2P generally satisfies user requirements all three crops, but underestimates LAI forests. In this study, fAPAR, fCOVER, products, SL2P, were validated 281 representative most North American forest ecozones also compared to Moderate Resolution Imaging Spectrometer (MODIS) Copernicus Global Land Service (CGLS) products. addition meeting Committee on Earth Observation Satellites Stage 3 areas, our study explores relationship between bias in products clumping provides empirical correction functions each variable. implemented within Landscape Evolution Forecasting Toolbox Google Engine both efficiency due bugs Sentinel Application Platform implementation. found underestimate by 20% 50% forests with > 2; agreement other studies comparisons MODIS CGLS fCOVER fAPAR transitions ∼0.1 low values ∼ − 0.1 high values. Precision error, one standard deviation, ∼0.5 slightly less than fAPAR. Total uncertainty dominated greater precision error Target satisfied 51% LAI, 37% 31% in-situ measurements. For variables, accuracy exhibited weak moderate linear relationships (r2 ≤0.52), scatter plots indicated larger negative biases northern latitude where canopies clumping. With exception evergreen broadleaf forests, data reduced 40% 57% and, 92% increased rate up 8%. Users recommended apply or consider recalibrating spatially heterogenous radiative transfer model simulations.

Язык: Английский

---

HemiPy: A Python module for automated estimation of forest biophysical variables and uncertainties from digital hemispherical photographs

Methods in Ecology and Evolution, Год журнала: 2023, Номер 14(9), С. 2329 - 2340

Опубликована: Авг. 9, 2023

Abstract Digital hemispherical photography (DHP) is widely used to derive forest biophysical variables including leaf, plant, and green area index (LAI, PAI GAI), the fraction of intercepted photosynthetically active radiation (FIPAR), vegetation cover (FCOVER). However, majority software packages for processing DHP data are based on a graphical user interface, making programmatic analysis difficult. Meanwhile, few natively support RAW image formats, while none incorporate propagation or provision uncertainties. To address these limitations, we present HemiPy, an open‐source Python module deriving uncertainties from images in automated manner. We assess HemiPy using simulated images, addition multiannual time‐series litterfall several forested National Ecological Observatory Network (NEON) sites, as well comparison against CAN‐EYE package. Multiannual PAI, FIPAR FCOVER demonstrate HemiPy's outputs realistically represent expected temporal patterns. Comparison reveals reasonable accuracies achievable, with RMSE values close error ~1 unit typically attributed optical LAI measurement approaches. good agreement CAN‐EYE. Consistent previous studies, when compared better observed derived gap near hinge angle 57.5° only, opposed over wider range zenith angles. should prove useful tool its nature means that it can be adopted, extended further refined by community.

Язык: Английский

---

Correcting confounding canopy structure, biochemistry and soil background effects improves leaf area index estimates across diverse ecosystems from Sentinel-2 imagery

Remote Sensing of Environment, Год журнала: 2024, Номер 309, С. 114224 - 114224

Опубликована: Май 28, 2024

High-spatiotemporal-resolution leaf area index (LAI) data are essential for sustainable agro-ecosystem management and precise disturbance detection. Previous LAI products were primarily derived from satellite with limited spatiotemporal or spectral resolutions, which could be overcome the use of Sentinel-2. While hybrid methods that integrate PROSAIL simulations machine learning offer advantages in extracting high-spatiotemporal-resolution Sentinel-2, they still face challenges due to confounding factors related canopy structure, biochemistry, soil background. To reduce impacts these confounders, we developed an efficient method Sentinel-2-based retrieval. Our approach consists random forest models trained on simulated datasets generated by PROSAIL-5B two refinements: variable fraction fully senescent leaves (FS) bidirectional reflectance factor (BRF) Brightness-Shape-Moisture (BSM) model. We corrected BRF using near-infrared vegetation (NIRV) cover within mixed pixels (VC). For validation, used ground measurements across different types Copernicus Ground Based Observations Validation (GBOV) Korea flux (KoFlux) sites during 2019–2023. results showed coupling BSM FS improved estimates, reducing RMSE 10.8%–73.8%. Utilizing NIRV VC correct better quantified most types, reduced 15.3%–64.8%. robust agreement validation GBOV (R2 = 0.88, 0.71) KoFlux 0.80, 0.75). Overall, our 0.58–0.93, 0.04–0.83) outperformed both benchmark Sentinel Application Platform 0.11–0.85, 0.28–1.67) data-driven 0.09–0.85, 0.29–0.93) algorithms producing seasonal at finer resolutions. findings underscore potential proposed retrieval diverse ecosystems.

