07nrm

Information about 07nrm

Published on January 22, 2008

Author: Susann

Source: authorstream.com

Content

Slide1:  Where, is the mother Wavelet; λ, represents the scale factor which is related to the frequency, and u represents the translation, associated with time. The WT, rescales the signal maintaining its structure[5]. ESTIMATING RAINFALL FROM THE NORMALIZED DIFFERENCE VEGETATION INDEX USING WAVELET TRANSFORM Christian Yarlequé1, Adolfo Posadas1,2, Roberto Quiroz1 1Centro Internacional de la Papa, Apartado Postal 1558, Lima 12-Perú, 2 Facultad de Ciencias Físicas, DAFI, UNMSM, Lima 1, Perú. [email protected] / [email protected] / [email protected] 1. Abstract A methodology to estimate rainfall from remotely sensed vegetation indexes is described. Daily time series of measured rainfall are jointly analyzed with the corresponding signal of the normalize difference vegetation index (NDVI). The Fourier Transform (FT) is used to estimate the magnitude and the phase lag between the two signals. The periodic and proportional events, already corrected for the magnitude and lag time were processed with the Wavelet Transform (WT). Using the high-frequency Symmlet2 wavelet filter, the base and the noise spectra for both signals were attained. It was shown that the base spectra (NDVI and rainfall) for the second level (D=2) decomposition were alike. The base spectra at D=2 from the NDVI time series was then combined with the noise from the rainfall time series to reconstruct the rainfall event. Once the appropriate lag time was determined, the determination coefficient (R2) for daily rainfall events was greater than 0.71. The lag time varied from 56 to 95 days, variation explained by differences in soil texture and land cover. The precision for determining rainfall from NDVI is superior to published data and comparable to studies conducted over higher aggregations such as monthly or seasonal rainfall. 2. Data and Data Processing The 10-days composite NDVI data were assumed to represent the average value of the dekad and thus repeated for the specific period to construct the daily series (Figure 3). The method presented in this paper is suitable for applications where rainfall data is required daily e.g. for the simulation of crop growth or for characterizing how land surface influences climate on different spatial and temporal scales. The precision of the method described (0.85>R2>0.71) is found in the literature only for rainfall aggregations over long periods of time such as months or the whole growing season. The application is recommended for areas where weather station networks are not sufficient to account for the spatial variability, which is a global problem. It is also suitable for applications in climate change studies. The study also shows that the lag time varied across weather stations. This is a reflection of the changes in water retention capacity and the influence topography and land cover exerts on runoff and water balances. Improvements on the conversion of dekadal to daily NDVI data as well as the use of other vegetation indexes are being implemented in our laboratory. 4. Conclusions 5. References [1] W. W. IMMERZEL, R. A. QUIROZ, S. M. DE JONG; “Understanding complex spatiotemporal weather patterns and land use interaction in the Tibetan Autonomous Region using harmonic análisis of SPOT VGT-S10 NDVI time series”. (2004) [2] MARIAN PRUTSCHER (1998), Series de Fourier, http://www.e-technik.uni-ulm.de/world /lehre/basic_mathematics/fourier/node2.php3 [3] GONZÁLES Rafael C. and Richard E. Woods, (1992), Digital Image Processing, Editorial Addison-Wesley Publishing Company INC,Firts edition. [4] POLIKAR ROBI, (1996), “The Wavelet Tutorial”, 329 Durham Computation Center Iowa State University. http://users.rowan.edu/~polikar/WAVELETS/WTtutorial.html [5] FOUFOULA-GEORGIOU Efi and KUMAR Praveen, (1994), Wavelets in Geophysics, editorial Academic Press inc, Primera edición. [6] NICHOLSON S. E. and FARRAR T. J., The influence of Soil type on the relationships between NDVI, Rainfall, and Soil Moisture in semiarid botswana. I. NDVI response to Rainfall, Remote Sens. Envirom. 50: 107-120, (1994). Where f(t) is the signal and t the time when data was collected. The ratio of the mean value of the two signals (n=0 in equation 3) or characteristic amplitude was used to bring both signals to the same scale (unit). 