Leaf Area Estimation for Sugar Beet yield prediction Using ERS SAR Data  

ABSTRACT

In Sugar Beet, previous studies have shown that forecasts of potential sugar yield can be provided through the growing season, based on satellite observations of crop canopy cover at various stages of development. These observations are used in conjunction with corresponding regional weather information, crop sowing dates and soil types to provide successively refined yield forecasts as harvest time approaches. 5 ERS-1 SAR and 3 ERS-2 SAR images were obtained of East Anglia. Field data were collected in two contrasting fields from June 11 to July 29, 1996, comprising radiometric and biophysical measurements of the crop canopy. A independently fitted version of the 'Cloud model' was inverted to calculate LAI from values of ERS-1 SAR backscatter and soil moisture samples. LAI estimates were in good agreement with measured values and were used to estimate canopy cover using a standard exponential relationship that has a well established coefficient for sugar beet. The study shows that Radar data can provide useful estimates of canopy cover for crop production modelling, especially in the case of loss of optical data due to cloud.

Keywords: ERS-1/2, SAR, Radar cloud model, Leaf area index (LAI), Crop cover (f).

1. Introduction

In Sugar Beet, previous studies have shown that forecasts of potential sugar yield can be provided through the growing season, based on satellite observations of crop canopy cover at various stages of development.Incorporation of radar data into the crop growth model ensures the supply of data to estimate crop cover (f ) when weather conditions prevent the acquisition of satellite data at optical wavelengths. Compared with optical remote sensing, radar has a more ambiguous relationship with crop canopy variables, but as shown by Xu and Steven, 1996, the relationship with leaf area in sugar beet is significant and strong enough to provide reasonable estimates of crop cover. However, the radar model used by (Xu and Steven, 1996) was calibrated and fitted to agronomic field data from a single date in 1994. This paper describes a independent study to evaluate the reliability of radar estimates of crop cover in Sugar Beet.

2. Approach

To test the robustness of the Xu and Steven model over time, ERS-1/2 SAR data were related to ground measurements of sugar beet for a sequence of dates through the growing season LAI was estimated from the model and used to estimate crop cover (f) for incorporation into the sugar beet growth model. The procedure used was as follows:

1. Collection of Ground truth data in selected two sugar beet fields on 10 dates with simultaneous acquisition of ERS-1/2 SAR image data.

2. Derivation of radar backscatter coefficient through ERS-1 SAR calibration. 3. Calculation of leaf area index from radar backscatter and soil moisture using the Cloud model

4. Crop canopy cover (f) estimation from LAI using SAR data

5. Comparison with Ground truth data

3. Data used

3.1 Ground truth data

The study area was chosen near Broom’s barn sugar beet Research Institute, East Anglia, UK (lat/long=52.2/0.46 degrees).The field data were collected weekly from June 11 to August 16, 1996 and SAR data were acquired for 10 ERS-1/2 overpass dates. A Parkinson radiometer with 2 bands (NIR and Red respectively at 650 and 850nm) and 2 heads (radiance for 30⁰ FOV and irradiance) was used for radiometric measurements of the crop canopy. In each field, 5 sample areas about 12 m apart used to collect biophysical measurements; namely row direction, plant population, canopy biomass, canopy and leaf water content, leaf area index (LAI), leaf inclination angle, soil moisture and soil roughness. The average value were used to represent each field. The detailed fieldwork and analysis procedures were as described in Xu and Steven (1996). The Ascending and Descending ERS-1/2 SAR overpasses for East Anglia were at approximately 2200 GMT and 1053 GMT, respectively. A larger set of sugar beet fields was also identified for the extension of image analysis with ERS, SPOT and ground data for a single date.

3.2 ERS-1/2 SAR data

The SAR precision images are multi-look, ground range, digital images generated from raw SAR image mode data using up-to-date auxiliary parameters and corrected for antenna elevation gain and range spreading loss. A full scene of ERS-1/2 SAR normally has a geographical coverage of 100 km in ground range and at least 102.5 km in azimuth (along track). Seven ERS-1 and 3 ERS-2 SAR scenes were used for this study.

4.ERS-1/2 SAR calibration

As advised by DRA (Bird, 1996), a single calibration constant was used for calibration of all ERS-1.SAR.PRI images processed by DRA based on the 131 corner reflector measurements over a period of 8 months. The calibration constant was given as 59.96 dB for ERS-1 and 55.61 dB for ERS-2 the mid-swath position. Therefore, to calculate the radar backscatter coefficient s ⁰ (dB) for ERS-2, the following equation was used :

where <I> is the mean pixel intensity as given in Equation (2) , and a is the local incidence angle.

The two sugar beet fields where the ground data collection campaign took place were identified on the ERS-1/2 SAR images. The area for each field has been defined on the ERDAS IMAGINE image processing system together with a statistical report on the number of pixels in the selected area, the mean value of pixel amplitude and the standard deviation etc. The average radar intensity for each field could then be generated as the sum of the square of mean and the square of standard deviation, i.e.

where DN_mean is the average amplitude for the area of interest and SD is the standard deviation of the pixel values.

The local incidence angle for each field was then computed according to the centre pixel column of field and the backscatter coefficient was derived subsequently.

5. Radar Model Fitting and Inversion

The relationship between backscatter and leaf area was as defined by Xu and Steven (1996) based on Leeuwen and Clevers’ (1994) version of the Cloud model (Attema and Ulaby, 1978).

s ⁰(m²/m²)=A·cosq (1-exp(-2B·L/cosq ))+C´·(1+0.0603·m_s)·exp(-2·B·L/cosq ) (3)

Where, A=0.3259, B=0.167, C´=0.0452 and D´ = 0.0603

A, B and C’ were coefficients fitted in previous experiment on nine fields in 1994 (Xu and Steven, 1994) and D’ was predetermined from published values on soil backscatter. Sigma 0 (s ⁰) is defined in power units in this equation.

