CA MARKOV MODELING OF LAND USE LAND COVER CHANGE PREDICTIONS AND EFFECT OF NUMERICAL ITERATIONS, IMAGE INTERVAL (TIME STEPS) ON PREDICTION RESULTS

In this study, land use land cover (LULC) map of 1987, 1997 and 2007 derived from digital satellite images of Landsat 5 TM of 1987, IRS-1C LISS III images of 1997 and IRS-P6 LISS III images of 2007, respectively, to monitor and predict future LULC scenario. Future LULC has been predicted using CA Markov Model for 2017, 2027 and 2050 by using LULC map of 1987 & 1997, 1997 & 2007 and 1987 & 2007. The period (image interval / time steps) between different predictions years (i.e., 2017, 2027 and 2050 using 1987 & 1997, 1997 & 2007 and 1987 & 2007 LULC) are 10, 20, 30, 43 and 53 years. The number of iteration was based on the time steps i.e., iterations 20 to predict LULC for 2017 (prediction from 1997 to 2017); iterations 30 for 2027 (prediction from 1997 to 2027); iterations 53 for 2050 (prediction from 1997 to 2050). The relationships (correlations) of quantity of predicted LULC found strong positive correlation for three time periods (0.981, 0.984, 0.966 for 2017, 0.981, 0.984, 0.975 for 2027, 0.977, 0.987, and 0.980 for 2050) and established that there are almost no effect in quantity of prediction results of different time steps (iterations) images and time intervals are used to predict future LULC. The predicting location of predicted LULC of 2017, 2027 & 2050 for the three cases showing positive correlation, where r are 0.728, 0.758 and 0.708 for 2017 when relatively less time steps used; r are 0.696, 0.761 and 0.674 for 2027 when medium time steps used; r are 0.599, 0.721, 0.574 for 2050 when more time steps used. The analysis established that although there are nearly no effect on quantitative prediction results but have small impact of iterations (time steps) and time intervals on spatial distribution of predicted LULC results. * Corresponding author


INTRODUCTION
CA (Cellular Automata) Markov is a combined of CA and Markov chain land use land cover (LULC) prediction procedure that adds an element of spatial contiguity as well as knowledge of the likely spatial distribution of transition to Markov change analysis. Markov analyses two qualitative land cover images from different dates and produces a transition matrix and CA adds the spatial distribution of the transition. Within CA Markov model, the transition areas file from a Markov Chain analysis of two prior LULC maps establishes the quantity of expected land use and land cover change from each existing category to each other category in the next time period. The basic LULC image (the later land cover image used in the Markov Chain analysis) is used as the starting point for change simulation. Suitability maps (here, evidence likelihood map) establish the inherent suitability of each pixel for each land cover type. However, a contiguity filter down-weights the suitability's of pixels far from existing areas of that class (as of that iteration), thus giving preference to contiguous suitable areas. The number of iterations chosen establishes the number of time steps that will be used in the simulation. Within each time step, each land cover is considered in turn as a host category. All other land cover classes act as claimant classes and compete for land (only within the host class) using the multi-objective land allocation (MOLA) procedure. The area requirements for each claimant class within each host are equal to the total established by the transition areas file divided by the number of iterations. The results of each MOLA operation are overlaid to produce a new LULC map at the end of each iteration. CA Markov automatically normalizes the filter kernel to force the values to sum equal to 1. This filter is passed over a Boolean image for each class from the current land cover image within each iteration. First, the transition areas file from a Markov chain analysis of two prior LULC maps establishes the quantity of expected land use land cover change (LULCC) from each existing category to each other category in the next time period, and then finally number of iterations chosen establishes the number of time steps that will be used in the simulation. In this study, the spatio-temporal CA Markov model of landscape change using multi-temporal satellite imagery has been used which enabled us to predict spatial pattern of future land use/land cover for the study area -Kamrup Metropolitan district of Assam state in India (Figure 1). For this purpose, land use/land cover maps of the study area have been extracted (digitally classified) from multi temporal satellite images i.e, Landsat MSS image of 1987, Landsat TM image of 1997 and IRS-P6 LISS III image of 2007 ( Figure 1 and Table 1). Land use land cover (LULC) maps derived from satellite images of 1987 and 1997 were used to predict future land use/land cover of 2007.

