HIGH RESOLUTION X-BAND SAR IMAGERY FOR PRECISE AGRICULTURE AND CROP MONITORING

Transcription

1 HIGH RESOLUTION X-BAND SAR IMAGERY FOR PRECISE AGRICULTURE AND CROP MONITORING Dan G. Blumberg Department of Geography and Environmental Development, Ben-Gurion University of the Negev, Beer- Sheva Israel 84105, & Lamont-Doherty Earth Observatory at Columbia University, Palisades, NY ABSTRACT In the next couple of years at least two SAR high resolutions payloads will be launched, and during the next decade many more. Optical remote sensing is being used routinely in agricultural management and studies. Radar data which is often used in geophysical studies is not utilized heavily in agriculture and specifically X- band data are not commonly used. In the following paper we had unique access to high resolution polarimetric X-band radar data from Elta Systems Inc test bed SAR flown onboard a Boeing 737 aircraft. Results show potential in using X-band SAR for agriculture and bio-geophysical studies. Hence the ability to map small agricultural fields, assess their relative soil water content and distinguish between young and mature tree plantations is demonstrated. INTRODUCTION Geophysical exploration encompasses numerous methods for better understanding the terrestrial and marine environments. These methods are often applicable to agricultural management. One of the most promising geophysical exploration tools for agriculture includes the use of remote-sensing. The use of remote sensing from air and space borne sensors has been around for several decades now. In essence, for many years aerial photogrammetry has been applied to several components of the agricultural production scheme and its management. However, the ability to use multispectral sensors capturing emitted and reflected light beyond that visible to the naked eye has been applied in agriculture in a serious manner only in the past two or so decades. The majority of the applications have used remote sensing to monitor photosynthetic processes as manifested through the absorption of red light and the reflection of near infrared light. The most common of these methods is that of the normalized difference vegetation index better known as NDVI or the normalized difference vegetation index. Modifications of the NDVI have been around too, that account for the soil background such as the SAVI or the WDVI index [6]. Several indices have also been proposed for mapping the Leaf Area Index (LAI) from multispectral data or the vegetation fraction (VF). These have been used mostly in research activities and not in agricultural practice. The thermal spectrum has been used for agriculture, primarily for detecting the thermal differences as a function of vegetative moisture. The use of microwave energy in remote sensing is available since the late 1970s but has become more widely used only for the past decade and half since the SIR-C project in 1994 and the launch of the European ERS in The current level of experience in operational use of SAR (Synthetic Aperture Radar) data is limited compared to the use of visible and infrared data. Several characteristics of SAR data taken together, may promote more extensive evaluation and use of SAR data for agricultural management. These include the unique information on surface roughness, physical structure, and electrical conduction properties; the high spatial resolution; and the ability to image 24-hours a day and during all-weather conditions. There are very few successful reports on the use of X-band radar (wavelengths on the order of 3 cm) for vegetation and crop monitoring. This is primarily because of sensor saturation at the relatively short wavelength (for microwavelengths) but yet, at least two such sensors will be launched in the next year. This paper reviews some of the potential uses of X-band radar data in agriculture and gives some preliminary testbed examples. Precise Agriculture and Remote Sensing The prime objectives for the use of remote sensing in precise agriculture are: a. increase farming yield b. reducing farming costs c. early detection of farming problems such as diseases or decrease in productivity d. harvest prediction At the macro-scale, remote sensing has been applied in recent years informally and formally to methods of subsidy control, monitoring, and farmer compensation for weather damages such as drought or hail damages primarily in Europe. For example, the European Union uses remote sensing to estimate the productivity of the member nations and compare this with their respective reported productivity to control subsidies. This project termed in the EU as MARS [9], started in 1988 and was

