After 50 years of development, satellite remote sensing has become one of the most effective tools to detect the earth on a regional and global spatial scale, providing a previously unimaginable amount of data. Satellite remote sensing can quickly monitor the atmosphere, land and ocean, and the non-contact observation process effectively avoids the limitations and dangers in field measurement. At present, satellite remote sensing data support a large number of practical applications. For example, it supports weather forecasting, agriculture, fisheries, mapping, urban planning and mineral resources exploration, monitoring environmental pollution, climate change, coastline dynamics, sea surface temperature and salinity, marine ecosystems and biomass, sea level change, vegetation, forests, etc., and provides important remote sensing data support for scientific research such as COVID-19.

Especially in the past two decades, satellite design has developed rapidly, and more advanced satellite payloads have been introduced. For example, MOPITT, GOSAT and OCO for monitoring carbon dioxide and methane in the environmental field, infrared sounders AIRS, TES, IASI, IMG and CRIS for weather forecasting and climate change by monitoring atmospheric conditions, as well as multi-purpose MODIS, MERIS and SGLI, as well as more specialized loads such as ATSR, AATSR, MISR and POLDER. In addition to the traditional passive observation, active observation loads such as CloudSat, CALIPSO and Aeolus are also launched to monitor the vertical structure of clouds, aerosols and wind fields.

The prosperity and development of space instrument technology and informatics have greatly weakened the restrictions on satellite observation in terms of hardware, data acquisition and processing, and academia has accumulated rich experience in managing and analyzing satellite data. continue to tap the potential of existing satellite data sets, and gradually understand how to improve future satellite data. These advances have helped people understand the true value and potential of satellite remote sensing, but also recognize its limitations. Academia has never stopped trying to solve the basic difficulties of remote sensing technology, such as the separation of signal and noise, accurate instrument calibration, less information of observation data and insufficient to fully describe geophysical, chemical and biological processes, and the increasing number of observable targets and objects. In the actual processing process, these fundamental problems of satellite remote sensing often need to be properly constrained by theoretical models, prior knowledge and auxiliary observation, which have not been completely solved to a certain extent.

At present, the main challenges faced by satellite remote sensing technology include the challenge of spatio-temporal coverage, the challenge of increasing the amount of information and cooperative observation, the challenge of remote sensing inversion algorithm, and the challenge of high-quality long-term data consistency and continuity. Looking to the future, it is necessary to sort out the challenges from the perspectives of satellite design, observation, data processing and monitoring applications.

1. The challenge of time and space coverage

One of the main advantages of satellite remote sensing is that it can quickly observe most areas of the earth, but the coverage of available satellite data has obvious limitations. Polar orbital loads in low orbits usually take at least one day or more to achieve global coverage, so many natural phenomena with high spatio-temporal variability can not be fully captured. Geostationary observation high orbit (GEO) loads make day and night observations of the same region with high frequency to solve this limitation. However, there is still a trade-off between spatial coverage and the resolution of satellite images (usually, higher coverage leads to lower spatial resolution).

The ideal situation is that satellite remote sensing has a wide range of high-resolution space-time coverage, but this is very challenging. For this reason, the design of satellite observation needs innovation and the synergy of auxiliary data and complementary observations to increase the coverage and resolution of spatio-temporal data records.

2. The challenge of increasing the amount of information and collaborative observation

Although the existing satellite observation performance is already high, the amount of information provided by these observations is still limited. In a complex environment, no single load can provide comprehensive information about the target, so it is still necessary to develop and deploy new sensors or their combinations with enhanced capabilities. For example, multi-angle polarization loads can provide the most appropriate data to characterize the detailed characteristics of atmospheric aerosols and clouds, but due to their limited sensitivity to vertical variations of aerosols and clouds, even the most advanced multi-angle polarization loads cannot guarantee a completely reliable three-dimensional characterization of aerosol characteristics. The active remote sensing loads represented by spaceborne lidar and radar can provide detailed information about the vertical variation of the atmosphere.

Therefore, while deploying satellite instruments with enhanced capabilities, it is necessary to explore complementary cooperative observations and integrate load observations with different sensitivities to targets on different time or space scales and in different spectral ranges. Satellite observation is combined with suborbital observation and chemical transport model results to increase the total amount of information of observation. As shown in the figure, when observing highly heterogeneous targets such as clouds, passive observation data (spectral, polarization and microwave) and active observations (lidar and radar) are combined to reduce observation uncertainty.

Cooperative observation of passive observation loads (spectrum, polarization and microwave) and active observation loads (lidar and radar)

3.The challenge of remote sensing inversion algorithm

The performance of the algorithm is a key factor affecting the quality of satellite remote sensing products. In fact, once a payload is deployed, the quality of the observed data can not be improved in hardware, but the inversion algorithm can still be continuously improved. In the past decade, the new remote sensing inversion algorithm has made remarkable progress. For example, the new algorithm tends to use a more accurate atmospheric state model (rather than a pre-calculated lookup table) to obtain aerosol and land surface and cloud characteristics at the same time, and to retrieve a large number of parameters at the same time. In the face of collaborative observation tasks, it is necessary to develop appropriate algorithms, apply them to different loads or their combinations, and analyze the results of multi-source data. More and more researchers take this as the goal of algorithm development, and the evolution process of the algorithm does show that the use of collaborative observation to improve the accuracy of inversion has great potential.

At the same time, the experience gained from the continuous application of machine learning methods in algorithm development shows that machine learning is very suitable for the analysis and interpretation of earth observation data. Machine learning can "learn" from data and make decisions with the least human intervention. In recent years, the emerging deep learning and deep neural network technology in this method has not only been applied to the processing and analysis of a large number of data in remote sensing research, and is expected to further improve the prediction and modeling of physical phenomena.

4. Challenges of consistency and continuity of high-quality long-term data

Long-term and high-quality observational records of basic variables are essential for monitoring and studying the Earth (for example, climate change, etc.). In order to ensure the continuity of high-quality observation, it is the key to correctly explain and explain the differences and consistency in multi-load data records. Therefore, the absolute calibration of loads and the mutual calibration of multiple related loads are very important. Calibration and calibration are challenging for many new instruments, especially for small satellite clusters themselves. After the problems of calibration and data authenticity verification have been solved, the continuity and compatibility of data generated by future satellite payloads and current payloads are equally important in order to achieve the continuity of satellite observations and high-quality long-term data, ensure that the accumulated records meet the needs of research on major global issues such as climate change.

In the more than half a century since the launch of the first satellite, satellite remote sensing has developed into a highly complex tool that provides large amounts of data to support all aspects of human activities, from basic science to daily life. These major challenges faced by satellite remote sensing also give birth to a huge development space in this field in the future. Here, we sincerely invite readers to contribute what they know and learn to solve these challenges together.