The first purpose of this research is to incorporate remote
sensing data into a water quality model as a more accurate initial
condition. The second purpose is to adapt a geographic information
system (GIS) to enhance the contribution of water quality modeling to
practical water quality forecasting. The water quality model (QUAL2E)
and an image processing and GIS package (ERDAS IMAGINE) were used in a
case study of the Te-Chi Reservoir in Taiwan. All water quality
variables from simulations are displayed on a geographically
registered map and in color to correspond with varying water quality
levels. The visualizing technique is helpful for rapid understanding
of water quality conditions.
Introduction
Eutrophication caused by the excessive growth of algae is a major
water quality problem in water resources. Environmental engineers have
been making efforts on monitoring, simulating, and controlling
eutrophication for a couple of decades. Various mathematical models
have been developed and applied on streams, lakes, and estuaries (e.g.
Lung, 1986; Thomann and Mueller, 1987; Kuo and Wu, 1991; Kuo et al.,
1994). All water quality models simulate the change based on an
initial condition and demands a comprehensive water quality sampling
program. However, the conventional measurement of water quality
requires in situ sampling and expensive and time-consuming laboratory
work. Due to these two limitations, the sample size often cannot be
large enough to cover the entire water body. Therefore, the difficulty
of overall and successive water quality sampling becomes a barrier to
water quality forecasting.
Remote sensing could overcome these constrains by providing an alternative for water quality monitoring over a range of temporal and spatial scales. A number of studies have shown that applications of remote sensing can meet the demand of a large sample over a spatial and temporal view of water quality. Imagery from satellite and aircraft remote sensing systems were used in the assessment of water quality, such as temperature, chlorophyll a, turbidity, and total suspended solids for lakes and reservoirs (Lillesand et al., 1983; Lathrop and Lillesand, 1989; Ritchie et al., 1991), for estuary water bodies (Verdin, 1985; Harding et al., 1995), and for tropical coastal areas (Ruiz-Azuara, 1995).
Most previous studies focused on the discovery of the relationship between remote sensing data and in-situ measurements. To move the application of remote sensing to a practical level, it still remains a task to incorporate water quality modeling with water quality monitoring. Moreover, integrating a geographic information system (GIS) in the system improves the display of the monitoring simulating results rather than using traditional numerical figures. This research is about integrating water quality modeling with remote sensing and GIS techniques to enhance the contribution of water quality modeling in practical water quality management. A short-term water quality forecasting system was developed by using SPOT satellite data, an image processing and GIS package ERDAS IMAGINE, and a water quality model QUAL2E. The specific objectives of this study were to:
Two decades ago, a part of the Te-Chi Reservoir watershed was developed into an agricultural area for economic purposes. A large amount of nutrients from fertilizers and farm wastes enter into the reservoir and stimulate an excessive algal growth because of the uncontrolled development of orchards and vegetable farms. A eutrophication problem becomes a big concern on the Te-chi Reservoir in the summer, specially in the upper Ta-Chia River. Water samples are taken regularly on a scheduled basis which is a part of long-term water quality monitoring project on Ta-Chia River. Chlorophyll a was measured for five of the 28 water samples taken during 30 and 31 August, 1994 and compared with a SPOT satellite data. The SPOT image were acquired on August 31, 1994 which has a time difference within 24 hours with the time of in situ sampling. There was no significant meteorological change during this 24-hour period. A 500 by 350 subarea image was subset from the original SPOT scene (see Figure 2).
Figure 1. -- Te-Chi Reservoir's location and watershed.
Figure 2. -- SPOT near infrared image of the Te-Chi Reservoir, which was
georeferenced in the UTM coordinate and subset from the nominal scene K299/J301
. The darkest area is water.
The first step, feature extraction, was to identify the water body
and extract this from the image by using unsupervised classification.
In this way the water pixels were differentiated from the land pixels.
An unsupervised classification was applied on the SPOT band 3 (NIR)
image and resulted in a total of 150 categories. By assigning the
first class to water throughout the entire image, the water area of
the reservoir can be identified. The procedure developed a binary mask
layer by renumbering the water pixels as "1" and the rest of the
pixels as "0". Afterward, the water body was extracted from the
other SPOT bands by multiplying by this mask layer.
With regard to the use of remote sensing in water quality
monitoring and modeling, the major concerns are: 1) the suitable
spectral channel to match the various characteristics of water quality
variables, and 2) the appropriate and efficient image processing to
convert image brightness to traditional water quality indices. To
cover distinctive spectral characteristics and distinguish the
brightness values reflected from plankton and suspended solids,
multispectral sensors would be a good choice (Robinson, 1985).