2 initially designed to apply emerging space technologies to provide independent and timely information on crop areas and yields. Since 2000 the MARS project team has also been applying their expertise in crop yields outside the EU. Their services support EU aid and assistance policies are aimed to provide building blocks for a European capability for global agricultural monitoring and food security assessment. In a technical specification document [4] the remote sensing unit for monitoring agriculture states that the majority of the European Union Member States, in co-operation with the European Commission, will use Remote Sensing in 2006 to control at least a part of the subsidies for the agricultural areas. This requires identification of all agricultural parcels on the holding, their area expressed in hectares to two decimal places, their location and, where applicable, their use and whether the agricultural parcels are irrigated. At the micros-scale extensive research has been conducted showing that multispectral remote sensing has the capability to provide information on crop health and status. Reference [12] for example demonstrated how multispectral data can assess the nitrogen availability or limitation in a cotton field. In an extensive review paper [11] provided an overview of the application of multispectral sensing in agriculture. While they conclude that spectral sensors can provide the necessary spectral and spatial resolution for assessing nutrients in plants such as nitrogen, this can be achieved only when the sensor is mounted at low altitudes. Furthermore, according to [11] the optical spectrum cannot provide information about soil conditions. HIGH RESOLUTION OPTICAL SENSORS In the past seven years there have been major advances with the launch of several high resolution commercial optical sensors starting with the Ikonos, launched in 1999 with a panchromatic sensor operating at a resolution of 1m and a multispectral sensor at a resolution of 4m. Other sensors with similar resolutions include Quickbird, the Israeli EROS1a, and the Cartosat. At these resolutions the individual trees can often be observed, the spatial variability within a field assessed and depending on the vegetation fraction at times bare soil can be observed. Reference [3] assessed the geometric corrections of three very high resolution optical sensors for use in agriculture and found them all to agree with the target specifications of the European Commission [4]. However, the availability of such imagery is often prohibited by clouds and coverage. MICROWAVE RADAR IMAGERY Synthetic aperture radar is an active form of remote sensing. The surface is illuminated by a beam of energy with a fixed wavelength that can be anywhere from 1 cm (K band) to approximately 70 cm (P-band). These long wavelengths penetrate clouds and atmospheric interferences common to optical imagery and therefore are not limited spatially or temporally because of solar illumination or atmospheric interferences. At these wavelengths the factors dominating the retroflection are slope relative to the incident wave, roughness at scales comparable to the wavelength, vegetation structure, and dielectric properties. Hence, surface geometry will dominate the radar scattering and its signature over agricultural fields. Reference [10] demonstrated that radar can be used to investigate vegetative crops. The response of the radar to the crops will depend on the radar wavelength but overall the literature has limited reports on the use of X-band imagery for crop monitoring because of saturation. Depending on the specific wavelength and the point at which saturation will occur, different components of the canopy can be assessed. The shorter radar wavelengths will tend to respond to the upper part of the canopy and as wavelength increases more components will be observed. Thus, X band with a 3 cm wavelength will see the upper part of dense crops, where as P-band may see the soil or the trunks. Radar sensors, being active, also control the polarization of the transmitted and received wavelength providing yet, additional information. Therefore, information on the crop volume can be asserted from the radar backscattering and where polarimetric data are available additional information on the crop structure can be inferred from the polarimetric radar signature of the crops. The EC suggests the use of RADARSAT radar data limited to the far north areas where cloud cover prohibits the use of optical data. Two issues have hindered the use of radar in many applications; firstly speckle. Speckle is very common to radar images and shows up as a salt and pepper effect in the image. For the most part speckle is the result of interactions between the radar energy and facets within the target area at the subpixel scale. Consequently, the radar echoes interfere with each other both constructively and destructively causing speckle and fading. These are suppressed by averaging the pixels with the backscatter in adjacent pixels using various filtering techniques such as the median or mean filtering or radar specific filters such as the Lee filter. While these filters reduce speckle and increase the SNR, the radar image is resampled to a reduced resolution and, therefore, is more difficult to use for many precise agricultural applications. The second reason many users do not

