Land Surface Phenology Products Research Articles

Seasonal Effect of the Vegetation Clumping Index on Gross Primary Productivity Estimated by a Two-Leaf Light Use Efficiency Model

Recently, light use efficiency (LUE) models driven by remote sensing data have been widely employed to estimate the gross primary productivity (GPP) of different terrestrial ecosystems at global or regional scales. Furthermore, the two-leaf light use efficiency (TL-LUE) model has been reported to improve the accuracy of GPP estimation, relative to the big-leaf MOD17 model, by separating the entire canopy into sunlit and shaded leaves through the use of constant clumping index estimation (Ω). However, ignoring obvious seasonal changes in the vegetation clumping index (CI) most likely results in GPP estimation errors since the CI tends to present seasonal changes, especially with respect to the obvious presence or absence of leaves within the canopy of deciduous vegetation. Here, we propose a TL-CLUE model that considers the seasonal difference in the CI based on the TL-LUE model to characterize general changes in canopy seasonality. This method composites monthly CI values into two or three Ω values to capture the general seasonal changes in CI while attempting to reduce the potential uncertainty caused during CI inversion. In theory, CI seasonality plays an essential role in the distribution of photosynthetically active radiation absorbed by the canopy (APAR). Specifically, the seasonal difference in CI values mainly considers the state of leaf growth, which is determined by the MODIS land surface phenology (LSP) product (MCD12Q2). Therefore, the one-year cycle (OYC) of leaf life is divided into two (leaf-off and leaf-on) or three seasons (leaf-off, leaf-scattering, and leaf-gathering) according to this MODIS LSP product, and the mean CI of each corresponding season for each vegetation class is computed to smoothen the uncertainties within each seasonal section. With these two or three seasonal Ω values as inputs, the TL-CLUE model by which the seasonal differences in CI are incorporated into the TL-LUE model is run and evaluated based on observations from 84 eddy covariance (EC) tower sites across North America. The results of the analysis reveal that the TL-LUE model widely overestimates GPP for most vegetation types during the leaf-on season, particularly during the growth peak. Although the TL-LUE model shows that the temporal characteristics of GPP agree with the EC observations in terms of general trends, the TL-CLUE model further improves the accuracy of GPP estimation by considering the seasonal changes in the CI. The result of GPP estimation from the TL-CLUE model shows a lower error (RMSE = 2.46 g C m−2 d−1) than the TL-LUE model (RMSE = 2.75 g C m−2 d−1) and somewhat decreases the eight-day GPP overestimation in the TL-LUE model with a constant Ω by approximately 9.76 and 8.970% when adapting three and two Ωs from different seasons, respectively. The study demonstrates that the uncertainty of seasonal disturbance in the CI, quantified by a standard deviation of approximately 0.071 relative to the mean CI of 0.746, is diminished through simple averaging. The seasonal difference in CI should be considered in GPP estimation of terrestrial ecosystems, particularly for vegetation with obvious canopy changes, where leaves go through the complete physiological processes of germination, stretching, maturity, and falling within a year. This study demonstrates the potential of the MODIS CI application in developing ecosystem and hydrological models.

A robust and unified land surface phenology algorithm for diverse biomes and growth cycles in China by using harmonized Landsat and Sentinel-2 imagery

Land surface phenology (LSP) is beneficial to understand ecosystem response to climate change, vegetation and crop type discrimination, and ecological modeling. However, the existing efforts based on coarse resolution data (≥500 m) cannot perform well in regions with higher spatial heterogeneity and multi-cropping system, such as China. Given that the majority of 10 m/30 m-based phenological research has focused on North America and Europe, developing spatiotemporally explicit LSP data in China is imperative. More importantly, the existing 30 m LSP products are mainly suitable for vegetation types with a single vegetation cycle, but cannot work well for biomes with complex seasonality (e.g., multiple growth cycles). Here we first harmonized three vegetation indices, i.e., the normalized difference vegetation index (NDVI), two-band enhanced vegetation index (EVI2), and land surface water index (LSWI) from Landsat-7/8 and Sentinel-2 imagery on the Google Earth Engine (GEE) platform. We then developed a new 30 m LSP algorithm that unified different phenological cycle-seeking processes per vegetation type and improved the existing algorithm. Furthermore, we used the algorithm to estimate the LSP product (LSP30CHN) for 2016–2019 across China, suitable for all vegetation types. The validation results showed a reasonably high accuracy (R2 > 0.6, RMSE < 15 days, mostly) of the LSP30CHN data against multi-sources in-situ observational (e.g., PhenoCam) and satellite-retrieved vegetation phenology data. Moreover, LSP30CHN data showed a consistent pattern but finer spatial details with the 500 m Moderate Resolution Imaging Spectroradiometer (MODIS) phenology product (MCD12Q2) at the homogenous area. We also found that phenological differences between LSP30CHN and MCD12Q2 increased with surface fragmentation, suggesting the potential of LSP30CHN to delineate phenological information on more fragmented landscapes. In contrast, the 500 m LSP data cannot provide such details in the regions with mixed cropping structures (e.g., corn, rice, and soybean) and multiple cropping index (e.g., single- and double-cropping systems). This study offers high accuracy of the LSP map for China, valuable for finer phenology-based services such as field-level crop management and agricultural phenology monitoring. It opens up new insights about exploring large-scale refined agricultural management and ecological assessment for other regions with complicated, fragmented landscapes and vegetation seasonality.

