Surface Freeze-Thaw Dataset Development of the Antarctic Ice Sheet Based on Multisource Data (1999-2019)

Liu, Y.^1,2  Zhou, C. X.^1,2*  Zheng, L.^1,2  Wang, Z. M.^1,2

1. Chinese Antarctic Center of Surveying and Mapping, Wuhan University, Wuhan 430079, China;

2. Key Laboratory of Polar Surveying and Mapping, Ministry of Natural Resources, Wuhan 430079, China

Abstract: Surface melting of the Antarctic Ice Sheet (AIS) is a factor that is sensitive to global climate changes. It constitutes a considerable contribution to the surface mass and energy balance of the AIS. In this study, we ranked the snowmelt determined by the radiometer, scatterometer, and climate model using categorical triple collocation (CTC), which can identify the most accurate observations with unknown true values. Then, a CTC fusion product was generated by adopting the best data source for each pixel during 1999–2019. By combining the advantages of the different observations, the fusion product reports the daily freeze-thaw status of the AIS, with a high spatial resolution of 4.45 km. The data used in this dataset are stored as integers in which 1 represents melting and –1 represents freezing. The dataset is archived in the .nc format, and the file size is 35.7 MB after compression.

Keywords: Antarctic Ice Sheet; freeze-thaw; data fusion; CTC; 1999–2019

Dataset Availability Statement:

The dataset supporting this paper was published and is accessible through the Digital Journal of Global Change Data Repository at: https://doi.org/10.3974/geodb.2020.05.01.V1.

1 Introduction

Surface melting of the Antarctic Ice Sheet (AIS) is a factor that is sensitive to global climate changes. It constitutes a substantial contribution to the surface mass and energy balance of the AIS^[1-2]. Due to to global warming, runoff generated by AIS melting has introduced large uncertainties into the prediction of changes in climate and sea levels. It is of great significance to explore the development and driving factors of surface melting in the AIS. However, early studies developed AIS freeze-thaw products mostly based on radiometer data, resulting in a low spatial resolution (25 km). In addition, there are several limitations in using a single sensor to detect snowmelt under various conditions (latitude, altitude, topography, etc.)^[2].

We employed categorical triple collocation (CTC) to estimate the rankings of snowmelt derived by the radiometer, scatterometer and climate model^[3], and a CTC fusion product over the period of 1999-2019 was generated by combining optimal snowmelt estimations^[4]. The fusion product has a spatial resolution of 4.45 km, which is higher than that of other compared products. Additionally, the fusion product combines the advantages of different observations by using multisource data.

2 Metadata of the Dataset

The metadata summary of the “Antarctic Ice Sheet freeze-thaw dataset (1999-2019)” is listed in Table 1, including the full name, short name, authors, geographical region, time, dataset files, data publisher, and data sharing policy of the dataset, etc.

Table 1  Metadata summary of the “Antarctic Ice Sheet freeze-thaw dataset (1999-2019)”

3 Methods

We first estimated the relative performances of snowmelt derived by the radiometer, scatterometer and climate model in the AIS. Then, the daily freeze-thaw product over the period of 1999-2019 was generated by combining optimal snowmelt estimations.

3.1 Data Sources

Surface melting in the AIS was detected by radiometer, scatterometer, and model. For the radiometer, the daily brightness temperature (T_b) in both ascending and descending orbits spans from 1987 to present with a spatial resolution of 25 km^[6-7]; the Special Sensor Microwave/Imager (SSM/I) and Special Sensor Microwave Imager/Sounder (SSMIS) Level-3 Southern Hemisphere EASE-Grid Brightness Temperature dataset were from the National Snow and Ice Data Center (NSIDC)^[8]. We utilized both passes of the SSM/I horizontally polarized 19.35 GHz T_b to determine surface melting characteristics.

For the scatterometer, we used the Quick Scatterometer (QSCAT), which was a quick recovery mission to replace the NASA Scatterometer. Since the QSCAT ceased operation in 2009, the Advanced Scatterometer (ASCAT) has been used to extend the QSCAT backscatter record for studies on snow and ice melting. ASCAT is a C-band (5.3 GHz) scatterometer with three vertically polarized antennas for the Exploitation of Meteorological Satellites (EUMETSAT) MetOp-A and MetOp-B. To improve the utility of the data, the Scatterometer Image Reconstruction (SIR) product was developed by the NASA Scatterometer Climate Record Pathfinder (SCP) project^[9]. The radar scatterometer-based time series achieved a balance between high resolution and coverage in the polar region and are publicly available (http://www.scp.byu.edu/). We utilized the daily SIR product to detect surface melting with a nominal pixel spacing of 4.45 km for both QSCAT and ASCAT.

Forced by the ERA-Interim reanalysis, the Regional Atmospheric Circulation Model (RACMO2) combines the atmospheric dynamics description from the High Resolution Limited Area Model (HIRLAM) with the physical processes output by the ECMWF global model. RACMO2 has been widely used in studies on surface mass balance in the Antarctic Ice Sheet, such as melt flux, snowdrift, precipitation and sublimation^[3]. Here, we utilized the RACMO2 liquid water content (LWC) to detect surface melting.

