Evaluation of the MODIS collections 5 and 6 for change analysis of vegetation and land surface temperature dynamics in North and South America

https://doi.org/10.1016/j.isprsjprs.2019.07.011Get rights and content

Abstract

The latest collection (C6) of MODIS data provides several algorithmic improvements and calibration adjustments that correct for sensor degradation, theoretically making the C6 MODIS products more accurate compared to previous collections. C6 adjustments also introduce several improvements in the vegetation index (VI) retrieval algorithms. With these improvements, we expect only minor differences between data from Terra and Aqua, but significantly different results between C5 and C6. In this paper, we investigate three different MODIS products to determine the extent that improvements made to C6 influence the overall trend results for time series between 2001 and 2017. We focus on these three products specifically, both to allow for a comparison of vegetation index products—NDVI and EVI from MOD13C1, and NDVI and EVI calculated based on surface reflectance from MCD43C4—and also to gain an understanding of the improvements on an entirely different product from the same sensor, namely Land Surface Temperature (LST) from MOD11C2. For the MCD43C4 dataset, we find that 17.9% and 16.4% of EVI and NDVI pixels, respectively, display trend discordance between C5 and C6. For the MOD13C1 vegetation indices, we found comparable rates of trend discordance between C5 and C6: 18.5% and 17.4% for the EVI and NDVI pixels, respectively. For both products the greatest changes between C5 and C6 are an overall increase in pixels exhibiting a significant greening trend and an overall decline in pixels exhibiting a significant browning trend. Moreover, the largest differences between C5 and C6 for the NDVI and EVI data appear in cropland areas and in regions with relatively little human influence. In the Land Surface Temperature product (MOD11C2), the discordance between C5 and C6 is much lower: only 3.2% of day and 5.0% of night LST trends exhibited discordance between C5 and C6. We analyze the complementary results of vegetation index and land surface temperature trends and demonstrate that combining the results from different products observed at different portions of the electromagnetic spectrum—but linked through the biogeophysical processes of surface energy balance—allows us to portray change with more confidence than when relying on vegetation index data alone.

Introduction

Satellites provide the ability to observe large spatial areas over long periods, enabling researchers the opportunity to reveal both abrupt and subtle changes in the vegetated land surface (de Beurs et al., 2015, de Beurs et al., 2018, Fan and Liu, 2016). To allow such analysis, it is important to carefully analyze and compare image time series so that the behaviors of the sensors and the algorithms that generate products can be well understood. With a better understanding of our data sources, researchers can be more confident that no false changes or trends are being reported, for example, due to operational error or degradation of the sensors (Wang et al., 2012, Zhang and Roy, 2016). There is now an abundance of freely available data from multiple remote sensing platforms, enabling comparative studies to analyze the consistency and accuracy of these data. It is important to assess the stability across senor platforms to create long standing archives (Fan and Liu, 2016) but as operational life rises (Belward and Skoien, 2015) it will become increasingly essential to understand how degradation over time impacts results and how degradation can be separated from other observed changes. The Moderate Resolution Imaging Spectroradiometer (MODIS), launched first on board the Terra spacecraft in December 1999 and later on board Aqua in May 2002, forms the basis for several clearly documented and freely available products at various spatial and temporal resolutions. The periodically renewed product “collections”, which result from algorithmic adjustments and improvements, provide strong argument for using the MODIS products.

