Mapping of Land Cover Change in Mato Grosso State, Brazil from Landsat Satellite Imagery

Posted: August 26th, 2021

Mapping of Land Cover Change in Mato Grosso State, Brazil from Landsat Satellite Imagery

Name of student

Institutional affiliation

Mapping of Land Cover Change in Mato Grosso State, Brazil from Landsat Satellite Imagery

Introduction

            Various regions around the world are currently going through rapid and wide-spreading changes in land cover. Generally, accurate monitoring of changes on the earth’s surface is paramount in building the understanding of the relationship that exists between man and nature. Brazil is among the tropical belt countries in which much of this activity is centered. These changes in land cover, especially the clearing of tropical rainforests within Brazil’s Mato Grosso State recently,has raisedconcerns among researchers and policymakers because of the potential effects they have on erosion, increased concentration of carbon dioxide in the atmosphere, increased run-off, and the loss of biodiversity (Pravalie, Sîrodoev & Peptenatu, 2014). For effective monitoring of these changes, therefore, the remote sensing technology has proven to be invaluable in the provision of data from which accurate and up to date land cover information can be extracted.

Due to the technology’s repetitive coverage at short intervals and consistent image quality, thus, change detection has become a significant application of remotely sensed GIS data on the land cover between 1990 and 2015, which is the focus of the paper.The basic principle in using remote sensing data for change detection is pegged on the fact that changes in land cover affect changes in the satellite’s radiance values. In turn, these changes in radiance values are usually caused by factors such as soil moisture differences, and differences in sun angles (Zhu & Liu, 2015). Other factors include vegetation diversity and land cover interspersion, both of which are high in the humid tropics that Brazil generally is. Therefore, identification of a robust change detection methodology is crucial in dealing with wide-ranging vegetation cover data in Brazil’s Mato Grosso State.

Methods and Study Area

The study area is Mato Grosso, which covers large sections of the Amazon state within Brazil. With a landmass of 906,000 square kilometers, the state of Mato Grosso is located in the southern part of the Legal Amazon. Although it has been occupied for a long time, its spatial inhabitation started in the 1930s. Between 1990 and 2015, Mato Grosso state has become a frontier in the sense that natural vegetation, which comprises of tropical rainforest and savanna, which is commonly referred to as Cerrado in the area, is being replaced with crop production. Mechanized farming had already begun in the region by themid-1990s. Major crops included the soybeans. Since then, crop production has been on the increase. This increase has largely been facilitated by cropland expansion and the transition from predominantly single cropping to double cropping. The state of Mato Grosso ratified the Soy Moratorium in July 2006 to reduce amazon deforestation for the sake of the production of soybeans. Geographically, the location of the Mato Grosso State in Brazil is as shown on the Google Earth map below.

Figure 1: Location of Mato Grosso State

Following the increase in deforestation of the tropical rainforest in the Mato Grosso State, several remote techniques have been discovered and applied in the production of land use maps. The three primary spatial-temporal dynamics monitored through these techniques include agricultural expansion, deforestation, and agricultural intensification (Rodrigues, Marcal, Furlan, Ballester & Cunha, 2013). For each of these dynamics, the analysis was used to produced GIS maps for 1990, 2005, and 2015 in order to analyze the land cover changes that have taken place in those years. The Landsat Images used in this study belong to the NALC triplicate dataset, and the EROS Digital Image Processing Center has geometrically and radiometrically corrected them.

In the course of processing, the 1990 map was restriped in order to compensate for the variations in the radiometric response of the individual detectors. By use of the cubic convolution, the maps were then corrected and resampled to a UTM output image that is comprised of 60m by 60 pixels whose root mean square error is less than 1 pixel. Two thousand forty 2040) columns by 1650 lines sample subsection corresponding to the Mato Grosso State and its surrounding was then extracted for 1990, 2005, and 2015 Landsat mappings. For digital image processing roles, the ERDAS Software was used on a SUN Workstation.