3.2 Categorical Triple Collocation (CTC)

Triple collocation (TC) is a general solution used to validate measurements with unknown true values by assuming that the measurements (M) are related to the true values (T) as follows^[10]:

                                                                                                                (1)

where i represents the different measurement systems, A and B are calibration parameters and ε indicates the random error. TC is now widely used in estimating the errors of various measurements, such as soil moisture, precipitation, snow depth, surface temperature, wind speed and leaf area index. It is assumed that errors of the measurement systems are not correlated with each other and with the true value. For binary wet/dry snow retrievals with only two values (1 and -1), Eqution (1) can be rewritten as:

                                                                                                                 (2)

ε only has three possible values: 0 indicates that the observation is correct, 2 indicates that dry snow has been mistaken for wet snow, and -2 indicates that wet snow has been mistaken for dry snow. ε is correlated with T for binary retrievals, which violates the key assumption of classical TC.

Instead, the balanced accuracy (α) of the measurement systems was introduced to measure the relative performances of the binary classifications and correlate the errors and truth^[11]:

                                                                                             (3)

where ψ and η are the sensitivity (the probability of the measurement being true when it is wet snow) and specificity (the probability of the measurement being true when it is dry snow), respectively.

The sample covariance matrix (Q) corresponds to the balanced accuracy α for the stationary variables^[12]:

                                                    (4)

For the non-stationary variables that show significant seasonal variations (e.g., wet/dry snow retrievals), Eq. (4) can be generalized as:

           (5)

where p(t) ≡ pM (M = 1| t) and t is time. The random errors ε for the different measurement systems are supposed to be conditionally independent of each other when applying CTC (i.e., Pr(ε_i, ε_j|T) = Pr(ε_i|T) Pr(ε_j|T), i ≠ j). There are three equations for the three measurement systems when i ≠ j. If we define W_i as follows:

                                                                         (6)

Then we have:

                                                                                           (7)

Since W_i is a monotonically increasing function of α_i, the sorting of W in descending order represents the rankings of the performances for the corresponding measurement systems^[3,11].

3.3 Methods

The flowchart outlining how the freeze-thaw dataset was developed is shown in Figure 1. We first obtained the time-series data of the different measurement systems. Then, the radiometer grids and RACMO2 grids were reprojected and resampled to the same spatial resolution as the scatterometer (4.45 km). The freeze-thaw dataset was generated by adopting the best source of CTC rankings.

In this study, surface melting detected by the radiometer and scatterometer was based on thresholding the T_b values and backscatter values σ⁰ time series. Wet snow was identified when the daily observations exceeded (or dropped below) a certain value. The general methods can be described as follows:

                                       (8)

Figure 1  Flowchart of outlining the development of the freeze-thaw dataset

where t is time and M (t) represents the wet/dry state, where M = 1 indicating wet snow and M = ‒1 indicating dry snow. For the radiometer, we utilized the horizontally polarized 19 GHz T_b to recognize surface melting, and T_wm was the winter (from May to July) mean T_b. Surface melting detected by this method agreed well with the positive air temperature (T_air) when b was set to 30 K. For QSCAT, b = 2 dB was empirically determined by comparison with T_air^[6,9,13]. The threshold for ASCAT was set to 2.7 dB. Surface melting derived from RACMO2 was identified when the LWC exceeded 0.4 mm (i.e., 0.4 kg·m^‒2).

Assuming conditional independence between the three different measurement systems, the 3 × 3 covariance matrix can be decomposed to estimate the rankings of the OLW products with respect to their balanced accuracies. The ranking of the OLW products based on the CTC algorithm can be summarized as follows: (i) calculate the covariance matrix Q from the scatterometer, radiometer and RACMO2 OLW estimations; (ii) estimate W from Q based on Eqution (7); and (iii) obtain the rankings by sorting W in descending order. A CTC fusion OLW product was generated by adopting the best source for each pixel.

4 Data Results and Validation

4.1 Data Files

The freeze-thaw dataset was named “DailyMeltingAntarctic_1999-2019”, and it is composed of 2 subsets named “Melt1999_2009.nc” and “Melt2010_2019.nc”. The naming convention of the dataset was “subject+study area+time”, and the subsets were named in the form of “subject+time”. The two subsets stored the freeze-thaw data of the AIS during 1999‒2009 and 2010‒2019.

4.2 Data Results

The dataset covers from 1 July 1999 to 30 June 2019. To capture continuous melt seasons, we defined a melt year as starting on 1 July and ending on 30 June in the following year. The dataset has a spatial resolution of 4.45 km, and its temporal resolution is 1 day. The data type of this dataset is integer in which 1 represents melting and –1 represents freezing. The dataset is archived in the .nc format, and the file size is 35.7 MB after compression.