Over the years several improvements have been made to the MODIS products from one collection to the next. After changes are approved for each new collection, the entire MODIS archive is reprocessed to ensure that users have access to the most consistent and accurate data possible. The latest MODIS collection (C6) provides several algorithm improvements and calibration adjustments that correct for sensor degradation, theoretically making the products more accurate. For example, one important aspect that affects all products based on MODIS Terra data is the correction for sensor degradation. There has been noticeable degradation in both the Terra and Aqua MODIS sensors, but previous research has shown there was substantially greater impact on the Terra sensor (Lyapustin et al., 2014). The sensors are launched with onboard equipment, such as a solar diffuser and a solar diffuser stability monitor to perform periodic calibration, but this equipment can also experience operational degradation (Wang et al., 2012, Xiong et al., 2001). Two documented events are reported to have had significant impact on Terra’s degradation. First, during a pre-launch thermal vacuum test, a portion of the door paint came off the nadir aperture door and coated parts of the optics and scanning mirror. The paint was cleaned but lasting residue or damage to protective coating appears to have impacted performance (Lyapustin et al., 2014). Second, in May 2003, the solar door was permanently opened, and the solar door screen closed. The fixed position of this equipment has caused the solar door plate to degrade at a quicker rate, decreasing the reflectivity (Lyapustin et al., 2014). The implication of decreasing reflectivity is a reduced capability to track sensor response over time (Lyapustin et al., 2014). These events lead us to expect significant differences between the data collected from the MODIS sensors aboard Terra and Aqua.

The collection 6 adjustments were implemented not only to correct sensor degradation impacts, but also to introduce several improvements in the vegetation index (VI) retrieval algorithms (Didan et al., 2015). With these improvements, we expect only minor differences between the Terra and Aqua products, but significantly different results between C5 and C6 (Zhang et al., 2017). Other papers have explored the differences between trends in VI data based on collections 5 and 6. For example, Detsch et al (2016) applied the Seasonal Kendall trend test to compare Terra and Aqua MODIS VI data from C5 and C6 for the period 2003–2010 in a study area surrounding the Kilimanjaro region of Tanzania. They found that at a seasonal scale, products created from Terra and Aqua in C5 and C6 compared well with each other. However, throughout the length of the time series, the negative impacts created by the degradation of MODIS Terra were noticeable; the NDVI collected from Terra in C5 displayed more browning and less intense greening in comparison to the NDVI product collected from Aqua. The NDVI collected from Terra in C6 now displays more greening compared to the Aqua product. The presence of more greening trends in collection 6 points to the calibration changes that were made to compensate for degradation in the Terra MODIS.

A global comparison of collection 5 and 6 VIs focused only on trends in the annual maximum vegetation index (Zhang et al., 2017). They noted sensor degradation in both Terra and Aqua: during the time span of 2003–2015, Aqua experienced an increase in NDVI of 0.03% year−1 and an increase in EVI of 0.11% year−1 from C5 to C6, respectively. We also expect to find at least some minor differences in MODIS land surface temperature data (LST) between C5 and C6. There are few comparative studies covering the LST products except those of (Duan et al., 2017, Duan et al., 2018, Duan et al., 2019), which describe the algorithm improvements for bare soil regions.

In our analysis, we investigate four MODIS products (MOD13C1, MCD43C4, MCD43A4 and MOD11C2) to determine the extent that changes made to C6 impact the overall trend results. We compare vegetation index products—the NDVI and EVI available from MOD13C1 with the NDVI and EVI calculated from MCD43C4 and MCD43A4. We also explore changes in LST from MOD11C2 to see if the changes in trends evident in the VIs are also apparent in a different product from the same sensor. Our study differs from previous studies (Detsch et al., 2016, Zhang et al., 2017) in that we focus less on the difference between Terra and Aqua and more on the comparison between C5 and C6. We also apply a nonparametric trend analysis that is less sensitive than simple linear regression to outliers, seasonality, and autocorrelation (de Beurs and Henebry, 2004). Unlike many related studies, we look beyond the regional scale and calculate the trend results for the entire Western Hemisphere (excluding Greenland because of its extensive ice cover). We have previously demonstrated that the use of more than a single index time series can significantly improve trend interpretation and attribution, and we have advocated for the use of complementary suites of multiple indicators as the new standard approach for change analysis (de Beurs et al., 2015). Here we will follow this approach and not only investigate the changes in the vegetation indices by themselves, but also link the vegetation changes with warming and cooling of the land surface as observed by the land surface temperature data. Combining the results from different products observed at different regions of the electromagnetic spectrum—but linked through the biogeophysical processes of surface energy balance—allows us to portray change with more confidence than when relying on vegetation index data alone.