Calibration of MODIS-Derived Cropland Growing Season Using the Climotransfer Function and Ground Observations

The global environment experienced notable changes in the recent past of planet Earth. Satellite remote sensing has played an increasingly important role in monitoring and characterizing these changes. Being recognized as a sensitive indicator of global climate change, land surface phenology (LSP) observations by satellite remote sensing have received much attention in recent years; however, much less attention has been paid to the calibration of these observations using standardized procedures. Here, we propose a new approach to calibrating the satellite LSP products by developing a climotransfer function (CTF) based on a polynomial regression of the satellite-ground observation difference in key crop phenophases against climatic factors. We illustrate the model development and evaluation process with a case study of the cropland growing season in Northeast China (NEC) from 2001 to 2010 using the MODIS LSP product MCD12Q2 Collection 6 and the ground-observed crop phenology and climatic data from 98 agrometeorological stations across the region. Our results showed that the start of the cropland growing season (SOS) derived from MODIS data compared well to the ground-observed SOS, whereas the MODIS-derived season end (EOS) was delayed by 15.5 d, relative to ground observation. The MODIS-derived EOS was, therefore, spatiotemporally calibrated using a CTF model fitted to the satellite-ground difference in EOS (∆EOS) versus two climatic factors, namely, the growing degree-days on the base temperature of 10 °C (GDD10) and cloud cover (CL). The calibrated MODIS data revealed that the cropland growing season in NEC tended to shorten at 4.5 d decade−1 during 2001–2010, mainly driven by a significant delay in SOS at a similar rate, whereas no trend was detected for EOS. The calibrated data also revealed a significant shortening gradient of 1.7 d degree−1 of latitude northward. These spatiotemporal patterns would have been erroneously characterized if calibration had not been applied. More attention is therefore called to the proper calibration of satellite LSP products prior to any meaningful applications.

Quantitative Assessment of the Spatial Scale Effects of the Vegetation Phenology in the Qinling Mountains

Vegetation phenology reflects the temporal dynamics of vegetation growth and is an important indicator of climate change. However, differences consistently exist in land surface phenology derived at different spatial scales, which hinders the understanding of phenological events and integration of land surface phenology products from different scales. The Qinling Mountains are a climatic and geographical transitional region in China. To better understand the spatial scale effect issues of land surface phenology in mountainous ecosystems, this study up-scaled vegetation start of season (SOS) and end of season (EOS) in the Qinling Mountains derived from three different Moderate Resolution Imaging Spectroradiometer (MODIS) normalized difference vegetation index (NDVI) products to four scales (i.e., 2 km × 2 km, 4 km × 4 km, 6 km × 6 km, and 8 km × 8 km) using the spatial averaging method. Then, similarities and differences between the up-scaled SOSs/EOSs were examined using the simple linear regression, cumulative distribution function, and absolute difference. Finally, the random forest model was used to reveal the major factors influencing the spatial scale effect of land surface phenology in Qinling Mountains. Results showed that the derived basic SOS/EOS datasets using the same filtering method from the 250 m and 500 m NDVI datasets were consistent in spatial distribution, while the results from the 1000 m NDVI dataset differed. For both the basic and the up-scaled datasets, the land surface phenology derived from the Savitzky-Golay-filtered NDVI showed an advance in SOS, but a delay in EOS, compared to those derived from the asymmetric Gaussian- and double logistic-filtered NDVI. The up-scaled SOS was greatly impacted by both NDVI resolution and the filtering methods. On the other hand, EOS was mostly impacted by the filtering methods. Moreover, up-scaled SOSs usually had larger differences compared to up-scaled EOSs. While different filtering methods sometimes amplified the absolute differences between different SOS/EOS across scales, the upscaling reduced the differences. Influence factor analysis showed that spatial variations observed in SOS in Qinling Mountains were mainly caused by forest cover, uneven distribution of spring precipitation, and annual precipitation, while spatial variations in aspect, winter temperature, and autumn precipitation all strongly influenced the observed EOS across scales in the study area. These findings enhance our understanding of the effects of observational scale on vegetation phenology in mountain ecosystems and provide a reference for phenology modeling in mountainous areas.