3. Results and Discussions This new signal was processed with the Wavelet Transform (WT)[4], Wf(λ,u), of the signal f(t): Figure 6 shows the similarity of the base signal when D=2. Therefore the reconstruction was initiated at this level by combining the base NDVI (2[1SB}BN) with the noise from the rainfall (2{1SB}AL ) to obtain 1SBR. To arrive at the final reconstruction (R=0), the reconstructed signal (R=1) is then combined with the rainfall noise for D=1 (1SAL). A pure regression analysis between daily NDVI and rainfall showed no relation-ship whatsoever. Nonetheless, reconstructed daily rainfall explained 71 to 85 percent of the measured rainfall and a s=2.46mm/day Remote sensing Vegetation indexes are spectral measures derived from remotely sensed data in the red and near infrared spectral regions. The red spectral response is inversely related to the chlorophyll density, and the near infrared response is directly related to scattering in individual leaves and between leaves in the canopy. Combining data from these two adjacent spectral regions can be used to estimate the intercepted fraction of the photosynthetically active radiation. The Normalized Difference Vegetation Index (NDVI) is the most commonly used index to estimate this photosynthetic capacity at large spatial scales (equation 1) Data and data processing A data set containing 197 10-day composite NDVI images derived from the SPOT-4 and SPOT-5 VEGETATION instrument was used, spanning the period January 1999 to September 2003. The images were geometrically corrected and stacked for the extraction of the NDVI time series data. Daily rainfall data recorded by the Peruvian Meteorology and Hydrology Service (SENAMHI) in 10 localities of the Puno high plateau were used. The corresponding weather stations were located in the image (Figure 1). The graphical results presented in this paper correspond to the Mazo Cruz weather station located at: X0= 70°14’5.78; Y0 = 16°3’55.36''; Yf = 17°26’42.10''; Xf = 68°51’19.04'‘. The periodical behavior of the two signals (frequency, periodicity and amplitude), was performed using the first 6 coefficients (n=0,1,…,6) [1][2][3]in Fourier Series (equation 2): Figure 1. The squares indicate "Bands of NDVI" each 10 days compared with the data of rain ("."). Figure 3. Daily NDVI, Mazo Cruz. Figure 2. Daily Rain, Mazo Cruz. Signals (rainfall and NDVI) were decomposed twice (second level decomposition) using the wavelet Symmlet2, a high frequency filter. The two spectra obtained for each signal –low frequency (base) and high frequency (noise) – were used in the reconstruction. The NDVI “base” signal duly re-scaled by the factor obtained from equation 3, was combined with the “noise” extracted from the rainfall signal from the respective weather station, to re-construct the daily rainfall (Figure 6). The lag between the signals (Figure 4) was determined for the aggregation of the data for different time periods and then subtracting the lag period (days) to the daily NDVI series and calculating the determination coefficient between the reconstructed and the measured rainfall (Figure 5 & Table 1). Figure 4. Waves with lag (Mazo Cruz) Figure 5. Data in Phase (Mazo Cruz) The level of predictability changed as a function of the lag time, which was calculated with aggregations at different time intervals (Table 1). The maximum relationship was obtained with lag times of 2 to 3 months. This is a reflection of the water holding capacity of the soils which explains how rainfall affects vegetation growth[6]. A visual representation of the goodness of fit is shown in Figure 7. Table 1. Determination coefficients and lags (d) corresponding to different aggregations for 10 weather station in the high plateau of Peru. The results obtained exceeded by far the previous studies where NDVI is used to assess rainfall based on pure statistical relationships. The literature reports no relationship on a daily basis and the relations become meaningful when the data is aggregated monthly or throughout the growing season. (1) (2) (3) (4) Figure 7. Measured Rainfall (----) and Reconstructed rainfall (----), for Mazo Cruz, with a lag = 56 days. Equation 1. IR - reflectance in the near infrred region (0.7 - 1.1 um) and R -reflectance in the red region (0.58-0.68 um), Figure 6. "MODEL FOR RAINFALL RECONSTRUCTION“.

