Scholarly article on topic 'Greening the Urban Environment Using Geospatial Techniques, A Case Study of Bangkok, Thailand'

Greening the Urban Environment Using Geospatial Techniques, A Case Study of Bangkok, Thailand Academic research paper on "Earth and related environmental sciences"

CC BY-NC-ND
0
0
Share paper
Keywords
{"Land Change Modeller" / "Ecological Sustainibility" / LULC / "Remote Sensing" / "Urban Sprawl" / "Quality of Life"}

Abstract of research paper on Earth and related environmental sciences, author of scientific article — Nargis Kamal, Muhammad Imran, Nitin Kumar Tripati

Abstract Urbanization is one of human induced activities causing land use changes. In recent years, various land usetransformationsin Bangkok influenced the city's ecological sustainability in all means, i.e., diminishing the city's cultivated land and greenery. This study investigates lack of green spaces due to extreme urban growth in the mega city. To do so, first, land use transitions are modelled through two different images; one from Landsat 5Thematic Mapper for the year 1994, and second fromHJ-1A CCD for the year 2012. Next, theMulti-Layer Perceptron Markov Model (MLP-Markov) is applied to predict land usechange for the year 2030. The MLP neural network is trained to modelland usetransitionsthrough creating transition maps. Markov Chain predictive model is applied with sufficient accuracy to process the transition maps for the prediction process. The results indicate that 348km2of green areas are transformed into built-up areas for the period 1994-2012,witha considerable loss of greenery (42%). The MLP model predictions show 4% increase in built-up and 6% decrease in greenery for the period 2012-2030. The study highly recommends conservation of green spaces and green corridors in the city. Future research can includeanalysing greenplot ratio for suitablegreen patches in vulnerable sites. The research output will benefit urban planners to implement long term planning strategies forsecuring natural environment in mega cities.

Academic research paper on topic "Greening the Urban Environment Using Geospatial Techniques, A Case Study of Bangkok, Thailand"

CrossMark

Available online at www.sciencedirect.com

ScienceDirect

Procedia Environmental Sciences 37 (2017) 141 - 152

International Conference - Green Urbanism, GU 2016

Greening the Urban Environment Using Geospatial Techniques,

A Case Study ofBangkok, Thailand Nargis Kamala*' Muhammad Imranb,Nitin Kumar Tripatic

aDepartment of Geography, University of Balochistan, Quetta, 83700, Pakistan, bCenter for Geographical Information Systems, University of the Punjab, Lahore, Pakistan, Asian Institute of Technology, P.O. Box 4, Klong Luang, Pathumthani 12120, Thailand c

Abstract

Urbanization is one of human induced activities causing land use changes. In recent years, various land use transformations in Bangkok influenced the city's ecological sustainability in all means, i.e., diminishing the city's cultivated land and greenery. This study investigates lack of green spaces due to extreme urban growth in the mega city. To do so, first, land use transitions are modelled through two different images; one from Landsat 5 Thematic Mapper for the year 1994, and second from HJ-1A CCD for the year 2012. Next, the Multi-Layer Perceptron Markov Model (MLP-Markov) is applied to predict land use change for the year 2030. The MLP neural network is trained to model land use transitions through creating transition maps. Markov Chain predictive model is applied with sufficient accuracy to process the transition maps for the prediction process. The results indicate that 348km2 of green areas are transformed into built-up areas for the period 1994-2012, with a considerable loss of greenery (42%). The MLP model predictions show 4% increase in built-up and 6% decrease in greenery for the period 2012-2030. The study highly recommends conservation of green spaces and green corridors in the city. Future research can include analysing green plot ratio for suitable green patches in vulnerable sites. The research output will benefit urban planners to implement long term planning strategies for securing natural environment in mega cities.

©2017 The Authors.PublishedbyElsevierB.V. Thisis an open access article under the CC BY-NC-ND license

(http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the organizing committee of GU 2016

* Corresponding author. Nargis Kamal. Tel.: 92-3342461808; fax:081-9211-277. E-mail address: nargiskamal786@gmail.com

1878-0296 © 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license

(http://creativecommons.org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the organizing committee of GU 2016

doi:10.1016/j.proenv.2017.03.030

Keywords: Land Change Modeller; Ecological Sustainibility; LULC; Remote Sensing; Urban Sprawl; Quality ofLife.

