Mapping the Spatio-Temporal Evolution of Irrigation in the Coastal Plain of Georgia, USA

This study maps the spatial and temporal evolution of acres irrigated in the Coastal Plain of Georgia over a 38 year period. The goal of this analysis is to create a time-series of irrigated areas in the Coastal Plain of Georgia at a sub-county level. From 1976 through 2013, Landsat images were obtained and sampled at four year intervals to manually detect Center-Pivot irrigation (CPI) systems in the analysis region. During the 38 year analysis period there was a 4,500 percent increase in CPI systems detected that corresponded to an approximate 2,000 percent increase in total acreage. The bulk of the total acreage irrigated is contained in southwest Georgia, as seven counties in the region contained 38 percent of the total acreage irrigated in 2013. There was substantial growth throughout the entire Coastal Plain Region, but southwest Georgia was identified as the most heavily irrigated region of the state. Introduction Agriculture has always been critical for sustaining human life on Earth. Improving technology and agricultural practices made it possible for world food production to double over a 31 year period between 1960 and 2000 (Tilman, 1999), which is part of a larger increased agricultural production in the 20th Century known as the Green Revolution (Evenson and Gollin, 2003). In the year 2000, approximately 15 million square kilometers of the global land cover was dominated by cropland (Ramankutty et al., 2008). With the current world population of 7.3 billion, which is expected to reach 11.2 billion by the year 2100 (UN Department of Economics and Social Affairs) and the growing demand for biofuel production (Evans, 2009) the need for agricultural landscapes could potentially increase in the future. One catalyst from the rapid improvement of agricultural production was the large expansion of irrigation (Tillman et al., 2001). Irrigation can be defined as land areas that receive full or partial application of water by artificial means to offset periods of precipitation shortfalls during the growing period (Ozdogan et al., 2010). In 2000, it was estimated that 2.8 million km2 were irrigated, with this number forecasted to increase 5.29 million km2 by 2050 (Tilman, 2001). Irrigation, much like urbanization, acts to alter the natural landscape properties such as partitioning latent and sensible heating at the surface of the Earth which can impact surface temperature and surface moisture transport. Understanding the extent and usage of irrigation is imperative in answering questions about future water resources, as it is estimated that irrigation uses over 70 percent of the world’s consumption of freshwater (Boucher, 2004; Velpuri et al., 2009). Irrigation accounts for approximately 60 percent of consumptive use of freshwater in the United States where estimates show that over 222,577 km2 of cropland are irrigated (Braneon, 2014; Minchenkov, 2009). For Georgia, it is estimated that approximately 5.5 billion gallons of water per day were withdrawn from surface and ground waters in 2004 (Barnes and Keyes, 2010). Agricultural water use during 2005 totaled 752 million gallons per day for irrigation, with the highest rate of irrigation occurring in the Coastal Plain region of Georgia. The primary crops irrigated in Georgia are maize, cotton and peanuts as they accounted for approximately 68 percent of the total irrigated acreage in 2002 (Braneon and Georgakakos, 2014). Agricultural water use in Georgia is also tied into the ongoing dispute between Georgia, Florida, and Alabama over water use in the Apalachicola-Chattahoochee-Flint (ACF) River Basins known as the “Tri-State” waters wars. Georgia is the upstream water user and the heavy agricultural water usage in southwest Georgia impacts the amount of fresh water that reaches Apalachicola Bay in Florida which supports a multi-million dollar shellfish industry. Research has shown that irrigated croplands can impact land-atmosphere interactions and fresh water supply. Various modeling and observational studies have demonstrated that irrigation influences climate at the local, regional, and global level by enhancing evapotranspiration, altering precipitation patterns, as well as impacting minimum temperature, maximum temperature, and diurnal temperature range (Barnston and Schickedanz, 1984; Greets, 2002; Adegoke et al., 2003; Boucher, 2004; Kueppers, 2007; Lobell and Bonfils, 2008; DeAngelis et al., 2010; Sen Roy et al., 2011; Cook et al., 2014 Shukla et al., 2014, Williams et al., 2015) These activities present a need for accurate and detailed geospatial information on irrigated croplands (Pervez and Brown, 2010). In the United States, most mapping efforts are focused primarily on the California and the Great Plain regions. To expand and contribute to existing knowledge on the spatial and temporal changes in irrigation in Georgia, this analysis maps center pivot irrigation systems (CPI systems) through visual interpretation of Landsat satellite imagery. This shape-based method of mapping irrigation is commonly done for local scale mapping efforts. CPI systems are easy to identify in Landsat imagery because of their distinct arc-like appearance. Landsat was preferred for this analysis because of its greater spatial coverage and the availability of imagery for more time periods. In Georgia, CPI systems are used to irrigate multiple crops, and an accurate estimate of the number of CPI systems in the state could lead to better estimation of water use (Boken et al., 2004) and help identify potential climatic impacts. The analysis herein is conducted on a regional scale, with a methodology normally used for local scale studies. The goal was to produce detailed spatial extent of areas equipped for irrigation over a 38-year Marcus D. Williams and Christie M.S. Hawley are with the USDA Forest Service, Southern Research Station, Center for Forest Disturbance; 320 E. Green Street, Athens, GA 30602 ([email protected]). Marguerite Madden and J. Marshall Shepherd are with the University of Georgia, Department of Geography, 210 Field St. No. 204, Athens, GA 30602. Photogrammetric Engineering & Remote Sensing Vol. 83, No. 1, January 2017, pp. 57–67. 