estimating actual evapotranspiration of lateral inflow receiving areas in rural landscapes across...

10
WIRADA Science Symposium Proceedings, Melbourne, Australia, 1–5 August 2011 189 Estimating actual evapotranspiration of lateral inflow receiving areas in rural landscapes across Australia Estimating actual evapotranspiration of lateral inflow receiving areas in rural landscapes across Australia McVicar TR 1 , Van Niel TG 2 , Li, LT 1 , Emelyanova, IV 2 , Donohue, RJ 1 , Marvanek, SP 3 , Van Dijk, AIJM 1 , Guerschman JP 1 and Warren GA 1 1 CSIRO Land and Water, P.O. Box 1666, Canberra 2601, Australia, [email protected]. 2 CSIRO Land and Water, Private Bag No. 5, Wembley 6913, Australia. 3 CSIRO Land and Water, Adelaide, Australia. Abstract: In this paper, we provide an overview of ongoing research using satellite estimates of actual evapotranspiration (AET ) from both moderate‑resolution imaging spectroradiometer and advanced very high‑resolution radiometer, and characterising the volumes and temporal patterns of AET for lateral inflow receiving areas (LIRAs) across Australia’s rural landscape. LIRAs are important to Australia’s water balance, as they transfer and store a disproportionately large amount of water relative to their area. Examples of LIRAs are irrigated areas, floodplains, off‑reach wetlands, off‑reach storages and groundwater‑dependent ecosystems. Often these areas do not have enough ground‑based metering to determine their water use with enough confidence to support national water accounts. For much of Australia’s rural landscape AET is limited by water availability, with precipitation (P) being the only source of water. However, due to inflows above or below‑ground, whether natural or anthropogenic, there are areas where AET exceeds P (i.e. the LIRAs). Using gridded P data, we can determine how much of the LIRA‑AET can be attributed to lateral inflow. Having access to this information assists in reconciling national water accounts. Keywords: actual evapotranspiration; inflow; remote sensing; water accounts. 1 INTRODUCTION Australia is the driest permanently inhabited continent, and more than 95 percent of the land area can be considered water‑limited (WL). This means on climatological timescales that actual evapotranspiration (AET) is less than potential evapotranspiration (ETp) as AET is limited by water availability. This contrasts with energy‑limited (EL) environments, where AET is limited by the amount of energy available, so AET approaches ETp. For a comprehensive introduction to Budyko’s EL‑WL framework see Donohue et al. (2007). A consequence of vast areas of Australia being WL is that efficient water resource management is imperative (Van Dijk and Renzullo, 2011). This is especially the case in the rural landscape, where lateral inflow receiving areas (LIRAs) transfer and store a disproportionately large amount of water relative to their area. Examples of LIRAs include (i) irrigated areas, (ii) floodplains, (iii) off‑reach wetlands, (iv) off‑reach storages, and (v) groundwater‑dependent ecosystems. While LIRAs are important for a range of ecological, hydrological, production, and cultural reasons, their dynamics are poorly known due to suboptimal ground‑based metering. This means knowledge of water use from LIRAs is insufficient to support national water accounts. To improve this situation, time series of remotely sensed AET estimates are being used with gridded meteorological data to characterise the role that LIRAs play in the rural water balance. Figure 1a shows the relative abundance of LIRAs in south‑east Australia; contrast this with the output from a water balance

Upload: independent

Post on 25-Nov-2023

0 views

Category:

Documents


0 download

TRANSCRIPT

WIRADA Science Symposium Proceedings, Melbourne, Australia, 1–5 August 2011

189Estimating actual evapotranspiration of lateral inflow receiving areas in rural landscapes across Australia

Estimating actual evapotranspiration of lateral inflow receiving areas in rural landscapes across Australia

McVicar TR1, Van Niel TG2, Li, LT1, Emelyanova, IV2, Donohue, RJ1, Marvanek, SP3, Van Dijk, AIJM1, Guerschman JP1 and Warren GA1

1CSIRO Land and Water, P.O. Box 1666, Canberra 2601, Australia, [email protected]. 2CSIRO Land and Water, Private Bag No. 5, Wembley 6913, Australia. 3CSIRO Land and Water, Adelaide, Australia.