3 utilize radar despite the value of the data is the complexity of the data. This is caused by the fact that the measurement units are very different than commonly used in optical imaging and lack intuitive interpretation. Furthermore, radar does not show chlorophyll content like optical systems but rather structural elements of the plant and dielectric properties of the underlying soils. There have been many papers demonstrating that SAR such as the ERS sensor responds to soil water content for example [2]. However, the resolutions and accuracies of the soil moisture are still far for being sufficient to be applied in operational monitoring activities of field scale soil water content. Moreover, in the absence of multi-paramater multi-wavelength freeflying SARs it is difficult to separate the affects of roughness from dielectric constant. HIGH RESOLUTION MICROWAVE RADAR SENSORS In the near future a new set of high resolution sensors will be launched. These include medium to high resolution synthetic aperture radar sensors. Most of these will be launched for dual useage; the Canadian Radarsat-2, for example, will have a 3 linear m/pixel resolution. This is opposed to the 8.3 m/pixel capability of Radarsat 1 and the 12.5 m/pixel capability of ERS and other sensors. The German Space Agency will be launching its own Terra SAR X. This system will be a high resolution X-band SAR free flying sensor with various modes of operation including a spot mode of 1 x 1 m resolution. Israel Aircraft Industries will be launching later this year the TECSAR, an X-band SAR system with similar resolutions. The TECSAR payload is designed to for a lifetime of 4 to 8 years. The system will provide several modes of operation yielding wide and spot imagery, mosaic modes, and an uncommitted possibility for multipolarization. The orbit inclination will be 143 and orbit altitude will be 550Km. This will yield orbits per day. The objective of this paper is to demonstrate some of the advantages in using high resolution SAR (<5 m/pixel) for agricultural research and agricultural management. EXAMPLES To demonstrate the utility of high resolution SAR two datasets were used. The first is an ERS SAR image. ERS is a European sensor operating at 5.3 Ghz with a 5.66 cm wavelength (C band) and a single polarization of VV (vertical transmit and vertical receive). The midswath incidence angle is 23 degrees with little variation to the edges. The resolution of the precision product (PRI) is 12.5 m/pixel. The second sensor was an X-band fully polarimetric sensor. This sensor was designed and built by Elta systems group which is a subsidiary of the Israel Aircraft Industries (IAI). Elta s multimode radar was flown onboard an airborne platform, a Boeing 737, and data were provided at a resolution of 1m/ pixel. This sensor is a testbed for the TECSAR satellite that is being developed by ELTA [8]. Corner reflectors of 1.5 m (side) were used in the field for calibration and geometric correction. Both images were acquired for an area in the western Negev, near Kibbuts Saad, where precise agriculture techniques are being tested. The SAR image was from early March, 2004 and the ERS image from February 26, Field Size This example is aimed at showing the ability to automatically identify and distinguish between individual fields in the SAR data. The use of SAR mapping field boundaries is important for management issues such as subsidies or insurance issues. In some areas this can be achieved from optical data but many areas are subjected to frequent cloud cover and thus, in these areas radar is very useful. The ERS and testbed data were all treated with a Lee speckle reduction filter. The Lee speckle reduction filter is designed to reduce speckle in radar imagery while simultaneously preserving texture information. The filter uses local statistics (coefficient of variation) within individual filter windows. Each pixel is put into one of three classes: homogeneous; heterogeneous; or point target. Each class type is treated differently. For the homogeneous class, the pixel value is replaced by the average of the filter window. For the heterogeneous class, the pixel value is replaced by a weighted average. For the point target class, the pixel value is not changed [7]. Following this process a 3x3 edge detection filter was applied from within Erdas Imagine software. Fig. 1 shows the original and processed Elta X-band imagery. The ERS image at 12.5 m/pixel could not distinguish between individual fields in this area what so ever but high resolution X-band image was found to be very useful and the boundaries of individual fields and were marked exceptionally well. Fields as narrow as 130 m with dirt ways on the boundaries 9 m wide were measured. The accuracy of the field sizes was assessed within 1-2 m accuracy. Plant Growth The type of crop can not be assessed using synthetic aperture radar. However, with respect to vegetation and trees SAR has the ability to penetrate the canopy and respond to the volume of dry mater within the canopy.

