Gaofei Yin

Satellite-observed solar-induced chlorophyll fluorescence (SIF) provides a novel observational constraint for monitoring photosynthetic phenology. However, phenological estimates derived from different SIF processing schemes remain difficult to compare directly, and the relative impacts of observation geometry, temporal aggregation, and canopy escape correction remain unclear. Using TROPOMI far-red SIF observations from 2019 to 2021, we constructed four 0.1° gridded products across vegetated regions of the Northern Hemisphere: daily canopy SIF, 8-day averaged canopy SIF, nadir-normalized daily canopy SIF, and escape-corrected 8-day SIF. Within a unified framework integrating Whittaker time-series reconstruction and double logistic phenology extraction, we quantified the impacts of angular effects, temporal aggregation, and canopy escape correction on the start of season (SOS), peak of season (POS), end of season (EOS), and length of season (LOS) through three paired comparisons. Results showed that angular effects and temporal aggregation caused relatively small phenological changes, with area-weighted mean changes generally within ±1.5 d and mean absolute differences generally ranging from 2 to 6 d. In contrast, canopy escape correction contributed the largest processing-induced differences: SOS and POS were delayed by 6.9 and 5.0 d, respectively, whereas EOS occurred 5.3 d earlier, and LOS was shortened by 12.2 d. The mean absolute difference in LOS reached 21.0 d and remained 20.6 d when comparisons were restricted to common valid pixels. This influence exhibited pronounced latitudinal and vegetation-type dependence, with the largest LOS shortening (39.2 d) occurring in low-latitude regions. Seasonal curve analysis further indicated that canopy escape correction primarily modified the curve structure near the peak; in some pixels, POS shifted by up to 18.0 d, whereas SOS and EOS changed by only 2.5 and 1.2 d, respectively. It should be noted that this study does not aim to validate the absolute accuracy of phenological estimates using ground observations; rather, it focuses on quantifying the relative impacts of different SIF processing schemes on photosynthetic phenology estimates. Our results demonstrate that canopy escape correction exerts the strongest influence on differences among TROPOMI SIF-derived phenological estimates, whereas angular effects and temporal aggregation play secondary roles. These findings provide a basis for comparing, interpreting, and assessing uncertainties in satellite SIF-based photosynthetic phenology products.

1. In the context of global climate change and ecological sustainability, it is critical to understand how natural disasters disturb and reshape terrestrial ecosystem functions (Attiwill 1994; Dobo...

Pixel-scale reference ”truth” serves as a critical link between in situ measurements and satellite observations. However, owing to surface heterogeneity, the upscaling of single-site observations to the satellite pixel scale has long been constrained by poorly understood mechanisms of heterogeneity effects, inadequate characterization of nonlinear responses, and insufficient utilization of spatial information. Existing methods, particularly machine learning approaches, generally adopt a direct point-to-pixel mapping strategy that implicitly assumes mutual independence among subpixels within a coarse pixel. This assumption conflicts with the spatial autocorrelation inherent in a spatially continuous land surface, thereby introducing systematic biases into upscaling results over heterogeneous land surfaces. To address this challenge, we propose a site-to-pixel upscaling method that integrates environmental attribute modeling with spatial context learning. The core of this method is a parallel dual-branch architecture comprising a multilayer deep neural network and a convolutional neural network. The former establishes a nonlinear mapping between environmental attributes and the target variable, whereas the latter extracts multiscale spatial heterogeneity patterns within the coarse pixel. The features from the two branches are then adaptively fused through a gating mechanism, followed by spatial aggregation using a point spread function. Evaluation across 416 globally distributed sites encompassing 15 land cover types shows that the model achieves R² values of 0.802–0.997 and RMSE values of 0.007–0.032. At sites jointly affected by spatial heterogeneity and topographic complexity, the median RMSE is more than 30% lower than that of the best-performing benchmark model. Ablation experiments show that the combined mechanism of per-subpixel modeling and residual learning is the primary source of model accuracy improvement, reducing RMSE by more than 60%. The complete dual-branch model accurately reproduces the spatial texture and spatial variability patterns within coarse pixels. Using MCD43A3 product validation as an example, the upscaled reference reduces validation RMSE by 10.2%–22.4%; at highly heterogeneous sites, Bias is reduced by 85.7%. This study demonstrates that the proposed method effectively enhances the ability of in situ observations to represent pixel-scale conditions over heterogeneous surfaces, providing a reliable and cost-effective solution for generating high-quality pixel-scale reference ”truth” and a fundamental basis for remote sensing product validation and the construction of pixel-scale training samples.

