This report studies the spatial distribution of X-band, high resolution and high grazing angle polarimetric sea clutter data. The K distribution usually provides a good fit for the distribution of the VV polarised data. The HH polarised data is spikiest and its distribution exhibits a sudden departure from the K distribution in the tail region, which usually requires the KA or the similar distributions to achieve a better fit in the tail region. Due to drawbacks of the KA distribution, this report proposes the KK and WW distribution models to fit the distribution of sea clutter with spikes. It is found that the KK distribution provides overall the best fit. Distributions of the sum of K and Weibull distributed samples are also presented. DGAD This report contributes to the delivery of Milestone 4.1.1.1.1: High grazing angle sea clutter and target signatures in the AIR 7000 S&T Plan (Annex C – Technical Support Plan). The outcomes of the analysis contained herein will also form a component of the model delivered for Milestone 4.1.1.1.2: Radar modelling capability development – maritime of the Technical Support Plan. These activities are aimed at better understanding the radar performance drivers for operation of High Altitude Long Endurance (HALE) unmanned aerial vehicles (UAVs) in the maritime surveillance role, and therefore reducing risk in any acquisition decision. Sea clutter distributions have been studied for many decades. However most of these studies are based on sea clutter data collected at low grazing angles and for applications of radar mounted on ships or at the coast. Very little analysis of high grazing angle sea clutter has been published in the open literature. The next generation of airborne maritime radar surveillance systems, such as high altitude UAVs, views the sea surface at much higher grazing angles. Sea clutter returns at low grazing angles are often dominated by multipath, shadowing and ducting mechanisms, whereas the Bragg scattering from rough surfaces and scattering from whitecaps often dominate at high grazing angles. These different scattering mechanisms mean that the nature and characteristics of sea clutter at high and low grazing angles are different. In addition, the resolution of the future radar tends to be higher. The finer the radar resolution, the more discrete sea spikes in sea clutter. Understanding and modelling of such spikes are important for the prediction of radar performance and for guidance in developing improved target detection algorithms. Therefore searching for distribution models which provide precise distribution agreement especially in the tail region is necessary in sea clutter distribution studies in order to improve radar performance. In support of Project AIR 7000, DSTO conducted a sea clutter collection trial in the Southern Ocean approximately 100 km south of Port Lincoln in South Australia in 2004 using the DSTO developed X-band, fully polarised airborne radar imaging system, Ingara. Data were collected with incidence angle varying from approximate 45o to 80o, on 8 separate days over an 18-day period. The wind and wave conditions were also recorded using a wave buoy deployed nearby and the information provided by the Royal Australian Navy’s Directorate of Oceanography and Meteorology, and the Australian Government Bureau of Meteorology. The data used in this report are real aperture high range resolution (0.75 m) data with the radar operated in a circular spotlight mode. The radar could therefore be considered to look at the same spot but with different incidence and azimuth angles. Each dataset used in the analysis consists of approximately 106 samples, corresponding to a span of 3.5o to 8o in incidence angle change, depending on nominal incidence angle, and a span of 5o in azimuth angle change. Since the nominal incidence angle is in the plateau region and the span of the azimuth angle is narrow, we can consider the data distribution to be as the spatial distribution. The size of the samples in each dataset provides a reliable data distribution up to 1-cdf equal to the 10-5 level. It is found that the mean clutter varies periodically in azimuth with the maxima and the minima in the upwind and crosswind directions, respectively, and the second peak in the downwind direction. The shape parameter of clutter distributions, however, does not show a noticeable azimuthal pattern correlating to wave/wind directions. The VV polarised data is with the lowest spiky level compared to the HH and HV data. In general the VV data can be fitted by a K distribution with the shape parameter varying from about 4 to 25. The HH polarised data is spikiest and its distribution exhibits a sudden departure from the K distribution in the tail region, often in the region of 1-cdf (cumulative distribution function) equal to 10-3 and beyond. The phenomenon of the sudden departure is believed to be attributed by sea discrete spikes. The finer the radar resolution, the severer is the phenomenon. This observation indicates that the traditional K distribution might not be precise enough to model the distribution of sea clutter with spikes. The KA distribution, which has been more recently proposed in the literature to model the distribution of sea clutter with spikes, has significantly improved the agreement between the data and model distributions in the tail region. However, the KA distribution cannot be expressed in closed form, so it is computationally very expensive. It also imposes a difficulty for the analysis of radar performance, as the analysis often involves the clutter distribution function. Aimed at simplifying the distribution function, this report proposes a KK distribution, which is a mixture of two K distributions of which one representing the distribution of Bragg/whitecap scatterers and the other for the distribution of sea spikes. It shows that the KK distribution is as good as the KA distribution in terms of agreement in the tail region. In addition, the KK distribution introduces the least distortion to the K distribution in the low and mid regions. Mathematically, a KK distribution is simply a sum of two K distributions. Since the Weibull distribution is very close to the K distribution, this report also proposes a WW distribution to improve the agreement between the data pdf and the modelled pdf in the tail region. A WW distribution is a mixture of two Weibull distributions. In general, a Weibull distribution converges a little faster than a K distribution for shape parameters normally found in sea clutter statistics, which often leads to a bigger discrepancy between the data pdf and the Weibull pdf in the tail region. The Weibull fit, even for the VV data is not as good as the K fit. This however can be compensated if a WW distribution is used, as the convergence of the WW distribution is tuneable. The results show that the fitness of the WW distribution in the tail region is comparable to the KK or KA distribution. However, in the low and mid region, the agreement between the data pdf and the WW pdf is not as good as that between the data pdf and the KK (or K) pdf. The use of the KA, KK and WW distributions improves the agreement between the data and fitted distributions in the tail region. It is shown that the difference between the data cdf and the K cdf for the HH polarised data at the 1-cdf equal to 10-5 level can be as big as about −7dB, but the difference can be reduced to about ±1dB if the KK distribution is used to model the data distribution. Since the KK distribution provides the least distortions to the K distribution in the low and mid regions, the fit improvement in the tail region does not worsen the agreement in the low and mid region. The report also proves that a Weibull distribution can be transformed to a Rayleigh or gamma distribution and vice versa through a non-linear but simple mapping. Therefore, in the case where clutter data is modelled as a Weibull distribution, the data may be first transformed accordingly and then treated as a Rayleigh or gamma distribution, as the Rayleigh or gamma distribution is much easier to be dealt with. For simulation, a Weibull distributed dataset can be easily generated from a transform of a Rayleigh distributed dataset. CFAR schemes often employ local statistics of clutter to adaptively set the threshold for target detection. This report also discusses the distribution of the sum of K or Weibull distributed samples. A formula in closed form approaches the distribution of the sum of Weibull distributed samples, which does not have close form, has been proposed. Its correctness has been numerically verified using both the convolution method and simulated data. No noticeable error between the values given by the formula and the values numerically computed from the convolution method or simulated data has been found.