Cloud top pressure retrieval with DSCOVR-EPIC oxygen A and B bands observation

13 An analytic transfer inverse model for Earth Polychromatic Imaging Camera (EPIC) 14 observation is proposed to retrieve the cloud top pressure (CTP) with considering in-cloud 15 photon penetration. In this model, an analytic equation was developed to represent the reflection 16 at top of atmosphere (TOA) from above cloud, in-cloud, and below-cloud. The coefficients of 17 this analytic equation can be derived from a series of EPIC simulations under different 18 atmospheric conditions using a non-linear regression algorithm. With estimated cloud pressure 19 thickness, the CTP can be retrieved from EPIC observation data by solving the analytic equation. 20 To simulate the EPIC measurements, a program package using the double-k approach was 21 developed. Compared to line-by-line calculation, this approach can calculate high-accuracy 22 results with a one-hundred-fold computation time reduction. During the retrieval processes, two 23 kinds of retrieval results, i.e., baseline CTP and retrieved CTP, are provided. The baseline CTP is 24 derived without considering in-cloud photon penetration, and the retrieved CTP is derived by 25 solving the analytic equation, taking into consideration the in-cloud and below-cloud 26 interactions. The retrieved CTP for the oxygen A and B bands are smaller than their related 27 baseline CTP. At the same time, both baseline CTP and retrieved CTP at the oxygen B-band are 28 larger than those at the oxygen A-band. Compared to the difference of baseline CTP between the 29 B-band and A-band, the difference of retrieved CTP between these two bands is generally 30 reduced. Out of around 10000 cases, in retrieved CTP between A-and B-bands we found an 31 average

to the CTH and the cloud geometrical thickness. The difference in the O2 A-and B-band cloud 84 centroid heights is resulted from the different penetration depths of the two bands. Compared to 85 the cloud height variability, the penetration depth differences are much smaller and the retrieval 86 accuracy from this method can be affected by the instrument noise (Davis et al. 2018a, b). 87 In this paper, to address the issue of in-cloud penetration, we proposed an analytic method 88 to retrieve the CTP by using DSCOVR EPIC oxygen A-and B-band observation. This analytical 89 method adopted ideas of the semi-analytical model (Kokhanovsky and Rozanov, 2004;Rozanov 90 and Kokhanovsky, 2004), and developed a quadratic EPIC analytic radiative transfer equation to 91 analyze the radiative transfer in oxygen A-and B-band channels. The structure of this paper is as 92 follows: section 2 describes the theory and methods, which includes several subsections, i.e., the 93 introduction of DSCOVR EPIC oxygen A and B bands filters, the theory of CTP retrieval based 94 on EPIC oxygen A-and B-band observation, and the detailed retrieval algorithm; section 3 95 describes the application and validation of the CTP retrieval method, which also includes several 96 subsections, i.e., case studies of CTP retrieval, validation of the retrieval method, and retrieval of 97 global observation; and section 4 states the conclusions of this study.   Thus, we can use the ratio of reflected radiance (or reflectance) at the top of atmosphere (TOA) 131 of oxygen absorption and reference bands (i.e., 764 and 779 , 688 and 680 ) to study the 132 photon path length distribution and derive the cloud information. Also, compared to any specific 133 EPIC oxygen absorption bands (i.e., 764   selections. Due to the complexity of cloud vertical distribution, whatever the accuracy of the 185 correlation coefficients is, the estimation will certainly bring in error.

186
With an estimated cloud pressure thickness, a multi-variable LUT searching method can 187 then be used to interpolate and obtain the CTP. It is worth noting that the reflectance ratio of 188 absorption/reference can be seen as a function of surface albedo, solar zenith and viewing angles, 189 COD, CTP, and cloud pressure thickness. Some atmospheric variables will have a non-linear 190 effect on the reflectance ratio. For example, the reflectance ratio is more sensitive to the variation 191 of COD when COD is small. Overall, the reflectance ratio varies monotonically and smoothly 192 with these variables (shown in Figure 3). With a relatively high-resolution simulated  respectively.