1. Introduction

Urbanization is one of major problems in all over the world. It is estimated that over 65% populations will be residing in urban areas in the year 2025 [1]. Moreover, there is a direct relationship between environment and land use change induced by human beings, e.g., built-up lands often occupy natural vegetations [2]. Urban sprawl is generally associated with the construction of new buildings, route planning, designing development plans around the suburb and central business districts, and emerging new towns around suburbs. Moreover, a poor land use planning influences the urban ecology. For instance, heat islands and air pollution within metropolitan cities are often associated with urban sprawl [3].

Rapid urbanization often influences a common's well-being and quality of life in metropolises. Among many factors, the common's well-being in urban areas is explicitly quantified with green or open spaces. Green infrastructure being unique among urban services expands the commons well-being through ecological sustainability. However, urban expansion often causes demise of existing vegetation around the cities, while natural greens rises mostly nearby the city centers, and suburb. To this end, a term urban public green spaces (UPGS) is used for facilities like various types of parks including, gardens, private garden and backyards, community gardens, street trees, unplanned vegetation, rock walls resting places, recreational areas providing active & passive relaxation, green roofs, vacant lots, school grounds, channel of storm water, rock walls, walking and cycling routes, shared vacant places around the apartments, riparian areas sporting fields. Such UPGS may overcome the challenges related to environmental sustainability in big cities [4, 5]. Nevertheless, many cities are scrambling for protecting the existing greenery, for instance, flood mitigation, food provision, cooling urban temperatures, suppressing noise and dust pollution, protecting urban wildlife, decreasing stress and nervousness [4].

To avoid reduction in greenery and to attain a sustainable environment, frequent assessment in land transformations in mega cities has become immense important. Trend assessment can be quantified through different models of spatiotemporal land use changes in cities [3, 4], for example, Cellular-Automata (CA) - Markov of land use change and predictive model have been designed to protect the environmentally sensitive areas. [6, 7]. Moreover, by applying CA the model can simulate urban growth at both local and regional levels, which is the most precious application for urban planners in building policies for future actions. Applying such models however is challenging, as it requires projecting complex and inter-connected processes in space and time [8].

The main objective of this study is to quantify the urbanized part of the city of Bangkok resulting from vegetation and cultivated lands. To do so, the study assesses the greenery losses irrespective of uncontrolled urban sprawl in Spatio-temporal domain. The results can be used to monitor greenery in adaptation patterns by long-term planning strategies at the local and regional level.

1.1. Study Area

Bangkok (Krung Thepmahanakorn) lies at East-West of the river Chao Phraya. People reside nearby the river along canals. Due to tourism and trading canals, the city is also famous as Venice of the East. Over time, development in road ways declined the waterways. These roadways along riverside became economic centers that widen the Southern part of the city. Migrants are another cause of abrupt population growth in Bangkok (i.e., 16-fold between 1944 and 2002) [10]. Trade and tourism activities in the city attracted people from all around the world for better opportunities ofjob, heath, education etc. Consequently, the city undergone through tremendous changes, particularly, increases in built-up areas caused decline in open green spaces that further affected the natural environment [12]. Built-up areas mostly emerged from the construction of new transportation networks, flyovers, link roads, air ports, and railways around small urban centers. The city of Bangkok is therefore a perfect case scenario to quantify the land use change

Figure 1: Bangkok and vicinity, (a) Bangkok and other neighbouring five provinces, (b) Bangkok

Province, (c) Thailand

that diminish the green environment, to understand the dynamics of land use change, and to predict the future land use change for developing better plans for quality life in mega cities. Since the city is exponentially expanding in space, therefore its five neighboring provinces (i.e. Pathumthani, Samutprakarn, Nakonpathom, Nonthaburi and Samutsakorn) were also included in the analysis (Figure 1).

2. MATERIALS AND METHODS

2.1 DATA

To detect land use changes, two remote sensing (RS) of spatial resolution 30m were acquired; one from Landsat 5 Thematic Mapper for the year 1994, and second from HJ-1A CCD for the year 2012. Both images were georeferenced and corrected for geometric and atmospheric errors. To minimize the seasonal effect on greenery, both images were

acquired for the same month of October. GIS vector data (shapes files) were collected from the department of Public Works and Town & Country Planning (BMA) that provides free of cost spatial data for the entire country.

