«All MODIS land products include quality assurance (QA) information designed to help users understand and make best use of the data that comprise each ...»
MODIS Land Products Quality Assurance Tutorial: Part-1
How to find, understand, and use the quality assurance
information for MODIS land products
NASA LP DAAC, USGS EROS Center, Sioux Falls, SD
(Created: March 1, 2012 | Last updated: April 4, 2014)
All MODIS land products include quality assurance (QA) information designed to help
users understand and make best use of the data that comprise each product. This Part-1
document contains material for beginners as well as intermediate MODIS land product users to help educate them on how to correctly use the QA information. It includes the
following five sections:
1. A general description of MODIS land products and QA
2. Why is it important for users to consult the QA information?
3. QA metadata sources in MODIS land products
4. Information on Land Data Operational Product Evaluation (LDOPE) tools
5. Links to MODIS QA-specific online information resources Part-2 of this document (in the near future) will provide a detailed delineation and examples of the pixel-level QA structure in three MODIS land product suites: land surface reflectance, vegetation indices, and BRDF and albedo.
Section-1: MODIS land products and their QA information The MODIS Adaptive Processing System (MODAPS) facility at the Goddard Spaceflight Center (GSFC) in Greenbelt, MD routinely produces MODIS land products from data derived from twin MODIS instruments aboard the Terra and Aqua platforms that were launched in December 1999 and May 2002 respectively. These data are archived at and distributed from the Land Processes Distributed Active Archive Center (LP DAAC) at the USGS EROS Center in Sioux Falls, SD. The MODIS Land Science Team (MODLAND) is responsible for the MODIS land products in terms of their QA and validation. They help evaluate and document the science quality of the products that are intended to constantly inform the user community. The Land Data Operational Product Evaluation (LDOPE) 1 MODIS Land Products Quality Assurance Tutorial: Part-1 [April 4, 2014] facility, collocated with MODAPS at GSFC, is responsible for the overall coordination of the QA activities in support of the MODIS Science Team. A fairly complex and laborious process, this includes the evaluation and documentation of the science quality of all MODIS land products, which is finally incorporated in the operational production code and carried within the products (at the pixel-level) and their metadata (at the file-level).
The MODIS land collections comprise over sixty-five products that include dailies to n-day composites. Over the 14+ years since Terra’s launch, MODIS data at the LP DAAC comprise over 31 million granules and ~650 Terabytes in volume (as of April 4, 2014).
Over this time, the NASA data discovery interfaces to search and procure data have also continued to evolve, and the current discovery interface called “Reverb” is the third incarnation, which replaces the Warehouse Inventory Search Tool (WIST). A number of other MODIS data search, access, and procurement methods exist as well.
Given the large number of MODIS land products (they include 16 daily, 1 four-day, 29 eight-day, 8 sixteen-day, 7 monthly, and 5 yearly products), the dependencies that exist between them, and the differences in the QA procedures that are applied to them, it is difficult to provide a generic description and approach that applies to all. Within the userdata interaction process, MODIS QA-related information has the potential to manifest itself at different times and locations. One of the primary goals of this document is to direct users to the best combination of QA information sources, and methods to tap them to help drive the data requirements for their research and applications.
NASA’s Earth Observing System (EOS) manages one of the largest science data production and applications enterprises in the world, of which MODIS datasets comprise a leading component. QA has always been identified as very essential to the success of the real-world applications that MODIS datasets help support but its complexity has discouraged widespread use. The mechanisms to generate, publish, access, communicate, and interpret QA for diverse suites of MODIS products are very elaborate. This document is intended to expose users to basic information that successfully helps initiate their interaction with the QA layers of MODIS land products.
Section-2: Why is it important for users to consult the QA information?
MODIS QA information provides vital clues regarding the usability and usefulness of the data products for any particular science application. Usability is the capability of being used for a particular purpose while usefulness refers to what extent something serves a purpose 2 MODIS Land Products Quality Assurance Tutorial: Part-1 [April 4, 2014] towards meeting a practical objective. Usability and usefulness address any of the following
requirements that MODIS QA information provide, which are not mutually exclusive:
Are sufficiently enough cloud-free data available to meet the requirements of a • particular science application?
Do sufficiently enough data meet the nominal output specifications as expected by • the product’s algorithm?
What proportion of data artifacts and anomalies present in the data are deemed to • exist within a satisfactory threshold to proceed with a particular science application?
Are there mitigating conditions under which we can rule certain science data layers • within a product as more or less useful than others?
Are the science data layers derived using the main algorithm deemed satisfactory • compared to the back-up algorithm or vice-versa?
Parts 2 through 4 of this document (that are available separately), provide details regarding the QA structure and implementation within MODIS Land Surface Reflectance, Vegetation Indices, and Bidirectional Reflectance Distribution Function and Albedo product suites.
Please note that the examples provided in these product suites demonstrate how the QA characterization propagates to the higher-level products, and underscores the need to understand the data fidelity at the very beginning stage when users contemplate use of a particular MODIS data product.
Given the fact that any particular MODIS land product is the result of a fairly complex process that involves a science algorithm, inputs that include MODIS level-1B data, ancillary data, lookup tables, auxiliary inputs, and possibly, other derived MODIS data products, users run a serious risk by not consulting the QA information. Some of the known sources of error that impact data quality include data loss due to instrument contact errors, striping in the land surface reflectance data, geolocation errors traceable to instrument maneuvers, effects of solar eclipse on the data, and problems stemming from the cloud mask, especially as a function of latitude. MODIS land data product users are therefore strongly encouraged to consult the QA information before they decide to use their data.