Abstract: In this paper, we provide an overview of ongoing research using satellite estimates of actual evapotranspiration (AET ) from both moderate‑resolution imaging spectroradiometer and advanced very high‑resolution radiometer, and characterising the volumes and temporal patterns of AET for lateral inflow receiving areas (LIRAs) across Australia’s rural landscape. LIRAs are important to Australia’s water balance, as they transfer and store a disproportionately large amount of water relative to their area. Examples of LIRAs are irrigated areas, floodplains, off‑reach wetlands, off‑reach storages and groundwater‑dependent ecosystems. Often these areas do not have enough ground‑based metering to determine their water use with enough confidence to support national water accounts. For much of Australia’s rural landscape AET is limited by water availability, with precipitation (P) being the only source of water. However, due to inflows above or below‑ground, whether natural or anthropogenic, there are areas where AET exceeds P (i.e. the LIRAs). Using gridded P data, we can determine how much of the LIRA‑AET can be attributed to lateral inflow. Having access to this information assists in reconciling national water accounts.

Keywords: actual evapotranspiration; inflow; remote sensing; water accounts.

1 INTRODUCTION

Australia is the driest permanently inhabited continent, and more than 95 percent of the land area can be considered water‑limited (WL). This means on climatological timescales that actual evapotranspiration (AET) is less than potential evapotranspiration (ETp) as AET is limited by water availability. This contrasts with energy‑limited (EL) environments, where AET is limited by the amount of energy available, so AET approaches ETp. For a comprehensive introduction to Budyko’s EL‑WL framework see Donohue et al. (2007).

A consequence of vast areas of Australia being WL is that efficient water resource management is imperative (Van Dijk and Renzullo, 2011). This is especially the case in the rural landscape, where lateral inflow receiving areas (LIRAs) transfer and store a disproportionately large amount of water relative to their area. Examples of LIRAs include (i) irrigated areas, (ii) floodplains, (iii) off‑reach wetlands, (iv) off‑reach storages, and (v) groundwater‑dependent ecosystems. While LIRAs are important for a range of ecological, hydrological, production, and cultural reasons, their dynamics are poorly known due to suboptimal ground‑based metering. This means knowledge of water use from LIRAs is insufficient to support national water accounts. To improve this situation, time series of remotely sensed AET estimates are being used with gridded meteorological data to characterise the role that LIRAs play in the rural water balance. Figure 1a shows the relative abundance of LIRAs in south‑east Australia; contrast this with the output from a water balance

www.csiro.au/WIRADA-Science-Sympsosium-Proceedings

190 McVicar TR, Van Niel TG et al.

model where no above or below ground lateral flow is accounted for (Figure 1b). The inherent sensitivity of remote sensing to provide greater resolution of moisture dynamics of the land surface is thus revealed.

Figure 1. Contrasting examples of assessing water availability of south‑east Australia. Part (a) is a remotely sensed‑based example identifying lateral inflow receiving areas (Van Dijk et al., 2010), whereas part (b) assumes that precipitation is the only source of water and that there is no lateral flow of water across the landscape; it shows a climatological estimate of water yield (or stream flow) using Budyko’s framework as implemented in Donohue et al. (2011)

This paper is based around five research areas being pursued in the assessment of LIRAs:

1. identifying LIRAs, both climatologically and dynamically

2. characterising landscape AET for use in river reach water balance modelling

3. estimating AET from on‑farm storages at both regional and property levels

4. identifying actively irrigated areas

5. ‘blending’ high‑resolution (Landsat) and high‑frequency (moderate‑resolution imaging spectroradiometer, MODIS) remotely sensed data to support water accounting for important LIRAs that cover small areas (e.g. on‑farm storage dynamics in irrigation areas, and wetland and floodplain dynamics).