An optimal method for validating satellite-derived land surface phenology using in-situ observations from national phenology networks

Satellite-based land surface phenology (LSP) products play an important role in understanding atmosphere-vegetation carbon and energy exchanges. These products have been widely calculated from various satellite observations from local to global scales. However, the quality and accuracy of LSP products are often poorly quantified due to spatial mismatch between satellite observed pixels and in-situ observations. In the present study, we demonstrate an optimal algorithm leveraging the scalability, consistency, and representativeness of rich in-situ observations from national phenology networks to validate LSP products. Specifically, we demonstrate two approaches for validating the phenological timing of greenup onset in the operational Visible Infrared Imaging Radiometer Suite (VIIRS) LSP product developed at NASA using in-situ observations collected from the Pan European Phenological database (PEP725, 9664 site-years) and the USA National Phenology Network (USA-NPN, 3144 site-years) spanning 2013–2020. The first approach assumes that in-situ data contain observations of phenological transitions (e.g., leaf-out) that are directly comparable with satellite detections. Accordingly, in-situ data were aggregated using four upscaling methods (mean, median, 30th percentile, and minimum bias) to directly compare with VIIRS LSP. The second approach assumes that species-specific phenological timing in in-situ data is basically impossible to spatially reconcile VIIRS LSP, but phenological events in a local area are driven by the same or very similar weather conditions. Therefore, interannual variations and long-term trends were applied to compare VIIRS LSP with in-situ data. The result shows first that the 30th percentile method is more promising in aggregating in-situ observations than the commonly used mean method. Second, direct comparison indicates that VIIRS greenup onset has a mean absolute difference of 13.9 ± 9.8 days with PEP725 in-situ observations and 12.3 ± 10.9 days with USA-NPN observations in well-selected deciduous forest sites. Third, the interannual comparison reveals that VIIRS greenup onset exhibits the same directions of multi-year anomalies and long-term trends as those of both PEP725 and USA-NPN observations in over 70% of sample sites. These findings improve our understanding of the scale mismatch and sample representativeness of species-specific phenology and the uncertainties of long-term LSP detections from remote sensing data.

A novel algorithm for the generation of gap-free time series by fusing harmonized Landsat 8 and Sentinel-2 observations with PhenoCam time series for detecting land surface phenology

Vegetation phenology is one of the most sensitive indicators to environmental and climate changes. In order to characterize the seasonal variation in relatively pure or homogenous vegetation types, fine spatial resolution satellite data (≤ 30 m), such as Landsat, Sentinel-2, PlanetScope, or Harmonized Landsat and Sentinel-2 (HLS), have been increasingly applied to detect land surface phenology (LSP). However, the most critical challenge in LSP detections is the gaps in temporal satellite observations caused by noise and persistent cloud/snow cover. Therefore, this study presented a novel algorithm for generating synthetic gap-free time series at the field scale (30 m) for LSP detections. Specifically, we first developed a framework to establish a large collection of temporal shapes of vegetation growth with as many as 100 grid-based Green Chromatic Coordinate (GCC) time series in a single PhenoCam site. For a given HLS pixel, the two-band Enhanced Vegetation Index (EVI2) time series was matched and fused with the most comparable temporal GCC shape selected from the collection of PhenoCam GCC time series to generate a synthetic gap-free HLS-PhenoCam EVI2 time series, which was used to detect the 30 m phenometrics. The detected phenometrics were evaluated using manually selected and spatially matched GCC observations as well as phenology detections from HLS alone. The result indicates that the HLS-PhenoCam phenometrics are very close to the observations from PhenoCam network with a correlation coefficient (R) of 0.82–0.97, a mean absolute difference (MAD) of 2.8–3.5 days, a root mean squared error (RMSE) of 3.5–4.0 days, and a mean systematic bias (MSB) of 0.1–2.2 days. The HLS-PhenoCam detections are significantly improved relative to the HLS phenometrics that have a statistic accuracy of R = 0.57–0.78, MAD = 6.4–9.3 days, RMSE = 8.8–13.9 days, MSB = -5.2–5.9 days. The difference between HLS-PhenoCam and HLS alone LSP detections over a HLS tile could be on average larger than two weeks if high-quality observation (HQO) proportion in the annual HLS time series is <10%, which exponentially reduces with the increase of HQO in HLS observations. The analyses in this study suggest that the gap-free HLS-PhenoCam time series is able to be generated for producing high-quality phenology datasets across a local and regional scale, to bridge near-surface PhenoCam observations with satellite observations data at various scales, and to be used as a scalable phenology dataset for the validation of global MODIS and VIIRS LSP products.