This model may be inverted for estimation of LAI using measured values of s ⁰ and values of soil moisture content:

The canopy cover fraction or the fraction of intercepted solar radiation was estimated using following formula:

f = 1 - exp (-K.L) (5)

Where f is the fraction of incident solar photosynthetic irradiance intercepted or absorbed by leaves and L is the leaf area index. K is normally treated as a constant for a given crop type, with a typical value of K=0.71 for sugar beet.

6. Results

6.1 Crop cover estimation by LAI using ERS SAR for two beet fields

Table 1 presents the measured and predicted radar backscatter s ⁰, LAI and crop cover for the two sugar beet fields on 10 ERS-1/2 SAR overpass dates. Correlation coefficients of 0.92 and 0.96 were found between measured and predicted Leaf area index (LAI) for field A and B, respectively (fig 1). Figure 2 shows the comparison of crop cover (%) using ERS-1/2 SAR data for the two sugar beet fields on 10 dates from June to August. Correlation coefficients of 0.95 and 0.92 were achieved for field A and B, respectively.

Table 1: Comparison between measured and predicted values of two sugar beet field for 10 ERS-1/2 SAR dates

Fig.1:Comparison between Measured LAI and predicted by SAR

Fig.2:Comparison between Crop cover estimates by measured LAI and SAR

6.2 Extension of Radar-optical-ground data comparison

As part of a field surveys done for water stress experiment on 12 and 13 august 96, a total of 9 sugar beet fields near Brooms Barn were identified on the 13 June and 16 August, 1996 ERS-1 images. The same fields were identified on a SPOT image of 15 June, 1996. Backscatter values were calculated for each of the fields and used to estimate crop canopy cover by equation 5. The average soil moisture from field data collection in fields A and B was used in the calculations for 15 June to compare with SPOT, and the 12-13 August field data were used as the reference for the August comparison with ERS-1. Crop cover (f) was also estimated with the SPOT data using the OSAVI relationships applied to the same 9 sugar beet fields as were identified on the ERS-1 SAR images (Rondeaux et. al, 1996). In June the comparison between SPOT and ERS shows a correlation coefficients of 0.88 (Figure 3). Figure 4 shows a correlation coefficients of 0.71 between crop cover estimates from ERS SAR and from Measured LAI.

Fig.3: Comparison between crop cover(%) by SPOT and ERS SAR for 9 sugar beet fields

Figure 4: Crop cover (%) for 9 sugar beet fields using ERS and measured LAI

7. Conclusions

Comparison of the predicted values and measured data together with their error analysis demonstrated that ERS-1/2 SAR data can be used in the Cloud model for estimation of leaf area index and canopy cover with an acceptable accuracy and thus has the potential for operational application. The ERS SAR data have been shown to be sensitive to leaf area throughout the growing season of Sugar Beet and are a viable alternative to the use of optical remote sensing in an operational sugar beet yield forecasting system. Further validation of the model with RADARSAT C-band HH polarization, which has potentially higher temporal resolution will be made in 1997.

8. References

Attema, E.P.W. and Ulaby, F.T., 1978, Vegetation modelled as a water cloud, Radio Science, 13(2), pp357-364.

Bird, P.J., 1996, Personal communication. Space and Communications Dept., Defence Research Agency, Farnborough, UK.

Leeuwen, H.J.C. van and Clevers J.G.P.W., 1994, Synergy between optical and microwave remote sensing for crop growth monitoring, Proc. Sixth Int. Symp. Physical Measurements and Signatures in Remote Sensing, 17-21 Jan. 1994, Val d’Isere, France, pp 1175-1182.

Rondeaux, G., Steven, M.D. and Baret, F., 1996. Optimization of Soil-Adjusted vegetation indices. Remote Sensing Environ., 55, 95-107.

Vyas, S.P. and Steven, M.D., 1995, ERS-1 SAR for leaf area index (LAI) prediction. Proc. Remote Sensing Society One Day Student Meeting, 29 March 1995, Dept. of Geography, Uni. of Leicester, UK. pp 146-152.

Vyas, S.P. and Steven, M.D., 1996. Sugar beet yield estimation by using ERS-1 SAR data, Proceedings of the Remote Sensing Society One Day Student Meeting, Salford University, April 4, 96, pp 33-38.

Vyas, S.P., Steven, M.D.and Jaggard, K.W., 1996. Comparison of SPOT and SAR estimates of canopy cover in sugar beet, Proceedings of the 22nd annual conference of the Remote sensing society, 11-14 september, 1996, Durham, UK, pp 614-622.

Vyas, S.P., Steven, M.D., Xu, H., Milnes, M. and Jaggard, K.W., 1995, ERS-1 SAR for sugar beet yield prediction. Proc. II ERS Applications Workshop, London, UK, 6-8 Dec. 1995, 403-405.

Xu, H. and Steven, M.D., 1996, Monitoring leaf area of Sugar Beet Using ERS-1 SAR Data. International Journal of Remote Sensing , 17,17,3401-3410.

Keywords: ESA European Space Agency - Agence spatiale europeenne, observation de la terre, earth observation, satellite remote sensing, teledetection, geophysique, altimetrie, radar, chimique atmospherique, geophysics, altimetry, radar, atmospheric chemistry