AT 652 – (Satellite) Atmospheric Remote Sensing - PowerPoint Presentation

1. AT 652 – (Satellite) Atmospheric Remote SensingCourse ObjectivesTo expose students to a broad spectrum of satellite remote sensing methods to derive geophysical parameters of interestTo provide students with the understanding necessary to distinguish between algorithm fundamentals and nuanced implementation detailsTo expose students to the latest thinking in remote sensing applications and agency plans going forward (this involves the instructor’s perception of current issues)

2.

3.

4. Remote SensingThe science and practice of deriving information about the physical environment without being in direct contact (in situ) with it.For satellite remote sensing, we measure time and E&M radiation (no sound in space)Requires a basic understanding of how the physical environment interacts with radiation (optical properties).Sensing of reflected/emitted radiation, decoupling of various components (e.g., surface/atmosphere).Examining the space/time/spectral properties of the scene constituents to identify unique characteristics.Can be done passively or actively. The focus of these lectures will be on the satellite platform.

5. Despite what kids are learning, this is not how Sounders work.

6.

7. Remote SensingThe science and practice of deriving information about the physical environment without being in direct contact (in situ) with it.For satellite remote sensing, we measure time and E&M radiation (no sound in space)Requires a basic understanding of how the physical environment interacts with radiation (optical properties).Sensing of reflected/emitted radiation, decoupling of various components (e.g., surface/atmosphere).Examining the space/time/spectral properties of the scene constituents to identify unique characteristics.Can be done passively or actively. The focus of these lectures will be on the satellite platform.

8. The Electromagnetic Spectrum

9.

10. The Forward ModelA model that determines the tracks that an animal is likely to make based upon the properties of the animal In the atmosphere, typically a radiative transfer model that predicts radiances from surface and atmospheric parametersRetrieval/Inversion Method?In the atmosphere, typically code that infers geophysical parameters from a set of observed radiances. The non-uniqueness problem refers to two animals that leave the exact same tracks – thus making it impossible to definitively specify the animal from the tracks alone.

11. DefinitionsLet y be the observation vector (i.e. channels 1, …. , n)Let x be the state of the Earth/Atmosphere system of interest, while b are the other parameters that affect y.Let f be the forward model that predicts y based upon xy = f(x, b) x = f -1 (y, b)If we carefully measured y, know b & f -1 and the problem is well constrained, then done.Remote Sensing• You don’t know f• You can’t sort out what is x and what is b• You sort of know f and b, but not f -1 • You know f and b, but x = f -1 (y, b) is ill-constrained

12. There are two general types of parametersThe continuous variable: Ozone concentration, rain rate, soil moisture, etcThe discrete variable: cloud/clear; land surface type

13. Remote sensing in Weather and Climate is dominated by the retrieval of continuous variablesThe easiest retrievals are for continuous variable for which there is training data that we believe, and there is a direct functional relationship between the observable, and the parameter of interest. These are perhaps the most common retrievals around todayRadar reflectivity (dBZ) & Obs. Sfc rainfall (mm/hr)Thermal IR (K) & Obs. Soil Moisture (g/kg)6.9 GHz brightness Temp. (K) & Obs. SST (K)

14. Simple Linear RegressionSoil Moisture Index

15. SSE = Want to find line y = a + b•x that minimizes Square of Sum of Residuals:

16. Pearson’s correlation coefficient Note: The PCC is not able to tell the difference between dependent variables and independent variables. For example, if you are trying to find the correlation between precipitation and soil moisture, you might find a high correlation of .8. However, you could also get the same result with the variables switched around. In other words, you could say that soil moisture causes high precipitation. While there is some feedback, the correlation is not that good. Therefore, as a researcher you have to be aware of the data you are plugging in.

17. Always be careful about the things you correlate

18. Channel naming conventions can be archaic

19. The “Cirrus” channelChannel 9: 1.36-1.39mm.

20. Classification Methodse.g.: Maxar’s Worldview-3 has 31cm resolution. (8 Vis & 8 NIR channels). Also CAVIS (Clouds Aerosols Water Vapor Ice and Snow) with 30m resolution . Ag. Conglomerate wants to estimate global acreage of planted rice in order to manage own crop.Solution: Pick out a manageable number of known rice fields and non-rice fields and teach the algorithm to classify scenes according to training data.

21. Rule based learningLet y be satellite observed radiation at three distinct wavelengths y = y1, y2, y3Minimum distance approach to learning. Start by quantifyingknown scenes by whether measurement is above or below a radiation thresholdi.e. y = (1, 1, 1) or (1, 0, 1) = Sfc. type is rice y = (0, 0, 0) = Sfc. type is something else1 1 1 0 0 0 01 1 0 0 1 1 0 00 1 0 1 0 1 0Truth 1 1 0Learn 1 1/0 1 0 0 Now build a “truth table”

22. Incoming channels are thresholded and compared to a known solution. Output is based on proximity to known solutions. These techniques have all been replaced by ML as they work really well. Training data is well known.> 0.6