A high spatial resolution land surface phenology dataset for AmeriFlux and NEON sites

Vegetation phenology is a key control on water, energy, and carbon fluxes in terrestrial ecosystems. Because vegetation canopies are heterogeneous, spatially explicit information related to seasonality in vegetation activity provides valuable information for studies that use eddy covariance measurements to study ecosystem function and land-atmosphere interactions. Here we present a land surface phenology (LSP) dataset derived at 3 m spatial resolution from PlanetScope imagery across a range of plant functional types and climates in North America. The dataset provides spatially explicit information related to the timing of phenophase changes such as the start, peak, and end of vegetation activity, along with vegetation index metrics and associated quality assurance flags for the growing seasons of 2017–2021 for 10 × 10 km windows centred over 104 eddy covariance towers at AmeriFlux and National Ecological Observatory Network (NEON) sites. These LSP data can be used to analyse processes controlling the seasonality of ecosystem-scale carbon, water, and energy fluxes, to evaluate predictions from land surface models, and to assess satellite-based LSP products.

GCI30: a global dataset of 30 m cropping intensity using multisource remote sensing imagery

Abstract. The global distribution of cropping intensity (CI) is essential to our understanding of agricultural land use management on Earth. Optical remote sensing has revolutionized our ability to map CI over large areas in a repeated and cost-efficient manner. Previous studies have mainly focused on investigating the spatiotemporal patterns of CI ranging from regions to the entire globe with the use of coarse-resolution data, which are inadequate for characterizing farming practices within heterogeneous landscapes. To fill this knowledge gap, in this study, we utilized multiple satellite data to develop a global, spatially continuous CI map dataset at 30 m resolution (GCI30). Accuracy assessments indicated that GCI30 exhibited high agreement with visually interpreted validation samples and in situ observations from the PhenoCam network. We carried out both statistical and spatial comparisons of GCI30 with six existing global CI estimates. Based on GCI30, we estimated that the global average annual CI during 2016–2018 was 1.05, which is close to the mean (1.09) and median (1.07) CI values of the existing six global CI estimates, although the spatial resolution and temporal coverage vary significantly among products. A spatial comparison with two satellite-based land surface phenology products further suggested that GCI30 was not only capable of capturing the overall pattern of global CI but also provided many spatial details. GCI30 indicated that single cropping was the primary agricultural system on Earth, accounting for 81.57 % (12.28×106 km2) of the world's cropland extent. Multiple-cropping systems, on the other hand, were commonly observed in South America and Asia. We found large variations across countries and agroecological zones, reflecting the joint control of natural and anthropogenic drivers on regulating cropping practices. As the first global-coverage, fine-resolution CI product, GCI30 is expected to fill the data gap for promoting sustainable agriculture by depicting worldwide diversity of agricultural land use intensity. The GCI30 dataset is available on Harvard Dataverse: https://doi.org/10.7910/DVN/86M4PO (Zhang et al., 2020).

Exploration of global spatiotemporal changes of fall foliage coloration in deciduous forests and shrubs using the VIIRS land surface phenology product

Fall foliage coloration is a phenological development process that has a significant impact on ecosystem carbon and water balance, climate change, and human sightseeing activities worldwide. Studies of land surface phenology concentrating on the onset timing of senescence and dormancy are insufficient to reveal the full details of the fall foliage coloration process that is particularly important for travel and the tourism industry. Thus, the aim of this study is for the first time to explicitly explore the fall foliage development in six coloration phases (little/no coloration, low coloration, moderate coloration, near-peak coloration, peak coloration, and post-peak coloration) and the related deciduous vegetation across the globe. The onset and duration of each coloration phase were estimated using the VIIRS (Visible Infrared Imaging Radiometer Suite) land surface phenology product. The results indicate spatial patterns and interannual variations in the onset and duration of six phases in foliage coloration. The coloration phases present regular shifts along latitudes in the Northern Hemisphere but are complex in the Southern Hemisphere. The rate of change along latitudes is faster during the later coloration phases than the early phases in the Northern Hemisphere, while the rates are similar in the Southern Hemisphere for all six coloration phases. Moreover, the interannual variations in the onset of a foliage coloration phase are smaller in forested areas than in shrub areas, and more so in dense tree/shrub cover areas than in sparse tree/shrub cover areas. In contrast, the interannual variations in the duration of a foliage coloration depend on the vegetation type. Finally, there were no significant trends in the foliage coloration phase during 2013–2019. The results from this study improve our understanding of fall foliage coloration development and provide valuable datasets to the communities of phenology study and tourism.

