Challenges of Spatiotemporal Data Application
Every technology is bound to face various challenges in its application. These challenges could arise from the technology itself, or from the scenarios that the technology is used in, or even from influences of natural environments. These challenges may bring limitations to the application of the technology, and what TerraQuanta needs to do is to solve the problem from a technical point of view on the one hand, and to find the best use scenario of the technology on the other hand.
Chart below is a simple analysis of challenges of spatiotemporal data faced on the application level by TerraQuanta
Challenges, Impact, and Solution
Challenges | Impacts | Our Thoughts and Solutions |
|---|---|---|
| Influences from Cloud | Cloud cover directly influence the observation result of optical remote sensing satellites, and most of the data products by TerraQuanta is based on these satellites. This means that a large amount of clouds in the area of the data service could impact negatively the performance of the data. | TerraQuanta uses a multi-source data fusion technique to fuse as much optical and radar satellite data as possible to minimize the impact of cloud cover on observations. At the same time, TerraQuanta is also doing research on replacing optical satellite remote sensing data with radar satellite remote sensing data, hoping to reduce the dependence of some data products on optical satellite remote sensing data. |
| Lacking of Spatial Resolution | The highest resolution of civil remote sensing satellites is 0.5 meters, and any monitoring demand in centimeter scales can not be met by satellite remote sensing. | The resolution of satellite remote sensing is both a commercial and a policy issue. For purely commercial reasons, satellite remote sensing might not be the best solution if there is a demand for higher resolution monitoring. Satellite remote sensing in general is more suitable for providing large-scale monitoring. Please click the data module list pages by TerraQuanta for specific scenarios. |
| Lacking of Monitoring Frequency | In an ideal situation, the current constellation of high-resolution satellites would already be capable of daily observations. However, except for geostationary satellites, remote sensing satellites can not achieve continuous real-time monitoring like video surveillance, and geostationary satellites usually face the problem of insufficient spatial resolution | The nature of satellites means that they do not provide the continuous monitoring capabilities of surveillance cameras. However, considering the advantages of satellite monitoring (such as large monitoring range, not easy to be interfered by human factors, low cost per unit area, traceable historical data, etc.), it can be used independently in many scenes, and can also be used as a supplement to video surveillance equipment in some scenes |
| Data Accuracy | In some quantitative monitoring scenarios, users will compare the accuracy of on-site IoT sensors with that of satellite remote sensing. Frankly speaking, from the perspective of accuracy alone, the data effect of satellite remote sensing cannot be compared with that of sensors | Improving the accuracy of satellite quantitative remote sensing data products will be the constant goal of TerraQuanta. But objectively speaking, it is not realistic to achieve the accuracy of ground sensor in a short time. The requirement of TerraQuanta on the accuracy of satellite remote sensing data is more to fully reflect the spatial distribution characteristics and historical evolution trend of some indexes, and provide information support from a macro perspective. A typical example is water environment monitoring products |
| Model Generalization | TerraQuanta’s data products rely heavily on algorithmic models based on deep learning. However, the spatial differences of remote sensing monitoring objects (for example, the different growth periods of maize in Henan and Heilongjiang) lead to the failure of the same model to solve the problem of the whole region | In essence, the generalization problem of the model is a sample problem, and the TerraQuanta needs to collect scattered and representative samples to help adjust the adaptability of the model |
| Huge Amounts of Computing | Spatiotemporal data naturally requires a huge amount of data, and the standardized data service provided by TerraQuanta demands an even larger amount of calculation. Computing pressure is relatively heavy in order to make sure that data is delivered on time. | In order to more efficiently complete the normalized product data calculation work, TerraQuanta independently built the Mammoth supercomputer Center, which has a number of A100 80G, A100 40G, A10, 3090,3080 Ti and other first-class GPU resources, as well as multiple AMD EPYC third-generation CPU resources including 7763. The hybrid cloud architecture has a total force close to 100 PFLOPS, which is enough to guarantee the demand for computing power. |
| Reliance on Samples | Models can not be established without samples. TerraQuanta provides rich data product for its clients, and naturally puts forward higher requirements on the type and quantity of samples. | TerraQuanta established self-controllable sample labeling and management system, with around 50,000 new samples added each month. |
| Result Verification | Manually verifying each data product is impossible due to the large amount of data products and vast serviceable area. | This is an important issue. To see how TerraQuanta solves this challenge, please check the Data Product Verification page. |