Magda S. Galloza 1 , Melba M. Crawford 2 School of Civil Engineering, Purdue University and

1. Exploiting Multisensor Spectral Data to Improve Crop Residue Cover Estimates for Management of Agricultural Water Quality Magda S. Galloza1, Melba M. Crawford2 School of Civil Engineering, Purdue University and Laboratory for Applications of Remote Sensing Email: {mgalloza1, mcrawford2}@purdue.edu July 28, 2011 IEEE International Geoscience and Remote Sensing Symposium

2. Outline • Introduction • Estimation of crop residue • Research Motivation • Evaluation of Hyperspectral/ Multispectral Sensor data for estimating residue cover • Investigation of approaches for large scale applications • Methodology • Experimental Results • Summary and Future Directions

3. Introduction • Residue Cover (RC): Plant material remaining in field after grain harvest and possible tillage - Nutrients - Organic material (soil) - Agricultural ecosystem stability - water evaporation - water infiltration - moderate soil temperature - Critical in sustaining soil quality - erosion - runoff rates • Ecosystem-based management approaches (monitoring and damage assessments)

4. Introduction • Manual methods of analysis • Statistical sampling of fields via windshield surveys • Costly, requires trained personnel • Line transect method • Time and labor intensive • Remote sensing based approaches • Capability for 100% sampling • Detect within field variability • GREATER coverage area • Potentially reduce subjective errors Landsat-7 ETM+ 185 Km EO-1 ALI 37 Km EO-1 Hyperion 7.5 Km Satellite Track

5. Research Motivation Transect Method vs. Remote Sensing based Method 66 ft 100 beads

6. Research Motivation • Land Cover Characteristics • Agricultural Cover Discrimination and Assessment

7. Research Motivation • Evaluate performance of multispectral and hyperspectraldata for estimating residue cover over local and extended areas • Evaluate performance of next generation sensors • Landsat 8 Operational Land Imager (OLI) • Investigate sensor fusion scenarios • Potential contribution of hyperspectraldata for improving (calibrating) residue cover estimates derived from wide coverage multispectral data • Contributions of multisensor fusion

8. Approach - NDTI • Band based indices • Based on the absorption characteristics (reflectance) of RC • Linear relationship between RC and indices exploited via regression models • Multispectral NDTI (Normalized Difference Tillage Index) • Empirical models developed and validated locally • Applicable to multiple sensors: ASTER, Landsat, ALI (EO-1) • Sensitive to soil characteristics • NDTI = (TM5 - TM7)/(TM5 + TM7) Where: - TM7: Landsat TM band 7 or equivalent - TM5: Landsat TM band 5 or equivalent

9. Proposed Approaches - CAI • Hyperspectral CAI - (Cellulose Absorption Index) • Related to the depth of the absorption feature (2100 nm) • Demonstrated to accurately detect estimate RC [Daughtry, 2008] • Robust to crop and soil types characteristics • Limited coverage and availability Estimate of the depth of the cellulose absorption feature 2000 2100 • CAI = 0.5 * (R2.0 + R2.2) – R2.1 2200 Where: - R2.0, R2.1, R2.2: average response of 3 bands centered at 2000 nm, 2100 nm and 2200 nm respectively

10. Study Location / Field Data

11. Remote Sensing Data (2008-2010) Landsat-7 ETM+ 185 Km EO-1 ALI 37 Km EO-1 Hyperion 7.5 Km Satellite Track

12. Linear Models 1- 1- 2- 3- Substitute in Model 1 3- 2-

13. Model 1 - CAI Index Watershed Scale Evaluation 0% - 25% 26% - 50% 51% - 75% 76% - 100% EO-1 Hyperion (30m) Resample SpecTIR (30m) SpecTIR (4m)

14. Model 2 – NDTI Index Watershed Scale Evaluation 0% - 25% 26% - 50% 51% - 75% 76% - 100% Model 1 – SpecTIR (4m) Model 2 - ALI Model 2 – Landsat TM

15. CAI (SpecTIR) vs. NDTI (Landsat/ALI) -85% - -80% -70% - -60% -59% - -40% -39% - -20% -19% - 0% 1% - 20% 21% - 40% 41% - 60% SpecTIR vs. ALI SpecTIR vs. Landsat TM

16. Little Pine Creek Model Applied to Darlington Region Model 2 0% - 25% 26% - 50% 51% - 75% 76% - 100% Little Pine Creek Data Watershed Scale Evaluation Darlington Data (ALI)

17. Little Pine Creek Model Applied to Darlington Region Model 1 0% - 25% 26% - 50% 51% - 75% 76% - 100% Little Pine Creek Data (Model 1) Watershed Scale Evaluation Darlington Data (SpecTIR)

18. Model 3 – Substitution in Model 1 Substitute in Model 1 0% - 25% 26% - 50% 51% - 75% 76% - 100% Model 3 - (Substitution Model) Watershed Scale Evaluation Model 2 – SpecTIR (30m)

19. Conclusions and Future Work • Multispectral – not sensitive enough to the low coverage • ALI multispectral sensor provides better residue cover estimates in comparison with Landsat TM • Pushbroom vs. whiskbroom • Radiometrically ALI – 12-bit (vs. 8 bit) • ALI SNR between four and ten times larger than SNR for TM • Potential improvement from next Landsat generation - Operational Land Imager (OLI) on the LandsatFollow- on Mission - will be similar to the ALI sensor Future Directions • Weighted least squares method for multisensor fusion • Effect of soil moisture • Assimilate RC information into a hydrologic model - The OLI design features a multispectral imager with pushbroom architecture of ALI heritage

20. Thank You This research is supported by the U.S. Department of Agriculture, the Agricultural Research Service, the Department of Agronomy and its Laboratory and the Laboratory for Applications of Remote Sensing (LARS) at Purdue University.

21. SpecTIR 30m vs. Hyperion 30m -60% - -40% -39% - -20% -19% - 0% 1% - 20% 21% - 40% 41% - 60% 61% - 70%