CALIBRATION OF CA MARKOV MODEL
In this study, future LULC has been predicted for 2017, 2027and 2050using 1987LULC image and 1997& 2007LULC image. LULC map of 1987 Table 2). Future LULC has been also predicted for 2017, 2027& 2050using 1987& 2007 image. The period between the first and second images is 10 years and the period to project forward from the second image is 20 years to predict LULC in 2017, 30 years to predict LULC in 2027 and 53 years to predict LULC in 2050 when using 1987 & 1997 LULC image. The number of time periods between the first and second images is 10 years and the number of time periods to project forward from the second image is 10 years to predict LULC in 2017, 20 years to predict LULC in 2027and 43 years to predict LULC in 2050when using 1997& 2007 LULC image to predict future LULC. The period between the first and second images is 20 years and the period to project forward from the second image is 10 years to predict LULC in 2017, 20 years to predict LULC in 2027and 43 years to predict LULC in 2050when using 1987& 2007

Validation of CA Markov LULC Prediction Results
The validation for CA Markov model land use/land cover prediction results calculated using various Kappa Index of Agreement (KIA or Kstandard) and related statistical variations on the KIA and showing Kno is 0.8347, Klocation is 0.859, Klocation Strata is 0.8591 and Kstandard is 0.7928.

RESULTS AND ANALYSIS
A comparative analysis and qualitative and quantitative assessment of all above LULC predictions was done. Relationships (correlations) between predicted quantities of all above LULC predictions have been established. Relationships (correlations) between predicted locations of all above LULC predictions have been also established.   Figure 3 shows a comparative graphical presentation of overall (total) area statistics of predicted LULC of 2017, 2027, 2050using 1987LULC image, using 1997& 2007LULC image and using 1987& 2007 image. Figure 9 also shows a comparative graphical presentation of overall (total) area statistics of the observed LULC of 1987, 1997 and 2007 as well as predicted LULC of 2017, 2027, 2050using 1987LULC image, using 1997& 2007LULC image and using 1987& 2007 image.  Figure 4).  Figure 5).

Correlations between area of Predicted LULC of 2050 using 1987 & 1997 LULC Image, Area of Predicted LULC of 2050 using 1997 & 2007 LULC Image and Area of Predicted LULC of 2050 using 1987 & 2007 LULC Image
The correlation between area of predicted LULC of 2050 Figure 6).   1987LULC images, using 1997& 2007LULC images and using 1987& 2007 images. Figure 8 shown allocation of predicted LULC of 2027using 1987LULC images, using 1997& 2007LULC images and using 1987& 2007 images. Figure 9 shown allocation of predicted LULC of 2050using 1987LULC images, using 1997& 2007LULC images and using 1987& 2007LULC images. Figure 7. Predicted LULC of 2017using 1987LULC image, using 1997& 2007LULC image and using 1987& 2007LULC image Figure 8. Predicted LULC of 2027using 1987LULC image, using 1997& 2007LULC image and using 1987& 2007 The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLIII-B3-2020, 2020 XXIV ISPRS Congress (2020 edition) Figure 9. Predicted LULC of 2050using 1987LULC image, using 1997& 2007LULC image and using 1987& 2007  The area statistics of different predicted LULC of 2017, 2027& 2050(using 1987LULC, using 1997& 2007LULC and using 1987& 2007 shows that there are not much of differences. But the spatial differences between predicted LULC of 2017using 1987LULC, using 1997& 2007LULC and using 1987& 2007 LULC are found different. Regression analysis of three pairs of images (predicted LULC of 2017using 1987LULC images and using 1997& 2007predicted LULC of 2017using 1997& 2007LULC images and using 1987& 2007predicted LULC of 2017using 1987LULC images and using 1987& 2007   The spatial differences between predicted LULC of 2027using 1987LULC, using 1997& 2007LULC and using 1987& 2007 LULC are found different. Regression analysis of three pairs of images (predicted LULC of 2027using 1987LULC images and using 1997& 2007predicted LULC of 2027using 1997& 2007LULC images and using 1987& 2007predicted LULC of 2027using 1987LULC images and using 1987& 2007 Figure  11.a, Figure 11.b, and Figure 11.c). It is also established that spatially they have some differences. It is established that there have some effects in spatial distribution of predicted LULC of 2027 when different time steps (iterations) images are used to predict future LULC. Figure 11.a. Relationships between predicted LULC of 2027 using 1987LULC image and predicted LULC of 2027using 1997& 2007Figure 11.b: Relationships between predicted LULC of 2027using 1997& 2007LULC image and predicted LULC of 2027using 1987& 2007Figure 11.c: Relationships between predicted LULC of 2027using 1987LULC image and predicted LULC of 2027using 1987& 2007

