<<

7th Int. Conf. on Data Science & SDGs EC - 022 December 18-19, 2019, pp 143-147  Dept. of Statistics, University of ,

Cropping Pattern mapping of - 1 Block using Machine Learning Technique Alamgir Shaikh1, Argha Ghosh2, Kajori Parial1 and Manoj K. Nanda2 1 State Council of Science and Technology Salt Lake, , West Bengal, 2Department of Agricultural Meteorology and Physics, Bidhan Chandra Krishi Viswavidyalaya, West Bengal, India

Abstract: With the global trend of increasing population, the demand for food is also expected to increase. The traditional way of increasing the crop production simply by adding up new lands for agricultural activities is no more feasible owing to the competition for land to facilitate other human activities. In this scenario, the alternative is to increase the produce from the same amount of or possibly from less amount of land through replacing monoculture farms with diversified farms. To increase the production through diversified agriculture, a proper crop planning is required. To carry out a proper crop planning, it is essential to know the existing cropping pattern. In this present study the cropping system of Suti-1 block of district has been assessed through multi dated satellite images. Spectral profiles of different cropping systems were generated to understand the conditions of the crops in this study area. Normalized Difference Vegetation Index (NDVI) was calculated for each and every cropping system to study the developmental cycle of the crops. Cropping systems could be successfully differentiated by studying the NDVI profile of the systems. The findings of the present study gave an overview of the cropping scenario of the study area. Keywords: Cropping pattern, Random forest, NDVI, Sustainable agriculture.

1. Introduction At present about 821 million people go to sleep hungry. With the global trend of increasing population, the demand for food is also expected to increase. Studies suggest that the world will need 70-100% more food by 2050 to feed the global population (Charles et al., 2010). However, the traditional way of increasing the productivity simply by adding up new lands for agricultural activities is no more feasible due to competition for land to facilitate other human activities like industries, urban facilities etc. On contrary, there is a decrease in existing agricultural lands owing to urbanization, other anthropogenic activities, desertification, soil erosion, water logging, unsustainable landuse, climate change etc. In this scenario, the alternative is to increase the produce from the same amount of, even possibly from less amount of, land. The solution to this problem may be achieved through replacing large monoculture farms with small and diversified farms. Researchers have observed that such small and diversified farms show greater productivity per area (Tscharntke et al., 2012) - a phenomenon known as ‘paradox of scale’ or the ‘inverse farm size-productivity relationship. To ensure food security such practice can be an effective solution. To increase the production through diversified agriculture, a proper crop planning is required. For a proper crop planning, it is essential to know the existing cropping pattern, proper information on natural resources at regional and global scale along with efficient management of natural resources like soils, vegetation, water etc. In this regard, satellite remote sensing has proved to be an important and reliable tool. Its capacity to provide precise information with high spatial as well temporal resolution covering a large area at once also makes it a strong contender amongst other conventional methods which are often field based and human labour intensive. A leap in application of remote sensing in the domain of agriculture could be noticed since the launch of European Space Agency’s (ESA) Copernicus programme. The Sentinel-2 data with a temporal resolution of 5 days is used globally for understanding the dynamics of cropping in field. In this paper, Suti-I block has been selected as the study area. The work was carried out during the period from October 2018 to April 2019 to assess the diversity of cropping pattern. The objectives of the study can be summed up as: a.to generate of spectral profile of different cropping systems from multi-dated Sentinnel-2 satellite images, b. to classify land cover from Sentinnel-2 images using supervised Random forest classification method and c. to determine area under different cropping system using multi satellite data.

