Remotely-Sensed Surface Temperature and Vegetation Status for the Assessment of Decadal Change in the Irrigated Land Cover of North-Central Victoria, Australia

: Monitoring of irrigated land cover is important for both resource managers and farmers. An operational approach is presented to use the satellite-derived surface temperature and vegetation cover in order to distinguish between irrigated and non-irrigated land. Using an iterative thresholding procedure to minimize within-class variance, the bilevel segmentation of surface temperature and vegetation cover was achieved for each irrigation period (Spring, Summer and Autumn). The three periodic proﬁles were used to deﬁne irrigation land covers from 2008–2009 to 2018–2019 in a key agricultural region of Australia. The overall accuracy of identifying farms with irrigated land cover amounted to 95.7%. Total irrigated land cover was the lowest (approximately 200,000 ha) in the 2008–2009 crop year which increased more than three-fold in 2012–2013, followed by a gradual decline in the following years. Satellite images from Landsat series (L-5, L-7 and L-8), Sentinel-2 and ASTER were found suitable for land cover classiﬁcation, which is scalable from farm to regional levels. For this reason, the results are desirable for a range of stakeholders.


Introduction
Irrigated agriculture is a vital source of food and fiber [1][2][3]. Recently there has been an increasing reliance on irrigated agriculture worldwide as the dryland agriculture has been adversely impacted by climatic changes [4]. Irrigated agriculture uses a large proportion of available freshwater globally, amounting for about 70% [2,5]. In the face of competing demands for limited water resources, there is a sharp focus on the management of irrigated agriculture [6]. The mapping and monitoring of accurate irrigated land cover is vitally important for effective irrigation management and judicious decision making.
The past few decades have witnessed a wide application of Remote Sensing for mapping cropland [3,7,8]. Several studies have adopted standard procedures of different classification techniques for mapping at various spatial scales, including regional [9][10][11][12] and continental [9,11,13,14] levels. There is an increasing interest in cropland mapping at global scale utilizing high frequency coarse resolution data sets from various sensors/satellites including MODIS, AVHRR, MERIS and SPOT VGT [15,16]. Mapping irrigated agriculture and assessing irrigation activities is not common at a global or regional level [17]. There are very few studies on operational approaches to map and monitor Land 2020, 9, 308 2 of 19 irrigated agriculture on a routine basis [18,19]. Some recent reviews have adequately summarized the current status of Remote Sensing applications to agricultural mapping [3,7,18]. Studies on irrigated agriculture are notably far more scarce as compared to agriculture in general. What most studies on irrigated cropland have done so far is to delineate the usual 'irrigation' areas, which is not the same as the actually 'irrigated' land within an irrigation period. In order to investigate periodic or in-season variation in irrigation, it is important to identify the land cover which is actually 'irrigated' with sufficient spatial detail, which has been attempted in this study.
Since the earliest attempts to develop vegetation indices in 1960s-1970s [20,21], many new and modified indices have been researched that relate to plant parameters [8]. Most of these vegetation indices use the combination of spectral responses between visible and the near-or mid-infrared range. However, the most widely used measure is the normalized difference vegetation index (NDVI), which is based on two spectral bands (near infrared and red). Vegetation indices including NDVI have been used in mapping agricultural land cover. However, for irrigated land cover, surface wetness is required as additional information. Surface temperature has been recognized as an indicator of surface moisture and crop water [22]. The differences in surface temperature are potential indicators of irrigation variations [7,13,19]. However, there is a lack of detailed studies using temperature information for irrigation mapping [7]. The objective of this study was: (a) to map irrigated areas for each season by using satellite-based surface temperature and NDVI; (b) to identify land cover classes by using the seasonal profile of irrigated areas; and (c) to evaluate the changes in irrigated land cover over ten years (2008)(2009) to 2018-2019) in a key irrigation region located in the northern part of Victoria, Australia. For the accuracy assessment of our mapped irrigated areas, we used information on irrigation water deliveries of a recent season to determine whether the farms were actually irrigated or not within that season.