Correlations between Location of Predicted LULC of 2050 using 1987 & 1997 LULC Image, Location of Predicted LULC of 2050 using 1997 & 2007 LULC Image and Location of Predicted LULC of 2050 using 1987 & 2007 LULC Image
The spatial differences between predicted LULC of 2050using 1987LULC, using 1997& 2007LULC and using 1987& 2007 LULC are found different. Regression analysis of three pairs of images (predicted LULC of 2050using 1987LULC images and using 1997& 2007predicted LULC of 2050using 1997& 2007LULC images and using 1987& 2007predicted LULC of 2050using 1987LULC images and using 1987& 2007 Figure 12.a, Figure 12.b, and Figure 12.c). It is also established that spatially they have some differences. It is established that there have some effects in spatial distribution of predicted LULC of 2050 when different time steps (iterations) images are used to predict future LULC.

CONCLUTIONS
This approach presented in this chapter examines if there is any effect in LULC prediction results when different time interval and time steps (iteration) are used. When the quantity of predicted LULC of 2017, 2017& 2050(using 1987LULC, using 1997& 2007LULC and using 1987& 2007 changes are analyzed, they are found to have also strong positive correlation for three time periods (i.e., 0.981, 0.984, 0.966 for 2017, 0.981, 0.984, 0.975 for 2027, 0.977, 0.987, and 0.980 for 2050). Firstly, these strong correlations established that there is almost no effect in quantity of prediction results different time steps (iterations) images and time intervals used to predict future LULC. Secondly, the predicting location of predicted LULC of 2017, 2027 & 2050 for the three cases showing positive correlation (0.728, 0.758 and 0.708 for 2017 -when relatively less time steps used; 0.696, 0.761 and 0.674 for 2027 -when using medium time steps used; 0.599, 0.721, 0.574 for 2050 -when using more time steps used). It proves that there have some effects in spatial distribution of predicted LULC when different time steps (iterations) images are used to predict future LULC. The results also indicate that relatively less time steps (iteration) produced spatially more accurate results. Whereas, more time steps (iteration) produce spatially less accurate results. The results also indicate less time intervals produced spatially more accurate results; whereas, more time intervals produce spatially less accurate results. The results also indicate that combination of less time steps (iteration) and less time intervals produced spatially most accurate results. Overall from the present study results, it is found that there is nearly no effect in quantity of prediction results when different time steps (iterations) and time intervals images are used to predict future LULC but there have some effect in spatial distribution of predicted LULC when we use different time steps (iterations) and time intervals images to predict future LULC. It is also established that this approach can be used examines if there is any effect in CA Markov LULC prediction results when different time interval and time steps (iteration) are used.