2.2 METHODS

Figure 2: Methodological framework ofthe study

Two major steps were carried out in the framework shown in Figure 2: 2.2.1 Image classification

Vegetation indices convert spectral bands into digital data. Such data are used to characterize the earth features, monitoring dynamic variation of biophysical parameters in remote sensing [13]. Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) offer comprehensive vision into biotic events [14]. NDVI depicts absorption of the red light energy by chlorophyll and trickling of the near-infrared energy by plants and, therefore can be formulated as [14],

NDVI = (NIR - RED) + (NIR + RED) (1)

where RED and NIR denote measurements for spectral reflectance in red and near infrared bands, respectively.

NDVI was used to distinguish vegetated from non-vegetated cells in image data. Theoretically, NDVI values in image pixels range from -1 to +1. Image pixels representing vegetation have NDVI values often ranging from 0.5 to 0.8 or higher, while NDVI values close to zero represent non-vegetated area. Classification of RS images groups cells to a relatively small set of Land cover categories [15]. Supervised classification technique based on maximum likelihood was used to classify RS images of Bangkok and vicinity into five classes, including built-up land, green areas, cultivated land, water bodies and barren land (See Table 1). In total 2063 km2 area of Bangkok were observed in the both concerned RS images of years 1994 and 2012. For ground truthing, the field points were gathered by means of Google Earth high resolution images of 2m. In total, 22 samples of each land feature were gathered for the recent RS image of2012. Kappa coefficient was calculated to assess accuracy ofland features.

Table 1: Descriptions ofland use classes used in image classification

Land Description

Cover Type

Built-up All residential, commercial, industrial areas, villages, settlements and transportation infrastructure

Green Area Trees, shrub lands and semi natural vegetation: deciduous, coniferous, and mixed forest, palms, orchard, herbs, climbers, gardens, inner-city recreational areas, parks and playgrounds, grassland and small property vegetable lands, Urban green space that includes contiguous vegetated areas like parks or forest stands, and isolated trees in street medians, trees growing along streets, minor fields around the built-up areas.

Cultivated Major rectangular cultivated fields,

Water Body River, permanent open water, lakes, ponds, canals and reservoirs

Barren land Inbuilt and uncultivated open dry land, play grounds, dry fields, inner-city open land across transportation network and water bodies, etc.

2.2.2 Land Cover Change Detection

Artificial Neural Networks (ANNs) work with input hidden nodes and output nodes forming a network of functions [16, 17]. Multi-Layer Perceptron Markov Model (MLP-Markov) models the space-time changes of many layers at once with a feed forward ANN through input, hidden and output layers [16, 18]. ANN applies forward and back propagation using data from training sites [16,18]. Nodes for input layers work as a feature vector of various elements, including image texture, zoning, elevation, slope, wavebands of datasets or ancillary data like soil type, etc. The hidden layer contains layer functions and unit weights. Hidden layer weights are multiplied by input neurons to generate a joint value [16], as

net] — Za>jiOi , (2) (weight of single node as input) where wij are the weights between ith and jth nodes and Oi is the input from the node i,

Oi — f(netj), (3) (resultofoutputfrom nodej)

where f is a non-linear sigmoid function that is applied to the weighted sum of inputs before the signal passes to the next layer.

Next to forward propagation, the output node activities are compared with expected ones. Generally, there is a difference or error between the network output and desired outputs, for example, unknown classes in training data sets [16, 18]. Through network the difference is again back propagated and forward until the error gets minimized during iterations [16, 17]. Thus, the connection between weights are corrected through the following formula,

Au>ji(n + 1) = ri(8j0i) + aAoiji(n) (4)

where "q is the learning rate parameter, a is the momentum parameter, Sj is an index of the rate of change of the error.

The main idea behind the training network is to get accurate weights among the layers in classifying unknown pixels [16, 18]. It is important to estimate correct number of nodes, training samples and iterations for generalizing a neural network [16, 17]. Number of nodes can be estimated as follows,

Nh = INT (jNi * N0 ) (5)

where Nh is the number of hidden nodes, Ni is the number of input nodes, and No is the number of output nodes.

