High-Resolution Enhanced Product Based on SMAP Active-Passive Approach using Sentinel 1A and 1B SAR Data

Transcription

1 High-Resolution Enhanced Product Based on SMAP Active-Passive Approach using Sentinel 1A and 1B SAR Data Narendra N. Das 1, Dara Entekhabi 2, Seungbum Kim 1, Scott Dunbar 1, Andreas Colliander 1 Simon Yueh 1, Thomas Jagdhuber 3, Jeff Walker 4 1 Jet Propulsion Laboratory, California Institute of Technology, Pasadena, USA 2 Massachusetts Institute of Technology, Cambridge, MA, USA 3 German Aerospace Center, Microwaves and Radar Institute, Oberpfaffenhofen, Germany 4 Monash University, Australia

2 Outline of the Presentation a) Introduction to the SMAP Active-Passive (AP) Algorithm and potential to use Sentinel SAR b) Sample of SMAP_Sentinel Active-Passive Product c) AP Algorithm Parameters Estimation d) Validation of the Algorithm Output Using the Airborne Data from SMAPEx e) Validation of the High Resolution Soil Moisture Product Using the Data from the SMAP Cal/Val Core Sites f) Conclusion

3 SMAP AP Algorithm Adaptation Heritage Requires change due to following developments: a) Estimate new parameters for C-Band and multi-angular b) Computing parameters using time-series regression is not possible due to low revisit interval of Sentinel and time mismatch of swath overlap of SMAP and Sentinel c) The Active-Passive Algorithm is modified to accommodate the snapshot and work in the emissivity space, as shown below. Radiometer Brightness Temperature and Algorithm parameters Surface Temperature SAR co-pol backscatter SAR x-pol backscatter

4 Data Flow: Baseline L2_SM_SP Algorithm SMAP (L-Band) Sentinel 1A/1B (C-Band) Sentinel co-pol and x-pol observations SMAP TB Gridded at 9 km Note: It is a moving window 33 km domain is used for disaggregation to conform with the L2_SM_P_E. Disaggregated TB at 3 km Retrieved Soil Moisture from Disaggregated TB at 3 km

5 Data Flow: Option-1 L2_SM_SP Algorithm SMAP (L-Band) Sentinel 1A/1B (C-Band) SMAP Soil Moisture Gridded at 9 km Note: It is a moving window 33 km domain is used for disaggregation to conform with the L2_SM_P_E. Sentinel co-pol and x-pol observations Disaggregated Soil Moisture at 3 km

6 With the present data stream of Sentinel 1A and 1B, it is expected to get a global coverage in every 12 days. Twelve Days Coverage of SMAP-Sentinel high resolution soil moisture data from 1 st May, 2017 to 12 th May, 2017

7 SMAP-Sentinel-1 AP Soil Moisture Results May 17 th, 2015 over Southern Canada in Manitoba 3 [km] SMAP Radiometer + Sentinel Radar SMAP Active-Passive Soil Moisture Product at 3 km The above image shows that SMAP-Sentinel does capture finer details and the overall pattern is similar to what observed by SMAP AP retrievals 9 [km] Gridded SMAP Radiometer-based Soil Moisture

8 Data Driven Beta Retrievals -- Retrieved Beta from the Radiometer and Radar Data -- Brightness Temperature of Radiometer (K) -- Surface Temperature (K) -- Vegetation Opacity parameter -- Albedo -- SAR co-pol data -- SAR x-pol data -- Parameter derived from SAR co-pol and x-pol data The physically-based retrieval procedure of β accounts for the effects of vegetation/roughness on emission as well as on backscatter. This approach does not require time series. T. Jagdhuber 1, M. Baur 2, N.N. Das 3, D. Entekhabi 4, M. Link 1,5, 2016, Physically-based Retrieval of Active-Passive Covariation Parameter β for the SMAP Mission. To be submitted to TGARS. 8

9 Beta Retrieval Evaluate against the 84 days of SMAP Active and Passive Data New Beta Retrievals Beta Retrievals from V and vv combination, averaged over 84 days [-] Beta derived from Time series From V and vv combination

10 SMAP-Sentinel BETA Evolution over Global Domain Mean Beta goes to zero with increasing density of vegetation [-/-] High CV indicates Seasonality in Beta Coefficient of Variation (CV) in [-/-]

11 Dependence of Beta on Incidence Angle of Sentinel Data Grassland Cropland Shrubland Savanna The mean Beta values show clear dependence on Sentinel data incident angle.

