GOES-R R AWG Land Team: Green Vegetation Fraction (GVF)

Transcription

1 GOES-R R AWG Land Team: Green Vegetation Fraction (GVF) Presented by Peter Romanov Task Lead CICS University of Maryland, College Park 1

2 Land-GVF Team Peter Romanov, CICS/UMD (Product lead) Felix Kogan, STAR/NESDIS Yuhong Tian, IMSG Bob Yu, STAR/NESDIS (Land Team lead) Dan Tarpley, Short and Assoc. Hui Xu, IMSG

3 Outline Executive Summary Algorithm Description Product Output Validation Results Summary

4 Executive Summary The GVF algorithm generates Option 2 GVF product from GOES-R R ABI. ATBD (8%), Software Version 3, and TRR will be delivered correspondingly in June, July and August 21, as scheduled The GVF algorithm uses NDVI estimates GVF algorithm has been developed and tested with MSG SEVIRI data. Validation studies indicate that the retrieved product is within specs.

5 Current Requirements: Vegetation Fraction: Green Product Observational Requirement Accuracy Precision Latency Horizontal Resolution GVF Full Disk.1 (LZA < 55).2 (LZA < 7).1 (LZA < 55).2 (LZA < 7) 6 min 2 km 5

6 Qualifiers GVF product has the following qualifiers. Solar zenith angle < 7 degrees 6

7 Algorithm Description

8 Algorithm Summary This ABI GVF Algorithm generates the Option 2 GVF product. GVF is defined as the fraction of area within the instrument footprint occupied by green vegetation GVF is derived from NDVI at TOA. Linear mixture approach is used. GVF estimates require daylight and clear sky conditions GVF is generated hourly 8

9 GVF Algorithm GVF is defined as: GVF = (NDVI NDVI min ) / (NDVI max NDVI min ) NDVI max and NDVI min are location independent All NDVIs are TOA values 9

10 NDVI Angular Anisotropy NDVI* Lat=16.2S Year: 27, Day: 78 Location: 16.2 S, 51.6 W Satellite view angle: 6 Lon=51.6W Backscatter NDVI Relative Azimuth Solar Elevation Forward Scatter Angle, deg NDVI* FS Lat=42.7N Year: 27, Day: 183 Location: 42.7 N, W Lon=3.6W Satellite view angle: 49 Backscatter NDVI Relative Azimuth Solar Elevation FS Angle, deg NDVI* Lat=8.7N Year: 27, Day: 32 Location: 8.7 N, E Lon=22.7E Satellite view angle: 28 Forward Scatter NDVI Relative Azimuth Solar Elevation Backscatter Angle, deg Local Time, h Local Time, h Local Time, h If NDVI is not corrected for angular anisotropy, GVF will also depend on the observation geometry MSG

11 GVF Algorithm: Major Tasks Establish NDVI angular anisotropy model - Needed to bring all NDVI to a reference geometry Establish NDVI max and NDVI min - Needed to calculate GVF using observed NDVI 11

12 NDVI Anisotropy NDVI anisotorpy model: NDVI(Ɵ S, Ɵ V, φ) = NDVI(,,) [1 + C 1 f 1 +C 2 f 2 ] f 1 =( tanɵ S + tan Ɵ V ) : change with zenith angle f 2 =( cosφ + 1 ) 2 (tanɵ S tanɵ V ) 1/2 : azimutal change where Ɵ S : solar zenith angle Ɵ v : satellite zenith angle φ : relative azimuth C 1 and C 2 have been established empirically from SEVIRI diurnal clear-sky NDVI data 12

13 Clear-Sky NDVI Dataset Over 8 diurnal clear-sky NDVI time series identified interactively Even coverage of SEVIRI domain and major surface cover types Dataset was used to establish kernel weights for NDVI anisotropy model (C 1 = -.723, C 2 = -.11) Sample of NDVI daily records from the dataset Locations where clear NDVI daily records were acquired 13