3 MODIS Land Products Quality Assurance Tutorial: Part-1 [April 4, 2014] Section-3: QA information sources in MODIS land products Typically, a MODIS HDF dataset contains several Science Data Sets (SDS), one or more QA data layers, and metadata. The QA data layers provide pixel-level QA for the science data, and the metadata describe summary statistics of certain attributes, and also a statement about the product QA. Three sources of QA information exist within MODIS
land products that serve specific purposes, and include the following:
★ File-level metadata ★ Pixel-level metadata ★ LDOPE Web information File-level metadata: File-level QA refers to metadata that summarizes the data quality within that file. Please note that this file-level assessment largely helps the search and discovery process, and users should not solely rely on them as a means to filter data for their science
application needs. They include the following:
1. Additional attribute metadata returned from a user search on a data discovery interface (this includes an overall percent quality, and a percent-based assessment of product-specific variables)
2. Granule-level QA Stats and QA Flags metadata returned from a user search on a data discovery interface (this includes percent-based assessments of cloud-cover, missing-, interpolated-, and out-of-bounds data)
3. The encapsulated metadata that exist in the header of the HDF file
4. The external xml metadata file Essentially, these metadata sources provide the same information that users encounter at different stages of the data search and acquisition processes. The first two are generally designed to help as part of the user search and screening processes, especially as users look for good quality, cloud-free datasets. Numbers 3 and 4 refer to information from acquired products that contain the same metadata represented in 1 and 2. This document’s major emphasis is on pixel-level metadata, discussed next.
Pixel-level metadata: QA metadata that reside at the pixel-level is most valuable for applications that rely on consistent use of particular MODIS land products. For instance, pixel-level QA metadata may help applications based on time-series analyses to ensure that 4 MODIS Land Products Quality Assurance Tutorial: Part-1 [April 4, 2014] their data inputs remain consistently of reliable quality. Two kinds of pixel-level metadata implementations exist in MODIS land products.
1. The first includes a QA SDS that contains multiple information sources accomplished through binary encoding.
2. The second involves a QA SDS that contains a single information source, such as pixel reliability in the Vegetation Indices products, or albedo quality in the BRDF/Albedo products.
All MODIS land products contain one or more SDS devoted to QA among the multiple HDF arrays. These SDSs are critical to understand, parse, and interpret pixel-level QA. As users open the MODIS HDF dataset in any image processing software system, the one or more QA-specific SDSs are identifiable through the inclusion of “QA”, “QC”, or “Quality” in their name. Table 1 identifies the QA SDS arrays for each of the MODIS land products.
Pixel-level QA varies between products and their levels. In general, there are two types of pixel-level QA metadata provided.
1. MODLAND-wide QA: The first is the MODLAND-wide QA bits that provide 1 or 2 generic QA (least-significant**) bits for each pixel of every product. Its purpose is to provide a consistent quality interpretation across all MODIS land products. Prior to Collection-5, a 2-bit QA was used to describe four potential conditions. Table 2 describes the codes and their interpretation for this product collection-wide QA.
Starting with Collection-5, some MODLAND products were implemented with a 1-bit generic product assessment (Table 3) rather than a 2-bit summary (Table 2) in part to reflect * A bit (short for Binary Digit) is the smallest unit of information/memory in digital computing that can represent two possible values, represented by 0 and 1.
** The least-significant bit is the lowest bit in a series of numbers in binary notation, located at the far right of a string; also referred to as the right-most bit.
Pixel-level QA code Interpretation 0 Pixel produced, good quality, not necessary to examine more detailed QA 1 Other quality (produced or not produced; if produced unreliable or unquantifiable quality, examination of more detailed QA is recommended) Table 3: Pixel-level QA across all MODIS land products implemented in Collection-5
2. Product-specific QA: The second type of pixel-level QA addresses product-specific attributes. This metadata may address a variety of characteristic conditions that constitute a product’s elements. For instance, “products that can have meaningful error estimates assigned to them store per-pixel uncertainty estimates and/or ranges: for example, the land surface temperature product stores emissivity and temperature error estimates. Information on external factors known to affect product quality and consistency is also stored for each product. These data include atmospheric conditions (e.g., cloud cover); surface type (e.g., ocean, coast, wetland, inland water); scan, solar and viewing geometry; and whether dynamic ancillary data or backup estimates have been used as input (e.g., aerosol climatology estimates used to replace missing observations in the MODIS aerosol product)” (Roy et al., 2002).
Given the variety of MODIS land products, we cannot elaborate on all the different pixellevel QA attributes in this document. Links to relevant sources that offer both distilled versions of this information as well as complete product file specifications maintained by the MODIS Science Team are provided in Section-5. Users are strongly encouraged to consult these sources to better understand and interact with their particular MODIS land products.
Generic description of the MODIS QA binary bits and bit-fields
Users often encounter problems with interpreting the binary encoded bitmap that represents useful product-specific QA metadata. This section provides a generic introduction to the basic elements of binary notation used to represent the pixel-level QA metadata.
The simplest way to parse the QA bits is by understanding the binary notation (built on base-2 rather than base-10) used to represent the values. A single bit represents two values
The QA-specific SDS for each MODIS land product (listed in Table 1) generally breaks down into four columns. The order of these columns as depicted under the “Layers” tab in the product documentation on the LP DAAC Web page is described below. This order and
The following four examples demonstrate how you handle a particular pixel-level QA value from four different products.