Noroozi, Aliakbar 
  
is needed. For a selected test area which had no obstruction by clouds or smoke and for which the groundcover was 
constant, for each image a linear radiometric correction was applied which mapped the histograms of the reference areas 
located in the Saudi desert to the same mean and standard deviation as a reference date. The script used as procedure is 
presented in [3, Annex F]. 
In order to reduce the amount of processing a selection of all data has been made on the following criteria: 
For the purpose of the detection and estimation of smoke and smoky clouds: 
- Channel 2, the near infrared must have sufficient incoming radiation. 
- Shadow and shading effects caused by clouds and mountains should be at a minimum. 
The selection criteria are met by Midday images. 
There were some limitations for generation of reference image for estimation of smoke density. Having no fully 
clouded free images for the time before the Gulf War in the available dataset was a major problem in generating the 
reference image. Bad quality of numbers of the available NOAA-AVHRR images also was one of the restrictions on 
selecting images that could be used in making the reference image. 
For making the reference image, some sub windows of the images from May and June period also were used, so 
because of smoke and soot deposition in those periods, the smoke density estimation would be already underestimated. 
22 Introduction to model based image analysis, the integration of sparse ship and ground observations with 
densely sampled NOAA data 
In the context of the problem of assessing the damage caused by the oil during the Gulf War, the role of the analysis of 
the NOAA satellite images is to solve a transportation problem. 
The role of image analysis is to calibrate the transportation models involved and to provide a spatial dense sample set of 
the estimated concentration of the pollutants. 
The available external data consists of sparse samples at the receiving side, sparse meteorological data for the 
initialization of the transport equations and some estimates of the material flow as a function of time at the sources. 
Model based image analysis relates the ground observations through an explicit model to the spatial patterns detected in 
the digital images, and transforms the image data to estimated parameters for the amount of oil derived pollutants. The 
ground observations are used to calibrate the models such that the models produce a minimum error / maximum 
likelihood estimate prediction of the ground observations at the given ground sample points. 
The quality of the model fitting is derived from the residual errors between model and ground observations after a best 
fit. 
The major transportation model is transport by air in the form of soot, oily or greasy smoke and mixtures of haze and 
cloud with oil derived pollutants. 
For aerial transport, the required wind field data at some isobaric intervals could not be made available. The alternative 
was to estimate the smoke density as a function of space and time. The spatial density of smoke is accumulated over 
observation time. The accumulated smoke density is calibrated against point samples at ground observations, providing 
a dense sample of predicted accumulated load of pollutants at and between the observations at ground stations. 
23 Model for estimation of the accumulated smoke concentration 
Goal: estimation of total amount of pollutants transported by air from the Kuwaiti oil fires to the territory of the Iran. 
The product to be delivered is an accumulated smoke density map calibrated to the source-input data and available 
ground observations. 
Pre-processing 
Before the transportation models can be inverted, it is necessary to correct the remotely sensed data for undesirable 
geometric and radiometric effects. An important criterion for the quality of the pre-processing is that objects with 
constant properties should show constant parameters of the statistical models (such as mean vector and covariance 
matrix) that relating objects properties to RS measurements. 
In view of the large amount of data a selection has to be made of data suitable for the purpose. This is treated under * 
pre-processing of data". 
The model used here links the reflectance in the near infrared channel to the estimated smoke concentration by 
assuming that photons over land are mostly absorbed by smoke particles, while the combination of smoke and water 
particles produces an additional reflectance over the high absorption of NIR radiation by water. 
smoke-fraction(t) = 1 — SQRT(Reflection(t)/Reflection. Scene Land). 
Over water bodies the approximate model is: 
smoke-fraction(t) = (Reflection(t) — haze x Reflection Cloud)/Reflection Smoke , 
  
1020 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 
  
Fo 
cal 
Th 
ob: 
In 
CO! 
his 
Co 
ope 
Est 
Ove 
bet 
sen 
wel 
the 
mov 
Apri