Study Area
The study area is located approximately between 35.14 • S and 36.71 • S latitudes, and between 143.31 • E and 146.03 • E longitudes in the north-central part of Victoria, Australia ( Figure 1). With the Murray River in the north, it is spread over the river catchments of Goulburn-Broken, Campaspe and Loddon. It covers an area of about 9950 sq km. About 75% of the land is irrigated. Of the total irrigated land, 87% is used for pastures, 4% for treed horticulture crops and the remaining 9% is used for other purposes including vegetables and grain crops. The climate is temperate, and the region is relatively dry with average annual rainfall of between 300 mm and 500 mm. Generally, winters (June to August) are wet, receiving most of the annual rainfall. Summers (December to February) are usually dry and are when the demand for supplemental water for crops is the highest. Irrigation demand during Spring (September to November) and Autumn (March to May) is variable. The major agricultural industries are dairy, and stone and pome fruit production. The flat terrain and shallow natural drainage of the region are overlaid by a network of irrigation channels. Multiple irrigation system configurations are used in the region including micro-irrigation, conventional sprinkler, flood and furrow.
For management purposes, the irrigation region is divided into six sub-regions or 'irrigation areas' [23] as shown in Figure 1B: Central Goulburn, Shepparton, Murray Valley, Campaspe, Pyramid-Boort and Torrumbarry.

Materials and Methods
In this study, we present an operational approach to use satellite-based surface temperature and vegetation status to map irrigated areas with enough spatial detail and to monitor land cover changes. We introduce an approach to incorporate the relative differences of surface temperature with NDVI to map the areas of irrigated agriculture [24]. The operational approach adopted in this study was firstly to process the satellite images and generate surface temperature (Ts) and NDVI. Secondly, to identify relative differences of Ts and NDVI, appropriate thresholds were determined. At this stage, seasonal matrices of individual pixel profiles based on Ts and NDVI were compiled.
Land 2020, 9, 308 3 of 19 As a third step, irrigated land cover classes were generated using seasonal profiles of pixels, and maps were created for each crop year (September-May). Finally, temporal changes in the irrigated land covers were evaluated. were created for each crop year (September-May). Finally, temporal changes in the irrigated land covers were evaluated.

Step 1: Data Preparation
Satellite images were collected to represent the three irrigation periods (Spring, Summer and Autumn) of every crop year during 2008-2009 to 2018-2019 seasons, except the three crop years (2010-2011, 2011-2012 and 2016-2017) when there was no suitable imagery available over the study area, which have been excluded from the study ( Table 1). Most of the data sets were acquired from three Landsat satellites (L5, L7 and L8) sourced from USGS (https://earthexplorer.usgs.gov/). Data gaps were filled using ASTER (https://search.earthdata.nasa.gov/) and, to a lesser extent, Sentinel-2 (https://scihub.copernicus.eu/) imagery. To have a complete coverage over the study area, multiple adjacent images were acquired for each season (Figure 2).