2 MATERIALS AND METHODS

2.1 Identifying lateral inflow receiving areas

Monthly time‑step satellite remotely sensed AET estimates were available from two sources. First, the Guerschman et al. (2009) CSIRO MODIS Reflectance‑based Scaling ET (CMRSET) algorithm data were available at a 250‑metre spatial resolution from February 2000 through January 2011 (132 months). Secondly, Normalised Difference Temperature Index (NDTI, Kalma et al., 2008; McVicar and Jupp, 1999; McVicar and Jupp, 2002) AET estimates were available at 1000‑metre spatial resolution from April 1992 through December 2005 (165 months). The

(a) (b)

WIRADA Science Symposium Proceedings, Melbourne, Australia, 1–5 August 2011

191Estimating actual evapotranspiration of lateral inflow receiving areas in rural landscapes across Australia

71 months of temporal overlap (February 2000 through December 2005) were used to define the relationship between CMRSET and NDTI AET. Using this relationship, oversampling and per‑pixel bias correction was applied to NDTI AET to simulate CMRSET AET, resulting in a continental 250‑metre remotely sensed based estimate of AET from April 1992 through January 2011 (226 months). Monthly precipitation (P) grids at 0.05° (~5 kilometres) spatial resolution were obtained from the Australian Bureau of Meteorology (Jones et al., 2009) for April 1992 through January 2011.

Using the monthly series (denoted i) of AET and P data, LIRAs were identified as 12 12

1 1i i

i iETa P

= =∑ ∑

> 1.24. We used 1.24 rather than 1.00 to allow for estimation error, both in AET and P. We decided on this threshold after visually inspecting its impact on defining known LIRAs across Australia. Three starting months were used to define a year. They are: (i) if i = 1 for January (i.e. defining a calendar year); (ii) if i = 1 for July (i.e. defining an Australian financial year and the time period used for Australian National Water Account reporting); and (iii) if i = 1 for May (i.e. defining a water year from May to the following April).

2.2 Characterising landscape actual evapotranspiration for river reach water balance modelling

The monthly remotely sensed estimates of AET from 1992 onwards were overlaid with vector (or geographic information system, GIS) datasets of land use in the Murray–Darling Basin (MDB). These datasets include (i) irrigation areas (Bureau of Rural Sciences, 2009), (ii) floodplains (Geoscience Australia, 2006), (iii) wetlands (Australian Federal Government Department of the Environment and Water Resources, 2007), and (iv) dryland, being the remainder of the catchment. We used these temporally static datasets as geographic masks to calculate the AET rate for all land uses for the 453 river reaches used in the Murray‑Darling Basin Sustainable Yields Project (CSIRO, 2008). The processing can be repeated to provide national coverage of river reach level land use‑specific AET using corresponding available national spatial datasets. Both land use average rates (millimetres per day) and AET volumes (megalitres [ML] per month) are calculated for use in river reach water balance models such as Australian Water Resources Assessment‑R (AWRA‑R), which is based on the eWater CRC ‘Source’ Product. Given that AET represents more than 90 percent of the extractive water balance in most parts of Australia, this output should help constrain river reach water balance modelling.

2.3 Estimating actual evapotranspiration from on‑farm storages

In the MDB there has been an intensive effort to map on‑farm storages (Murray–Darling Basin Authority (MDBA) and Geoscience Australia (GA), 2010). Using the monthly remotely sensed estimates of AET from 1992 onwards minus gridded P data, we have been able to characterise both the average rates (millimetres per day) and volumes (ML per month) of AET from on‑farm water storages. This analysis was performed at an annual time‑step (a water year; defined here as 1 May to the following 31 April). As most of Australia is WL for most of the time, we have subtracted the P experienced in the water year from the AET estimate. This analysis was performed at both a regional level (i.e. for six catchments in the north‑east of the MDB), and to assess the utility of continental remote sensing for water accounting at the property level. In the MDBA/GA database, storages are not assigned to properties. Therefore, for six properties we linked the on‑farm storages to properties using ancillary geo‑information, including licensed information reporting the size of the on‑farm storage. These six properties, ranging from 141 hectares (storing 5460 ML for Comilaroy) to 10793 hectares (462,000 ML for Cubbie Station) are important irrigation developments in the north‑east MDB where there is generally insufficient ground‑based metering. The remote sensing AET estimates are therefore an important source of information for closing the river reach water balance. Only AET losses from on‑farm storages are assessed; no account is taken of agricultural water use efficiency (Ag WUE), as we are not accounting for the production term from these farms (see McVicar et al., 2002, for discussion of