219
When clouds exist, the incident solar radiation is reflected to TOA in three primary ways. (1) Above Cloud: the reflected solar radiation is determined by the oxygen absorption optical 252 depth above the cloud and air mass directly. 253 Here, 0 is a weight coefficient.

256
(2) Within Cloud: the reflected solar radiation is not only determined by oxygen absorption 257 optical depth above cloud and in-cloud, but also by penetration related factors, e.g., COD. Due to 258 photon penetration, oxygen parameter 2 influences the enhanced path length absorption: Equivalence theorem (Irvine, 1964;Ivanov and Gutshabash, 1974; van de Hulst 1980) is used to 261 separate absorption from scattering: ( 2 ) is determined by two absorption dependences: strong (~ √ 2 ) and weak (~ 2 ).
When these parameters (i.e., A, B and C) are obtained from EPIC observation data and 307 other data source, we can easily solve the quadratic equation to retrieve cloud top O2 absorption 308 depth, and then CTP.  Where is the line intensity, and are the line center wave number and half width, 327 respectively; 0 and 0 are standard atmospheric pressure and temperature, respectively.

328
In the simulation of EPIC measurements, the atmospheric layer at a given layer-average 329 pressure can have drastically different temperature depending on the atmospheric profile in use.

330
To ensure the accuracy of simulation, we need to use the LBLRTM package to calculate oxygen 331 absorption coefficients for each pressure/temperature profile, which is a time-consuming process.

332
Our goal has been to find a simple and fast method to calculate oxygen absorption coefficients oxygen B-band, the relative difference between LBL and double-k approach is much smaller 404 than that of the high resolution spectrum, which is less than 0.1% for clear day. Compared to 405 clear sky situation, the relative difference for cloud situations can be bigger. As shown in Table   406 1, the relative difference is -0.06% and -0.32% for typical high level optical thin cloud and low-407 level thick cloud situations, respectively. The comparison of simulated narrowband measurement 408 at EPIC oxygen A-band channel (764 nm) is also shown in Table 1

483
As shown in Figure 5, the baseline CTP value at A-band is slightly higher than the effective 484 CTP from NASA ASDC L2 data. But the baseline CTP value at B-band is substantially higher  During the CTP retrieval, with the exception of the previously mentioned analytic 500 equation coefficients, we can get the surface albedo data from GOME, obtain reflectance data, 501 solar zenith and view angles, COD, etc. from the NASA ASDC data file. Another very important 502 step in the retrieval processing is the acquisition of cloud pressure thickness data, which has a 503 substantial impact on the retrieval results. We currently use a statistical approach (i.e., cloud 504 pressure thickness (mb) = 2.5* COD +23) to estimate the cloud pressure thickness based on

522
We also used the LUT based method to do the retrieval for the same observation data, 523 because both methods share the same EPIC simulation package and the same simulated data 524 only pixels with total cloud cover (i.e., EPIC Cloud mask = 4), surface albedo less than 0.05, and 537 liquid assumed COD larger than 3 are considered. As shown in Figure 7a, there are a series of 538 pixels (around 400 cases) from EPIC and CALISPO measurements can be used for the validation 539 analysis. For the convenience of reading, we perform the analyses by using the case number as x 540 axis. Figure 7b shows the comparisons of cloud layer top pressure from CALIPSO and different 541 CTPs (i.e., effective CTP, baseline CTP, and retrieved CTP) from EPIC measurements. Figure   542 7c shows the cloud layer number measured by CALIPSO. According to Figures 7b and 7c   effective CTP is shown in Figure 8d. The A-band retrieved CTP is overall smaller than A-band 584 effective CTP, which difference is within 100 mb. The highlighted (brown or red) areas are 585 located in the high level clouds areas or large COD areas. This indicates that the complexity of 586 cloud system has significant impact on the CTP retrieval. Figure 8g and 8h show the baseline 587 and retrieved CTP in B-band respectively, which are similar to, but greater than the A-band. As 588 shown in Figure 8i, the retrieved CTP at EPIC B-band is overall significantly larger than the 589 retrieved CTP at EPIC A-band, which mean difference is up to 200 mb. CTPs, but for the complex cloud system with multiple-layer clouds, the CTPs derived from EPIC 605 A-band measurements may be larger than that of high level thin-clouds.