Satellite-based remote sensing allows cryospheric scientists to better understand how the highly dynamic behavior of sea ice affects the climate system. The presence of sea ice influences atmospheric and oceanic temperature and circulation patterns and reduces the amount of solar radiation absorbed by the ocean. In addition, sea ice and its associated snow cover act as a strong insulator that restricts exchanges of heat, mass, momentum, and chemical constituents between the atmosphere and the ocean.

Multi-spectral radiometers such as MODIS can be used to determine sea ice extent, surface temperature, age, and other characteristics based on reflectance criteria and thermal data. For example, reflectance for snow-covered, opaque, white sea ice, thick first-year ice, and multiyear ice typically peaks between 0.4 µm and 0.8 µm and again at 1.9 µm. Young sea ice has a lower spectral albedo than older sea ice, when measured in this spectral range, by some 10-40 percent. Sea ice in the process of ablation and forming melt ponds exhibits decreasing reflectance between 0.6 µm and 0.8 µm, followed by a steady and uniformly low reflectance to approximately 1.6 µm (Riggs et al., 2015).

Ice surface temperature (IST) measurements offer scientists a useful tool that can be used to estimate radiative and turbulent heat fluxes for large-scale climate studies and as an additional discriminatory variable for identifying sea ice cover. For example, studies of MODIS Airborne Spectrometer (MAS) images in the Beaufort Sea, near St. Lawrence Island, Alaska, show that 271.4 K represents a typical threshold between the surface temperature of water and saline ice (Hall et al. 1998). In addition, because first-year ice has an emissivity of about 0.92 and multiyear ice has an emissivity of about 0.84—a difference that impacts recorded surface temperatures—researchers can use IST to distinguish the relative age of ice and infer relative ice thickness (Hall and Martinec, 1985).

The MODIS Science Team develops the algorithms used to detect snow cover and sea ice. The MODIS Data Processing System (MODAPS) generates the MODIS data sets and transfers them to NSIDC. The following sections outline the approach that the algorithm uses to generate the sea ice extent and ice surface temperature maps. Users seeking a fuller description should consult the MODIS Sea Ice Products User Guide to Collection 6.

Derivation Techniques and Algorithms

MYD29 is produced from the MODIS/Aqua Calibrated Radiances 5-Min L1B Swath 1km product (MYD021KM), the MYD03 geolocation product, and the MYD35_L2 MODIS Cloud Mask (click here to view a flowchart). The output file contains sea ice extent, ice surface temperature (IST), quality assessment (QA) checks, and latitude and longitude SDSs. To ensure complete global coverage, data are generated for all swaths acquired during a 24-hour day. However because the algorithm applies criteria tests based on visible and near-infrared wavelengths, swaths that were acquired completely in night mode contain only the temperature-based SDSs.

Data inputs to the sea ice algorithm are listed in Table 4:

Sea Ice

Sea ice analysis is constrained to daylight ocean pixels that have visible reflectance above 10 percent and are not obstructed by clouds. The new Version 6 land/water mask in MYD03 is used to guide processing on oceans. The mask's three ocean categories—deep, moderate and shallow—are all processed as ocean with no distinction between the categories.

Top-of-the-atmosphere (TOA) reflectance is read from MYD021KM and screened for missing and unusable data. Unusable data results from processing at L1B when the sensor radiance data fails to meet acceptable criteria. The algorithm maps missing MYD021KM data as missing in the output. No action is taken in the algorithm to analyze missing data. If other anomalous conditions are encountered in the input data, the algorithm makes no decision for that pixel, except for a few cases where a value indicating the source of the anomaly is written.

Land and inland water bodies are masked using the land/water mask in MYD03. Clouds are masked using the MODIS Cloud Mask (MYD35_L2) unobstructed field-of-view flag. If the flag is set to certain cloud then the pixel is set to cloud. If the cloud flag is set to clear, or any level of possible cloud, the pixel is interpreted as clear. This logic aims to increase the number of retrievals and balance the cloud conservative nature of the cloud mask, given the potential ice/cloud confusion and cloud contamination in IST.

Sea ice detection is achieved by using grouped criteria tests for sea ice reflectance characteristics in the visible and near-infrared regions. The algorithm assumes that sea ice is snow-covered and that the normalized difference snow index (NDSI) can be used to detect sea ice. The NDSI is computed from MODIS band 4 (0.55 µm) and band 6 (1.6 µm) as follows:

• NDSI = (Band 4 - Band 6) / (Band 4 + Band 6)

Pixels that pass all of the following criteria tests are identified as sea ice and passed through for additional tests:

• NDSI > 0.4
• Band 2 (visible) reflectance > 0.11
• Band 1 reflectance > 0.10

The algorithm also conducts intermediate checks based on theoretical bounds of 0 to 100 percent for reflectance and -1.0 to +1.0 for the NDSI. However, the sea ice test is performed regardless of violations of these limits and the QA data is set to indicate an unusual condition with low quality result.

Sea Ice QA

The QA SDS provides an indication of the quality of the sea ice by reflectance algorithm output for each pixel. QA performed within the algorithm checks that the input data and calculated intermediate values lie within theoretical or expected ranges. The default QA value is good (0); if any of the data checks fail, the value set to "other" (1). If the algorithm was not applied, for example to a land pixel, then the mask value is output. Unless the input data is unusable or missing, the data quality will usually be set to good.