Related presentations


Other presentations created by Susann

Athletic Footwear Industry
31. 03. 2008
0 views

Athletic Footwear Industry

radioactivity 1
07. 02. 2008
0 views

radioactivity 1

Africa Unit 3
02. 04. 2008
0 views

Africa Unit 3

Market Research
03. 03. 2008
0 views

Market Research

module 1 intro
08. 05. 2008
0 views

module 1 intro

zly
07. 05. 2008
0 views

zly

031016 inhale montreal
02. 05. 2008
0 views

031016 inhale montreal

new leg
02. 05. 2008
0 views

new leg

JM UWA Luncheon Oct2007
30. 04. 2008
0 views

JM UWA Luncheon Oct2007

Tutorial
24. 04. 2008
0 views

Tutorial

NSW China Briefing
22. 04. 2008
0 views

NSW China Briefing

Weber Fall06
18. 04. 2008
0 views

Weber Fall06

SharksPP
17. 04. 2008
0 views

SharksPP

Dennill Managing Your Risk
09. 01. 2008
0 views

Dennill Managing Your Risk

AAS Presentation
10. 01. 2008
0 views

AAS Presentation

GANG Powerpoint
11. 01. 2008
0 views

GANG Powerpoint

885
15. 01. 2008
0 views

885

Chapt 05
16. 01. 2008
0 views

Chapt 05

Class11
09. 01. 2008
0 views

Class11

Mercantilism
18. 01. 2008
0 views

Mercantilism

ppe p sp
20. 01. 2008
0 views

ppe p sp

training docs
21. 01. 2008
0 views

training docs

eukaryoticorgs
22. 01. 2008
0 views

eukaryoticorgs

LimitedBrandsPresent ation
04. 02. 2008
0 views

LimitedBrandsPresent ation

MLA outreach presentation
15. 01. 2008
0 views

MLA outreach presentation

CenturyTheatre Ad Sept06 gg
15. 01. 2008
0 views

CenturyTheatre Ad Sept06 gg

LutzWalter
22. 01. 2008
0 views

LutzWalter

secondary aluminum
12. 02. 2008
0 views

secondary aluminum

2 3 Ben Sekamatte
25. 01. 2008
0 views

2 3 Ben Sekamatte

Clara Qualizza
19. 01. 2008
0 views

Clara Qualizza

Session 4 Oper vs Aux
28. 01. 2008
0 views

Session 4 Oper vs Aux

Ch5 6 10 14
29. 01. 2008
0 views

Ch5 6 10 14

Rites of Passage
29. 01. 2008
0 views

Rites of Passage

Jonah Presentation
30. 01. 2008
0 views

Jonah Presentation

tjea
07. 02. 2008
0 views

tjea

firesafety
07. 02. 2008
0 views

firesafety

Wiriya Suwannet
10. 01. 2008
0 views

Wiriya Suwannet

s Oceans
13. 02. 2008
0 views

s Oceans

Freeman
20. 02. 2008
0 views

Freeman

Forage Diseases
27. 02. 2008
0 views

Forage Diseases

SlaneP constraints
22. 01. 2008
0 views

SlaneP constraints

EPD2 Present and Future
28. 02. 2008
0 views

EPD2 Present and Future

INSOMNIA DrJeanGrenier Nov 2007
28. 02. 2008
0 views

INSOMNIA DrJeanGrenier Nov 2007

SAAS Gianessi
28. 01. 2008
0 views

SAAS Gianessi

5th Grade FCAT Review
14. 03. 2008
0 views

5th Grade FCAT Review

talent engine discussion guide
16. 03. 2008
0 views

talent engine discussion guide

Lecture14 TheEarth
11. 03. 2008
0 views

Lecture14 TheEarth

GEO205 powerpoint 12
27. 03. 2008
0 views

GEO205 powerpoint 12

Parasitology Basic 07
28. 03. 2008
0 views

Parasitology Basic 07

world geography
15. 04. 2008
0 views

world geography

Irwin
16. 04. 2008
0 views

Irwin

BuildingWordnets
05. 02. 2008
0 views

BuildingWordnets

Douglas Morgan
16. 01. 2008
0 views

Douglas Morgan

Wien
05. 03. 2008
0 views

Wien

NSO 2007
14. 01. 2008
0 views

NSO 2007

GeneralOutreachKansas
29. 01. 2008
0 views

GeneralOutreachKansas

Rita Tucker
25. 01. 2008
0 views

Rita Tucker

Bob Moseley Kawagebo
05. 02. 2008
0 views

Bob Moseley Kawagebo

michigan0314b
15. 01. 2008
0 views

michigan0314b

Canning Fruits
06. 02. 2008
0 views

Canning Fruits

English Labofa EGO Nordic
05. 03. 2008
0 views

English Labofa EGO Nordic

SlidesWithNotes
14. 02. 2008
0 views

SlidesWithNotes

GMkk
24. 01. 2008
0 views

GMkk

20030909152346 EVI
13. 01. 2008
0 views

20030909152346 EVI

praca5
03. 03. 2008
0 views

praca5

majingweili
14. 04. 2008
0 views

majingweili

Module3Presentation
25. 02. 2008
0 views

Module3Presentation

Jones NDDA
20. 01. 2008
0 views

Jones NDDA

Thornton CreatingAClimate
17. 01. 2008
0 views

Thornton CreatingAClimate

FirstHalfStudyGuide
15. 02. 2008
0 views

FirstHalfStudyGuide