4 Looking at high resolution SAR is also extremely useful in studying the growth of trees. Near Saad, in southern Israel, the radar response of two different types of tree plantations was reviewed. These plantations included a mature grapefruit orchid and a relatively young Persimmon orchid (similar to grapefruit). Interestingly, the younger orchid (Fig. 2a-b) had higher backscatter values with low backscattering in between trees, whereas the older orchid had lower backscatter but was more consistent across the field. The higher backscatter from the younger tress is attributed to double bounce and direct retroreflection from the tree trunks and volume scattering within the canopy in the older orchid. index of refraction that is commonly used in visible remote sensing. The complex term is therefore a summation of the imaginary part that determines the skin depth and the real part that determines the new wavelength for a volume penetrating wave. The Fresnel reflection coefficient for a nadir looking antenna can be calculated as: R= εs εa εs + εa [1] in which R is the reflection coefficient and s is the soil medium, and a is the air medium, the real dielectric constant. This equation can be further shown for polarization where side looking radar is used. The dielectric constant of water at microwave frequencies is orders of magnitude greater than that of dry soils. This contrast between the dielectric constant allows the construction of sensitivity models to soil water content [2]. Fig. 1. The above figure shows the images from ERS Cband and X-band Elta systems Inc. The top image is the ERS full scene whereas the bottom left image displays a zoom in on the agricultural fields after speckle reduction. The bottom right image shows the X-band testbed image. Clearly, the ERS image at 12.5m/pixel cannot properly distinguish between fields in this area, whereas the high resolution X-band distinguishes well and maintains sufficient contrast within the field. Soil Water Content Of major concern in agricultural management is the distribution of water within the upper soil layers and irrigation of fields. The issue of estimating soil water content using radar backscatter has been studied often. X-band is probably not an ideal waveband for this purpose because of the shallow penetration in the soils but the high resolution has its advantages in being able to map the spatial distribution precisely. Radar is used to estimate soil moisture because microwave energy responds directly to changes of the complex dielectric constant of the soil-water mixture caused by the changing presence of water within the soil. The changes in water content modify the complex dielectric constant (ε), or relative permittivity as opposed to the complex Fig. 2. The above figures show two orchids near Saad in the Western Negev. These showed up in the radar imagery as bright for the young orchid (a) and dark for the mature orchid (b) A. is the young Persimmon orchid, and B. is the mature grapefruit orchid.

5 Because geometric reflection dominates the radar backscatter the surface roughness needs to be considered especially when using short wavelength SAR (such as X-band.) Hence, short wavelength radar will be most useful only for relatively smooth areas such as sandy regions and some desert areas. Blumberg et al., (2005) suggested using the soil-water saturation percentage, Θ as a way to estimate the availability of water in soils from radar data. This parameter (Θ) is less sensitive to soil texture than the volumetric or gravimetric water content parameters and can be calculated through a time series of measurements as demonstrated by [1]. The parameter represents: Θ= W Wr Ws Wr [2] In which W is the gravimetric soil water content and the subscript r denotes the residual water content at air dry or following a long dry spell and s denotes the gravimetric water content at the point of saturation. This can then be related to the radar backscatter coefficients as: W Wr σ 0 σ r0 =Θ= 0 0 W s Wr σ s σr Fig. 3. The above image shows in irrigated field near Ashalim south of Beer Sheva. The yellow area is saturated whereas the brown patches are dry soil. Hence the line irrigation unit moves from right to left. [3] Fig. 3 shows an image taken over a bare field in the area of