A deep learning method to predict soil organic carbon content at a regional scale using satellite-based phenology variables

Obtaining the spatial distribution information of soil organic carbon (SOC) is significant to quantify the carbon budget and guide land management for migrating carbon emissions. Digital soil mapping of SOC at a regional scale is challenging due to the complex SOC-environment relationships. Vegetation phenology that directly indicates a long time vegetation growth characteristics can be potential environmental covariates for SOC prediction. Deep learning has been developed for soil mapping recently due to its ability of constructing high-level features from the raw data. However, only dozens of predictors were used in most of those studies. It is not clear that how deep learning with long term land surface phenology product performs for SOC prediction at a regional scale. This paper explored the effectiveness of ten-years MODIS MCD12Q2 phenology variables for SOC prediction with a convolutional neural network (CNN) model in Anhui province, China. Random forest (RF) was applied to compare with CNN using three groups of environmental variables. The results showed that adding the land surface phenology variables into the pool of the natural environmental variables improved the prediction accuracy of CNN by 5.57% of RMSE and 31.29% of R2. Adding phenology variables obtained a higher accuracy improvement than adding Normalized Differences Vegetation Indices. The CNN obtained a higher prediction accuracy than RF regardless of using which group of variables. This study proved that land surface phenology metrics were effective predictors and CNN was a promising method for soil mapping at a regional scale.

Long-term, medium spatial resolution annual land surface phenology with a Bayesian hierarchical model

Land surface phenology (LSP) is a consistent and sensitive indicator of climate change effects on Earth's vegetation. Existing methods of estimating LSP require time series densities that, until recently, have only been available from coarse spatial resolution imagery such as MODIS (500 m) and AVHRR (1 km). LSP products from these datasets have improved our understanding of phenological change at the global scale, especially over the MODIS era (2001-present). Nevertheless, these products may obscure important finer scale spatial patterns and longer-term changes. Therefore, we have developed a Bayesian hierarchical model to retrieve complete annual sequences of LSP from Landsat imagery (1984-present), which has medium spatial resolution (30 m) but relatively sparse temporal frequency. Our approach uses Markov Chain Monte Carlo (MCMC) sampling to quantify individual phenometric uncertainty, which is especially important when considering long time series with variable observation quality and density, but has rarely been demonstrated. The estimated spring LSP had strong agreement with ground phenology records at Harvard Forest (R2 = 0.87) and Hubbard Brook Experimental Forest (R2 = 0.67). The estimated LSP were consistent with the recently released 30 m LSP product, MSLSP30NA, in its time period of 2016 to 2018 (R2 of 0.86 and 0.73 for spring and autumn phenology, respectively). Our Bayesian hierarchical model is an important step forward in extending medium resolution LSP records back in time as it accomplishes both critical goals of retrieving annual LSP from sparse time series and accurately estimating uncertainty.

A new framework to map fine resolution cropping intensity across the globe: Algorithm, validation, and implication

Accurate estimation of cropping intensity (CI), an indicator of food production, is well aligned with the ongoing efforts to achieve sustainable development goals (SDGs) under diminishing natural resources. The advancement in satellite remote sensing provides unprecedented opportunities for capturing CI information in a spatially continuous manner. However, challenges remain due to the lack of generalizable algorithms for accurately and efficiently mapping global CI with a fine spatial resolution. In this study, we developed a 30-m planetary-scale CI mapping framework with the reconstructed time series of Normalized Difference Vegetation Index (NDVI) from multiple satellite images. Using a binary crop phenophase profile indicating growing and non-growing periods, we estimated pixel-by-pixel CI by enumerating the total number of valid cropping cycles during the study years. Based on the Google Earth Engine cloud computing platform, we implemented the framework to estimate CI during 2016–2018 in eight geographic regions across continents that are representative of global cropping system diversity. Comparison with PhenoCam network data in four cropland sites suggests that the proposed framework is capable of capturing the seasonal dynamics of cropping practices. Spatially, overall accuracies based on validation samples range from 80.0% to 98.9% across different regions worldwide. Regarding the CI classes, single cropping systems are associated with more robust and less biased estimations than multiple cropping systems. Finally, our CI estimates reveal high agreement with two widely used land surface phenology products, including Vegetation Index and Phenology V004 (VIP4) and Moderate Resolution Imaging Spectroradiometer Land Cover Dynamics (MCD12Q2), meanwhile providing much more spatial details. Due to its robustness, the developed CI framework can be potentially generalized to produce global fine resolution CI products for food security and other applications.

Development and evaluation of a new algorithm for detecting 30 m land surface phenology from VIIRS and HLS time series