Ice Surface Temperature (IST)

Radiance data for bands 31 and 32 are read from MYD021KM and converted to brightness temperatures using a version of the Plank function:

• T = c₂v / ln(1 + ((ec₁v³)/E))

where:

• T = brightness temperature (K)
• c₁ = 1.1910659 × 10^-5 mW·m^-2·sr·cm^-4
• c₂ = 1.438833 cm·°·K
• v = central wavelength in cm^-1
• E = radiance from sensor in mW·m^-2·sr·cm^-4
• e = emissivity

ISTs are estimated using a split-window technique and a version of Key's equation for the Advanced Very High Resolution Radiometer (AVHRR) that was adapted for use with MODIS Channels 31 and 32 (Key et al., 1997):

• IST = a + bT₃₁ + c(T₃₁ - T₃₂) + d[(T₃₁ - T₃₂)(sec(q) - 1)]

where:

• T₃₁ = brightness temperature of MODIS channel 31 (11 µm)
• T₃₂ = brightness temperature of MODIS channel 32 (12 µm)
• q = sensor scan angle from nadir
• a,b,c, and d are coefficients determined by a multilinear regression of brightness temperatures to estimated surface temperatures.

Coefficients for the IST equation were derived using MODIS spectral response functions and radiative transfer calculations. Different coefficients sets are used for each of three temperature ranges in the Northern and Southern hemispheres as shown in Table 5:

The IST algorithm is constrained to ocean pixels that are not obstructed by clouds and is run for daytime and nighttime data. The new Version 6 land/water mask is used to guide processing on oceans. The mask's three ocean categories—deep, moderate and shallow—are all processed as ocean with no distinction between the categories.

The MYD021KM data is screened for missing data and for unusable data. Unusable data result from the processing at L1B when the sensor radiance data fails to meet acceptable criteria. MODIS data may be unusable for several reasons. Specifics of L1B processing and criteria can be found at the MODIS Calibration Support Team (MCST) web page and in supporting documentation. In the case of missing data, the algorithm maps the data as missing in the output product. No action is taken in the algorithm to make an analysis for missing data. If other anomalous conditions occur with the input data, the algorithm makes no decision for that pixel, except for a few anomalous conditions where a value indicating the source of the anomaly is written. Quality checks are applied to the input data and at intermediate processing steps, quality value is lowered if checks are failed, and the final quality value associated with a pixel is output. Clouds are masked using the MODIS Cloud Mask data product (MYD35_L2) unobstructed field-of-view flag for both daytime and nighttime data. If the cloud flag is set to clear or any level of possible cloud the pixel is interpreted as clear. That logic is used to increase the number of retrievals balanced against the cloud conservative nature of the cloud mask, and accepts potential ice/cloud confusion and cloud contamination in IST.

ISTs are stored in the SDSs as calibrated data. To convert these values to kelvins, use scale_factor = 0.01 and add_offset = 0.0 in the following equation:

• IST = scale_factor × (calibrated data - add_offset)

The values for scale_factor and add_offset are also stored as Local Attributes with the IST SDSs. The valid range for ISTs is 210 K to 313.20 K.

Ice Surface Temperature (IST) QA

The IST QA data provides an indication of the quality of the output data. The algorithm checks that the input data from MODIS bands 31 and 32 and calculated intermediate values lie within theoretical or expected ranges. The default QA value is good (0); if any of the data checks fail, the value set to "other" (1). If the algorithm was not applied, for example to a land pixel, then the mask value is output. Unless the input data is unusable or missing, the data quality will usually be set to good.

Error Sources

Anomalies in the input data can propagate to the output. Refer to Table 4 for a list of MODIS products that are input to the MYD29 algorithm.

The MODIS instrument provides 12-bit radiometric sensitivity in 36 spectral bands ranging in wavelength from 0.4 µm to 14.4 µm. Two bands are imaged at a nominal resolution of 250 m at nadir, five bands at 500 m, and the remaining bands at 1000 m. A ±55 degree scanning pattern at an altitude of 705 km achieves a 2330 km swath with global coverage every one to two days.

The scan mirror assembly uses a continuously rotating, double-sided scan mirror to scan ±55 degrees, and is driven by a motor encoder built to operate 100 percent of the time throughout the six year instrument design life. The optical system consists of a two-mirror, off-axis afocal telescope which directs energy to four refractive objective assemblies, one each for the visible, near-infrared, short- and mid-wavelength infrared, and long wavelength infrared spectral regions.

The MODIS instruments on the Terra and Aqua space vehicles were built to NASA specifications by Santa Barbara Remote Sensing, a division of Raytheon Electronics Systems. Table 4 contains the instruments' technical specifications:

Calibration

MODIS has a series of on-board calibrators that provide radiometric, spectral, and spatial calibration of the MODIS instrument. The blackbody calibrator is the primary calibration source for thermal bands between 3.5 µm and 14.4 µm, while the Solar Diffuser (SD) provides a diffuse, solar-illuminated calibration source for visible, near-infrared, and short wave infrared bands. The Solar Diffuser Stability Monitor tracks changes in the reflectance of the SD with reference to the sun so that potential instrument changes are not incorrectly attributed to changes in this calibration source. The Spectroradiometric Calibration Assembly provides additional spectral, radiometric, and spatial calibration.

MODIS uses the moon as an additional calibration technique and for tracking degradation of the SD by referencing the illumination of the moon since the moon's brightness is approximately the same as that of the Earth. Finally, MODIS deep space views provide a photon input signal of zero, which is used as a point of reference for calibration.

For additional details about the MODIS instruments, see NASA's MODIS | About Web page.