Mapping Evapotranspiration in the Sacramento San Joaquin Delta using simulated ECOSTRESS Thermal Data: Validation and Inter-comparison

Transcription

1 Mapping Evapotranspiration in the Sacramento San Joaquin Delta using simulated ECOSTRESS Thermal Data: Validation and Inter-comparison Andy Wong 1, Yufang Jin 1, Eric Kent 1, Kyaw Tha Paw U 1, Jay Lund 1 ; Ruyan He 2 ; Joshua B. Fisher 3, Glynn Gulley 3, Gerardo Rivera 3, Christine Lee 3, Simon Hook 3 ; Josué Medellín-Azuara 4 ; Feng Gao 5 1 University of California, Davis, CA, USA 2 China University of Mining & Technology (Beijing), Beijing, China 3 Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Dr., Pasadena, CA, USA 4 University of California, Merced 5 U.S. Dept. of Agriculture - Agriculture Resource Service, Beltsville, MD, USA April 3, 2017.

2 Motivation Evapotranspiration (ET) is perhaps the most difficult hydrological flux to measure or model, especially at regional scales and greater. Hydrology 2020: An Integrating Science to Meet World Water Challenges (IAHS Press, 2006) Land Surface Temperature (LST) Single Channel Method Brightness Temperature from Landsat 8 Thermal Infrared Sensor: > Spatial Resolution: 100 m/pixel [resampled or sharpened to 30m/pixel] > Temporal Resolution: 16 Days Surface energy balance remote sensing ET models: λλλλ = RR nn GG HH TT λλλλ = αα aa (RR nn GG) TT aa +γγ TT aa Interpolation or Data Fusion Daily ET

3 Motivation Evapotranspiration (ET) is perhaps the most difficult hydrological flux to measure or model, especially at regional scales and greater. Hydrology 2020: An Integrating Science to Meet World Water Challenges (IAHS Press, 2006) Land Surface Temperature (LST) Single Channel Method Brightness Temperature from Landsat 8 Thermal Infrared Sensor: > Spatial Resolution: 100 m/pixel [resampled or sharpened to 30m/pixel] > Temporal Resolution: 16 Days Surface energy balance remote sensing ET models: λλλλ = RR nn GG HH TT λλλλ = αα aa (RR nn GG) TT aa +γγ TT aa Interpolation or Data Fusion Daily ET

4 ECOSTRESS > Ground Sample Distance (m): ~68.5 x 38.5 > 5 TIR bands > Temporal Resolution: ~4 Days > Spatial Coverage: Continental US & ECOSTRESS Projects sites > Launch Date: Jun 6, 2018 > Nominal mission lifetime: 1 year > Irregular overpassing time due to ISS orbit. Help reduce uncertainty in Remote Sensing ET estimates New uncertainty Sensor Overpassing time at the SSJ Delta VIIRS Landsat

5 Research Objectives How ECOSTRESS will reduce uncertainty in remote sensing ET estimates? 1) What is the uncertainty of Landsat data driven remote sensing ET estimates? Compare Landsat remote sensing ET against ground measurement and model inter-comparison. 2) How does uncertainty in LST propagate in ET model? 3) How does increased temporal frequency of LST data reduce uncertainty in ET estimate? Generate simulated ECOSTRESS LST data. Compare remote sensing Rn, ET driven by simulated ECOSTRESS LST data against ground measurement and Landsat driven ET estimates.

6 Remote Sensing ET models > METRIC + ETrF Interpolation λλλλ = RR nn GG HH Use Landsat LST to compute instantaneous Rn & H Estimate 30m instantaneous ET during Landsat overpassing date. Derive fraction of reference evapotranspiration (ETrF) by dividing instantaneous ET by reference ET. Assume ETrF is constant throughout the day and Interpolate overpassing ETrF to estimate 30m daily ET. > PT-UCD + ETrF Interpolation λλλλ = αα TT aa TT aa +γγ TT aa Use Landsat and MODIS LST to compute daily Rn. PT Coefficient, αα, partition available energy (RR nn GG) to λλλλ (RR nn GG) Use Landsat derived LAI and NDMI to estimate αα. The relationship is crop specific and calibrated with ground measurement. Interpolate ETrF and estimate 30m daily ET.