Alamgir Shaikh, Argha Ghosh, Kajori Parial and Manoj K. Nanda

2. Study area Suti- I is a Community Development (CD) block that forms an administrative division in Jangipur of . Suti I CD Block lies in the in Murshidabad district. The Rarh region is undulating and contains mostly clay and lateritic clay based soil. The area experiences tropical climate with average annual rainfall of 1362 mm. The Bhagirathi River flows in the east of this block. Suti I CD Block is bounded by Suti II CD Block in the north, Chapai Nawabganj Sadar Upazila in Chapai Nawabganj District of Bangladesh, across the , in the east, I and Raghunathganj II CD Blocks in the south, and Maheshpur and Pakuria CD Blocks in of in the west. Figure 1 shows the map of the study area.

3. Methodology 3.1. Data used Sentinnel-2 data was collected from the data archive of ESA. The clear sky (<20% cloud cover) imageries acquired for the post-monsoon period starting from November to April had been used for the study. Table 1 provides the details of the images used for the study. A field survey was conducted to collect sample locations/fields to train the samples from Suti I block. The GPS based survey was done for different types of land use systems e.g., a) Tree + Settlement, b) Water body, c) Fallow, d) Fallow + settlement, e) Crop (paddy and pulses) and f) Vegetable.

7th Int. Conf., December 18-19, 2019, Dept. of Statistics, RU Page | 144

Alamgir Shaikh, Argha Ghosh, Kajori Parial and Manoj K. Nanda

3.2. Image processing and classification The uncorrected raw images (Level-1) of Sentinel-2 were pre-processed for atmospheric correction (haze removal) and radiometric correction using the in-built pre-processing menu in the Semi-Automatic Classification Plugin of Quantum GIS (QGIS 2.18). The geometric correction was done for the processed images with reference to Survey of India toposheet. The corrected images were used for generating spectral profile of different cropping systems with respect to Visible and Near Infrared (NIR) bands with reference to the ground truth survey coverage over the post-monsoon season.

3.3. Normalized Difference Vegetation Index Profiles of Normalized deviation vegetation index (NDVI) were computed for the selected fields from the reflectance in visible and NIR bands for different cropping systems over the post-monsoon period using the following empirical formula:

=

푅푁퐼푅− 푅푅 푁퐷푉퐼 푅푁퐼푅+ 푅푅 3.4. Random forest classification The classification was done using Random forest classification. It is an ensemble learning method for classification, regression and other tasks that operates by constructing a multitude of decision trees at training time and outputting the class that is the mode of the classification or mean prediction (regression) of the individual trees. In this study the classification was performed using the SAGA software.

4. Results and discussion The field survey in the Suti-1 block during the study period showed nine distinct cropping systems. They were a) Paddy – fallow - fallow (monocrop), b) Paddy (Kharif) –fallow - Paddy (Boro), c) Paddy – Pulses –fallow, d) Paddy – Mustard - fallow and e) Fallow - Pulses – fallow, f) Fallow – Mustard – fallow, g) Fallow – Vegetable – fallow, h) Vegetable – vegetable – fallow, i) Vegetable – fallow – vegetable.

4.1. Temporal NDVI pattern of different cropping system The comparison among these nine cropping system done on the basis of Normalized Difference Vegetation Index (NDVI) are given in Figure 2. The paddy – fallow – fallow system showed a high NDVI value (0.75) on October due to the presence of fully grown paddy in field. Towards the end of December, the NDVI value showed a decrease (0.26) as by then the paddy harvest was over. The paddy – pulse – fallow and paddy – mustard – fallow systems also recorded a similar pattern. The highest NDVI value of 0.67 for paddy-pulse-fallow system was recorded in January when pulses were in fully active vegetative phase. The lowest value of 0.22 was observed in the end of November when paddy harvesting was done. The paddy- mustard-fallow system showed the highest peak of NDVI (0.74) in February when growth of mustard was active. The paddy-fallow-paddy system showed a sharp decrease in NDVI during the span of end of January to mid of February due to the presence of puddles and standing water condition during crop establishment phase. This phase was followed by rapid increase of NDVI during early March and April that coincided with active vegetative growth phase of Boro rice. The fallow-pulse-fallow system showed the highest peak of NDVI value (0.70) during February, when the pulse crop was in full growth phase. NDVI decreased at the end of March which was can be attributed to the harvesting of pulses. The fallow-mustard-fallow system also showed a pattern similar to the fallow-pulse-fallow system with the highest peak of NDVI in the month of January. The fallow-vegetable-fallow system showed higher NDVI during December to end of January which coincided with the peak vegetative stage of the winter vegetable. The vegetable-fallow-vegetable showed higher NDVI during the month of October and NDVI started to decline afterwards. Very low values of NDVI were recorded from the month of December to the end of February and an increase in NDVI was observed from the middle of March due to begging of summer vegetable. The vegetable-vegetable-fallow system showed the peak NDVI during the mid of January due to the active vegetative growth of winter vegetable while the lowest NDVI was recorded during the middle of March when the crop terminated.

