Effectiveness of Data Reduction and Transmission Techniques

We first calculate the effective compression ratio (ECR) required to support various imaging resolutions of an EO mission for a given downlink capacity. ECR, in this case, is the data reduction ratio achieved by combination of early discard and image compression. Let’s optimistically assume that sufficient downlink capacity exists for 3 m-1 d resolution RGB imagery of all of Earth – Planet’s current Dove constellation (Table 1) provides 3 m-1 d resolution RGB imagery of Earth’s land. Fig. 6 shows the ECR needed to support various resolutions using this downlink capacity. The results show that fine resolutions require ECRs in the the thousands to hundreds of thousands. Such ECRs are likely unachievable in most settings.

Table 3 shows achievable rates of early discard and their associated effective compression ratios (ECR) for several types of early discard. These rates have been calculated using gross Earth characteristics (50% images correspond to night, 70% images correspond to ocean, 10% of Earth is uninhabited, built-up areas - those areas which contain vertical construction - account for 2% of Earth’s area, 3 2 of Earth is covered by cloud) for optimistic assumption orbital dynamics (e.g., a non-dawn/dusk circular orbit for night data, equatorial orbit instead of polar orbit for habitation data, etc.). As the results show, the achievable early ECRs are far lower than the required ECRs reported in Figure 6. Note that some forms of early discard may be combined (e.g., imaging only built-up areas during the day) to achieve higher ECRs. However, this is limited by conditional dependencies (e.g., cloud distribution is dependent on land vs ocean, uninhabited implies non-built up, etc.).

We similarly estimate realistic ECR values for EO data when data compression algorithms are used. We ran different data compression algorithms on the Crowd AI Mapping Challenge [105] dataset of satellite imagery of built-up areas of Earth. One thousand images were randomly selected from the dataset after removing images which did not display a full scene 1 . Analysis of compression of SAR imagery used the xView3 validation dataset [112]. Table 4 shows the results. The results show that achievable compression ratios using lossless image compression are limited to < 4× for RGB imagery. This is in line with previous studies on lossless compression of satellite imagery [152]. High-quality ‘quasi-lossless’ lossy compression, also results in compression ratios of only 10 − 20× [29]. These numbers are off from the required ECRs by several orders of magnitude.

Assuming independence of early discard and compression (e.g., by discarding images of non-built-up areas and images at night), the ECR of combined compression and early discard is ≤ 4 × 100 = 400. This best case ECR is still up to 3.5 orders of magnitude short of the ECR needed for some of the fine resolution targets. These results show that compression and early discard have limited effectiveness at addressing the problem of too much data that will generated by future high resolution EO missions.

Another way to decrease the downlink deficit is to increase the amount of information which can be moved from space to Earth. While number of ground stations is anticipated to double from 2021 to 2026 [153], doubling the number of ground stations leads to no more than proportional increase in downlink capacity.

Satellite designers can also increase downlink capacity by modifying the design of their satellites RF communications. RF downlinks are modeled as additive white Gaussian noise communication channels [136], and are thus subject to the Shannon-Hartley theorem [133], which relates channel capacity, C, (in bit \(s^{-1}\)) to the channel bandwidth, B, (in Hz), and the signal-to-noise ratio (SNR) at the ground-station.

\(C = B \\log_{2}(1 + \\text{SNR})\)

Note that \(\\frac{\\partial C}{\\partial B} = \\log_{2}(1 + \\text{SNR})\) and \(\\frac{\\partial C}{\\partial \\text{SNR}} \\propto \\frac{1}{\\text{SNR} \\log(2) + \\log(2)}\). Thus, when \(SNR >> 0\), C scales linearly with B, but with the reciprocal of the SNR. This is called a ‘bandwidth limited’ regime, and satellite downlinks are squarely in this regime (e.g., Dove’s ground stations experience \(SNR \\sim 19\)[55]).

As the electromagnetic spectrum is a limited natural resource, satellites cannot simply scale their bandwidth, which is allocated to them by national and international governing bodies, such as the Federal Communications Commission (USA), and the International Telecommunication Union. Thus, satellite designers can only increase RF channel capacity by increasing signal strength. This can be achieved in one of two ways: 1) increase the power output by the antenna, and 2) increase the directionality, or gain, of the antenna. Increasing antenna output power requires increasing the input power, while increasing the antenna gain requires increasing the antenna’s aperture size, and thus increasing the physical size of the antenna for common satellite antenna types (i.e., patch antennas, helical antennas, and parabolic antennas).

Fig. 7 shows the infeasibility of meeting fine spatial resolution targets through scaling of RF downlinks in a bandwidth limited communications regime. Both a 2 kW antenna input power and a 30 m antenna fall far short of meeting the downlink capacity requirements of a 1 m resolution target, let alone \(< 1\) m resolutions.

The number of channels needed to be supported on the ground may also be unrealistic. Fig. 4b shows that the number of channels needed is many orders of magnitude greater than the number of channels which can be supported by current or near future ground stations.