www.csiro.au/WIRADA-Science-Sympsosium-Proceedings

192 McVicar TR, Van Niel TG et al.

how Ag WUE can be defined). As to ‘efficiencies in scale’, these six properties may produce food and fibre at high Ag WUE.

2.4 Identifying areas actively irrigated in irrigation schemes

For those river reaches containing major irrigation developments, it is common practice for researchers to scale the area under active irrigation to reach specific diversion data for input to water balance modelling. To improve river reach water balance models, we have inspected whether it is possible to determine a dynamic estimate of actively irrigated area over major irrigation schemes by assessing the remote sensing‑based AET time series. For southern Australia, P is mainly received in the winter, so when summer AET is higher than surrounding areas, then it is likely to be a LIRA, and if located within a known irrigation scheme, is likely to be caused by active irrigation. In northern Australia, P is predominantly received in summer, so inspection of wintertime AET in northern irrigation schemes would be best. Our research performed an initial test case at the Coleambally Irrigation Area (CIA) using the 1992–2011 CMRSET and NDTI bias‑corrected AET discussed above. Based on knowledge of the site and inspection of the AET time series, the summer growing season LIRA (active irrigation area) was determined inside the CIA as anywhere with AET ≥ 95 millimetres per month in January. This threshold in January resulted in an area approximately equal to the long‑term (i.e. 1992–2011) average area of active irrigation and to the current static value used for river reach modelling (393.6 kilometres2). To avoid bias related to approximating the threshold rather than optimising it, the estimated active irrigated area was normalised to the static value.

Figure 2. Conceptual diagrams illustrating (a) the ‘blending concept’ and (b) how blending will be used to estimate actual evapotranspiration (AET) water losses for small LIRAs in WIRADA. In (a) two cloud‑free Landsat‑MODIS () image pairs (from 5 October 2000 and 30 March 2001) are blended with a cloud‑free MODIS image (9 January 2001) to simulate a Landsat‑like image for that day, as shown in the yellow square. The simulated image is a function of the relationships a, b, c and d shown in Part (a). Part (b) shows how the reflectance‑blended images will be used as input to both an AET algorithm and an inundation algorithm to ultimately estimate AET water losses at high spatial (Landsat) and high temporal (MODIS) resolution for small LIRAs in Australia’s rural landscape

2.5 ‘Blending’ high‑resolution and high‑frequency remotely sensed data

For many small LIRAs across Australia (e.g. intensively managed irrigation areas, smaller on‑farm storages filled by direct pumping or runoff harvesting, wetland dynamics) the 250‑metre pixel resolution of the remotely sensed AET products (see Section 2.1) are too coarse. Overcoming the spatial resolution issue requires that high‑spatial resolution remotely sensed data, such as Landsat 25 metres, should be used when mapping AET and inundation for such areas.

MODIS(t1)

Landsat(t1)

t1t

a

tp

MODIS(tp)

t2

d

c

MODIS(t2)

Landsat(t2)

5 October 2000 9 January 2001 30 March 2001

b

MODIS+ daily observation+ no data gaps+ high quality (clouds, calibration)- 500 m resolution

Landsat+ 25 m resolution- twice a month- lower quality

Reflectanceblending

CMRSETETa algorithm

(mm/day)

Inundation Productalgorithm

(m2)

ETa Water Losses(m3/day)

(a) (b)

(a) (b)

WIRADA Science Symposium Proceedings, Melbourne, Australia, 1–5 August 2011

193Estimating actual evapotranspiration of lateral inflow receiving areas in rural landscapes across Australia