Land surface phenology (LSP) provides critical information for investigating vegetation growth and development, studying ecosystem biodiversity, modeling terrestrial carbon and surface energy budgets, detecting land cover and land use change, and monitoring climate change. Although operational 500 m LSP products have been produced from coarse resolution data observed from Moderate Resolution Imaging Spectroradiometer (MODIS) and Visible Infrared Imaging Radiometer Suite (VIIRS), an LSP product is also needed at the Landsat scale (30 m) to enhance the environmental monitoring and modeling. However, temporal frequency of 30 m satellite data is always inadequate for reliable LSP detection, despite enrichment by the operational harmonized Landsat and Sentinel-2 (HLS) product. In this study, we propose a new algorithm of LSP detection for the generation of a 30 m LSP product using routinely produced HLS and VIIRS surface reflectance products. Specifically, the new algorithm compares a HLS EVI2 (two-band enhanced vegetation index) time series at a given 30 m pixel with the set of 500 m VIIRS EVI2 time series neighboring the HLS pixel and selects the most similar temporal shape of VIIRS time series even though the amplitude and/or phase between HLS and VIIRS EVI2 time series may be mismatched. The shape of the selected VIIRS EVI2 time series is then used to match to the given HLS EVI2 time series to generate a synthetic HLS-VIIRS time series. The HLS-VIIRS time series is subsequently processed using the hybrid piecewise logistic model to detect the phenological transition dates and to quantify the confidence of LSP detection. This new algorithm is evaluated by implementing 30 m LSP detection in eight HLS tiles in the northeastern (forests), central (croplands), and western (shrublands) United States. Evaluation finds that the new-algorithm-detected greenup onset (1) agrees well with the standard VIIRS LSP product without bias, (2) closely correlates to PhenoCam observations with a slope close to one, and (3) compares well with both PhenoCam and field species-specific observations with a mean absolute difference of 8 days and a difference less than 10 days in more than 70% of the validation samples. This implementation suggests that the new algorithm could be implemented for regional and global LSP product generation at a 30 m resolution.

Long-term continuity in land surface phenology measurements: A comparative assessment of the MODIS land cover dynamics and VIIRS land surface phenology products

Vegetation phenology contributes to, and is diagnostic of, seasonal variation in ecosystem processes and exerts important controls on land-atmosphere exchanges of carbon, water, and energy. Satellite remote sensing provides a valuable source of data for monitoring the phenology of terrestrial ecosystems and has been widely used to map geographic and interannual variation in land surface phenology (LSP) over large areas. The Visible Infrared Imaging Radiometer Suite (VIIRS) land surface phenology product provides global data sets characterizing the annual LSP of terrestrial ecosystems, and is designed to support long-term continuity of LSP measurements from the Moderate Resolution Imaging Spectroradiometer (MODIS). We used data from VIIRS and MODIS to evaluate the agreement and characterize the similarities and differences between LSP data from each instrument. Specifically, we compare data from the Collection 6 MODIS Land Cover Dynamics (LCD) product with data from the newly developed VIIRS LSP product over the most common land cover types in North America. To do this, we assessed the overall agreement between time series of vegetation indices from VIIRS and MODIS, evaluated the correspondence between retrieved phenometrics from each instrument, and analyzed sources of differences between phenometrics from the each product. As part of this analysis, we also compared phenometrics from MODIS and VIIRS with phenometrics derived from Landsat Analysis Ready Data and PhenoCam time series. Results show that two-band enhanced vegetation index (EVI2) values from VIIRS and MODIS are similar (R2 > 0.81; root mean square deviation < 0.062), but that VIIRS EVI2 time series show more high frequency variation than time series from MODIS. Further, even though the VIIRS and MODIS products are generated using different instruments and algorithms, phenometrics from each product are similar and show only minor differences within and across land cover types. Systematic differences between phenometrics from the two products were generally less than one week (absolute bias 4.8 ± 3.0 days), and RMSDs were less than two weeks for most phenometrics across different land cover classes (10.7 ± 4.3 days). Comparison of VIIRS and MODIS LSP data with corresponding metrics estimated from Landsat and PhenoCam data consistently showed high agreement among the data sets. Overall, results from this analysis indicate that the VIIRS LSP product provides excellent continuity with the MODIS record. However, studies attempting to create high-fidelity long-term LSP time series by merging these products should exploit the overlap period of MODIS and VIIRS to estimate land cover-specific corrections for modest systematic bias in the MODIS LCD product relative to the VIIRS LSP product.

Generation and evaluation of the VIIRS land surface phenology product

Vegetation phenology is widely acknowledged to be a sensitive indicator of the response of ecosystems to climate change, and phenological shifts have been shown to exert substantial impacts on ecosystem function, biodiversity, and carbon budgets at multiple scales. Therefore, long-term records of the phenology of the vegetated land surface are critical in exploring the biological response to environmental change at regional to global scales. Land surface phenology (LSP) from satellite observations has been extensively used to monitor the dynamics of terrestrial ecosystems in the face of a changing climate. Here we introduce and describe the global land surface phenology (GLSP) product derived from the Visible Infrared Imaging Radiometer Suite (VIIRS) data at a gridded spatial resolution of 500 m. This new product will provide continuity for the Moderate Resolution Imaging Spectroradiometer (MODIS) GLSP product that has been produced on an operational basis since 2001. The VIIRS GLSP algorithm uses daily VIIRS Nadir BRDF (Bidirectional Reflectance Distribution Function)-Adjusted Reflectance (NBAR) data as the primary input to calculate the two-band enhanced vegetation index (EVI2) at each 500 m pixel. The temporal EVI2 trajectory is modeled using a hybrid piecewise logistic function to track the seasonal vegetation development, detect phenological transition dates, calculate the magnitude of vegetation greenness development, and characterize the confidence of phenology detections. The VIIRS GLSP algorithm has been implemented across the contiguous United States, and the resulting phenological metrics have been evaluated through comparisons with species-specific field phenological observations, Landsat phenology retrievals, and the MODIS phenology detections. The results demonstrate that the VIIRS GLSP metrics are of high quality and are in a good agreement with the other independent satellite and field observations. The results also indicate that the uncertainty in the VIIRS GLSP retrievals is primarily associated with missing high quality observations in VIIRS EVI2 time series.

