A New Optical Trapezoid Model for Remote Sensing of Soil Moisture - PowerPoint PPT Presentation

A New Optical Trapezoid Model for Remote Sensing of Soil Moisture Morteza Sadeghi Ebrahim Babaeian Scott B. Jones Markus Tuller Dept. Plants, Soils, and Climate, Dept. Soil, Water & Environmental Science, Utah State University The University of Arizona 1

 EM radiation in various wavelengths is correlated to soil moisture. Microwave [0.5-100 c m]  High penetration depth. ⨯ Low spatial resolution. Downscale Optical [0.4-2.5 μ m] Thermal [3.5-14 μ m] ⨯ Low penetration depth.  High spatial resolution. 2

Thermal-Optical Trapezoid Model (TOTRAM)  Linear LST - θ relationship: θ − θ − LST LST = = d d W θ − θ − LST LST w d d w  Linear dry and wet edges: = + LST i s NDVI d d d = + LST i s NDVI w w w  TOTRAM: + − i s NDVI LST = d d W ( ) − + − i i s s NDVI d w d w 3

Two main limitations of TOTRAM: 1) TOTRAM cannot be used for satellites with no thermal band (e.g. Sentinel-2). 2) Beside soil moisture, LST depends on ambient environmental factors (e.g. air temperature, wind speed). TOTRAM needs to be parameterized for each individual image. 4

Core idea? Reflectance-soil moisture relationship is not significantly affected by environmental factors. So, a universal parameterization is feasible. Can we resolve both limitations by proposing an “Optical” Trapezoid model? 5

Optical Trapezoid Model Sadeghi et al. 2015. A linear physically- based model for remote sensing of soil (OPTRAM) moisture using short wave infrared bands. Remote Sensing of Environment . 164:66-76. OPTRAM is based on a linear 6 physically-based model: Aridisol Andisol 5 θ − θ − Mollisol STR STR = = d d Entisol W θ − θ − STR STR 4 w d w d STR 3 ( ) − 2 1 R = 2 SWIR where: STR 2 R SWIR 1 R SWIR : Reflectance at SWIR 0 0.0 0.2 0.4 0.6 0.8 STR : Transformed reflectance at SWIR Soil water content ( ฀ ) 6

Optical Trapezoid Model (OPTRAM)  Linear STR - θ relationship at a given NDVI : θ − θ − STR STR = = d d W θ − θ − STR STR w d w d  Linear dry and wet edges: = + STR i s NDVI d d d = + STR i s NDVI w w w  OPTRAM: + − i s NDVI STR = d d W ( ) − + − i i s s NDVI d w d w 7

Traditional Model New Model + − + − i s NDVI LST i s NDVI STR = = d d W d d W ( ) ( ) − + − − + − i i s s NDVI i i s s NDVI d w d w d w d w 8

Study Area Arizona Oklahoma 1 SCAN site; 15 rain-gauge stations 17 USDA-ARS micro-net stations 9

Satellite Imagery Landsat-8 Sentinel-2 NASA (11 February 2013) ESA (23 June 2015) 9 Optical and 2 thermal bands 13 optical bands Spatial resolution: 30-100 m Spatial resolution: 10 to 60 m Temporal resolution: 16 days Temporal resolution: ~10 days 17 images in WG 12 images in WG 4 images in LW 5 images in LW 2015-2016 2015-2016 10

Model Parameterization  Feasibility of universal parameterization TOTRAM: was tested incorporating all images. + − i s NDVI LST = d d W ( )  Two scenarios were considered: − + − i i s s NDVI d w d w 1) Local calibration: OPTRAM: Edges were determined visually. + − i s NDVI STR W was calibrated with θ data. = d d W ( ) − + − i i s s NDVI d w d w 2) No local calibration: Edges were determined by fitting. Normalized soil moisture: W was converted to θ using measured min θ − θ = d and max θ . W θ − θ w d 11

Traditional Trapezoid  A nearly trapezoidal shape is formed: LST is sensitive to θ in a broad range of fractional vegetation covers.  Integrated trapezoid consists of several separate smaller trapezoids: LST depends on ambient environmental factors besides soil moisture. 12

New Trapezoid  A nearly trapezoidal shape is formed: STR is sensitive to θ even in densely vegetated soils.  Trapezoids are visually similar: Universal calibration is feasible. NDVI NDVI 13

0.5 Estimated Soil Moisture (cm 3 cm -3 ) OPTRAM, Sentinel-2, WG OPTRAM, Sentinel-2, LW Overall Accuracy 0.3 MAE = 0.033 MAE = 0.024 0.4 RMSE = 0.042 RMSE = 0.031 R 2 = 0.500 R 2 = 0.886 (with local calibration) 0.3 0.2 0.2 0.1 0.1 0.0 0.0  TOTRAM and OPTRAM showed similar 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.4 0.5 0.5 Estimated Soil Moisture (cm 3 cm -3 ) OPTRAM, Landsat-8, WG OPTRAM, Landsat-8, LW accuracy. 0.3 MAE = 0.026 MAE = 0.027 0.4 RMSE = 0.033 RMSE = 0.037 R 2 = 0.608 R 2 = 0.785 0.3 0.2  Both models, when calibrated, yield 0.2 reasonable estimates (error < 4%) 0.1 0.1 0.0 0.0 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.4 0.5 0.5 Estimated Soil Moisture (cm 3 cm -3 ) TOTRAM, Landsat-8, WG TOTRAM, Landsat-8, LW 0.3 MAE = 0.033 MAE = 0.018 0.4 RMSE = 0.045 RMSE = 0.026 R 2 = 0.543 R 2 = 0.897 0.3 0.2 0.2 0.1 0.1 0.0 0.0 0.0 0.1 0.2 0.3 14 0.0 0.1 0.2 0.3 0.4 0.5 Measured Soil Moisture (cm 3 cm -3 ) Measured Soil Moisture (cm 3 cm -3 )