To analyze the extent of deforestation, the maps I used were produced through the INPE Technique basing on the Landsat images for the Brazilian Amazon biome under the PRODES Project frame. While focusing our study only on the Mato Grosso State cropped out areas covered by Pantanal biomes and regions that have been cleared in the Cerrado. Also, the State Secretary of Environment in Mato Grosso (SEMA-MT) produced deforestation maps basing on visual interpretation of CBERTS and Landsat images. Deforested areas were regarded to be so is at least one data source (SEMA-MT or PRODES) marked it as deforested in the tropical rainforest area, and when SEMA’s data source marked it as deforested in the Cerrado biome. Primarily, therefore, both data sources worked in a complementary way to each other in order to produce maps.

For change detection, the study made use of several methodologies to identify the environmental changes that have taken place in the Mato Grosso State for the years under consideration. For accuracy’s sake, three main change detection methods were used. These are image enhancement, multi-date data classification, and comparison of two independent land cover classifications. Closely related is the Vegetation Index Differencing technique, which relied on data depicting land cover changes that were shown to be relating to green biomass. Thus, the Normalized Difference Vegetation Index (NVDI) was calculated as follows:

NDVI = (NIR – RED)/ (NIR + RED)

Where, NIR – Near Infrared band response for a given pixel, MSS band 4

RED – Red response, MSS band 2

Also, the study made use of the selective principal components analysis (SPCA) technique in which only two bands of the MultiMate image were used as inputs instead of all the bands. Thus, the information that was common to both bands was mapped to the first component, whereas information that was unique to either of the two bands, which is the detected land cover change, was mapped to the second component. After that, selective standardized principal components analysis was performed using bands 2 and 4.

Results

For the radiometric normalization technique, the computation was made for the statistics of the mappings over the years under consideration. Of importance to note is the observation that in the 1990 mapping, there was more cloud cover and significant radiometric difference for the other years. Here,  the values for 1990 band 2 were regressed against 2005 band 2 using the least-squares regression method. The predicted values of 2005 and 2015 were obtained from the regression line and then were compared with the 1990 band 2 values by subtraction. If the difference was higher than the threshold value, the pixels are regarded as having changed, and thus, hence they were excluded from the histogram calculation. Say for a threshold value of 10, isolation of 5 percent of the pixels is allowed, which presented more paramount spectral variation in between the years under consideration. Then, it follows a histogram matching by making use of the histograms based only on the pixel values. After normalization, the changes in land cover for Mato Grosso state over the years under consideration came out clearly. These changes are illustrated in the maps shown below. The thematic map for Mato Grosso state for the year 1990 is shown first.

Figure 2: Thematic Map for 1990

            From the above map, it is clear that in 1990, rainforest cover in the Mato Grosso state was very dense. Bare land occupied only a small portion of the total landmass. With such extensive coverage of rainforest, the region received large amounts of rainfall, and in turn, rivers and other water bodies had sufficient amounts of water. Also, the region’s biodiversity had good standing. All these benefits, however, changed somewhere from the mid-1990s when mechanized farming was introduced in the state. From the study, the corresponding changes in land cover are shown in the GIS Remote map below.

Figure 3: Map for 2005

From the above map, it is apparent that the dense rainforests that had covered most parts of the Mato Grosso state in 1990 had significantly been reduced by the year 2005. Due to the intense production of soybeans in the region, most of the trees had been cut to allow for the creation of extensive farmlands. As such, the region’s biodiversity has significantly reduced over the years (Lillesand, Kiefer & Chipman, 2015). Equally, the Landsat imagery revealed a reduced cloud cover in 2005 when compared to the amount that was present in 1990. The reduction can be attributed to the reduced forest cover. It is a well-known fact that trees contribute immensely to the production of moisture and the waffling of winds, which in turn create clouds. Therefore, reduction in the tropical rainforest cover between 1990 and 2005 had a large hand in the reduced cloud cover that was captured by the Landsat mapping in 2005.