Step 1: Data Preparation
Satellite images were collected to represent the three irrigation periods (Spring, Summer and Autumn) of every crop year during 2008-2009 to 2018-2019 seasons, except the three crop years (2010-2011, 2011-2012 and 2016-2017) when there was no suitable imagery available over the study area, which have been excluded from the study ( Table 1). Most of the data sets were acquired from three Landsat satellites (L5, L7 and L8) sourced from USGS (https://earthexplorer.usgs.gov/). Data gaps were filled using ASTER (https://search.earthdata.nasa.gov/) and, to a lesser extent, Sentinel-2 (https://scihub.copernicus.eu/) imagery. To have a complete coverage over the study area, multiple adjacent images were acquired for each season ( Figure 2). For vegetation status, NDVI was calculated using reflectance of near infrared (NIR) and red spectral bands [25]: Standard procedures were used to calculate reflectance from Landsat series [26,27] and ASTER images [28]. The following formula was used to calculate at-sensor top of the atmosphere (TOA) reflectance: The downloaded Sentinel-2 images were already converted to reflectance [29]. Ts from the Landsat series was calculated using the standard procedures [26,27]. Ts as at-sensor brightness temperature was calculated using the following formula:  For vegetation status, NDVI was calculated using reflectance of near infrared (NIR) and red spectral bands [25]: Standard procedures were used to calculate reflectance from Landsat series [26,27] and ASTER images [28]. The following formula was used to calculate at-sensor top of the atmosphere (TOA) reflectance: The downloaded Sentinel-2 images were already converted to reflectance [29]. Ts from the Landsat series was calculated using the standard procedures [26,27]. Ts as at-sensor brightness temperature was calculated using the following formula: Surface temperature from ASTER was calculated using band TIR4 (10.25-10.95 µm) [28]. NDVI from ASTER and Sentinel-2 as well as Ts from ASTER were re-sampled to a 30 m resolution using bilinear interpolation method and were adjusted to be comparable to Landsat equivalents. The adjustment factors used in this study were taken from our previous investigations [24,30].
Ts provides information on vegetation water status [31]. In well-watered irrigation situations, the Ts is often lower than the surrounding air temperature (Ta). In other situations, when vegetation is water-stressed, there is less transpiration and often vegetation surface temperature rises above the surrounding Ta [32]. Therefore, it is considered useful to use the difference between surface and air temperatures (Ts -Ta) to assess vegetation water status in this study. Half-hourly Ta data, close to the time of satellite overpass, was sourced from the Bureau of Meteorology (www.bom.gov.au) for 13 weather stations across and close to the study area. Ta point data sets were rasterized, using the inverse distance weighted (IDW) method to match the Ts extent and spatial resolution. Then surface-air temperature difference (Ts -Ta) at a pixel level was calculated.

Step 2: Thresholding Process
It is well known that an 'irrigated' crop has high vegetation and low temperature as compared to other land covers within a region. To take this concept further, temperature (Ts -Ta) and vegetation (NDVI) were segmented into two classes each. Temperature classes were referred to as 'irrigated' (low Ts -Ta) and 'non-irrigated' (high Ts -Ta) and vegetation classes were 'crop' (high NDVI) and 'non-crop' (low NDVI). An iterative thresholding method was used to achieve the binary classification by minimizing within-class variance (σ 2 Within ) [33]: Ti is the threshold which varies by iteration i, ω 0 and ω 1 are the weights (proportion of pixels) of the two classes, and σ 2 0 and σ 2 1 are the variance of the two classes of NDVI and Ts -Ta each. For operational purposes, the thresholding procedure utilized the relationship of σ 2 Within with between-class variance (σ 2 Between ) and total variance (σ 2 Total ) [33]: The initial NDVI threshold (α) was taken as 0.4. All pixels >α were considered as 'crop'. Initial temperature threshold (β) was the median value of Ts -Ta. All pixels <β were taken as 'irrigated'. The iteration interval was set at 0.005 within the limit of ±0.025 of the initial NDVI threshold. For temperature, the iteration interval was set 0.1 within the limit of ±0.5 • C of the initial Ts -Ta threshold (i.e., the median). Altogether, 11 iterations each for NDVI and Ts -Ta were performed for each image. Thresholds with minimum σ 2 Within were used for binary classification. Binary classes were assigned to each pixel with four combinations, which we termed as pixel identification (PID) as shown in Figure 4. S1 denotes dry conditions with no or low vegetation; S2 denotes wet conditions with low or no vegetation; S3 indicates some vegetation, possibly crops, without irrigation; and S4 denotes vegetation with wet conditions indicating 'irrigated crop/pasture'. All pixels in an image of a season were assigned a PID. Thus, each pixel profile had three PID assignments representing three irrigation periods (Spring, Summer, Autumn) in each crop year.
For accuracy assessment, we used the water supply data of a recent season (2018-2019 summer) as a 'reference'. The mapped 'irrigated crop/pasture' PID (S4) of the same season was taken as an 'estimate'. Information on irrigation water supplies was sourced from the Victorian Water Register (VWR), a state-wide irrigation water database (https://waterregister.vic.gov.au/).  Binary classes were assigned to each pixel with four combinations, which we termed as pixel identification (PID) as shown in Figure 4. S1 denotes dry conditions with no or low vegetation; S2 denotes wet conditions with low or no vegetation; S3 indicates some vegetation, possibly crops, All pixels in an image of a season were assigned a PID. Thus, each pixel profile had three PID assignments representing three irrigation periods (Spring, Summer, Autumn) in each crop year.
For accuracy assessment, we used the water supply data of a recent season (2018-2019 summer) as a 'reference'. The mapped 'irrigated crop/pasture' PID (S4) of the same season was taken as an 'estimate'. Information on irrigation water supplies was sourced from the Victorian Water Register (VWR), a state-wide irrigation water database (https://waterregister.vic.gov.au/).