However, Landsat is collected with low temporal frequency (having a 16‑day repeat cycle), which contrasts with the MODIS daily repeat cycle. Operational systems relying upon optical remote sensing to provide information about the land‑surface are constrained by cloud presence, with the number of days between cloud‑free images typically being 3–4 times the repeat cycle (Landsat = 48–64 days; MODIS = 3–4 days). To overcome these limitations, we have developed and tested algorithms that blend Landsat and MODIS imagery to output Landsat‑like spatial resolution obtained at MODIS‑like temporal frequency (Figure 2). We are testing several blending algorithms at two water‑using sites with contrasting temporal dynamics: (i) the CIA, a rice‑based irrigation district with flood and furrow irrigation practices in southern New South Wales; and (ii) the Lower Gwydir Catchment (LGC), an area of cotton cropping in northern New South Wales. For CIA, we have data covering two summer growing seasons: (i) 13 cloud‑free Landsat‑MODIS pairs from 5 October 2000 to 10 May 2001 for summer growing season 1; and (ii) 17 cloud‑free Landsat‑MODIS image pairs from 8 October 2001 to 4 May 2002 for summer growing season 2. For LGC, we have 14 cloud‑free Landsat‑MODIS image pairs from 16 April 2004 to 3 April 2005, covering a period of major flooding. Using dense time‑series of Landsat‑MODIS cloud‑free pairs developed at each site, we are assessing the sensitivity of various blending algorithms in reflectance units (Emelyanova et al., 2011a) and in AET units of millimetres per day (Emelyanova et al., 2011b).

3 RESULTS AND DISCUSSION

3.1 Identifying lateral inflow receiving areas

By using a climatology of AET/P, LIRAs can be identified that are suitable for initial production of an operational AET surface for the Australian continent. In such a system, AET for LIRAs will be defined by the CMRSET AET estimate, and for areas that do not receive inflow, AET will be estimated from the AWRA‑L water balance (King et al., 2011). We statically define LIRAs using the longer temporal extent (i.e. the combined AVHRR‑MODIS period of observation from 1992–2011), as the spatial coherence of the identified LIRAs is slightly higher than if using the shorter MODIS‑only observation period (i.e. from February 2000 onwards); cf. Figure 3a and b. This is due to the influence of drought conditions that were experienced in south‑east Australia over the past 10–15 years.

Figure 3. Climatologically identifying lateral inflow receiving areas across Australia. Part (a) uses both the Normalised Difference Temperature Index actual evapotranspiration (NDTI AET) and CSIRO moderate‑resolution imaging spectroradiometer (MODIS) Reflectance‑based Scaling evapotranspiration (CMRSET) AET data products, whereas part (b) uses only the MODIS‑based CMRSET AET from February 2000 onwards

(a) (b)

www.csiro.au/WIRADA-Science-Sympsosium-Proceedings

194 McVicar TR, Van Niel TG et al.

We know that LIRAs are dynamic features in the landscape, and we are seeking to capture their dynamics by identifying LIRAs at an annual or subannual time‑step. Due to ‘observational error’ in the AET/P ratio and residual soil moisture stores, this has proven problematic (Figure 4). Annual water year results show that vast areas of inland Australia are incorrectly being defined as LIRAs, due to very low P. In contrast, when some P occurs (as between May 2005 and April 2006), the AET/P logic works well (Figure 4b). To overcome this issue, thereby enabling us to resolve LIRAs at short time‑steps, we are currently: (i) recasting the logic of the time series analysis, moving from threshold analysis of AET/P ratio (output grids have a unitless ratio) to assessing differences in AET–P (output grids have units of millimetre per integration period); (ii) using AWRA‑L output, which accounts for AET from soil stores, but does not consider lateral inflow to allow the issue of residual soil stores to be accounted for; and (iii) temporally decomposing the 250‑metre MODIS NDVI) time series into its recurrent (~ = grass) and persistent (~ = tree and shrub) components (Donohue et al., 2009). From the latter, we are determining the timing of maximum recurrent cover in the landscape. Our conceptual model is that the vegetation dynamics will provide a means to differentiate between irrigated crops and persistent wetlands.