Bridging the point-to-pixel scale mismatch remains a central challenge in soil moisture (SM) product validation when only a limited number of in situ sites are available. Conventional arithmetic averaging gives equal importance to site observations without explicitly accounting for site-specific error magnitudes, intersite error dependence, or uncertainty in the resulting pixel-scale reference. Rather than pursuing higher point-estimate accuracy, this study emphasizes the explicit statistical characterization of observation errors. We propose a Bayesian upscaling framework that explicitly characterizes these error structures through a regularized observation error covariance matrix incorporated into the likelihood function. The contributions of calibrated observations to the posterior estimate are thereby determined jointly by their error variances and intersite covariances, while the framework simultaneously yields a pixel-scale posterior estimate and a quantified measure of its uncertainty. The method was evaluated using 56,950 valid three-site combinations derived from 72 sites in the Qinghai Lake Basin. Results show that the method reduces reference bias for site combinations with low representativeness by approximately 66.3%. Across 24 SM time series from eight products, the deviations in correlation coefficients and error metrics relative to the benchmark reference are reduced by approximately 58%, while propagation of posterior reference uncertainty enabled the uncertainty in the validation metrics to be explicitly quantified. Without adding sites, this framework improves the statistical properties of the upscaling process by reusing the error information inherent in existing observations, thereby providing a practical Bayesian bridge from sparse point measurements to defensible pixel-scale reference with quantifiable confidence under routine, site-limited validation conditions.

The validation of soil moisture (SM) products has long been constrained by the point-to-pixel scale mismatch, limiting their utility for hydrological applications. In sparse in situ networks, single-site measurements lack sufficient representativeness at the pixel scale, and their direct use for validation inevitably introduces non-negligible representativeness errors. Existing upscaling approaches either rely on densely distributed observations, limiting their generalizability, or directly extrapolate machine-learning-based spatial estimates, which risks propagating model biases into validation metrics. To address this, we propose a single-site upscaling approach constrained by intra-pixel spatial variability. The core idea is to use spatial variability information to guide the estimation of upscaling conversion coefficients. Specifically, we integrate multi-source remote sensing auxiliary variables with automated machine learning (AutoML) to construct a high-resolution SM trend surface that characterizes intra-pixel spatial variations and quantifies site representativeness. We further design a progressive estimation strategy, from OLS to AOLS and SW-AOLS, to optimize the point-to-pixel conversion stepwise. Results show that upscaling performance is closely tied to site representativeness. SW-AOLS reduced median representativeness errors by approximately 57% and 64% for moderate- and low-representativeness sites, respectively. Validation across 24 SM products, including L/C/X-band, active-passive fusion, and reanalysis products, demonstrates that the upscaled single-site references closely approximate pixel-scale ground "truth" from dense networks. For moderate- and low-representativeness sites, the upscaling approach reduced validation errors by 72% and 89%, and uncertainty by 42% and 60%, respectively. Without requiring additional ground observations, this study enhances validation reliability under sparse conditions and supports more reliable SM information for hydrological applications.

Decametric-resolution leaf area index (LAI) is an essential parameter for fine-scale crop growth monitoring and ecosystem modeling. Prior-guided approaches using existing hectometric-resolution LAI products have demonstrated potential in large-scale decametric-resolution LAI estimation. However, within such approaches, the impacts of algorithm selection and band combination on retrieval accuracy remain insufficiently quantified, and the lack of model interpretability limits methodological transferability. To address these challenges, a multi-source data integration (MSDI) framework is developed to systematically assess the sensitivity of prior-guided LAI estimation to retrieval algorithms and spectral bands using Sentinel-2 imagery. In addition, Shapley Additive Explanations (SHAP) is employed to quantify the contributions of individual bands and interpret model behavior. The MSDI LAI was evaluated using ground LAI measurements and compared with Simplified Level 2 Product Prototype Processor (SL2P)-derived LAI and MODIS LAI products. The results indicated that Support Vector Regression (SVR) achieved the best performance in LAI estimation among six machine learning algorithms, likely due to its robustness in modeling nonlinear relationships across different training samples. Band optimization further reduced estimation uncertainty by >24% and increased R² by >44% for SVR-derived LAI estimates. Moreover, MSDI outperformed SL2P, especially at 20 m resolution, with Bias, RMSE, and R² values of 0.26, 0.76, and 0.71, respectively. Meanwhile, MSDI LAI exhibited a similar spatial distribution to MODIS LAI while providing substantially enhanced spatial detail and accuracy. SHAP analysis revealed that red-edge (RE) and shortwave-infrared (SWIR) bands contributed the most to LAI prediction, consistent with their sensitivity to vegetation canopy biophysical properties. Overall, this study highlights the importance of retrieval strategy optimization and model interpretability for improving prior-guided decametric-resolution LAI estimation and offers practical guidance for generating consistent LAI estimations across various scales.

