LAI Retrieval and Yield Modeling for Winter Wheat with TerraSAR-X Data Compared to Using Optical Data
Dotzler, Sandra1; Bach, Heike1; Spannraft, Katharina1; Migdall, Silke1; Hank, Tobias2; Frank, Toni2; Mauser, Wolfram2
1VISTA Remote Sensing in Geosciences GmbH, GERMANY; 2Department of Geography, Ludwig-Maximilians-University Munich, GERMANY

Introduction:

Crop growth and yield modeling is a powerful tool for farmers to validate their own crop management, organize the harvest and can be a helpful basis to negotiate with wholesale traders. Yield modeling can also improve the efficiency in operating biogas plants, through providing yield figures in advance.

From optical remote sensing data (e.g. RapidEye data) plant parameters can be retrieved spatially and temporally distributed over the whole vegetation period. Plant parameters from remote sensing data improve yield modeling, as they supply up to date information on crop growth, as has been demonstrated successfully with the land-surface model PROMET (Mauser & Bach 2009; Migdall et al. 2009; Hank et al. 2012). However the availability of optical products for biomass development depends on weather, since it requires cloud free conditions. This problem offers an opportunity for SAR data, since the SAR signal penetrates clouds and as active systems can observe even by night.

In this study the possibility of using TerraSAR-X for mapping crop growth and its accuracy of LAI retrieval was compared to optical approaches. It is evaluated whether X-band can substitute or complement optical data in the data assimilation of crop growth modeling and yield prediction. The question of adequate spatial aggregation of the SAR signal for agricultural monitoring is addressed and whether the spatial details seen with TerraSAR still allow site-specific applications.

Results:

For a test site in Saxony-Anhalt, Germany, the sensitivity of TerraSAR-X data to green LAI has been analyzed using a data set, which includes five pairs of TerraSAR-X and RapidEye images with a maximum time difference of three days. The data pairs cover the vegetation period 2011 from March to July. It is shown, that an increase of biomass leads to a higher attenuation of SAR signal. The backscatter in VV polarization showed high correlation to green LAI for wheat, retrieved from RapidEye data using radiative transfer simulations with SLC (Verhoef & Bach 2007). VV polarization outperformed cross polarization VH and the ratio of VH/VV. A correlation could also be found for barley. However this was not the case for rapeseed, indicating that a sensitivity of the X-band signal to green LAI may be limited to small leaved crops.

The spatial resolution of the SAR data has been modified iteratively to find a resolution, where both speckle is reduced sufficiently, and structures in the fields are still visible. For this, spatial correlations of the LAI patterns seen in RapidEye and TerraSAR-X were used. As a result, a ground resolution of 25 m was identified to be the best resolution for retrieving LAI from TSX backscatter signal. This resolution is adequate to identify heterogeneities within the fields and allows site-specific applications. Thus field-wise averages are not required for wheat and barley.

Sigma-naught values in VV polarization have been transformed to dB, because then a linear relationship between green LAI and sigma-naught could be established for wheat and barley. Integrating all available images of the vegetation period 2011 in the regression model, a coefficient of determination (R2) of 0.58 was achieved with an RMS error of 2.65 m2/m2. The regression model between green LAI and sigma-naught in Saxony-Anhalt has been used to calculate LAI values from SAR signal. The LAI from 5 TSX images in the harvest year 2011 has been assimilated in the dynamic plant growth model PROMET, to adjust the modeled biomass throughout the vegetation period with up to date information from earth observation. This assimilation procedure has been established and validated using optical data. PROMET simulates plant development and finally the yield on the harvest date in 2011. The modeled yield was compared to measured values from a combine harvester (mean yield: 6.88 t/ha, minimum: 1.57 t/ha, maximum: 11.93 t/ha). If optical RapidEye data from 5 images is used, for all 25m pixels (n = 8288) an RMSE = 1.51 t/ha for the modeled yield is achieved. If solely SAR data is used as assimilation input the RMSE amounts 1.96 t/ha, which is an increase of 0.45 t/ha. A combination of both sensors (3 RapidEye images, 2 TerraSAR-X images) leads to an RMSE of 1.57 t/ha which is only 0.06 t/ha more than without any SAR image.

The analysis showed that the modeled wheat yield using SAR data only cannot reach the quality of using optical remote sensing data in the assimilation. However the achieved yield is in the right range and the difference in the RMSE is not very large. Validation maps show, that TerraSAR-X data allow to correctly reproduce the spatial variation of yield in the fields for fields with very heterogeneous yield distribution. Though modeled yield from optical sensors correlates better with measured yield, data from TerraSAR-X can complement optical remote sensing data due to independency from daylight and stability in cloudy conditions and therefore improve the data basis for yield modeling.
The research shown here was conducted in the frame of the Project 'RapidSAR - Integrative use of RapidEye and TerraSAR-X through data assimilation into models of agricultural production (grant code numbers 50 EE 0920/22), supported by the Space Agency of DLR through funding by the German Federal Ministry of Economics and Technology (BMWi).

Literature:

Hank, T.; Bach, H.; Spannraft, K.; Friese, M.; Mauser, W. (2012): Improving the process-based simulation of growth heterogeneities in agricultural stands through assimilation of earth observation data. In: Proc. IEEE Internat. Geosci. And Rem. Sens. Symp. IGARSS ’12, p. 1006 - 1009

Mauser, W. & H. Bach (2009): PROMET - Large scale distributed hydrological modelling to study the impact of climate change on the water flows of mountain watersheds. In: Journal of Hydrology (367), p. 362 - 377

Migdall, S.; Bach, H. ; Bobert, J.; Wehrhan, M.; Mauser, W. (2009): Inversion of a canopy reflectance model using hyperspectral imagery for monitoring wheat growth and estimating yield. In: Precision Agriculture, p. 508 - 524

Verhoef, W. & H. Bach (2007): Coupled soil-leaf-canopy and atmosphere radiative transfer modeling to simulate hyperspectral multi-angular surface reflectance and TOA radiance data. In: Remote Sensing and Environment