Ashalim (south of Beer Sheva) that was being irrigated at the time of the image acquisition. The bright areas are close to saturation whereas the dark areas are still dry. Field measurements of saturation at this site suggested that saturation occurs when the upper 5 cm reach a 16% gravimetric soil water-content. The dry areas, appearing dark in this image has a soil water content of 3-4% gravimetrically. Fig. 4 shows a radar image of a field with distributed sprinklers. Using the high resolution SAR individual sprinklers were identified in the field, spaced 12m apart. In contrast, these were not identified in aerial photographs. Their unique response in radar imagery makes them easily identified in the X-band image. Management of agricultural fields requires knowledge concerning the distribution of water sprinklers. This is mostly important in arid and semi-arid areas. Fig. 4. shows the location of sprinklers in an agricultural field. These are spaced some 12.5 m apart along line pipes. Their precise location can be identified in the SAR image and can not be seen at all in visible data. SUMMARY In the above paper it has been demonstrated that X-band radar imagery that has been ignored for the most for bio and geophysical studies, compared to longer wavelengths is very useful when utilized in high resolution. The current experiments with Elta Systems Inc. testbed polarimetric SAR show great potential for

6 agricultural management and precise agriculture. Here the utility in estimating field boundaries and sizes and relative soil water content and differences between tree plantation types and maturity was shown. X-band free flyers are smaller than their longer wavelength counterparts and have utility in bio and geophysical studies. Further work is required to develop the tools for using this type of SAR prior to the launch of Terra SAR X and other high resolution X-band SAR free flyers. Acknowledgments The author wishes to acknowledge the generous support of the Katif Research center for coastal deserts and that of the BMBF under the GLOWA Jordan River Project. ELTA Inc. are acknowledged for the provision of some high resolution imagery. [9] Roebeling, R. A., E. Van Putten, et. al., Application of Meteosat Derived Meteorological Information for Crop Yield predictions in Europe, International Journal of Remote Sensing, 25(23): , [10] Saich, P. and M. Borgeaud, Interpreting ERS SAR Signatures of Agricultural Crops in Flevoland, , IEEE Transactions on Geoscience and Remote Sensing, 38(2): , [11] Scotford, I. M. and P. C. H. Miller, Applications of Spectral Reflectance Techniques in Northern European Cereal Production: A Review, Biosystems Engineering, 90(3): , [12] Sui, R., J. B. Wilkerson, W. E. Hart, L. R. Wilhelm, and D. D. Howard, Multi-spectral Sensor for Detection of Nitrogen Status in Cotton, Applied Engineering in Agriculture, 21(2): , REFERENCES [1] Blumberg, D. G., G. Ronen J. Ben-Asher, V. Freilikher, L. D. Vulfson, and A. L. Kotlyar, Utilizing a P-band Scatterometer to Assess Soil Water Saturation Percent of a Bare Sandy Soil, Journal of Hydrology, submitted, [2] Blumberg, D. G., V. Freilikher, et al., Subsurface Microwave Remote Sensing of Soil-Water Content: Field Studies in the Negev Desert and Optical Modeling. International Journal of Remote Sensing, 23(19): , [3] Chmiel, J., S. Kay and P. Spruyt, Orthorectification and Geometric Quality Assessment of Very High Spatial Resolution Satellite Imagery for Common Agricultural Policy Purposes, XXth ISPRS Congress, Istanbul, Turkey paper 492, [4] European Commission, Guidelines for Quality Checking of Ortho Imagery, [5] European Commission, Common Technical Specifications for The 2005 Campaign of Remote- Sensing Control of Area-Based Subsidies, 2004 JRC Ipsc/G03/P/Hke/Hke D(2004)(3486) [6] Hall R.J., D.P. Davidson and D. R. Peddle, Ground and remote estimation of leaf area index in Rocky Mountain forest stands, Kananaskis, Alberta, Canadian Journal Of Remote Sensing, 29 (3): , [7] Lopes, A., R. Touzi, and E. Nezry, Adaptive Speckle Filters and Scene Heterogeneity, IEEE Transactions on Geoscience and Remote Sensing, 28(6): , [8] Naftaly, U., TECSAR Performances, Design and Status, Proceedings of the EUSAR 2004, 5 th European Conference on Synthetic Aperture Radar, p , 2004.