The MLP model was applied to detect land use changes in the study area from the year 1994 to the year 2030. The model was applied in two steps. In the first step, overall change was detected within the past land covers. The MLP model was calibrated and simulated to measure changes in built-up areas that were resulted at the cost of water bodies, vegetation, low and fallow lands. To do so, transformation from all land use types were individually considered in the simulation. Second, the transformed areas were considered around the built-up regions [16]. Finally, a group of explanatory variables was created as a driving force of change from all land use to built-up, including distances from water bodies, cultivated land, greenery and barren land. Each variable is tested for potential explanatory power through Cramer's V [20]. Lower V values (i.e. lower than 0.15) were discarded, whereas high V values (i.e. 0.4 or higher) were considered significant. The MLP neural network was trained from 10,000 iterations [16].

In the second step, predictions were made for likelihood of changing all types of land use into built-up. Whereas, the empirical change likelihood incorporated the categorical variables during the change analysis. Moreover, it determined the relative frequency of different land use types occurred in the transition from year 1994 to year 2012. Higher values of likelihood would result higher number of pixels changed into built-up area.

The training accuracy is often affected from the quantity of training samples used in the MLP model. Selecting effective sample size is important in this regard, as a small sample size may not represent the situation while a large number of samples may cause overlapping within each land use category. Similarly, over training of the model (i.e. too many iterations) may lead to a poor generalization of the Neural Network. [16, 18, 19]. Overtraining of the model may cause error by initial discontinuing of training. In MLP neural- networks, logistic regressions are used to model transitions from other land use to built-up. In doing so, the model was calibrated using Root Mean Square Error (RMSE) equal to 0.2033 in the training phase and RMSE equal to 0.2193 in the resting phase through applying 1000 iterations. Whereas, the RMSE was estimated as,

RMS = S^2/n = ^ ~ a% (6)

where N is the number of elements, i is the index for elements, e; is the error of the ith element, t; is the target value (measured) for ith element, and a; is the calculated value for the ith element.

3- RESULTS AND DISCUSSIONS

Figure 3(a, b) shows the classified images for the years 1994 and 2012, respectively. The classified image for the year 1992 indicates 32% of total land ofBangkok and its vicinity areas (i.e. total 2063km2' is covered by built-up which is in equal to 652km2. Whereas, the greenery, cultivated land, water and barren land cover 58% (1198km2', 7% (145km2), 2% (47km2) and 1% (21km2), respectively. The classified image for the year 2012 (Figure 3b) shows a significant change in land use. The built-up is increased with 208km2 followed by the losses in greenery about 294km2, the cultivated land is increased by 97km2 area with a decrease of 4km2 in water areas, whereas a decline of 7km2 is observed in barren lands (See Figure 4).

Figure 3: Classified images ofBangkok and vicinity for the year 1994 (a), and the year 2012 (b)

Total extent of transformation encompasses 2063km2 of Bangkok province and other neighboring provinces. The major built-up is increased within the provinces of Pathumthani, Nonthaburi, Samut Sakhon, and Samut Prakan. The maximum transformation, after the Bangkok province, is realized in the provinces of Samut Sakhon and Samot Prakan. Overall 825km2 of area within the vicinity is changed, which is double the change within the Bangkok city. This indicates that the city is spreading and occupying its territory rapidly; thus, creating new sub-centers in the vicinity areas. Moreover, during last 18 years, green areas of 348km2 are transformed into built-up areas with a considerable loss of greenery (42%).

Overall 81% accuracy in classifying RS images is assessed through confusion matrices. Some misclassification between fallow fields and bare lands is seen due to off-seasonal cultivation. As cultivated land class (see Table 1) considers only major cultivations around the area, greenery is seen with higher producer's and user's accuracy i.e. 86 % and 84 %, respectively.

Area in km2

Barren land Water

Cultivated land Greenery Built up -400

0 200 400 600 800 1000 1200 1400

■ AreaDifference «2012 I 1994

Figure 4: Land use changes from the year 1994 to the year 2012

Figure 5: Distance maps to built-up, cultivated land, greenery, water, and bare land

Figure 5 shows distance maps calculated for various explanatory variables as a driving force for change from all land use to built-up. Overall V distance (i.e. Cramer's V matrix) from all land use to built-up is 0.30, with distances from water, cultivated land, greenery, roads, empirical likelihood, barren land equal to 0.24, 0.14,0.31,0.17, 0.29, and 0.26, respectively. The values of each variable are significant according to Crammer's threshold value of 0.15-0.4. Smooth and satisfactory curve is found with all limitation. The accuracy rate of 70% is observed during calibrating the MLP-model.