The increasing frequency of drought, heat stress and extreme precipitation is intensifying risks to agricultural systems in the Southern Slopes of the Himalaya (SSH) under climate change. Understanding crop loss responses across cropping systems is critical for regional food security. Yet, which extremes dominate productivity loss and how their risk patterns differ across systems remain unclear. Across the three climate extremes considered, loss probability dominated crop productivity risk in 77% of the study area. Double cropping showed a probability dominated vulnerability pattern, whereas single cropping exhibited a magnitude amplified pattern. Drought remained the dominant regional risk source, with its contribution declining as extreme precipitation and heat stress became more important. Furthermore, loss probability had lower thresholds than loss magnitude, indicating a nonlinear two stage process from risk triggering to damage amplification. Thus, agricultural risk in the SSH is structurally driven by cropping systems and multiple climate extremes.

Cross-evaluation results of multi-source satellite solar-induced chlorophyll fluorescence (SIF) products depend critically on the analytical choices made during evaluation, including product combinations, statistical models, and spatiotemporal preprocessing. However, this ”conditional dependence” has long lacked systematic quantification, severely limiting the interpretability of evaluation conclusions. This study constructs a diagnostic framework to systematically assess the methodological sensitivity of cross-evaluation. Using four representative SIF products from 2018 to 2021, we isolated the independent effects of product combination, error model (Three-Cornered Hat, TCH, vs. Quadruple Collocation Analysis, QCA), spatial resolution, and seasonal-cycle treatment through controlled experiments. Direct comparison shows that the two TROPOMI products exhibit the highest consistency, whereas GOME-2 SIF shows the largest discrepancies. A key finding is that when a product combination includes two products that share observations from the same sensor, TCH produces a structural bias due to violation of the error-independence assumption: it systematically underestimates the errors of the shared-source products while overestimating the errors of the independent product. By allowing non-zero error cross-covariance between a pre-specified product pair, QCA effectively corrects this bias. Based on QCA estimates, over 97% of valid vegetated pixels for OCO-2 SIF, TROPOSIFCaltech, and TROPOSIFESA exhibit RMSE values below 0.2 W·m-2·sr-1·μm-1, whereas over 77% of GOME-2 SIF pixels are concentrated in the 0.2-0.4 W·m-2·sr-1·μm-1 range. All four products show the highest errors in evergreen broadleaf forests and the lowest errors in closed shrublands. Spatially, coarse-resolution products are more sensitive to spatial aggregation; temporally, removing the common seasonal cycle substantially reduces error estimates for all products, with OCO-2 SIF showing the largest reduction, approximately 52%. Collectively, these findings reveal that error estimates from cross-evaluation are not intrinsic product attributes but conditional outcomes of the analytical framework. The proposed diagnostic framework is transferable to uncertainty analyses of other satellite-retrieved geophysical variables, offering a rigorous methodological foundation for remote sensing product validation and fusion.

The optimal temperature for photosynthesis (Topt), the temperature at which photosynthesis peaks, is crucial for estimating gross primary production (GPP). Most models, however, apply biome‐specific Topt, overlooking the spatial and temporal variability driven by the genetic adaptation and interannual thermal acclimation of plants, thus introducing systematic error in GPP estimation. We mapped Topt of Tibetan Plateau (TP) grassland averaged over 1982–2018, using satellite‐derived near‐infrared reflectance of vegetation as a proxy of photosynthesis, and projected future Topt under different scenarios of shared socioeconomic pathways (SSPs) using space‐for‐time substitution. We assessed the effects of Topt adaptation on current GPP estimation and acclimation on projected GPP using a light‐use efficiency model (i.e., EC‐LUE model). The spatial heterogeneity of Topt was pronounced across the TP grassland, with higher values in the northeast and lower values in the southwest, averaging 9.32°C ± 3.15°C. Topt tended to increase from 2040 to 2080 across all SSPs, with the slowest increase under SSP1‐2.6 (0.015°C y ⁻¹ ) and the fastest increase under SSP5‐8.5 (0.061°C y ⁻¹ ). Incorporating Topt adaptation into the EC‐LUE model improved the accuracy of GPP estimation, with R ² increasing by 7.19% and RMSE decreasing by 11.40%, compared to the model with a fixed Topt. In contrast, ignoring the adaptation of Topt led to systematically lower GPP values across 77.9% of the TP grassland, resulting in an overall reduction of estimated GPP by 7.07% (0.07 g C m ⁻² d ⁻¹ ). Ignoring the acclimation of Topt led to systematically higher projected GPP on the TP, ranging from 2.87% ± 8.96% (SSP1‐2.6) to 3.83% ± 5.97% (SSP3‐7.0) across different scenarios, covering more than 70% of TP regions. These findings highlight the necessity of incorporating both the adaptation and acclimation of Topt into GPP models to enhance the monitoring of the carbon cycle.