Global Monthly Vegetation Cover Reprint of Original Document

Introduction Page,
Reprint - [ Next Subsection, Acronyms , References, Home Page ]

---

3. Generic processing of AVHRR data

This section and section 4 are pertinent to AVHRR data use in general, and the GVI dataset in particular. The present section proposes an "ideal" processing scheme, which indicates the data flow and production, based on our experience with the GVI project. It should not be confused with the present accomplishments of the GVI project described in section 5. Although the introduction of a new structure in this review paper may appear as a complication, the lack of consensus on processing and nomenclature motivated us to present our views on these important issues here. The structure, given below, facilitates general overviewing of the main levels in data production and evaluating the present status of the GVI data products, as well as those within other projects.

Data for climate studies are usually subdivided into three general levels: A) primary instrument readings; B) retrieved geophysical parameters after appropriate quality assurance; C) fields prepared from level B data. Our experience with GVI data suggests that one more level D level, should be added to distinguish between generating complete fields and their statistical analysis. The currently proposed structure complies with the above categorization as well as with the data nomenclature in the International Satellite Cloud Climatology Project (Rossow and Schiffer 1991).

a. The A level
This level involves preprocessing of the acquired satellite sensor readings. This level is described in detail by NOAA's documentation, such as the NOAA Polar Orbiter Data Users Guide (Kidwell 1995), and therefore is not addressed here.

b. The B level
This level includes calibration and radiometric corrections of the A-level data; identification of the pixels that can be used for land studies; and transformation of the bidirectional reflectances, and , and brightness temperatures, T₄ and T₅, into climate-related land variables (surface albedo and temperature, fraction of vegetation, LAI, photosynthetically active radiation, etc.). The B-level data are mostly remote sensing oriented. They include auxiliary information, such as observation-illumination geometry, calibration coefficients, quality/cloud (QC) flags, that allow tests and development of methods to improve the quality of data products.

Steps in the B-level processing are shown in Fig. 1. The AVHRR data are supplied with information on navigation and prelaunch calibration appended but not applied, and are usually referred to as the NOAA 1B-level -- B1 in our notation. The data mapped daily into a regular geographical grid are denoted as B2. Temporal sampling is usually done by compositing procedures. The composite maps are referred to as B3. This stage provides users with BX.0 data (X=1,2,3): B1.0 (orbits), B2.0 (daily maps), B3.0 (composite maps). Any of them may be used as a starting point for further processing. BX.0 data contain original raw counts in AVHRR channels and information on calibration, navigation and observation-illumination geometry.

The first step of B-level data processing involves calibration -- conversion of sensor counts to physical quantities (, T₄, T₅)-- leading to BX.1 data products. At the next step, BX.2, quality/cloud identification of each pixel is done, with the results of different tests being packed in QC flags. The generated QC maps are appended to BX.1, making it the BX.2 data. Corrections for atmosphere and surface anisotropic and diurnal variabilities are made at the next two steps, BX.3 and BX.4, respectively. Transformation to land surface parameters is carried out at the final step BX.5. The last three steps are made by means of look-up tables so that the whole image, independent of QC flags, is transformed with efficient use of computer time. At the next level, it is decided which pixels are to be retained for land studies. Note that most of the boxes have two arrows coming in and two coming out, which implies that the processing sequence is nonunique and that the derived product (at each step) can be obtained in more than one way. The path that is used to derive the product thus should be added in a more detailed version of the proposed nomenclature.

To identify the time-space resolution of data products, we propose a flexible nomenclature: BX.Y/TS, where X=1,2,3 and Y=1,2,3,4,5 denote the stage and step of the B level at which the product was generated, and TS stands for the temporal and spatial resolutions, respectively. For example, T can be H, D, W, Dk, M, S, or Y for hourly, daily, weekly, dekadal, monthly, seasonal and yearly, respectively. It is implied that at the B level the original dataset resolution is not degraded (e.g. S=1, 4, 8, 15 standing for 1, 4, 8, and 15 km, respectively).

c. The C level
The data are for the users who need products for, for example, initialization and validation of models and do not want to spend time and resources on processing the remotely sensed data. At this level (Fig. 2), the Bdata are used as input, all auxiliary information is omitted, the QC flags are applied and complete fields of the variables are produced by spatial/temporal averaging, interpolation, and smoothing. The areas with missing data resulting from QC flags application are filled in by interpolation. Averaging/smoothing is done to filter out some residual noise due to imperfections in the original data (e.g. sampling) and the B-level processing. Some of the signal (e.g. surface spatial variability) may be lost as a result of smoothing so that the effective product resolution becomes lower. Using the TS mnemonics described above, the C-level data products can also be presented as CX.Y/TS with temporal/spatial scales specified. The latter could be already different (i.e. with degraded resolution) from the B-level data products that were used to generate the C-products.

d. The D level
This level involves statistical analysis of the C-level data. The D products may include climatology in terms of multiannual means and standard deviations with the original or reduced spatial resolution, relationships between various parameters on different spatial/temporal scales, and other statistics. The D-level products are most useful for climate analysis. Combining the D- and C-level products can be used for monitoring purposes, as described below. The nomenclature introduced above is applicable to the D products as well.