Step 3: Identifying Irrigated Land Cover Classes
Seasonal profiles of pixels were based on NDVI and Ts -Ta binary classes. These pixel profiles provided an indication for each season as to whether or not irrigation was actually applied, and the crop was actively growing. A set of rules were developed to identify irrigated land cover classes ( Table 2). Pixels with 'S4' PID were classed as irrigated. However, when some isolated pixels with 'S3' PID were located within or on the border of S4 cluster, those were also recognized as irrigated. Otherwise all pixels with S1, S2 and S3 were recognized as non-irrigated. A contiguity analysis was carried out on the land cover raster layer of each crop year to identify and filter out the small isolated groups of pixels, which were considered 'noise'. The groups of six pixels or less (<0.5 ha) were considered unlikely to belong to any managed irrigated crop or pasture.

Step 3: Identifying Irrigated Land Cover Classes
Seasonal profiles of pixels were based on NDVI and Ts -Ta binary classes. These pixel profiles provided an indication for each season as to whether or not irrigation was actually applied, and the crop was actively growing. A set of rules were developed to identify irrigated land cover classes (Table 2). Pixels with 'S4 PID were classed as irrigated. However, when some isolated pixels with 'S3 PID were located within or on the border of S4 cluster, those were also recognized as irrigated. Otherwise all pixels with S1, S2 and S3 were recognized as non-irrigated. A contiguity analysis was carried out on the land cover raster layer of each crop year to identify and filter out the small isolated groups of pixels, which were considered 'noise'. The groups of six pixels or less (<0.5 ha) were considered unlikely to belong to any managed irrigated crop or pasture. Table 2. Land cover classification criteria.

Step 4: Evaluating the Irrigated Land Cover Changes
To evaluate the temporal changes, maps and diagrams were generated using the pixel level land cover classes of each crop year. Maps of irrigated land cover classes were prepared for visual evaluation. Illustrations were created to show the changes of land cover classes for the total region as well as for the six individual irrigation sub-regions.

Results
Irrigation is the application of supplemental water to crops in order to maintain and enhance crop growth. Therefore, in the management of irrigation due consideration is given to the amount of irrigation water that is to be applied to meet the crop water demand. As the crop water demand varies due to multiple factors including crop type, phenology and evapotranspiration [34], the irrigation application also varies. In the North-Central Victoria, we have identified seven types of irrigated land cover ( Table 2). The land which is actively irrigated throughout all seasons is designated as 'perennially active', which refers to either perennial pasture (e.g., perennial ryegrass) or perennial horticulture crop. Other land areas are seasonally irrigated in one or two seasons (Table 2), which refer to either annual pasture, annual horticulture or seasonal crop.
To check the accuracy of whether a land parcel was 'irrigated' or not, we used the 2018-19 summer season data sets. An accuracy assessment was carried out at the farm level (irrigation water delivery unit). Figure 5 shows the spread of farms selected for accuracy assessment. Farms receiving ≤10 mL of water within this season were considered as not actively managed for plant production and were excluded from accuracy assessment. All farms that received irrigation water deliveries during this period were treated as 'irrigated' and others as 'non-irrigated'. These were taken as a 'reference' indicating 'actual' irrigation occurrence. The farms with an area (≥1 ha) identified as actively irrigated ('S4 ) were considered as 'estimates'. Altogether, 5650 farms were selected for accuracy assessment. Producer's and user's accuracies were calculated as per standard procedure [35,36]. Table 3 shows the results of the accuracy assessment. The values in Table 3a are the number of farms. The values in Table 3b are the proportion (%) of farms. The overall accuracy was 95.7 percent.  In the sub-sections below, we present the temporal changes in irrigated land cover at regional as well as sub-regional levels. Also, we present the changes in an important land cover class (i.e., 'perennially active') that has significant implications for dairy pastures and perennial horticulture.   In the sub-sections below, we present the temporal changes in irrigated land cover at regional as well as sub-regional levels. Also, we present the changes in an important land cover class (i.e., 'perennially active') that has significant implications for dairy pastures and perennial horticulture.

