The SASS Data

The SASS Sensor Data Record (SDR) files used in this study are described in [2]. While the SASS SDR files do not include the integrated cell corners, we developed an algorithm for reconstructing the cell corner positions from data available in the SDR files (see Fig. 1). The SDR files contain measurements spanning an incidence angle range of to and include estimates of the normalized standard deviation () for each measurement. We have chosen to exclude measurements with excessive (%) in the experiments described below. The discarded measurements represent less than 5%of the measurements, most of which occurred at large incidence angles. Orbits (revs) for which the measurements exhibited excessive error were not included (see [4] for a detailed description of the SASS data anomalies and the rationale for selecting and excluding orbits with excessive error).

To illustrate the variability in the scatterometer data as well as demonstrate the validity of the model, all the vertically-polarized measurements over the three month mission from a region in north-eastern Brazil were plotted as a function of the measurement incidence angle (see Fig. 2). Only measurements with incidence angles in the range of to with %were used. This region is a portion of the Amazon tropical rain forest, an area previously noted for its homogeneous radar response. Linear regression was used to compute the and coefficients and the corresponding line shown. As evident from this figure, the linear model fits the data well, though there is significant data scatter about the best-fit line. This scatter is due to: 1) thermal noise in the instrument, 2) errors in the computation of the radar parameters used to compute from the measured power (this is known as ``retrieval error''), 3) variations in the radar backscatter of the forest canopy, and 4) variations in the calibration of the SASS instrument. The latter effect probably accounts for less than 0.1 dB [4]. The additive error due to the retrieval error and the thermal noise are both subsumed into (see LHW), though for the ranges of observed over land, is dominated by the retrieval error. The standard deviation of the canopy backscatter is estimated to be 0.15 dB [4].

The temporal variation in the radar response can be observed with the aid of Fig. 3 which contains plots of the vertically-polarized and versus Julian day (1978) for the study region defined above. The same measurements used in Fig. 2 were employed in creating these plots. To generate this plot a moving 10 day window was used to compute a time-averaged and value by linear regression of the versus incidence angle within the temporal window. The corresponding time-averaged and coefficients were plotted as lines in the upper portion of the plot. To illustrate the frequent changes in the measurements as a function of time, the time-averaged value from the upper plot was used to remove the incidence angle dependence of each measurement. The resulting estimate was then plotted in the lower portion of Fig. 3 as individual symbols. A windowed time average of these points was plotted as a solid line on this plot. Gaps in the time series are due to missing data. We note that both and remain relatively constant through out the mission for this region of tropical forest. As discussed later, some seasonal variation in savanna areas was noted. In effect, the scatter in the estimates in the lower plot is treated as measurement noise by the resolution enhancement algorithm.