7 Ground Measurements Crop Type Site Counts Crop Type Site Counts Alfalfa 7 Corn 6 Pasture 4 Rice 1 Beardless Wheat Tomato/ Watermelon 1 1 ID Project/Network System Research Group D1-14 Crop Consumptive Water Use Estimation in the Sacramento-San Joaquin Delta Surface Renewal Dr. Kyaw Tha Paw U US-Twt,Tw3 US-Bi1-2 Ameriflux network Eddy Covariance Dr. Dennis Baldocchi RRT1-2 ECOSTRESS calibration and validation sites Eddy Covariance JPL

8 Model Comparison: Daily ET 1) What is the uncertainty of Landsat data driven remote sensing ET estimates? [Preliminary Results] Compare Landsat remote sensing ET against ground measurement and model inter-comparison. > EToF = ET/ETo (on Landsat overpassing date) > ET (on Landsat overpassing date) = interpolated EToF x ETo

9 Model Comparison: Daily ET 1) What is the uncertainty of Landsat data driven remote sensing ET estimates? [Preliminary Results] Compare Landsat remote sensing ET against ground measurement and model inter-comparison. > EToF = ET/ETo (on Landsat overpassing date) > ET (on Landsat overpassing date) = interpolated EToF x ETo

10 Model Comparison: Daily ET PTUCD + Landsat METRIC + Landsat Crops All Alfalfa Pasture Corn Rice Wheat Watermelon (mm/day) PTUCD + METRIC + Landsat Landsat RMSE R MAE RMSE R MAE RMSE R MAE RMSE R MAE RMSE R MAE RMSE R MAE RMSE R MAE

11 Model Comparison: Inst. ET > Estimate and measuring Instantaneous ET is challenging. > Uncertainty in METRIC s instantaneous ET estimates contributes to the scatering of METRIC s daily ET estimates. > METRIC assume EToF is constant throughout the day, which may not be applicable in some region. METRIC + Landsat Crops All Alfalfa Pasture Corn Rice Wheat Watermelon METRIC + Landsat RMSE 0.18 R MAE 0.15 RMSE 0.12 R MAE 0.10 RMSE 0.12 R MAE 0.11 RMSE 0.29 R MAE 0.24 RMSE 0.28 R MAE 0.25 RMSE 0.12 R MAE 0.10 RMSE 0.16 R MAE 0.13 (mm/hr)

12 Estimating Daily ET Crops All Alfalfa Pasture Corn Rice Wheat Watermelon (mm/day) Station inst. PTUCD + METRIC + EToF x ETo Landsat Landsat RMSE R MAE RMSE R MAE RMSE R MAE RMSE R MAE RMSE R MAE RMSE R MAE RMSE R MAE Station Inst. EToF x ETo

13 Estimating Daily ET > The constant EToF assumption has less uncertainty in the afternoon. > ECOSTRESS LST in the afternoon might improve daily ET estimates for METRIC. Landsat Crops (mm/day) 9:45 am 11:15 am 12:45 pm 2:15 pm 3:45 pm Alfalfa Cotton Dryland grain sorghum Bare soil Sensor Overpassing time at the SSJ Delta VIIRS RMSE R MAE RMSE R MAE RMSE R MAE RMSE R MAE Source: P. D. Colaizzi, P.D., S. R. Evett, S.R., T. A. Howell, T.A., J. A. Tolk, J.A., Comparison of Five Models to Scale Daily Evapotranspiration from One-Timeof-Day Measurements. Trans. ASABE 49, doi: /

14 Simulated ECOSTRESS data 2) How does uncertainty in LST propagate in ET models? 3) How does increased temporal frequency of LST data reduce uncertainty in ET estimate? VIIRS VNP09 NDVI (375m) ASTER GED NDVI (100m) NDVI (70m) Emissivity (100m) VIIRS VIAE L1 TIR Radiance (375m) Atmospheric & Emissivity Correction LST (375m) Sharpening LST (70m) MERRA2 Compute Atmospheric Correction Parameters with RTTOV LST (K) Image Date: 07/13/2016

15 PTUCD w/ Simulated data: Rn PTUCD + Landsat PTUCD + Landsat PTUCD + ECOSTRESS RMSE (W/m 2 ) R Absolute Bias (W/m 2 ) PTUCD + ECOSTRESS

16 PTUCD w/ Simulated data: ET PTUCD + Landsat PTUCD + Landsat PTUCD + ECOSTRESS RMSE (mm/day) R Absolute Bias (mm/day) PTUCD + ECOSTRESS

17 Upcoming Works 1) Quantify the uncertainty in Landsat and simulated ECOSTRESS LST data. 2) Improve the simulation process to get a more realistic LST. 3) Investigate how uncertainty in simulated LST propagate in METRIC and PTUCD. 4) Increase temporal frequency of simulated data and run the models to quantify the improvement of daily ET estimation as result of increasing temporal frequency in LST data. 5) Replicate the study for DisALEXI and PTJPL.

18 Acknowledgement Funding and Research Support from: North Delta Water Agency, Central Delta Water Agency, South Delta Water Agency. Dataset Contributors Dr. K.T. Paw U Dr. D. Baldocchi

19 Thank you!