Changes in Sub-Regions
The six sub-regions experienced changes in the area of irrigated land cover with moderate differences (Figure 8). Pyramid-Boort and Torrumbarry, located in the western part, appeared to have recovered from drought sooner than the rest of the region. Irrigated land in Pyramid-Boort sub-

Changes in Sub-Regions
The six sub-regions experienced changes in the area of irrigated land cover with moderate differences (Figure 8). Pyramid-Boort and Torrumbarry, located in the western part, appeared to have recovered from drought sooner than the rest of the region. Irrigated land in Pyramid-Boort

Changes in Perennially Active Class across Sub-Regions
A large proportion of 'perennially active' land cover class is occupied by perennial pasture (Lolium perenne L.) followed by perennial horticulture. Changes in irrigation over the last 10 years

Changes in Perennially Active Class across Sub-Regions
A large proportion of 'perennially active' land cover class is occupied by perennial pasture (Lolium perenne L.) followed by perennial horticulture. Changes in irrigation over the last 10 years have been driven by several factors including seasonal extremes, commodity prices and policy reforms in agriculture. However, these changes in irrigated land cover are not uniform across the region. Figure 9 shows the level of changes that occurred in the surface area occupied by the perennially active class of land cover in the six sub-regions during the study period.
Land 2020, 9, x FOR PEER REVIEW 14 of 19 have been driven by several factors including seasonal extremes, commodity prices and policy reforms in agriculture. However, these changes in irrigated land cover are not uniform across the region. Figure 9 shows the level of changes that occurred in the surface area occupied by the perennially active class of land cover in the six sub-regions during the study period.
In Central Goulburn, the total area of perennially active class in 2008-2009 was approximately 10,000 ha, which increased to over 40,000 ha in 2013-2014. Thereafter, there has been a consistent decline, reaching down to approximately 22,000 ha in 2018-2019 ( Figure 9A). The decline may be attributed to the trend of transition from dairy to annual horticulture in this sub-region. Still a large proportion of irrigation water here is used by dairy farmers [38]. In Central Goulburn, the total area of perennially active class in 2008-2009 was approximately 10,000 ha, which increased to over 40,000 ha in 2013-2014. Thereafter, there has been a consistent decline, reaching down to approximately 22,000 ha in 2018-2019 ( Figure 9A). The decline may be attributed to the trend of transition from dairy to annual horticulture in this sub-region. Still a large proportion of irrigation water here is used by dairy farmers [38].
In Shepparton Sub-region, total area of perennially active class in 2008-2009 was a little over 5000 ha, which increased to approximately 8000 ha in 2009-2010 and to 14,000 ha in 2012-2013. The changes in the subsequent years have not been uniform though area was reduced to approximately 10,000 ha in 2018-2019 ( Figure 9B). There has been a reduction in dairy in the last five years, transitioning to non-dairy activities including mixed farming and cropping [38].
In Murray Valley, the area of perennially active class in 2008-2009 was approximately 7,000 ha, which increased in the subsequent years, reaching up to approximately 22,500 ha in 2014-2015. Thereafter there was a decline in area, reaching down to approximately 14,500 ha in 2018-2019 ( Figure 9C). There has been a reduction in dairy and perennial horticulture industries during the last five years. A transition from dairy to cropping has occurred in this sub-region as well [38].
In  Figure 9D). The decline is related to the reduction in dairy industry in recent years. This sub-region is more known for cropping and mixed farming than for dairying [38].
The Pyramid-Boort Sub-region has the highest extent of cropping land use in the region. Other dominant land use is mixed farming. Dairying is limited with notable reduction especially found in the south-east in the recent past [38]. This is reflected in the relatively low area of perennially active class, which was approximately 7000 ha in 2017-2018 and 5000 ha in 2018-2019 ( Figure 9E).
In the past few years, there has been some decrease in dairy industry in the central and south-eastern half of Torrumbarry. This did not create dramatic changes in the total area of perennially active class from 2012-2013 until 2017-2018, ranging between 17,000 ha and 19,000 ha. However, in 2018-2019, the area was reduced to approximately 12,500 ha ( Figure 7F). Recently, mixed farming and grazing have increased in this sub-region [38].