Evaluation of land surface phenology from VIIRS data using time series of PhenoCam imagery

Land surface phenology (LSP) has been widely retrieved from time series of various satellite instruments in order to monitor climate change and ecosystem dynamics. However, any evaluation of the quality of LSP data sets is quite challenging because the in situ observations on a limited number of individual trees, shrubs, or other plants are rarely representative of the landscape sampled in a single satellite pixel. Moreover, vegetation indices detecting biophysical features of vegetation seasonality are different from (but related to) the specific plant life history stages observed by humans at ground level. This study is the first comprehensive evaluation of the LSP product derived from Visible Infrared Imaging Radiometer Suite (VIIRS) data using both MODIS LSP products and observations from the PhenoCam network across the Contiguous United States during 2013 and 2014. PhenoCam observes vegetation canopy over a landscape at very high frequency, providing nearly continuous canopy status and enabling the estimate of discrete phenophase using vegetation indices that are conceptually similar to satellite data. Six phenological dates (greenup onset, mid-greenup phase, maturity onset, senescence onset, mid-senescence phase, and dormancy onset) were retrieved separately from daily VIIRS NDVI (normalized difference vegetative index) and EVI2 (two-band enhanced vegetation index) time series. Similarly, the six phenological dates were also extracted from the 30-min time series of PhenoCam data using GCC (green chromatic coordinate) and VCI (vegetation contrast index) separately. Phenological dates derived from VIIRS NDVI and EVI2 and PhenoCam GCC and VCI were generally comparable for the vegetation greenup phase, but differed considerably for the senescence phase. Although all indices captured green leaf development effectively, performance discrepancies arose due to their ability to track the mixture of senescing leaf colors. PhenoCam GCC and VCI phenological observations were in better agreement with the phenological dates from VIIRS EVI2 than from VIIRS NDVI. Further, the VIIRS EVI2 phenological metrics were more similar to those from PhenoCam VCI than from PhenoCam GCC time series. Overall, the average absolute difference between the VIIRS EVI2 and PhenoCam VCI phenological dates was 7–11 days in the greenup phase and 10–13 days in the senescence phase. The difference was smaller in forests, followed by grasslands and croplands, and then savannas. Finally, the phenological dates derived from VIIRS EVI2 were compared with MODIS detections, which showed a good agreement with an average absolute difference less than a week except for the senescence onset. These results for the first time demonstrate the upper boundary of uncertainty in VIIRS LSP detections and the continuity to MODIS LSP product.

Determining the Start of the Growing Season from MODIS Data in the Indian Monsoon Region: Identifying Available Data in the Rainy Season and Modeling the Varied Vegetation Growth Trajectories

In the Indian monsoon region, frequent cloud cover in the rainy season results in less valid satellite observations during the vegetation growth period, making it difficult to extract land surface phenology (LSP). Even worse, many valid but humid observations were misidentified as clouds in the MODIS cloud mask, causing severe gaps in the LSP product. Using a refined cloud detection approach to separate clear-sky and cloudy observations, this study found that potentially valid observations during the vegetation growth period could be identified. Furthermore, the varied vegetation growth trajectories cannot be well-fitted by a global curve-fitting approach, but can be modelled by using the locally adjusted cubic-spline capping approach, which performed well for any seasonal patterns. Applying this approach, the start of growing season (SOS) was determined with 9.18% of vegetation growth amplitude between the maximum and minimum NDVI to generate the SOS product (2000–2016). The valid percentage of this regional product largely increased from 29.30% to 69.76% compared with the MCD12Q2 product, and its reliability was approximate to that of deciduous broadleaf forest in North America and Europe. This product could serve as a basis for understanding the response of terrestrial ecosystems to the changing Indian monsoon.