The evolution of surface melting of the AIS in the summer of 2012/13 is shown in Figure 2. In austral summer, surface melting first occurred in the northern part of the Antarctic Peninsula (AP) and the coastal areas of the AIS and then gradually expanded towards higher latitudes. In mid-December, large-scale surface melting occurred in the Larsen and Wilkins ice shelves. The melting area reached its peak in early January. By this time, most of the ice shelves had experienced melting, and the groundling ice sheet had also experienced sporadic melting. After the melting peak, the melting area shrank rapidly. In February, only the AP and a small number of ice shelves were still melting and then gradually froze.

Figure 2  Evolutions of surface melting of the AIS in the summer of 2012 and 2013

4.3 Data Validation

Suffering from a lack of in situ measurements of snow liquid water, melting products derived from satellites or models are usually evaluated when the recorded air temperature (T_air) from automatic weather stations (AWSs) exceeds 0 ºC. We utilized the Zhongshan Station T_air data from July 2016 to June 2018 to validate the accuracy of the data product. Melt signals derived from our product and the maximum T_air showed a high overall accuracy of 0.922.

5 Discussion and Conclusion

Surface melting derived from the scatterometer, radiometer, and RACMO2 data showed a complementary nature^[2]. The scatterometer showed a higher sensitivity to liquid water than the radiometer^[7]. The radiometers showed the best performance in places with long melt seasons^[3]. The RACMO2 model was the most applicable in mountainous areas with rock outcrops where it is difficult for satellites to work^[2]. Therefore, our product is more accurate than using a single data source product because it combines the advantages of various observations.

By merging the freeze-thaw products derived from scatterometer, radiometer and climate model, we developed a surface freeze-thaw dataset with a high temporal and spatial resolution for the AIS based on multisource data. The data validation was conducted using the AWS T_air data, and the validation results demonstrated the high accuracy of the product.

Author Contribution

Zhou, C. X., Liu, Y., and Wang, Z. M. developed the total design of the experiment and final dataset. Liu, Y. contributed to the data collection and processing. Zheng, L. and Liu, Y. designed the algorithms of the dataset. Liu, Y. conducted the data validation. Zhou, C. X. and Liu, Y. wrote the data paper.

References

[1]      Wang, X. D. Antarctic Ice Sheet freeze-thaw detection based on Active and Passive Microwave Remote Sensing [D]. Changsha: Central South University, 2013.

[2]      Zheng, L., Zhou, C., Liang, Q. Pan-Antarctic snowmelt detected by microwave remote sensing and the multi-scale driving forces [D]. Wuhan: Wuhan University, 2019.

[3]      Zheng, L., Zhou, C., Liang, Q. Variations in Antarctic Peninsula snow liquid water during 1999–2017 revealed by merging radiometer, scatterometer and model estimations [J]. Remote Sensing of Environment, 2019, 232: 111219.

[4]      Liu, Y., Zhou, C. X., Zheng, L., et al. Antarctic Ice Sheet Freeze-thaw Dataset (1999–2019) [J/DB/OL]. Digital Journal of Global Change Data Repository, 2020. https://doi.org/10.3974/geodb.2020.05.01.V1.

[5]      GCdataPR Editorial Office. GCdataPR data sharing policy [OL]. https://doi.org/10.3974/dp.policy.2014.05 (Updated 2017).

[6]      Steffen, K., Abdalati, W., Stroeve, J. Climate sensitivity studies of the Greenland ice sheet using satellite AVHRR, SMMR, SSM/I and in situ data [J]. Meteorology and Atmospheric Physics, 1993, 51(3/4): 239‒258.

[7]      Tedesco, M. Assessment and development of snowmelt retrieval algorithms over Antarctica from K-band spaceborne brightness temperature (1979‒2008) [J]. Remote Sensing of Environment, 2009, 113(5): 979‒997.

[8]      SSM/I Level-3 EASE-Grid Data Product. US National Snow and Ice Data Center-NSIDC.

https://nsidc.org/data.

[9]      Long, D. G., Hardin, P. J., Whiting, P. T. Resolution enhancement of spaceborne scatterometer data [J]. IEEE Transactions on Geoence & Remote Sensing, 1993, 31(3): 700‒715.

[10]   Stoffelen, A. Toward the true near-surface wind speed: error modeling and calibration using triple collocation [J]. Journal of geophysical research: oceans, 1998, 103(C4): 7755‒7766.

[11]   McColl, K. A., Roy, A., Derksen, C., et al. Triple collocation for binary and categorical variables: application to validating landscape freeze/thaw retrievals [J]. Remote Sensing of Environment, 2016, 176: 31‒42.

[12]   Parisi, F., Strino, F., Nadler, B., et al. Ranking and combining multiple predictors without labeled data [J]. Proceedings of the National Academy of Sciences, 2014, 111(4): 1253‒1258.

[13]   Jay, Z. H., Fiegles, S. Extent and duration of Antarctic surface melting [J]. Journal of Glaciology, 1994, 40(136): 463‒475.