Table 2 Transition probabilities grid for Markov chain (1994-2012)

Built-up Greenery Cultivated land Water Barren land

Built-up 0.6477 0.2925 0.0430 0.0116 0.0051

Greenery 0.3542 0.4863 0.1409 0.0107 0.0079

Cultivated land 0.2511 0.3557 0.3603 0.027 0.0058

Water 0.2623 0.2153 0.1645 0.3546 0.0033

Barren land 0.3726 0.4623 0.1447 0.0128 0.0076

Table 2 shows the transition probabilities matrix for Markov chain, i.e. based on probabilities for mutual transitions ofland use from year 1994 to year 2012. These values are used as fuzzy set membership degree (i.e. Oto 1 as sigmoid activation function) for the corresponding Land use category. The matrix indicates that urbanization is increasing and replacing the open, vegetation and cultivated lands, water bodies, and particularly greenery in the study area. Leftovers in the Western areas and Southern edges of Bangkok are emerged from the cultivated land; even some urbanized areas cover the cultivated lands of year 2012.

Table 3 shows differences in various land use areas from year 1994 to year 2012. This indicates a significant increase (10%) in built-up and a decrease in greenery (14%) during the period 1994-2012. Figure 8 shows the MLP predictions of land use change from year 1992 to year 2030. These predictions indicate 4% increase in the built-up and 6% decrease in greenery for the period 2012-2030 shown in table 3. This predicted decline in greenery is observed slow, compared to the observed for the period 1994-2012 (i.e. 405km2 predicted Vs 294km2 observed) (see Figures 8 and Table 3).

The results indicate that the change is a slow and continuous process in all scenarios, excluding the natural hazards or noticeable replacements of any land features. Comparing multi-date RS images is useful in detecting land use changes to built-up in Mega cities like Bangkok. The projected land use change can help implement short or long term developmental plans, whereas the observed land use change can indicate failure of various planning resulting from both in design or in implementations. Based on the GIS data obtained from BMA, the observed land use change may be interpreted as human induced change, resulting from population growth, urbanization, development of new suburban hubs, and tourism.

Figure 6: Projected landcover of Bangkok and vicinity for 2030

Table 3: Land use statistics predicted from year 1992 to year 2030

Percentage

Classes Area km2 1994 Percentage 1994 Area km2 2012 Percentage 2012 Difference 1994-2012 Area km2 2030 Percentage 2030 Percentage 2012-2030

Built up 652 32 860 42 10 954 46 4

Greenery 1198 58 904 44 -14 793 38 -6

Cultivated land 145 7 242 12 5 261 13 1

Water 47 2 43 2 0 43 2 -1

Barren land 21 1 14 1 0 13 1 0

Barren land

Cultivated land

Greenery

Built up

-200 0

Figure 7: Projected area differences in various land use from year 2012to year 2030

Barren land Water

Cultivated land Greenery Built up

-600 -400 -200 0

Figure 8: Projected area differences in various land use from year 1994 to year 2030

Future studies could include environmental data to analyze the impact of extreme climatic events on land use change both in space and time, for example, catastrophes, floods, earthquake, tsunamis, and, cyclones. Conversely, these extreme climatic events may also be due to land use change like deforestation, consumption of cultivated land, wounding trees, herbs, and shrubs of extensive parts round the urban cores.

The study recommends designing sustainable infrastructures for the airports of Suvarnabhumi and Don Muang airports, and developing sustainable township plans, e.g. for secure and quick transport, for mitigating floods around Bangkok and neighboring provinces. Subsiding green infrastructure and assisting grey infrastructure will leave adverse influences on the development of sustainable township plans.

Conclusions and recommendations

This study, firstly, quantifies the land use change in Benkok, Thialand from year 1992 to year 2012 through using remote sensing and GIS data. Secondly, it investigates various factors of the observed land use change in the city. Finally, it applies the Multi-Layer Perceptron Markov Model (MLP-Markov) model to predict the land use change from year 2012 to year 2030 and highlights various upcoming risks of urban sprawl.

12012-2030 2030

600 2012

200 400 600

■ 1994-2030 2030 «1994

The results show a high rate of conversion of greenery into built-up land in the observed land use change from the RS data, compared to the predicted from MLP model (i.e.294km2 observed Vs 405km2 predicted). Major pockets of new built-up areas are observed at the suburb of the city, followed by buildups areas across the transportation routes. The prediction maps highlight a tremendous replacement of cultivated land into built-up. The new built-up areas may result from new constructions nearby Central business districts, at Western ages of transportation routes, and on North-South banks of the river Chao Phraya. This can further induce river meandering by 2030. These urban expansions will definitely cause a significant decrease in greenery.