0.5 Estimated Soil Moisture (cm 3 cm -3 ) OPTRAM, Sentinel-2, WG OPTRAM, Sentinel-2, LW Overall Accuracy 0.3 MAE = 0.036 MAE = 0.048 0.4 RMSE = 0.045 RMSE = 0.059 R 2 = 0.316 R 2 = 0.596 (No local calibration) 0.3 0.2 0.2 0.1  Without local calibration, both 0.1 models still yield reasonable 0.0 0.0 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.4 0.5 0.5 Estimated Soil Moisture (cm 3 cm -3 ) estimates (error ~ 4-5%) OPTRAM, Landsat-8, WG OPTRAM, Landsat-8, LW 0.3 MAE = 0.032 MAE = 0.040 0.4 RMSE = 0.042 RMSE = 0.051 R 2 = 0.296 R 2 = 0.569 0.3  Scattering is due to approximations: 0.2 Linear LST - θ relationship at a given NDVI . 0.2 1) 0.1 Linear STR - θ relationship at a given NDVI . 2) 0.1 Linear LST - NDVI relationship at a given θ . 3) 0.0 0.0 Linear STR - NDVI relationship at a given θ . 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.4 0.5 4) 0.5 Estimated Soil Moisture (cm 3 cm -3 ) TOTRAM, Landsat-8, WG TOTRAM, Landsat-8, LW 0.3 MAE = 0.030 MAE = 0.036 0.4 RMSE = 0.041 RMSE = 0.045 R 2 = 0.489 R 2 = 0.677 0.3 0.2 0.2 0.1 0.1 0.0 0.0 0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3 0.4 0.5 15 Measured Soil Moisture (cm 3 cm -3 ) Measured Soil Moisture (cm 3 cm -3 )

Soil Moisture Maps  TOTRAM yielded W in a narrow range.  OPTRAM maps better match the DEM. They show river network. 16

1.0 RMSE : RMSE : RMSE : RMSE : OPTRAM = 0.12 Date-by-Date Comparison OPTRAM = 0.13 OPTRAM = 0.16 OPTRAM = 0.19 TOTRAM = 0.12 TOTRAM = 0.15 TOTRAM = 0.14 TOTRAM = 0.15 Estimated W 0.5 WG (1-Nov-15) WG (17-Nov-15) WG (3-Dec-15) WG (20-Jan-16)  TOTRAM failed in predicting spatial 0.0 1.0 RMSE : RMSE : RMSE : RMSE : OPTRAM = 0.21 OPTRAM = 0.12 OPTRAM = 0.10 variability of soil moisture: OPTRAM = 0.10 TOTRAM = 0.14 TOTRAM = 0.11 TOTRAM = 0.10 TOTRAM = 0.07 Estimated W Universal parameterization is not 0.5 feasible. WG (5-Feb-16) WG (21-Feb-16) WG (24-Mar-16) WG (9-Apr-16)  OPTRAM successfully captured 0.0 1.0 RMSE : RMSE : RMSE : RMSE : spatial variability of soil moisture: OPTRAM = 0.12 OPTRAM = 0.07 OPTRAM = 0.08 OPTRAM = 0.19 TOTRAM = 0.08 TOTRAM = 0.04 TOTRAM = 0.04 TOTRAM = 0.16 Estimated W Universal parameterization is feasible. 0.5 WG (25-Apr-16) WG (11-May-16) WG (27-May-16) LW (2-Dec-15) 0.0 1.0 RMSE : RMSE : RMSE : RMSE : OPTRAM = 0.19 OPTRAM = 0.18 OPTRAM = 0.16 OPTRAM = 0.20 TOTRAM = 0.16 TOTRAM = 0.17 TOTRAM = 0.15 TOTRAM = 0.17 Estimated W 0.5 LW (11-May-16) LW (18-Dec-15) LW (4-Feb-16) LW (23-Mar-16) 0.0 0.0 0.5 1.0 0.0 0.5 1.0 0.0 0.5 1.0 0.0 0.5 1.0 17 Measured W Measured W Measured W Measured W

Conclusions:  OPTRAM resolves two limitations of TOTRAM.  OPTRAM and TOTRAM overall accuracy is comparable. Future Work:  More extensive evaluations.  Improving model accuracy and parameterization. 18

Reference: Sadeghi, M., E. Babaeian, M. Tuller, S. B. Jones. 2017. The Optical Trapezoid Model: A Novel Approach to Remote Sensing of Soil Moisture Applied to Sentinel-2 and Landsat-8 Observations. Remote Sensing of Environment , Accepted. Acknowledgement: Funding from National Science Foundation awarded to USU and UofA. 19