14 NDVI Angular Anisotropy Model Testing Latitude: 47.4 N Longitude: 27.2 E 8 Observed NDVI Predicted NDVI Solar Elevation 9 8 Hourly NDVI values for the whole day were predicted using one value of NDVI observed close to local noon time. NDVI* Solar Elevation, deg Local Hour NDVI observed vs predicted 14

15 Effect of NDVI Angular Correction 4 Anisotropical correction reduces NDVI diurnal variation Still it can not eliminate the diurnal change completely Number of cases 3 2 Without correction With correction Before angular correction After angular correction Statistics of RMSD in daily NDVI records: before and after correction (Cloud clear daily NDVI time series used) Daily NDVI NDVI RMSD scatter 15

16 Establishing NDVI for max GVF 95 th percentile Weekly max NDVI obtained from SEVIRI data for Frequency, % NDVI* Cumulative Frequency, % NDVI values brought to the reference geometry of observations (Ɵ s = 45, Ɵ v = 45, φ = 9 ) NDVI max was assumed equal to the value of 95 th percentile of NDVI frequency distribution NDVI max =.59 Weekly max NDVI frequency distribution from SEVIRI data for

17 Establishing NDVI for min GVF 95 th percentile 2 1 Frequency, % Cumulative Frequency, % NDVI*1 Statistics of observed NDVI was generated for a desert site NDVI min was assumed equal to 95 th percentile of NDVI frequency distribution in Sahara desert (NDVI min =.13) 17

18 GVF Retrieval Acquire ABI NDVI (including cloud flag) Bring observed NDVI to a reference geometry Estimate GVF with linear mixture model 18

19 Algorithm Summary GVF is derived from GOES-R NDVI with a linear mixture model NDVI is corrected for angular anisotropy using a kernel-driven model All parameters in the GVF and in the NDVI model are determined empirically from SEVIRI data 19

20 Product Output

21 GVF Algorithm Output The primary output is a binary full disk file of scaled GVF values: (GVF+1)*1 Output file is generated every 6 min The algorithm also generates two bytes of QC flags» Day / Night flag» Land / Ocean flag» Snow / No snow flag» Cloud flag 21

22 Example of GVF Product GVF from instantaneous images (date: 27141) GVF from weekly composite data(date ) Light gray: clouds, dark gray: solar zenith angle above 7 deg 22

23 Validation Approach

24 GVF Validation: General Accuracy No GVF ground truth is available, hence real evaluation of GVF product accuracy is not possible Precision: Spatial and temporal change in GVF should be consistent with variation of the state of the vegetation cover Precision criterion: Small high-frequency intra day changes in estimated GVF 24

25 GVF Validation: Details Method: Evaluate diurnal variation of derived GVF Validity criteria: RMSD of diurnal GVF should fit GVF precision requirement.1 for LZA <55 deg.2 for 55< LZA <7 deg 25

26 Validation Results

27 GVF Validation Results Frequency, % Original GVF Angle Corrected GVF Observations taken at large view zenith angle Statistics of GVF diurnal variation for MSG clear sky data. Red: Angular corrected GVF Green: Non-corrected GVF RMSD of GVF*1 GVF*1 RMSD NDVI angular correction reduces spurious diurnal variations in GVF Diurnal variation in GVF exceeds.1 in about 15% of all cases. These are mostly observations taken at large (over 55 deg) satellite zenith angle. 27

28 Next Steps to Reach 1% Metadata output will be added Quality flags will be added Continue validation of the NDVI angle correction model - Use TOA NDVI modeled with 6S - Use SEVIRI-observed NDVI Extensive validation of the GVF product with the common dataset Scheduled delivery is in September

29 Summary GOES-R GVF algorithm development is on schedule. MSG SEVIRI is used as GOES-R ABI prototype The algorithms uses NDVI and a linear-mixture approach to convert NDVI into GVF. Preliminary estimates show that the algorithm meets performance requirements Further validation studies will be conducted with GOES-R simulated data/products generated by AIT. Tailored products such as daily and weekly products and mostly cloud-free GVF products are needed and have to be developed 29