Therefore, a band ratio regression model was developed for delineating
the water quality conditions by remote sensing monitoring. A band
ratio between the near infrared (NIR) and red was suggested to detect
chlorophyll in water, due to a positive reflectivity of chlorophyll in
the NIR and an inverse behavior in the red (Rundquist et al., 1996).
Concordant with the theoretical basis, the correlation matrix shows
that the ratio of the red-band to the infrared-band has the highest
correlation to chlorophyll. Also, a green/red (XS1/XS2) ratio shows a
significant correlation with Secchi depth and phosphorus
concentration. These ratios were adopted to derive these three water
quality variables from satellite data. The regression model was
calibrated by in situ samples, which were collected from the water
below 1 m from surface. These samples were correlated to the
remote sensing data from SPOT using an average digital value from a
3x3 pixel window centered at the sampling location. A natural
logarithmic regression model was developed for chlorophyll a (CHLA),
Secchi depth (SD), and phosphorus (PO4) as:
lnCHLA = 9.37 + 10.10 ln XS3/XS2 (1)
lnSD = -3.32 + 4.39 ln XS1/XS2 (2)
lnPO4 = 10.85 - 10.01 ln XS1/XS2 (3)
The regression model is statistically significant (R2 = 0.95, P =
0.005) between the band ratios with chlorophyll and Secchi depth. A
lower correlation (R2 = 0.827, P = 0.032) was found for the phosphorus
variable. The coefficients in the above equations could vary
with different weather and illumination conditions. Also,
the regression model could have a bias because chlorophyll was
examined only for 5 of 23 samples. Te-Chi Reservoir is more than 40 m
depth and the water has a high turbidity, so the reflectance from the
bottom sediment was ignored. Atmospheric corrections were not
performed on the raw data, due to the clear sky conditions with no
visual atmospheric distortions.
The reservoir was divided into 22 reaches, depending on hydraulic
characteristics. Each reach consists of small and homogenous
elements
which are assigned to one of seven different types: including
headwater elements, standard elements, elements just upstream from a
junction, junction elements, last element in system, input elements,
and withdraw elements. Once the initial conditions and environmental
parameters were given, QUAL2E calculates the transportation and
interactions of pollutants and consequently predicts the changes in
algae, nutrient, and temperature for a given time step. Among these
initial water quality variables, chlorophyll and nutrient
concentrations are the most important concern and were derived from
the SPOT image.
In the study, QUAL2E was run for a one-dimensional dynamic
simulation that needs meteorological, hydrological, and biological
data inputs. Water flow was obtained by hydrologic monitoring
stations; cell area and volume were defined in advance for the
simulation; and death rate and settling rate of phytoplankton were set
as a constant value because of their steady influences on the model.
Thus, the factor dominating algal growth is the growth rate of algae
which is limited by several environmental factors, such as light
intensity, nutrients, and temperature. The meteorological data of the
Te-Chi Reservoir, including temperature, wind, and atmospheric
pressure, which was recorded by the Taiwan Power Company, were input
to QUAL2E to simulate the environment for algal growth.
Because a significant stratification occurred in the Te-Chi
Reservoir during the summer, only about a 10-m depth layer of water
from the surface was embraced in the one-dimensional dynamic
simulations. The water body was divided into 93 computational
elements, which took a lateral average of all water quality variables.
All 93 elements belong to one of 22 reaches in which the computational
elements have the same biological rates and the same hydrogeometric
properties, such as stream slope, cross section, flow, and velocity.
Each element has a length of about 100 m (5 pixels on the SPOT image)
and a cross width the same as the river width ranging from 80 m to 600
m. Each element was identified as a specific type, depending on its
position and function in the model. The upstream section of the Ta-
Chia River located near the Shuon-Mou Creek was treated as a headwater
source element, because of its major contribution of flow. The rest of
the seven tributaries were treated as point sources, which are input
elements in the model. The attributes of each reach, such as cross
width, surface area, and constants in hydraulic functions were
tabulated in IMAGINE. A six-day water quality forecast resulted from
the simulation by giving a steady pollution source with a chlorophyll
a of 340 mg/L and phosphorus of 35 mg/L, which was the water quality
condition at the head water derived from the SPOT image.
These simulated results were converted to images and displayed in
IMAGINE. The images were smoothed by a 7x7 low pass filter to remove
the sharp difference between elements, then were overlaid with a land-
only background layer. Consequently, geo-registered thematic images
were created for water quality variables. The final product
from this study also includes a movie-like sequence of images.
The remote sensing-derived water quality thematic layers were
overlaid with a gray-scale SPOT image as a geographic background.
Figure 3 shows that high chlorophyll concentrations
appeared in the upper stream reaches and reduced in the lower reaches of Te-Chi
Reservoir. This was caused by a large amount of nutrients released
from the watershed having an intense farming activity. In
Figure 4,
Secchi depth had an inverse distribution with chlorophyll a. Same as
our understanding, a high concentration of nutrient is accompanied by
a low Secchi depth (or high turbidity) and algae bloom. The eutrophic
state of the reservoir is clearly illustrated by these thematic maps.