12 Evaluation: Comparison with Airborne Data

13 Airborne L-band SMAPEx during 2015 in the Southeastern Australia The SMAPEx Domain Urban area Waterbody Urban area Urban area Waterbody Urban area

14 SMAPEx Domain 2015, TBv, [K] 1 km, EASE GRID Expected error in the SMAP TBv is ~3 to 4 [K] due to incidence angle correction The Airborne L-band radiometer shows minimal impact due to manmade structures

15 Manmade Structures Urban Areas Sentinel vh Urban Areas Urban Areas Small Ponds and Manmade Structures Significant Impact of urban areas and water bodies are found in the high-resolution Sentinel (vv and vh) observations Sentinel vv

16 Approach to Mitigate Impact of Manmade Structures and Water Bodies ( ) [-] [-] with Median Filter of [3X3] Sentinel vv Sentinel vv 1 km, EASE GRID 1 km, EASE GRID 1 km, EASE GRID [-] [-] Sentinel vh Sentinel vh 1 km, EASE GRID 1 km, EASE GRID

17 Examples of Median Filtering ( ) Mitigation of Outliers in the Sentinel Observations Before Filtering After Filtering Limited Mitigation of Outliers in the Sentinel Observations

18 SMAP-Sentinel AP and L2_SM_P_E ( ) SMAPEx TBv Obs at 3 km SMAP_Sentinel TBv at 3 km L2_SM_P_E Gridded at 9 km Minimum Performance Average VWC in the SMAPEx Domain is ~1 kg/m2 [K]

19 Summary of Validation Statistics of SMAP_Sentinel L2_SM_SP vs SMAPEx Date: L2_SM_SP vs SMAPEx (slope) L2_SM_SP vs SMAPEx (R) L2_SM_SP vs Min-Perf (slope) L2_SM_SP vs Min-Perf Total at 3 km Total at 9 km (R) Date: L2_SM_SP vs SMAPEx (slope) L2_SM_SP vs SMAPEx (R) L2_SM_SP vs Min-Perf (slope) L2_SM_SP vs Min-Perf Total at 3 km Total 9 km (R)

20 SMAP_Sentinel Product Comparison with Core Validation Sites Minimum 3 number of in situ stations required to be a Core Validation Site at 3 km BL: Baseline Opt-1: Option 1 MP_BL: TB without Disaggregation MP_Op: SM without Disaggregation

21 Date: Core Site: TxSON

22 Climate class: Temperate (Cfa) Landcover: Grasslands Soil texture: S-%: 41 C-%: 27 BD: 1.48 TxSON (Core Pixel) Alg ubrmse Bias RMSE R BL Opt-1 MP_BL MP_Op Black: Use recommended [Retrieval Quality Flag bit(0)=0] Gray: Retrieval attempted and succeeded but use not recommended [bit(0)=1, bit(1)=0, bit(2)=0] Green: Retrieval attempted but failed [bit(0)=1, bit(1)=0, bit(2)=1] Cyan: Retrieval not attempted [bit(0)=1, bit(1)=1] Not for public release or redistribution. This document has been reviewed and determined not to contain export controlled technical data.

23 Date: Core Site: Valencia

24 Climate class: Arid (BSk) Landcover: Croplands Soil texture: S-%: 19 C-%: 49 BD: 1.40 Valencia (Core Pixel) Alg ubrmse Bias RMSE R BL Opt-1 MP_BL MP_Op Black: Use recommended [Retrieval Quality Flag bit(0)=0] Gray: Retrieval attempted and succeeded but use not recommended [bit(0)=1, bit(1)=0, bit(2)=0] Green: Retrieval attempted but failed [bit(0)=1, bit(1)=0, bit(2)=1] Cyan: Retrieval not attempted [bit(0)=1, bit(1)=1] Not for public release or redistribution. This document has been reviewed and determined not to contain export controlled technical data.

25 Date: Core Site: Yanco

26 Climate class: Arid (BSk) Landcover: Grasslands Soil texture: S-%: 46 C-%: 31 BD: 1.44 Yanco (Core Pixel) Alg ubrmse Bias RMSE R BL Opt-1 MP_BL MP_Op Black: Use recommended [Retrieval Quality Flag bit(0)=0] Gray: Retrieval attempted and succeeded but use not recommended [bit(0)=1, bit(1)=0, bit(2)=0] Green: Retrieval attempted but failed [bit(0)=1, bit(1)=0, bit(2)=1] Cyan: Retrieval not attempted [bit(0)=1, bit(1)=1] Not for public release or redistribution. This document has been reviewed and determined not to contain export controlled technical data.

27

28 Conclusions Comparisons with available airborne field experiment (SMAPEx) shows that L2_SM_SP does positively use Sentinel-1 information to provide highresolution brightness temperature. Minimum Performance (in TB and soil moisture) confirm conclusions about impact of Sentinel information on soil moisture retrievals. Core Site comparisons at 3 km and 9 km are severely limited in number of sites and temporal samples at each site. They nonetheless show that the L2_SM_SP performs to target accuracy of the original L1 requirement. More spatial resolution (with respect to L2/3_SM_P_E) comes in a trade-off with accuracy. Utility is dependent on the application needs. SMAP-Sentinel Active-Passive Product (L2_SM_SP) is ready for Beta Release in couple of months.

29 Thank you Acknowledgements: SMAP project, JPL Science Data System, and all the SMAP Cal/Val partners.