Section snippets

C5 and C6 data for comparison

The focus of this study relies on trend results derived from vegetation index data (VI) and land surface temperature data (LST). VIs are spectral transformations of two or more bands designed to enhance the contribution of vegetation properties. Although many vegetation indices exist, two of the most widely used are the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI), and these are the primary datasets analyzed in this study.

We use four different products

Results

An overview of all general trends for all products can be found in Table 3. We will discuss the differences in the individual products and the different collections below.

Trends in MODIS vegetation indices in collections 5 and 6

As early as 2012, Wang et al. (2012) reported on the impact of sensor degradation on MODIS collection 5 NDVI time series from Terra. That paper demonstrated a nearly threefold difference in the percentage of negative trends derived from Terra compared to those derived from Aqua for the period 2002–2010 (17.4% vs. 6.7%). Here we have presented a thorough and careful analysis of the consistency between MODIS collections C5 and C6 for the western hemisphere for three different products: a combined

Conclusions

There have been few papers discussing the discrepancy between C5 and C6 vegetation index data (Lyapustin et al., 2014, Zhang et al., 2017), and the major trend discordances between the two collections have received limited attention. As a result, even in 2018, there are still papers published based solely using C5 NDVI data (e.g., Fang et al., 2018), while others do not specify the collection of the data used (e.g., Browning et al., 2018, Murthy and Bagchi, 2018).

The aim of this study was to

Acknowledgements

This research was supported, in part, by the NASA Science of Terra & Aqua project NNX14AJ32G entitled Change in our MIDST: Detection and analysis of land surface dynamics in North and South America using multiple sensor datastreams, and the Center for Global Change and Earth Observations at Michigan State University. All MODIS data products used in this study were retrieved from the online Data Pool, courtesy of the NASA Land Processes Distributed Active Archive Center (LP DAAC), USGS/Earth

References (51)

  • X. Fang et al.

    Large-scale detection of vegetation dynamics and their potential drivers using MODIS images and BFAST: a case study in Quebec, Canada

    Remote Sens. Environ.

    (2018)
  • R. Fensholt et al.

    Evaluation of earth observation based global long term vegetation trends—comparing GIMMS and MODIS global NDVI time series

    Remote Sens. Environ.

    (2012)
  • P. Fu et al.

    Variability in annual temperature cycle in the urban areas of the United States as revealed by MODIS imagery

    ISPRS J. Photogramm. Remote Sens.

    (2018)
  • N. Pan et al.

    Increasing global vegetation browning hidden in overall vegetation greening: Insights from time-varying trends

    Remote Sens. Environ.

    (2018)
  • J. Walker et al.

    Land surface phenology along urban to rural gradients in the US Great Plains

    Remote Sens. Environ.

    (2015)
  • Z. Wan

    New refinements and validation of the MODIS land-surface temperature/emissivity products

    Remote Sens. Environ.

    (2008)
  • Z. Wan

    New refinements and validation of the collection-6 MODIS land-surface temperature/emissivity product

    Remote Sens. Environ.

    (2014)
  • D. Wang et al.

    Impact of sensor degradation on the MODIS NDVI time series

    Remote Sens. Environ.

    (2012)
  • Z. Wang et al.

    Capturing rapid land surface dynamics with Collection V006 MODIS BRDF/NBAR/Albedo (MCD43) products

    Remote Sens. Environ.

    (2018)
  • S.N. Williamson et al.

    Spring and summer monthly MODIS LST is inherently biased compared to air temperature in snow covered sub-Arctic mountains

    Remote Sens. Environ.

    (2017)
  • H. Zhang et al.

    Landsat 5 Thematic Mapper reflectance and NDVI 27-year time series inconsistencies due to satellite orbit change

    Remote Sens. Environ.

    (2016)
  • Y. Zhang et al.

    Reanalysis of global terrestrial vegetation trends from MODIS products: browning or greening?

    Remote Sens. Environ.

    (2017)
  • A. Bastos et al.

    Was the extreme Northern Hemisphere greening in 2015 predictable?

    Environ. Res. Lett.

    (2017)
  • P.S. Beck et al.

    Satellite observations of high northern latitude vegetation productivity changes between 1982 and 2008: ecological variability and regional differences

    Environ. Res. Lett.