Figure 3. -- Satellite-derived thematic map of chlorophyll distribution
at the Te-Chi Reservoir on August 31, 1994.
Figure 4. -- Satellite-derived thematic map of Secchi depth distribution
at the Te-Chi Reservoir on August 31, 1994.
Furthermore, the water quality model QUAL2E was linked with the
monitoring system as a water quality forecasting system. The system is
a "linked-model" due to the complexity of numerical computations in
water quality modeling. Water quality variables were transferred from
IMAGINE to QUAL2E as initial conditions, and the simulated results
were transferred back for display in IMAGINE. Every single water
quality variable was presented in an individual thematic layer and in
color (in gray tone in this paper) to correspond with varying
different levels. Figure 5 shows the algae variation during the six
days following August 31, 1994. In the fourth and
sixth-day images, it
was discovered that the algae bloom was extending to the down stream
reaches under a steady pollution source. Moreover, these images can be
displayed as movie-like dynamic pictures by time sequence. The new
visualization technology helps users to perceive the progression of
water quality easily and impressively.
Harding, L. W., E. C. Itsweire, and W. E. Esalas, 1995. Algorithm
Development for Recovering Chlorophyll Concentrations in the
Chesapeake Bay Using Aircraft Remote Sensing, 1989-91.
Photogrammetric Engineering & Remote Sensing, 61(2):177-185.
Kuo, J. T., J. H. Wu, and W. S. Chu, 1994. Water Quality Simulation of
Te-Chi Reservoir Using Two-dimensional Models. Wat. Sci. Tech
30(2):63-72.
Kuo, J. T., and J. H. Wu, 1991. A Nutrient Model for a Lake with Time-
variable Volumes. Wat. Sci. Tech 24(6):133-139.
Lillesand, T. M., W. L. Johnson, R. L. Deuell, O. M. Lindstrom, and D.
E. Meisner, 1983. Use of Landsat Data to Predict the Trophic
State of Minnesota Lakes. Photogrammetric Engineering & Remote
Sensing, 49(2):219-229.
Lathrop, R. G., and T. M. Lillesand, 1989. Monitoring Water Quality
and River Plume Transport in Green Bay, Lake Michigan with SPOT-1
Imagery. Photogrammetric Engineering & Remote Sensing, 55(3):349-
354.
Lung, W. S., 1986. Assessing Phosphorus Control in the James River
Basin. Journal of Environmental Engineering, 112(1):44-60.
Ritchie, J. C., and C. M. Cooper, 1991. An Algorithm for Estimation
Surface Suspended Sediment Concentrations with Landsat MSS
Digital Data. Water Resources Bulletin, 27(3):373-379.
Robinson, I. S., 1985. Satellite Oceanography: An Introduction for
Oceanographers and Remote-sensing Scientists. Ellis Horwood
Limited, Chichester, England, 455pp.
Poiani, K. A. and B. L. Bedford, 1995. GIS-based Nonpoint Source
Pollution Modeling: Considerations for Wetlands, Journal of Soil
and Water Conservation, 49(6):613-619.
Rundquist, D. C., L. Han, J. F. Schalles, and J. S. Peake, 1996.
Remote Measurement of Algal Chlorophyll in Surface Waters: The
Case for the First Derivative of Reflectance Near 690 nm,
Photogrammetric Engineering & Remote Sensing, 62(2):195-200.
Ruiz-Azuara, P., 1995. Multitemporal Analysis of "Simultaneous"
Landsat Imagery (MSS and TM) for Monitoring Primary Production in
a Small Tropical Coastal Lagoon. Photogrammetric Engineering &
Remote Sensing, 61(2):877-198.
Verdin, J. P., 1985. Monitoring Water Quality Conditions in a Large
Western Reservoir whit Landsat Imagery. Photogrammetric
Engineering & Remote Sensing, 51(3):343-353.
Thomann, R. V. and J. A. Mueller, 1987. Principles of Surface Water
Quality Modeling and Control. Harper & Row, Publishers, Inc., New
York, 644 pp.
2.
Carolyn J. Merry
3.
Robert M. Sykes
Methodology
A GIS provides a suitable environment to incorporate water
quality models with remote sensing systems. Two alternatives
integrating GIS with other systems are the "linked-model" approach and
the "modeling-within" approach (Poiani and Bedford 1995). In a linked-
model approach, water quality models are separate from GIS, but the
models can fully stretch their simulation ability by using this
approach. A modeling-within approach, which is simpler, builds a model
within the GIS, but the model is confined by the limitations of the
GIS numerical capabilities. In this study, a modeling-within model as
a water quality monitoring system was established in ERDAS IMAGINE to
generate images of the distribution of chlorophyll, Secchi depth, and
phosphorus derived from SPOT data. Also, the linked-model approach was
used in the water quality forecasting system due to the complexity of
water quality simulation.