Discussion
The study area is part of the Murray-Darling Basin (MDB). In MDB, there are highly regulated water management provisions, which are used to provide a reliable and equitable supply of water for irrigation and other uses. Recent regulations have strengthened the legal, market and price aspects of water supply. This has incentivized irrigators to achieve irrigation efficiencies. Irrigators have water entitlements of nominal volume of water to use. However, to accommodate the year-to-year variability in the amount of water available, annual allocations are changed. As a result, in years of drought, the nominal volume of water entitlement may be very low. During the millennium drought annual allocations fell to as low as 10% of the pre-drought entitlements [39]. The effect of low water availability was reflected in the low area of the 2008-2009 irrigated land cover estimated in this study.
This study presented the application of relative differences in temperature and vegetation to identify and monitor actual irrigated areas. Relating low surface temperature and high vegetation cover to irrigated crops is not a new concept [31,[40][41][42]. However, it was only after the space-borne thermal sensors became available that regional scale studies became possible. Temperatures derived from the Landsat series (L5, L7 and L8) and ASTER are suitable for farm level agricultural studies, both in terms of large regional coverage and appropriate spatial resolution, as presented in this study. The coarse resolution thermal bands (e.g., MODIS and AVHRR), though not ideal for farm scale studies, are useful to provide information at regional and landscape levels. More recently, surface temperatures from the Ecosystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) at 60 m spatial resolution have offered increased opportunities for studies in irrigated agriculture [43].
Vegetation cover (NDVI) at medium to fine resolution is readily available from several satellites (e.g., Sentinel-2, SPOT, RapidEye, QuickBird and WorldView). However, synchronous delivery of vegetation cover and surface temperature is available only from Landsat and ASTER. In some circumstances, vegetation cover (NDVI) alone may provide some indication of irrigated land but to investigate in-season changes in irrigation, temperature differences are needed. If the study area is large with heterogenous climatic conditions, surface temperature may not provide the desired distinction between irrigated and non-irrigated lands across the whole region. In such a situation, using the surface-air temperature difference is a better option.
The operational approach to map irrigated land cover used in this study is most suited to arid or semi-arid areas where rainfall is not a confounding factor. In this study area, the rainfall totals varied from year to year and from season to season ( Figure 10). Occasional spikes in rainfall were unlikely to influence the results of this study because the seasonal totals were not high enough to meet the crop water demand.
Land 2020, 9, x FOR PEER REVIEW 16 of 19 vegetation cover and surface temperature is available only from Landsat and ASTER. In some circumstances, vegetation cover (NDVI) alone may provide some indication of irrigated land but to investigate in-season changes in irrigation, temperature differences are needed. If the study area is large with heterogenous climatic conditions, surface temperature may not provide the desired distinction between irrigated and non-irrigated lands across the whole region. In such a situation, using the surface-air temperature difference is a better option. The operational approach to map irrigated land cover used in this study is most suited to arid or semi-arid areas where rainfall is not a confounding factor. In this study area, the rainfall totals varied from year to year and from season to season ( Figure 10). Occasional spikes in rainfall were unlikely to influence the results of this study because the seasonal totals were not high enough to meet the crop water demand. The initial threshold of 0.4 was adopted for the bilevel segmentation of NDVI in this study on the assumption that any vegetation