7th Int. Conf., December 18-19, 2019, Dept. of Statistics, RU Page | 145

Alamgir Shaikh, Argha Ghosh, Kajori Parial and Manoj K. Nanda

4.2. Crop coverage mapping The land use map of the study area for four different dates during the study period is provided in Figure 3.a), b), c) and d). The temporal changes in the areas under different types of land use cover as showed that the area under crop areas increased from October to end of February as pulses were growing. Crop area decreased in November when kharif paddy was harvested and increased again in February due to large scale cultivation of pulse crop and boro paddy. Fallow land increased gradually from November to March. The areas under marshy land decrease during November to March.

Figure 2 Normalized Difference Vegetation Index profile for the study area

Figure 3.a) and b) Landuse map of the study for 2nd October, 2018 and 11th November, 2018

7th Int. Conf., December 18-19, 2019, Dept. of Statistics, RU Page | 146

Alamgir Shaikh, Argha Ghosh, Kajori Parial and Manoj K. Nanda

Figure 3 Landuse map of the study for c) 4th February, 2019 and d) 16th March, 2019

The confusion matrix for the image classification was prepared. It was observed that the overall accuracy of the land use/ land cover classification increased from the image 2-October, 2018 (86.57%) to the image for 16 November, 2018 (90.78%). Accuracy of classification was very high for 4-February, 2019 (98.75%) while the accuracy of classification for 16-March, 2019 was estimated as 95.71 %. The producer’s accuracy of the class ‘settlement/fallow’ was lower as compared to the other class due to class mixing.

5. Conclusion The present study concluded that multi-dated Sentinel-2 MSI data with its fine spatial resolution and high temporal frequencies has good potentiality for assessing the cropping system of an area. Spectral response of different crops and cropping system was successfully analyzed with its good spectral resolutions. NDVI pattern of the cropping systems helped to identify the duration of different crops. The peak vegetative phase or grand growth period of the crops could be detected by studying the temporal NDVI profile of the corresponding cropping systems. Supervised classification of Sentinel-2 data using random forest approach was proved to be efficient in assessing the seasonality of land use and land cover of the study area. Land use maps of different dates over the study period indicated the spatiotemporal variation of crops, fallow, water bodies and other land use features in the block. Overall, the outcomes of the present study are expected to be useful to have an overview of the cropping systems dynamics in the study area.

References Godfray, H. C. J., Beddington, J. R., Crute, I. R., Haddad, L., Lawrence, D., Muir, J. F., Pretty, J., Robinson, S., Thomas, S. M. and Toulmin, C. (2010), Food Security: The Challenge of Feeding 9 Billion People, Science, 327, 812-818. Tscharntke, T., Clough, Y., Wanger, T. C., Jackson, L., Motzke, I.,Perfecto, I., Vandermeer, J. and Whitbread, A. (2012). Global food security, biodiversity conservation and the future of agricultural intensification, Biological Conservation 151, 53–59.

7th Int. Conf., December 18-19, 2019, Dept. of Statistics, RU Page | 147