Water Quality Monitoring System
A water monitoring model was established to extract the water
bodies from SPOT images, calculate band ratios, carry out regression
models, and display chlorophyll, Secchi depth, and phosphorus in a
geographic map. This "modeling-within" model was conducted by the
spatial modeler module in ERDAS IMAGINE.
Water Quality Forecasting System
In water quality models, a water body is subdivided into many
finite segments. All variables are represented as averaged values
within a cell, much like a pixel of digital imagery. Chlorophyll a is
a representative variable proportional to the concentration of
phytoplankton biomass. For each water cell, a mass balance equation
for the chlorophyll a is calculated for the changes of inflow,
outflow, growth, death, and settling. Because of the complexity and
detail requirement of water quality models, a stream numerical model
QUAL2E was used to simulate water quality conditions as a "linked-model"
and linked with IMAGINE.
Results and Discussion
To convert digital values of the SPOT images to recognizable
water quality variables, a regression model was developed in the water
quality monitoring system. The relationships between satellite digital
values and water quality variables were studied to reveal the best fit
of band ratios to the optical characteristics of a water body's
constituents. Concordant with the theoretical deduction, the
correlation matrix showed that band ratios and water quality indices
to eutrophic status are highly correlated, even though a high bias
could occur due to the small sample size in this study. A ratio of
XS3/XS2 was adopted for chlorophyll and XS1/XS2 was selected for
Secchi depth and phosphorus. A time discrepancy between satellite
detection and concurrent surface reference sampling can cause a small
error. This problem cannot be completely avoided because in situ
sampling is time-consuming, particularly for a large water body.
However, the Te-Chi Reservoir is a slow-moving river (ranging from
0.01 to 0.005 m/s) and a stable meteorological condition occurred
between image acquisition and water quality sampling, so bio-chemical
reactions could have only a slight variation within 24 hours.
Besides the regression model, other GIS operations, such as
feature extraction, band ratio calculation, and image creation were
developed into an automatic water quality monitoring system in
IMAGINE. This automatic system was executed on an Silicon Graphics
work station and took only 1 minute and 45 seconds, compared with
several days of laboratory work for traditional water quality
monitoring approach. In the future, as long as the image can be
acquired from satellite transmission, the water quality condition
could be instantaneously reported by the system.
Figure 5a. -- The scond-day forecast of algal distribution
Figure 5b. -- The fourth-day forecast of algal distribution
Figure 5c. -- The sixth-day forecast of algal distribution
Figure 5. -- Sequential pictures of simulated algal distribution
at the Te-Chi Reservoir since August 31, 1994.
Conclusion
Conventional field investigations of water quality takes a long
time, particularly for pollution of non-point sources over a large
area. Therefore, comprehensive field sampling is used very
infrequently. To fill the gap when no samples are taken and to
maintain a periodic real-time water quality forecast, SPOT satellilte imagery
was shown to be an efficient tool in the study. Moreover, by integrating
GIS with the combined system, the simulated results were presented in
two-dimensional thematic maps and in dynamic sequential images. The
visualizing technique is helpful for rapid understanding of water
quality conditions. Since a complete model has been built, it is low-cost
to quickly answer "what-if" questions when meteorological
condition or pollution sources change.
References
Acknowledgements
We appreciate the suggestions for water quality modeling at the
Te-Chi Reservoir from Dr. J. T. Kuo in the Department of Civil
Engineering, National Taiwan University. Purchase of the SPOT imagery
was made possible by the Department of Civil and Environmental
Engineering and Geodetic Science and the College of Engineering, The
Ohio State University.
Authors
1.
Ming-Der Yang
Graduate student in the Department of Civil and Environmental Engineering and
Geodetic Science
The Ohio State University
Room 416, Hitchcock Hall, 2070 Neil Avenue
Columbus, OH 43210
Email:
myang@magnus.acs.ohio-state.edu
Proffesor in the Department of Civil and Environmental Engineering and
Geodetic Science
The Ohio State University
Room 416, Hitchcock Hall, 2070 Neil Avenue
Columbus, OH 43210
Email: merry@cfm.ohio-state.edu
Proffesor in the Department of Civil and Environmental Engineering and
Geodetic Science
The Ohio State University
Room 416, Hitchcock Hall, 2070 Neil Avenue
Columbus, OH 43210
home page
Maintainer: AWRA Webserver Team
Last modified: 24Nov99 gaw
Copyright(c) 1996, American Water Resources Association