under 0.4 NDVI is not a managed crop or pasture. However, the initial threshold for Ts-Ta was the median value of Ts-Ta, which varied as per the surface and air temperatures at the time of image acquisition. The aim was to identify pixels on the basis of 'relative differences' in Ts-Ta. The assumption was that the relatively cool pixels indicated 'irrigation' and the relatively warm pixels indicated the absence of irrigation. This assumption holds good if a large proportion of area under study is agricultural land with no major features of climatic extremes as is the case in this study. The relative differences in Ts-Ta were meant to seek distinction between two situations i.e., (1) Actively irrigated land, and (2) Land with no active irrigation and/or rainfed area. It is desired that the Ts-Ta threshold and its range is revisited if significant changes occur in the land cover composition in the study area.
The study area is dominated by irrigated pastures (87% of irrigated land).
The key assumption made about irrigated pasture, unlike dryland pasture, is that the greenness is maintained throughout The initial threshold of 0.4 was adopted for the bilevel segmentation of NDVI in this study on the assumption that any vegetation under 0.4 NDVI is not a managed crop or pasture. However, the initial threshold for Ts -Ta was the median value of Ts -Ta, which varied as per the surface and air temperatures at the time of image acquisition. The aim was to identify pixels on the basis of 'relative differences' in Ts -Ta. The assumption was that the relatively cool pixels indicated 'irrigation' and the relatively warm pixels indicated the absence of irrigation. This assumption holds good if a large proportion of area under study is agricultural land with no major features of climatic extremes as is the case in this study. The relative differences in Ts -Ta were meant to seek distinction between two situations i.e., (1) Actively irrigated land, and (2) Land with no active irrigation and/or rainfed area.
It is desired that the Ts -Ta threshold and its range is revisited if significant changes occur in the land cover composition in the study area.
The study area is dominated by irrigated pastures (87% of irrigated land). The key assumption made about irrigated pasture, unlike dryland pasture, is that the greenness is maintained throughout the season to almost a uniform level, by applying irrigation as required. It is therefore sufficient to use one satellite image to represent an irrigation period, as done in this study. Similarly, for irrigated horticulture crops, the key assumption is that irrigation is used to maintain an optimal level of root zone moisture throughout period of canopy growth. Therefore, the use of one image per season is considered adequate to capture irrigation activity. Our decision to use a single image per irrigation period was in accordance with previous studies. In a study conducted previously on the perennial horticulture crops in the same region, it was found that the use of single 'midseason' image is adequate to assess maximum crop cover because of strong temporal stability in NDVI response (O'Connell 2011, p. 61) [44]. Quoting multiple studies, Velpuri et al. (2009Velpuri et al. ( , p. 1384 [45] reported that "single date fine-resolution imagery, acquired at critical growth stages, is sufficient to identify irrigation". However, minor temporal fluctuations in vegetation cover within a season are possible due to certain factors including over-grazing of pastures, onset of crop disease or extreme weather events. Severe cases of temporal fluctuations may warrant more than one image per season for land cover analysis.

Conclusions
The synchronous measures of surface temperature and vegetation cover based on the satellite images from the Landsat series (L-5, L-7 and L-8) and ASTER were found to be suitable to distinguish between irrigated and non-irrigated pasture/crop to an overall accuracy of 95.7% in the North-Central Victoria, Australia. The irrigation profiles of spring, summer and autumn seasons defined by the combined binary classes of the two measures (Ts -Ta and NDVI) were useful for land cover classification,