    (2011)
  • N. Benas et al.

    Trends of urban surface temperature and heat island characteristics in the Mediterranean

    Theor. Appl. Climatol.

    (2017)
  • Cited by (33)

    • Multiscale spatial analysis of headwater vulnerability in South-Central Chile reveals a high threat due to deforestation and climate change

      2022, Science of the Total Environment
      Citation Excerpt :

      We estimated the absolute change of day land surface temperature between the period 2000–2005 and 2015–2020. This estimation was generated using the MOD11A2.006 product by the MODIS sensor, which provides an estimation of surface temperature based on land cover emissivity and radiance at 1000 m pixel resolution, which is considered a good indicator of energy balance on the Earth's surface (Bateni and Entekhabi, 2012; Castelli et al., 2018; Heck et al., 2019). Finally, we quantified the absolute change in the median snow cover during winter (accumulation season) between the period 2001–2005 and the period 2016–2020.

    • 9.12 - Weather, Climatic and Ecological Impacts of Onshore Wind Farms

      2022, Comprehensive Renewable Energy, Second Edition: Volume 1-9
    • Statistical inference for trends in spatiotemporal data

      2021, Remote Sensing of Environment
      Citation Excerpt :

      The NDVI3g data have been corrected to minimize the adverse effects of various factors such as sensor calibration loss, orbit drift, and volcanic eruption compared to previous AVHRR NDVI data (Pinzon and Tucker, 2014). The NDVI3g data are highly consistent with an earlier version of the comparable MODIS product (Collection 5) but span more years and have a longer history (Fensholt and Proud, 2012), which is why we analyzed NDVI3g instead of MODIS Collection 6 vegetation data (Heck et al., 2019; Zhang et al., 2017). NDVI3g has been widely used to investigate vegetation productivity and phenological change, and to map land cover (e.g., Chen et al., 2014; Fensholt and Proud, 2012; He et al., 2017).

    • A practical approach to reconstruct high-quality Landsat NDVI time-series data by gap filling and the Savitzky–Golay filter

      2021, ISPRS Journal of Photogrammetry and Remote Sensing
      Citation Excerpt :

      Normalized Difference Vegetation Index (NDVI) time-series data have been widely used in recent decades for monitoring vegetation dynamics in terrestrial ecosystems. For example, with the release of the NDVI time-series products provided by several satellite sensors with high revisit frequency, such as the Advanced Very High Resolution Radiometer (AVHRR) and Moderate-resolution Imaging Spectroradiometer (MODIS), many studies have successfully used these NDVI products to estimate vegetation productivity (Myneni et al., 2002), detect vegetation change (Cao et al., 2018a; Delbart et al., 2006; Gao et al., 2017; Heck et al., 2019), and map land cover and its changes (Friedl et al., 2002; Jia et al., 2014; Hu et al., 2021). However, the current NDVI time-series products, such as MODIS NDVI time-series data, have a relatively coarse spatial resolution (i.e., hundreds of meters), which greatly hinders their application in geographically heterogeneous areas (Gao et al., 2006; Hwang et al., 2011; Li et al., 2017; Rao et al., 2015).

    • Estimating daily full-coverage near surface O<inf>3</inf>, CO, and NO<inf>2</inf> concentrations at a high spatial resolution over China based on S5P-TROPOMI and GEOS-FP

      2021, ISPRS Journal of Photogrammetry and Remote Sensing
      Citation Excerpt :

      Datasets from S5P-TROPOMI and GEOS-FP in China during 01 June 2018 to 31 March 2020 are considered as the main variables of the inputs in the proposed framework. Some other datasets are employed as the covariates of inputs, such as meteorological factors (e.g., wind speed and specific humidity), Normalized Difference Vegetation Index (NDVI) (Fan and Liu, 2016; Heck et al., 2019), and land cover map (Liu et al., 2020; Xiong et al., 2017), to increase the robustness of the trained models. Meanwhile, measurements from the China National Environmental Monitoring Center (CNEMC) are regarded as the ground truth-values, including the NSC of O3, CO, and NO2.

    View all citing articles on Scopus
    View full text