Agricultural Monitoring Using Envisat Alternating Polarization SAR Images

In agricultural remote sensing, applied images should be acquired frequently enough in order to monitor important crop growth stages. Thanks to the cloud penetrating and flexible swath-positioning capabilities of space-borne SAR at present, images can be acquired even at the interval of few days during a growing season. In this study, dual-polarization (VV/VH) Envisat SAR images with high a temporal resolution were used in association with limited ancillary data to monitor crop growth and to classify crop species. It was noticed that the high temporal resolution enabled nearly continuous monitoring, but it also caused problems because of the varying incidence angles. Moreover, to carry out field surveys rapidly enough for research purposes was observed as a problem. An R of 0.55 was obtained for estimating the crop growth, when average crop height in parcels was used to describe the amount of biomass. An overall accuracy of 74.7 percent was achieved for crop species classification. Envisat VH polarization appeared to be useful in the estimation, even though, the noise equivalent  was too high to detect early crop growth. Field-based averaging was required, thus, for example for precision farming purposes a better spatial resolution would be needed to detect biomass variations within parcels. Introduction In general, the utilization of agricultural remote sensing can be categorized into the following application areas: (a) mapping of yield losses caused by lodging, flooding, pests, etc., (b) estimation or prediction of crop yield, (c) assessment of the area under cultivation, (d) crop species interpretation, (e) precision farming, where maximum yield is sought with minimum fertilization, and (f) control of agricultural subsidies (Henderson and Lewis, 1998; Lillesand and Kiefer, 1994; Cramp, 2003). A characteristic feature of all these application areas is that the time window for an appropriate image acquisition is very narrow compared with, for example, topographical mapping applications. In Finland, the Ministry of Agriculture and Forestry (MAF) publishes crop yield predictions three times a year, i.e., in June, July, and August. Nowadays, these predictions are based on the reports acquired from experts working in the Finnish Rural Development Centers (Yield estimation seminar of the MAF, unpublished, 2004). On a small scale, the yield estimates are relatively reliable since these experts are specialists in crop husbandry, and they know the weather and growing conditions of the previous few months of the growing season Agricultural Monitoring Using Envisat Alternating Polarization SAR Images Mika Karjalainen, Harri Kaartinen, and Juha Hyyppä in question. However, at present the predictions and estimations lack geographical details, and furthermore, there is great variability between municipalities in the estimates due to the subjective nature of the yield estimation. The objective is to ensure that in the future equally good predictions will be available for each municipality in Finland; this means that to improve the present estimation system more objective and frequent data about the crop growth during a growing season is needed. In principle, these requirements could be satisfied using satellite images. The most common instruments used in agricultural remote sensing are optical cameras and synthetic aperture radars (SAR) if ground based close-range remote sensing methods are excluded. For optical images, there are wellestablished methods to obtain vegetation-related information from the intensity of the image, because films and digital sensors are sensitive to wavelengths that overlap the region of the photosynthetically active radiation of the vegetation (Lillesand and Kiefer, 1994). Nonetheless, the major problem in the use of optical images in agriculture is cloudiness. For example, in Finland, there are roughly only up to four possible cloud-free periods in a growing season. Thus, a crop yield estimation based only on optical images would be rather unreliable. SAR overcomes the problem of cloudiness by using microwaves that have wavelengths from a few centimeters to one meter. Basically, the SAR sensor sends a pulse of electromagnetic radiation, and then records the amplitude and phase of the radiation coming back from the target. The backscattering coefficient, , is a measure describing the strength of the recorded radar signals from the target per unit area (Henderson and Lewis, 1998). Despite the advantages of SAR over optical images, the exploitation of SAR in real-time agricultural applications has been almost non-existent. There are two reasons for this. First, the cost of SAR image data is high, and second, crop information retrieval from the SAR images has proved to be a complicated inverse problem (Ulaby, 1998). With respect to agricultural fields the inverse problem means that the recorded SAR backscattering is a function of several physical properties, such as soil surface moisture and roughness, vegetation biomass and moisture, crop type, land slope and the orientation of seed rows with respect to the SAR look direction. In crop yield estimation, one would like to estimate the biomass of the vegetation, but its inversion from the recorded SAR backscattering is very PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING J a n u a r y 2 0 0 8 117 Finnish Geodetic Institute, Department of Remote Sensing and Photogrammetry, Geodeetinrinne 2, 02430 Masala, Finland ([email protected]). Photogrammetric Engineering & Remote Sensing Vol. 74, No. 1, January 2008, pp. 117–126. 0099-1112/08/7401–0117/$3.00/0 © 2008 American Society for Photogrammetry and Remote Sensing