The estimated transitions may compel policy makers for precise solutions to decreasing greenery in such industrial metro cities that are much attractive for tourists. Though urban areas consist of a minor portion of globe, their effect on global environment however is significant. To mitigate these effects for eco-city and to acquire quality of life, planning more green corridors in the study area (e.g. public parks across the roads and river) is highly recommended.

References

1. Schell, Lawrence M., and Stanley J. Ulijaszek. Urbanism, health and human biology in industrialised countries. Vol. 40. Cambridge University Press, 1999...

2. Fichera, Carmelo Riccardo, Giuseppe Modica, and Maurizio Pollino. "Land Cover classification and change-detection analysis using multitemporal remote sensed imagery and landscape metrics." European Journal ofRemote Sensing 45.1 (2012): 1-18.

3. Han, Yu, and Haifeng Jia. "Simulating the spatial dynamics of urban growth with an integrated modeling approach: A case study of Foshan, China."Ecological Modelling (2016).

4. Byrne, Jason, and Neil Sipe. "Green and open space planning for urban consolidation-A review of the literature and best practice." Issues Paper 11 (2010).

5. Harnik, Peter. Urban green: Innovative parks for resurgent cities. Island Press, 2012.

6. Mitsova, Diana, William Shuster, and Xinhao Wang. "A cellular automata model of land cover change to integrate urban growth with open space conservation." Landscape and Urban Planning 99.2 (2011): 141-153.

7. Barredo, José I., et al. "Modelling dynamic spatial processes: simulation of urban future scenarios through cellular automata." Landscape and urban planning 64.3 (2003): 145-160.

8. Ashraf, Ather, Muhammad Imran, and Anam Shahbaz. "Protecting Agricultural Land in Developing Countries: A Case Study from Lahore, Pakistan." International Journal ofAdvanced Remote Sensing and GIS 4.1 (2015): pp-1181.

9. Thaiutsa, Bunvong, et al. "Urban green space, street tree and heritage large tree assessment in Bangkok, Thailand." Urban Forestry & Urban Greening7.3 (2008): 219-229.

10. Angel, Shlomo, et al. "The dimensions of global urban expansion: Estimates and projections for all countries, 2000-2050." Progress in Planning 75.2 (2011): 53-107.

11. Taniguchi, M. "Anthropogenic effects on subsurface temperature in Bangkok." Climate of the Past Discussions 2.5 (2006): 831-846.

12. Zhou, X., & Wang, Y.-C. (2011). Spatial-temporal dynamics of urban green space in response to rapid urbanization and greening policies. Landscape and Urban Planning, 100(3), 268-277.

13. Lee, R., et al. "Evaluating vegetation phenological patterns in Inner Mongolia using NDVI time-series analysis." International Journal of Remote Sensing23.12 (2002): 2505-2512.

14. Running, S. W., et al. "Terrestrial remote sensing science and algorithms planned for EOS/MODIS." International journal of remote sensing 15.17 (1994): 3587-3620.

15. Mather, Paul, and Brandt Tso. Classification methods for remotely sensed data. CRC press, 2016.

16. Ahmed, Bayes, and Raquib Ahmed. "Modeling urban land cover growth dynamics using multi-temporal satellite images: A case study of dhaka, bangladesh." ISPRS International Journal ofGeo-Information 1.1 (2012): 3-31.

17. Atkinson, Peter M., and A. R. L. Tatnall. "Introduction neural networks in remote sensing." International Journal of remote sensing 18.4 (1997): 699-709.

18. Eastman, J. Ronald. "IDRISI Taiga guide to GIS and image processing."Clark Labs Clark University, Worcester, MA (2009).

19. Karul, C., and S. Soyupak. "A comparison between neural network based and multiple regression models for chlorophyll-a estimation." Ecological Informatics. Springer Berlin Heidelberg, 2003. 249-263

20. Cramér, H. Methods of estimation. In Mathematical Methods of Statistics, 19th ed.; Chapter 33; Princeton University Press: Princeton, NJ, USA, 1999; pp. 497-506