Figure 4. Annual water‑year identified lateral inflow receiving areas across Australia using the threshold actual evapotranspiration/precipitation (AET/P) ratio approach. Data are CSIRO moderate‑resolution imaging spectroradiometer (MODIS) Reflectance‑based Scaling evapotranspiration (CMRSET) AET and Bureau of Meteorology component of Australian Water Availability Project (BAWAP) P. The legend of Figure 3 also applies to this figure

3.2 Characterising landscape actual evapotranspiration for river reach water balance modelling

Using the monthly 250‑metre remotely sensed‑based estimates of AET (specifically MODIS CMRSET AET from February 2000 onwards and the bias corrected AVHRR NDTI AET from January 1992 to January 2000) with the static land use geoinformation, we have characterised the land use‑specific AET for the 453 reaches that comprise the MDB. Both land use average daily rates and monthly volumes are calculated.

3.3 Estimating actual evapotranspiration from on‑farm storages

Using the monthly 250‑metre remotely sensed‑based estimates of AET from 1992 to 2011, as outlined in Section 3.2, and knowing the locations of on‑farm storages, we were able to calculate the water losses that are occurring only from on‑farm storages. Figures 5a and 5b show the annual regional rate and volume of AET loss, respectively (adjusted for within‑year P). Figures 5c and 5d represent the corresponding water losses from the on‑farm storages in these regions. At a property level, we see that the annual rate of AET loss (adjusted for within‑year P) contains both spatial variability (across properties) and temporal variability (climatic variability; see Figure 5c). When the annual rate is multiplied by the on‑farm storage area to provide a volume of AET loss, this clearly shows – as expected – the influence on the storage surface area (Figure 5d). The on‑farm storage AET losses from Cubbie Station constitute 41 percent of the on‑farm storage losses from the entire Condamine‑Balonne catchment (compare Figure 5b and 5d).

(a) May 2004 – April 2005 (b) May 2005 – April 2006 (c) May 2006 – April 2007 (d) May 2007 – April 2008

WIRADA Science Symposium Proceedings, Melbourne, Australia, 1–5 August 2011

195Estimating actual evapotranspiration of lateral inflow receiving areas in rural landscapes across Australia

Figure 5. Characterisation of AET from on‑farm storages. Part (a) is the daily rate and (b) is the total volume for the regional‑level assessment, with (c) and (d) being the same analysis at the property level, respectively

3.4 Identifying areas actively irrigated in irrigation schemes

The sub‑annual AET/P characterisation is shown in Figure 6a. In red areas, AET/P is high in summer; in green areas, AET/P is high in winter; and in white areas, AET/P is uniformly high all year. Irrigation schemes found in the red parts of Figure 6a (see e.g. the CIA inset) should be assessed in summer, whereas irrigation schemes in the green parts of Figure 6a should be assessed in winter. The results of the dynamic estimation of active irrigation area are shown in Figure 6b for the CIA. The severe drought experienced in south‑east Australia is clearly evident by the below‑average area of active irrigation from 2003 onwards. River reach water balance estimation would likely be different from the dynamic estimation, whether the area under irrigation is estimated from the dynamic red line, or scaled from the maximum assumed area (the black line, as is currently done) in Figure 6b.

(a) (b)

(c) (d)

www.csiro.au/WIRADA-Science-Sympsosium-Proceedings

196 McVicar TR, Van Niel TG et al.

Figure 6. Subannual actual evapotranspiration/precipitation (AET/P) characterisation for Australia is shown in part (a). In part (b), the dynamic estimation of summer irrigated area (red line) is compared to the static estimate (black line) for the 950 kilometres2 Coleambally Irrigation Area (CIA)