Intercomparison and evaluation of spring phenology products using National Phenology Network and AmeriFlux observations in the contiguous United States

Many remote sensing based spring phenology products have been developed to monitor and study vegetation phenology at regional and global scales. It is important to understand how these products perform relative to each other and to ground observations. In this study, we extracted spring green-up onset dates (GUD) over the contiguous United States (CONUS) from six major land surface phenology (LSP) products: (1) Moderate Resolution Imaging Spectroradiometer (MODIS) Land Cover Dynamics Phenology (MCD12Q2); (2) Vegetation Index and Phenology Multi-sensor Phenology (VIPPHENEVI2); (3) Global Long-Term Climate Modeling Grid Land Surface Phenology (CMGLSP); (4 and 5) North American Carbon Program (NACP) Phenology (MOD09Q1PEVI and MOD15PHN); and (6) USGS/EROS advanced very high resolution radiometer (AVHRR) phenology (AVHRRP). We characterized and compared the GUD data in these LSP products, and evaluated their accuracy using ground-based phenology observations [i.e., human observations of first leaf and sensor readings of gross primary productivity (GPP)] from the USA National Phenology Network (USA-NPN) and AmeriFlux. The results revealed the consistencies and discrepancies of GUD estimates among LSP products. Intercomparison of the six products indicated that the root mean square error (RMSE) of these products range from 17.8days to 31.5days, whereas AVHRRP GUD has the lowest correlation and largest RMSE (∼30days) relative to other products. When compared to ground observations, GUD estimates in six LSP products generally have RMSE values of ∼20days and significant correlations (p<0.001). For the products (MCD12Q2, AVHRRP, MOD09Q1PEVI, and MOD15PHN) available for comparisons in the short-term period (from 2001–2007), AVHRRP GUD presented relatively weaker correlations and a lower index of agreement (IOA), however, MCD12Q2 GUD showed overall slightly better consistencies with ground observations. In the two long-term products (CMGLSP and VIPPHENEVI2 from 1982–2013), CMGLSP exhibited stronger correlations, lower RMSE, and higher IOA with ground observations of the first leaf dates than VIPPHENEVI2 did. To our knowledge, our study provides the first comprehensive evaluation of phenology products using two independent ground-based datasets.

Object-based Dimensionality Reduction in Land Surface Phenology Classification

Unsupervised classification or clustering of multi-decadal land surface phenology provides a spatio-temporal synopsis of natural and agricultural vegetation response to environmental variability and anthropogenic activities. Notwithstanding the detailed temporal information available in calibrated bi-monthly normalized difference vegetation index (NDVI) and comparable time series, typical pre-classification workflows average a pixel’s bi-monthly index within the larger multi-decadal time series. While this process is one practical way to reduce the dimensionality of time series with many hundreds of image epochs, it effectively dampens temporal variation from both intra and inter-annual observations related to land surface phenology. Through a novel application of object-based segmentation aimed at spatial (not temporal) dimensionality reduction, all 294 image epochs from a Moderate Resolution Imaging Spectroradiometer (MODIS) bi-monthly NDVI time series covering the northern Fertile Crescent were retained (in homogenous landscape units) as unsupervised classification inputs. Given the inherent challenges of in situ or manual image interpretation of land surface phenology classes, a cluster validation approach based on transformed divergence enabled comparison between traditional and novel techniques. Improved intra-annual contrast was clearly manifest in rain-fed agriculture and inter-annual trajectories showed increased cluster cohesion, reducing the overall number of classes identified in the Fertile Crescent study area from 24 to 10. Given careful segmentation parameters, this spatial dimensionality reduction technique augments the value of unsupervised learning to generate homogeneous land surface phenology units. By combining recent scalable computational approaches to image segmentation, future work can pursue new global land surface phenology products based on the high temporal resolution signatures of vegetation index time series.

An approach for evaluating the impact of gaps and measurement errors on satellite land surface phenology algorithms: Application to 20 year NOAA AVHRR data over Canada

Land surface phenology (LSP) is directly related to seasonal variations in the condition of vegetated land surfaces and hence is an important indicator for studying the effect of climate change on ecosystems. Satellite data is widely used to monitor LSP over regional to global extents. Currently, there are many algorithms for generating satellite based LSP products. The Committee of Earth Observing System identifies performance assessments of these products as a pressing need. Performance assessment includes both validation in comparison to independent reference data and evaluation by comparison against estimates of noise free phenology indicators. This study proposes a novel methodology for evaluation of satellite based LSP algorithms and applies this methodology to evaluate three well-known LSP algorithms, Savitzky–Golay based filter (SGF), asymmetric Gaussian fitting (AGF) and logistic function fitting (logistic), using 20years of daily Normalized Difference Vegetation Index composites for Canada and Northern USA. The three LSP methods were evaluated in terms of accuracy for start of season, maximum of season and end of season. Signal to noise ratio (SNR) and the temporal gap frequency (gaps) of the NDVI time series are shown to be proportional to the confidence level for clear sky labeling. The SGF and logistic methods were found to be more sensitive to the SNR than AGF. However, AGF require a minimum proportion of temporal samples for reconstruction. These results highlight the importance of evaluating the trade-off between gap frequency and SNR impacts on LSP methods and indicate that the performance of these methods will depend on both land surface condition and the clear sky identification approach adopted. The developed evaluation approach serves as a basis for assessing the uncertainty in satellite LSP products due to radiometric noise and missing data.