0099-1112/17/57–67 © 2017 American Society for Photogrammetry and Remote Sensing doi: 10.14358/PERS.83.1.57 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING J anuar y 2017 57 time frame in the Georgia Coastal Plain. The following sections include a discussion on previous mapping efforts at the global, regional, and local levels followed by information on the study area, data and methods used in our analysis, followed by the results, conclusions, and a summary section. Background/Previous Literature Irrigation is mapped at three distinct scales; local, regional, and global. As defined by Ozdogan et al. (2010), local scales refers to one or more irrigation basins and they are typically on the order of several square kilometers in size. Regional scale studies are defined as studies that include large river basins to continental areas that extend from tens to thousands of square kilometers in area, while global scale refers to studies that attempt to map irrigation worldwide. Most mapped irrigation studies take place at the local scale, as methods developed for one location may not be appropriate for other locations (Ozdogan et al., 2010). The methodology for local scale studies includes visual interpretation of satellite imagery or digital image classification. Manual identification of irrigated areas is often conducted for visual interpretation studies while automated classification techniques are often used for digital image classification studies. One technique to automatically detect irrigated versus non-irrigated vegetation is through digital image processing to calculate the Normalized Difference Vegetation Index (NDVI). The NDVI is a normalized ratio of the near-infrared bands and red bands (Ustin, 2004) and the greater amount of healthy vegetation present in the sensor, the greater the NDVI value (Jensen, 2005). Pervez and Brown (2010) noted that automated techniques such as using NDVI to identify irrigated areas in humid locations can be problematic as there is little spectral difference between irrigated and non-irrigated landscape. NDVI is calculated in this study to assist in the manual detection of irrigated areas, but was not used as a stand-alone automated classification technique. Prior irrigation mapping studies (Doll and Siebert 1999; Ozdogan and Gutman, 2008; Siebert et al. 2005) conducted at the global and regional scales were performed with very coarse resolution (pixel sizes of 500 m to10 km). Many of the studies produced maps that represented irrigated areas as a percentage of the pixel unit area, which does not provide information on the sub-pixel location of irrigated areas. While this process is sufficient for national and global applications, this level of detail is not adequate for regional analysis. Boken et al. (2004) stated that sub-county, high-resolution irrigation mapping would lend better understanding to agricultural water use. Pervez and Brown (2010) attempted to make improvements on the prior irrigation maps by assimilating U.S. Department of Agriculture (USDA) National Agricultural Statistic Service (NASS) data with Moderate Resolution Imaging Spectroradiometer (MODIS) imagery. Their analysis produced maps of irrigated lands at the 250 m cell size across the conterminous US for 2002. They were unable to conduct a quantitative accuracy assessment for the Eastern US stating that humidity made it difficult for the NDVI to distinguish between irrigated and non-irrigated agricultural areas. A joint effort conducted by the Georgia Environmental Protection Division EPD and the University of Georgia (UGA) mapped irrigated areas in the analysis region for 2007 to 2008 using National Agriculture Imagery Program (NAIP) imagery to serve as a baseline for water resource management purposes (Braneon, 2014). Our mapping analysis serves to update and provide historical context to the mapping efforts of the Georgia EPD. The research herein has a goal to quantify the temporal and spatial evolution of areas irrigated in the Southeastern US Coastal Plain study region of southwestern Georgia, mainly by using satellite imagery obtained from the long-term US Landsat Program. Accurate, detailed, geospatial information on irrigated croplands is essential for answering many Earth science systems, climate change, and water supply questions (Ozdogan et al., 2010). Irrigated areas are estimated through the use of time-series remote sensing data to map center pivot irrigation systems. The analysis is conducted from ten dates of imagery acquired as early as 1976 and as current as 2013 in order to assess long-term trends in irrigation construction within the analysis region. Study Area, Data, and Methods Study Area Georgia, located in the Southeastern United States, has a climate that is classified as fully humid with hot temperatures and warm summers (Kottek et al., 2006). Georgia has a yearly average temperature of 17.4°C (63.4°F) and on average receives 1, 267 mm (49.89 in) of precipitation annually (SERCC, 2015). Georgia receives an adequate amount of rainfall to support agricultural crops such as maize. The sporadic nature of rainfall during the growing season (defined as March through October for this study) requires farmers to rely on irrigation to supplement rainfall. Using the Irrigated Fields with Sources in the Georgia Water Planning Region (WPR) dataset (Hook, 2010), initial analysis showed that 99 percent of the identified center pivot irrigation acreage occurs below the Georgia Fall Line that is approximately 60 percent of the total land area of Georgia. With this information, the study area was narrowed to the Georgia Coastal Plain (Figure 1). Figure 1. Physiographic Regions of Georgia. Covering a total area of 92, 333 km2, the Georgia Coastal Plain landscape is characterized by relatively gently rolling to level topography with elevations ranging from approximately 228 meters to sea level. At higher elevations, there is little level terrain except for the occasional marshy flood plain or narrow steam terrace. Soils are generally productive, well-drained, and moderately permeable. However, in areas of nearly level terrain, i.e., closer to the coast of Georgia, the soils become restrictive for agriculture and pasture (Hodler and Schretter, 1986). The primary agricultural crops grown in the region are cotton, maize, and peanuts. 58 J anuar y 2017 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING Data and Methods This study used a visual interpretation-based approach to identify areas equipped for irrigation from a time-series of Landsat satellite imagery. It should be noted that areas equipped for irrigation were mapped instead of acres irrigated. This distinction is necessary as there is no concrete way to determine if active irrigation coincided with the passing of the Landsat satellite every 16 days and recording of the images. The cloud-free images used in the analysis were captured at various points during the growing season (March through October) over a 38 year period; a particular pivot could be in between the planting or harvest stage when the image was collected. Although band combinations and spectral enhancements can be applied to the images to highlight areas that were recently wetted, without historical ground truth data documenting actual irrigation, efforts to verify that a center pivot was operational at the time of image acquisition were not possible. Therefore, the distinction that is made in this study is areas equipped for irrigation are documented and reported as total acreage. The visual interpretation used a shape-based approach to identify center pivot irrigation (CPI) systems in the Landsat imagery. Center pivot irrigation systems are easily identified in Landsat imagery due to their arc-like appearance (Figure 2). This approach is easily transferable to other locations as Rundquist et al. (1989) used similar methods to create a 15 year time series of CPI systems in Nebraska from Landsat imagery. Optical sensors of the Landsat satellite program began collecting images of the Earth’s surface starting in 1972 with the launch of Landsat 1. In the study area, quality images were available starting with the year 1976, Landsat scenes were selected at four-year intervals until 2008. There was a five-year interval between 2008 and the final year of 2013 due to scan line correction issues with the Landsat-7 satellite. In total, data from ten dates covering the four Landsat satellite missions were selected. Those were Landsat-1 (1976), Landsat 2(1980), Landsat-5 (1984 to 2008; sampled every four years), and Landsat-8 (2013). Table 1 provides information on the sensors and bands for all of the Landsat missions used in our analysis. The criteria for image selection were that the images must be at least 90 percent cloud free with no visible cloud obstruction in areas with CPI systems. The indexed path and row numbers of selected scenes were consistent for the four satellite missions. The primary Landsat scenes analyzed were paths 17 to 19 and rows 37 to 39. The approximate size of each Landsat scene is also consistent among the four satellites with each scene size being 185 km × 185 km. Additional information about the Landsat satellite program can be found at the US Geological Survey website (USGS, 2015). The identification of the CPI systems in the Landsat images consisted of several steps: 1. Load a Landsat scene into ArcGIS®; 2. Load additional shapefiles into ArcGIS that contain spatial reference data about Georgia and the Georgia Coastal Plain (including geographic coordinate system used and Universal Transverse Mercator (UTM) zones ); 3. Find the best combination of bands or other spectral enhancements to highlight the CPI systems; and 4. Manually digitize the CPI systems through a process called heads-up digitizing. Two vector shapefiles for the state of Georgia including counties, and physiographic provinces were obtained from the Esri database of US map data. ArcGIS 10.1 Desktop was used to compile all images into a single geodatabase, perform basic image processing, visual interpretation, and heads-up digitizing of CPIs. The image and vector data were georeferenced to the geographic coordinate system (GCS) (also referred to as Latitude Table 1. Table Providing Information on the Landsat Missions Used in Analysis Landsat 1 Landsat 2 Landsat 5 Landsat 8 Year(s) used 1976 198