For 2015, the situation was even worse. The reason is that there has never been a reduction in the appetite for crop production in the Mato Grosso region, and as such, the tropical rainforest cover in the region has kept diminishing. Consequently, the area no longer enjoys the high rainfall amounts it used to receive in the early 1990s. Similarly, the amount of water in the region’s river has significantly reduced due to the interference with its hydrological cycle (Hasmadi, Pakhriazad & Shahrin, 2017). The 2015 mapping for this state also depicts an enlarged bare land. All these changes in the land cover as of 2015 are captured in the map shown below.

Figure 4: Highlight changein Mato Grosso’s land cover  as of 2015

Discussion

The unsupervised classification was conducted by the use of the eight bands of the multi-date images to categorize the mapped region into approximately 30 change clusters. As seen from the above maps in the results section, the resulting categorized image displayed the land cover changes that corresponded to each of the 30 clusters for each of the years under consideration. Thus, at level detection, the kappa coefficient was 0.3, with a global accuracy of 80 percent of the pixels that were correctly classified.

From each of the given maps, it is also clear that the highest level of accuracy was obtained, and the researcher made the post-classification comparison basing on the supervised classification of the three maps. The excellent performance of this technique is attributable to the high classification accuracy of the 1990 and 2005 maps. Additionally, the single band analysis of band 2 proved to be better than band 4 when it comes to the detection of land cover changes. The reason for the ineffectiveness of band 4 data lies in using the high infrared return from the herbaceous understorey of the cleared regions (Almeida at al., 2016). According to band 2, the SPCA technique offers better accuracy than the image differentiating procedure in the calculation of changes in the vegetation index changes. As such, it is apparent that after carrying out radiometric normalization, the SPCA Method removed the inter-images variations that are usually occasioned by the sensors and atmospheric conditions.

Conclusion

The Mato Gross is regarded as one of the last frontiers in the world. Both economic and political reasons drive the continued clearing of the tropical rainforest cover in the region. It is, therefore, necessary that these changes in land cover are effectively monitored. It is for this reason that GIS Remote sensing techniques have been developed to map the changing land cover between 1990 and 2015. The resulting maps have highlighted the rapidity of the expansion in the bare landmass, mainly due to the expansion of soybean production from the mid-1990s. Going forward, therefore, it is necessary to institute effective environmental governance that regulates expansion of the deforested area and, at the same time, promotes ecological intensification practices for Brazil’s Mato Grosso state.

References

Almeida, C. A. D., Coutinho, A. C, Esquerdo, J. C. D. M., Adami, M., Venturieri, A., Diniz, C. G., … & Gomes, A. R. (2016). High spatial resolution land use and land cover mapping of the Brazilian Legal Amazon in 2008 using Landsat-5/TM and MODIS data. Acta Amazonica, 46(3), 291-302.

Hasmadi, M., Pakhriazad, H. Z, & Shahrin, M. F. (2017). Evaluating supervised and unsupervised techniques for land cover mapping using remote sensing data. Geografia-Malaysian Journal of Society and Space, 5(1).

Lillesand, T. M., Kiefer, R. W., & Chipman, J. W. (2015). Remote sensing and image interpretation (6th Edition). John Wiley & Sons Ltd, Chichester

Pravalie, Sîrodoev, I., & Peptenatu, D. (2014), Detecting climate change effects on forest ecosystems in Southwestern Romania using Landsat TM NDVI data. Journal of Geographical Sciences, 24(5), 815-832.

Rodrigues, A., Marcal, A, R. S., Furlan, D., Ballester, M. V., & Cunha, M. (2013). Land cover map production for Brazilian Amazon using the NDVI SPOT VEGETATION time series. Canadian Journal of Remote Sensing, 39(4), 277-289.

Zhu, X., & Liu, D. (2015). Improving forest aboveground biomass estimation using seasonal Landsat NOVI time-series. ISPRS Journal of Photogrammetry and Remote Sensing, 102, 222-231.