Язык: Английский

---

Advancements in high-resolution land surface satellite products: A comprehensive review of inversion algorithms, products and challenges

Science of Remote Sensing, Год журнала: 2024, Номер 10, С. 100152 - 100152

Опубликована: Июль 27, 2024

For many applications, raw satellite observations need to be converted high-level products of various essential environmental variables. While numerous are available at kilometer spatial resolutions, there few global high resolutions (10–30 m), which also referred fine or medium in the literature. To facilitate development more resolution products, this paper systematically reviews state-of-the-art progress on inversion algorithms and publicly regional products. We begin with an inventory high-resolution data, then present different for determining cloud masks, estimating aerosol optical depth, performing atmospheric correction topographic land surface reflectance retrieval. The majority existing 18 variables four major categories: 1) Land radiation, including broadband albedo, temperature, all-wave net radiation; 2) Terrestrial ecosystem variables, leaf area index, fraction absorbed photosynthetically active fractional vegetation cover, forest tree height, above-ground biomass gross primary production, agricultural crop yield; 3) Water cycle cryosphere, soil moisture, evapotranspiration, snow cover; 4) types, such as impervious surface, inland water, type, fire. Since over large regions usually spatially discontinuous due contamination, data fusion assimilation some producing seamless temporally continuous presented. In end, we discuss a variety challenges generating

Язык: Английский

---

Evidence of a bias-variance trade off when correcting for bias in Sentinel 2 forest LAI retrievals using radiative transfer models

Remote Sensing of Environment, Год журнала: 2024, Номер 305, С. 114060 - 114060

Опубликована: Март 1, 2024

Forest canopies exhibit spatial heterogeneity that impacts the relationship between essential climate variables such as leaf area index (LAI) or fraction of absorbed photosynthetically active radiation (fAPAR) and bi-directional surface reflectance, subsequently estimation these from satellite measurements. The Simplified Level 2 Prototype Processor (SL2P) allows global LAI fAPAR mapping at 20 m resolution using Sentinel imagery. Previous validation studies over forests indicate SL2P underestimates by up to 50% in comparison in-situ reference Our study tests hypothesis bias can be reduced replacing spatially homogenous SAILH canopy radiative transfer model used calibrate with heterogenous 4SAIL2 model, together a shoot clumping parameterization. We also hypothesized additional parameters involved this new version (SL2P-CCRS) would lead an increase precision error bias-variance trade off. SL2P-CCRS 65%, SL2P, during direct 1107 absolute ∼0.5 3 ∼1 6. 31% compared but <0.05 on basis. Bias reduction was accompanied so overall uncertainty, quantified root mean square difference measurements, only 6% for 5% fAPAR. These findings support updating heterogeneous RTM reduce forests. results there is trade-off LAI, lesser extent fAPAR, when increasing complexity accounts heterogeneity. Nevertheless, increased agreement rate Global Climate Observing System uncertainty requirements 52% 58% 32% 40% suggesting worthwhile, algorithms SL2P-CCRS, use should applied Sentinel-2

Язык: Английский

---

HiQ-LAI: a high-quality reprocessed MODIS leaf area index dataset with better spatiotemporal consistency from 2000 to 2022

Earth system science data, Год журнала: 2024, Номер 16(3), С. 1601 - 1622

Опубликована: Март 26, 2024

Abstract. Leaf area index (LAI) is a crucial parameter for characterizing vegetation canopy structure and energy absorption capacity. The Moderate Resolution Imaging Spectroradiometer (MODIS) LAI has played significant role in landmark studies due to its clear theoretical basis, extensive historical time series, validation results, open accessibility. However, MODIS retrievals are calculated independently each pixel specific day, resulting high noise levels the series limiting applications regions of optical remote sensing. Reprocessing predominantly relies on temporal information achieve smoother profiles with little use spatial may easily ignore genuine anomalies. To address these problems, we designed spatiotemporal compositing algorithm (STICA) reprocessing products. This method integrates from multiple dimensions, including quality information, correlation, original retrieval, thereby enabling both “reprocessing” “value-added data” respect existing products, leading development high-quality (HiQ-LAI) dataset. Compared ground measurements, HiQ-LAI shows better performance than product root-mean-square error (RMSE) or bias decrease 0.87 −0.17 0.78 −0.06, respectively. improvement capturing seasonality phenology reducing abnormal time-series fluctuations. stability (TSS) index, which represents stability, indicated that smooth expanded 31.8 % 78.8 (HiQ) globally, this more obvious equatorial where sensing cannot usually good performance. We found demonstrates superior continuity consistency compared raw perspectives. anticipate global generated using STICA procedure Google Earth Engine (GEE) platform, will substantially enhance support diverse applications. 5 km 8 d datasets 2000 2022 available at https://doi.org/10.5281/zenodo.8296768 (Yan et al., 2023).

Язык: Английский

---