Global 500 m seamless dataset (2000&ndash;2022) of land surface reflectance generated from MODIS products

Liang, Xiangan; Liu, Qiang; Wang, Jie; Chen, Shuang; Gong, Peng

doi:https://doi.org/10.5194/essd-2023-314

Preprints

https://doi.org/10.5194/essd-2023-314

Preprints

22 Aug 2023

| 22 Aug 2023

Status: this preprint is currently under review for the journal ESSD.

Global 500 m seamless dataset (2000–2022) of land surface reflectance generated from MODIS products

Xiangan Liang, Qiang Liu, Jie Wang, Shuang Chen, and Peng Gong

Abstract. The Moderate Resolution Imaging Spectroradiometer (MODIS) is widely utilized for retrieving land surface reflectance to reflect plant condition, detect ecosystem phenology, monitor forest fire, and constrain terrestrial energy budget. However, the state-of-art MODIS surface reflectance products suffer from temporal and spatial gaps due to atmospheric conditions (e.g., clouds and aerosols), limiting their use in ecological, agricultural, and environmental studies. Therefore, there is an urgent need for reconstructing spatiotemporally seamless (i.e. gap-filled) surface reflectance data from MODIS products. In general, there are two challenges in reconstructing seamless MODIS surface reflectance product. First, the intrinsic inconsistency of observations due to various sun/view geometry. Second, the prolonged missing values resulting from polar night or heavy cloud coverage, especially in monsoon season. To address these challenges, we built a framework for generating the global 500 m daily seamless data cubes (SDC500) based on MODIS surface reflectance dataset, which contains the generation of a land cover-based a priori database, BRDF correction, outlier detection, gap filling, and smoothing. The first global spatiotemporally seamless land surface reflectance at 500 m resolution was produced, covering the period from 2000 to 2022. Preliminary evaluation of the dataset at 12 sites worldwide with different land cover demonstrated its robust performance. The quantitative assessment shows that the SDC500 gap-filling results have a root-mean-square error (RMSE) of 0.0496 and a Mean Absolute Error (MAE) of 0.0430. The SDC500 BRDF correction results showed a RMSE of 0.056 and a bias of -0.0085 when compared with MODIS NBAR products, indicating the acceptable accuracy of both products. From a temporal perspective, the SDC500 eliminates abnormal fluctuations while retaining the useful localised feature of rapid disturbances. From a spatial perspective, the SDC500 shows satisfactory spatial continuity. In conclusion, the SDC500 is a well-processed global daily surface reflectance product, which can serve as the fundamental input for large-scale ecological, agricultural, environmental applications and quantitative remote sensing studies. The SDC500 is available at: http://data.starcloud.pcl.ac.cn/resource/27 or https://doi.org/10.12436/SDC500.27.20230701 (Liang et al., 2023).

Received: 03 Aug 2023 – Discussion started: 22 Aug 2023

Download & links

Xiangan Liang et al.

Status: open (until 18 Oct 2023)

Post a comment Subscribe to comment alert

RC1:
'Comment on essd-2023-314', Anonymous Referee #1, 27 Sep 2023 reply
[Overview]
This study attempts to create a gap-filled (seamless) surface reflectance datasets from the corresponding MODIS products. The work is important, the methodology is largely OK (but with caveats indicated below), and the writing of the paper is clear. Overall I think the paper and the data products have the potential to be published. My main concern with the current paper is that it lacks a quantitative description of the uncertainties associated with the gap-filling method and thus the final results. At least these uncertainties should be discussed in the paper.
[Major]
Line 167-170 (Section 3.1, Generation of the a priori database). Predicting missing values at a location from sample values at other locations is a typical problem of spatial analysis or geostatistics. There are rich methodologies to address the problem (e.g., Kriging, Sequential Gaussian Simulation, etc.) but you appear to use a very simple home-made approach. My concern with your approach is that it lacks a way to quantify the uncertainties associated with the predicted values or, more rigorously, the random spatial field. Take the example of the grid cell in a tropical desert described here, there are simply no adjacent pixels to provide information of snow, crop, or forest. Therefore the information you fill in is purely random with large uncertainties (compared to other cases). And based on my understanding, this random information (time series) might be used in the next step (Landcover-based BRDF correction), which can be problematic! Even though this could be an extreme example, the point is that your methodology is too simple and lacks necessary mathematical/statistical rigor. There are numerous curves that can make the data fitting “looks” good (e.g., artists can do it) but, for science, we must make our choice based on solid quantitative reasoning.

For the same Section 3.1, Land cover type is not the only local factor that regulates BRDF. Topographic characteristics of the terrain (e.g., slope structure) can have significant influence on BRDF as well. It should be at least discussed.

Line 175-179. I don’t fully understand why you’d use all the 17 curves to fit the observed time series of surface reflectance. Why not just pick the curve of the same land cover type? Or broad categories (e.g., forests or vegetation)?

[Minor]
Line 55-57. “Urgent” may not be the best word to use here. After all, the MODIS instruments are > 20 years old already.

Line 146. “BRDR” should be “BRDF”.

Line 157-158. The last sentence “They are roughly independent … ” is confusing and can be removed.

Line 185. The Greek letter “phi” on the left-hand-side of the equation should be “phi 0”.

Line 305 and Fig. 5. The gray thin lines (spline-filled) show spikes at some of the raw observations. How is it possible, for the spline base functions are smooth low-frequency functions?

Figs. 5-16. What is the relationship between the SDC500 curves and the original MODIS observations? It doesn’t seem to be the time-smoothed mean at locations. Again, is there a way to quantify the uncertainties associated with the smoothed curves versus the raw observations?

Reply
Citation: https://doi.org/10.5194/essd-2023-314-RC1

Xiangan Liang et al.

Viewed

Total article views: 384 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
291	84	9	384	6	10

HTML: 291
PDF: 84
XML: 9
Total: 384
BibTeX: 6
EndNote: 10

Views and downloads (calculated since 22 Aug 2023)

Month	HTML	PDF	XML	Total
Aug 2023	153	48	3	204
Sep 2023	137	36	6	179
Oct 2023	1	0	1

Cumulative views and downloads (calculated since 22 Aug 2023)

Month	HTML	PDF	XML	Total
Aug 2023	153	48	3	204
Sep 2023	137	36	6	179
Oct 2023	1	0	1

Viewed (geographical distribution)

Total article views: 381 (including HTML, PDF, and XML) Thereof 381 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 02 Oct 2023

Short summary

The state-of-art land surface reflectance products suffer from temporal and spatial gaps, which make it difficult to characterize the continuous variation of the terrestrial surface. We proposed a framework for generating the first global 500 m daily seamless data cubes (SDC500), covering the period from 2000 to 2022. We have demonstrated its robust performance at 12 sites worldwide. The SDC500 can serve as the fundamental input for long-term large-scale ecological studies.


Total:	0
HTML:	0
PDF:	0
XML:	0