The frozen/non-frozen state of the landscape is closely linked to numerous components of the climate system, including changes in the surface energy budget and evapotranspiration, vegetation growth and phenology, snow-melt dynamics, permafrost extent and stability, terrestrial carbon budgets, and land-atmosphere trace gas exchanges. Satellite-borne passive microwave sensors are particularly well-suited to monitoring global F/T status of the landscape because they are strongly sensitive to changes in dielectric properties at the surface that correspond to frozen and thawed states, are relatively insensitive to atmospheric contamination, and do not require solar illumination. The following sections outline the approach used to infer F/T state changes from remotely sensed T_b. For a complete description, see Kim et al. 2011.

For Version 4, F/T state is determined using a modified seasonal threshold algorithm (MSTA) that identifies F/T transition sequences in radiometric T_b time series. This approach exploits the dynamic temporal response of T_b to changes in the aggregate landscape dielectric constant that occur as the landscape transitions between predominantly frozen and non-frozen conditions (Kim et al., 2017). One advantage the MSTA offers relative to the seasonal threshold algorithm F/T in Version 3 (Kim et al., 2011) is that the T_b threshold selection does not depend on frozen and non-frozen reference states, derived by averaging T_b measurements over respective winter and summer periods. As such, F/T status is less sensitive to T_b data gaps during these reference periods.

Annual F/T thresholds were defined for each grid cell using an empirical linear regression between satellite T_b retrievals and daily ERA-Interim surface air temperature (SAT) estimates; separate AM and PM thresholds were derived using the corresponding daily SAT minimum (SATmin) and maximum (SATmax) values. SAT values closer to 0° C were more heavily weighted when selecting the corresponding T_b based F/T threshold for each grid cell; this weighting was derived using a cosine function within a temperature range from -60.0° C to 30.0° C which represented 99% of the SAT frequency distribution defined from the 36-year ERA-Interim SAT global climatology (Kim et al., 2017).

Satellite ascending and descending orbital data time series were processed separately to produce AM and PM F/T states; these states were then combined to provide a composite daily status which identifies transitional (AM frozen and PM thawed) and inverse transitional (AM thawed and PM frozen) conditions.

The global reanalysis data were also used to define the global F/T ESDR domain (see Figure 2) using a simple SAT-driven bioclimatic index (Kim et al. 2011) that identified all land areas where seasonally frozen air temperatures are a major constraint to ecosystem processes and land surface water mobility. The global F/T domain encompasses unmasked vegetated land areas where low temperatures significantly constrain annual vegetation productivity as defined from climatological reanalysis data. Masked areas include permanent ice and snow, barren land, open water, and regions unconstrained by freezing temperatures.

Quality Assessment

F/T daily accuracy was independently verified using daily SAT observations from global in situ World Meteorological Organization (WMO) weather station measurements. The landscape surface was classified as frozen or non-frozen using a simple 0° C SAT threshold and the results were then compared with the F/T ESDR daily classification results for every grid cell. The resulting spatial classification accuracy from all WMO stations was then summarized on a daily basis.

Global QA maps were constructed for each year of the record to provide a discrete, grid cell-wise indicator of the F/T ESDR's relative quality. These maps account for potential negative impacts from open water cover, terrain complexity, length of the F/T transitional season, and uncertainty in the MSTA F/T threshold that influences assessing mean annual F/T accuracy from WMO station comparisons. The annual QA map for 2012 in Figure X shows regions of relative high to low quality. The QA values were stratified into a smaller set of discrete categories ranging from low quality (estimated mean annual F/T accuracy < 70%) to best (> 95%). Mean proportions of the four QA categories encompass 54.1% (best), 36.0% (good; 85-95% agreement), 6.6% (moderate; 75-85% agreement), and 3.3% (low) of the global F/T domain.

F/T accuracy is primarily assessed in relation to daily maximum and minimum air temperature measurements from the global World Meteorological Organization (WMO) weather station network (3,160 ±487 stations); mean annual F/T spatial classification accuracies are approximately 91.8 ±1.03 and 84.2 ±1.00 percent for respective F/T ESDR PM and AM retrievals over the global domain and long-term record. The AMSR-E portion of the F/T ESDR has similar spatial classification accuracy.

The F/T ESDR accuracy shows strong seasonal and annual variability and is reduced during active F/T transition periods when spatial heterogeneity in landscape F/T processes is maximized in relation to the relatively coarse (~25 km) satellite footprint (Kim et al. 2011). A daily F/T spatial classification accuracy data quality (QA) metric is included with each daily F/T global grid, defined from pixel-wise comparisons of F/T classification accuracy in relation to co-located global weather station network daily air temperature (minimum and maximum) measurements (Kim et al. 2011, Kim et al. 2012); spatial classification accuracy is defined as the proportion of global stations where the daily F/T classification is consistent with station air temperature measurement based F/T estimates. Additional data quality (QA/QC) metrics are included that provide more spatially explicit information on algorithm performance, including potential negative impacts of temporal gaps in sensor data time series, precipitation, open water and dry soil effects, terrain and land cover heterogeneity, and uncertainty associated with use of global reanalysis temperature data to define per grid-cell frozen and non-frozen reference state thresholds for the seasonal threshold algorithm (STA) based F/T classifications. The resulting F/T ESDR database provides a consistent and continuous multiyear record of daily (AM and PM) F/T dynamics for the global biosphere.

 

Quality control (QC) metrics provide an indicator of F/T product quality for each grid cell within the F/T ESDR domain. Refer to Figure 3. The dynamic QC information includes the number of days per year with SSM/I radiometric brightness temperature data gaps and flagged precipitation events (Ferraro et al. 1996). The static QC information includes the potential effects of fractional open water cover (MODIS 17-class IGBP Global Land Cover Product, Friedl et al. 2002), complex topography (GLOBE, 1999) and heterogeneous land cover conditions defined from the 1 km resolution global land cover classification and Digital Elevation Model (DEM), and the average number of days (first year) with global model reanalysis air temperatures within ± 3 Celsius degrees of 0.0 Celsius degrees (Kim et al. 2011).

Version History