3.5 ‘Blending’ high‑resolution and high‑frequency remotely sensed data

The root mean squared difference (RMSD) between real Landsat images and their corresponding simulated Landsat‑like images resulting from the blending procedure are shown for all dates in Figure 7a, in both reflectance and water balance units, when each mid‑date was simulated using its two closest bounding temporal neighbours at the CIA. The Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM, Zhu et al., 2010) and the Global Empirical Image Fusion Model (GEIFM, Emelyanova et al., 2011a) showed similar magnitudes and temporal patterns of RMSD for the red band and the near infrared band (red and black lines, respectively in Figure 7a) when considering reflectance. This was also true for CMRSET AET derived from real Landsat images when compared to AET calculated from the blended images (blue lines, Figure 7a). The spatial representation of CMRSET AET was also similar when calculated from real Landsat imagery (Figure 7b), compared with when CMRSET AET was calculated from simulated Landsat using the GEIFM algorithm (Figure 7c). These initial results suggest that the computationally efficient GEIFM model might be a practical alternative to the computationally expensive ESTARFM algorithm for use in an operational water assessment and accounting system.

CIA

(a) (b)

(a) (b)

4 CONCLUSIONS AND FUTURE RESEARCH DIRECTIONS

This research shows that continental remotely‑sensed based estimates of AET are valuable for assessing water loss from LIRAs across the sparsely gauged areas of Australia. Combining these estimates with suitable conceptual models, AWRA‑L output, ancillary datasets (e.g. gridded P, land use vector data of river reach, on‑farm storages) and vegetation dynamics allows LIRA dynamics to be characterised. Given the variability of Australia’s precipitation and the disproportionately large amount of water transferred and stored by LIRAs relative to their area, this characterisation is an essential component to improving Australia’s water accounts – especially for rural landscapes. While continental AET estimates with a spatial resolution of 250 metres are of great value, we acknowledge that this spatial resolution is inadequate for operational monitoring of key water using areas in some situations. Given this, we are developing and testing methods of blending

WIRADA Science Symposium Proceedings, Melbourne, Australia, 1–5 August 2011

197Estimating actual evapotranspiration of lateral inflow receiving areas in rural landscapes across Australia

Figure 7. Blending results for the Coleambally Irrigation Area (CIA) for the 2000–2001 summer growing season. Part (a) shows the root mean squared difference (RMSD) across the entire image in both reflectance units (for the red and near infrared (NIR) bands) and actual evapotranspiration (AET) units for the time series for two blending methods. Part (b) shows the AET estimated using the CSIRO moderate‑resolution imaging spectroradiometer (MODIS) Reflectance‑based Scaling evapotranspiration (CMRSET) algorithm for real Landsat data acquired on 9 January 2001 for ~20 x ~20‑kilometere subset of the CIA. Part (c) is the same, except that simulated Landsat data, produced from the Global Empirical Image Fusion Model (GEIFM) algorithm were used as input to the CMRSET algorithm

Real Landsat

GEIFM Blended Landsat

0.0 7.03.5

ETa (mm/day)

(a) (b) (c)

(a) (b) (c)

high spatial‑resolution (Landsat) and high temporal‑resolution (MODIS) imagery to simulate Landsat‑like imagery as often as cloud‑free MODIS data are collected. This simulated imagery will be input into suitable AET and inundation algorithms, from which we can ultimately monitor AET losses at high spatial and temporal resolution.

Our future research directions are fourfold:

1. We aim to improve identification and characterisation of LIRA dynamics by use of (i) the AET–P logic, (ii) AWRA‑L output, and (iii) persistent and recurrent vegetation dynamics.

2. Based on 1) above, we will be able to provide more realistic measures of LIRA AET water losses by considering the dynamic nature of LIRAs, as opposed to considering them to be temporally static.

3. The outputs from components 1) and 2) above will be summarised for use in river reach water balance modelling across Australia. Such models are expected to greatly benefit from the inclusion of dynamics as observed by remote sensing systems.

4. Our research in blending Landsat and MODIS is progressing through difference characterisation in reflectance units to characterisation in both units of millimetres per day and metres2 of inundated area. These will be combined to estimate LIRA AET at high spatial and high temporal resolutions for the two case study sites. In discussion with water information users and producers (e.g. in the Bureau of Meteorology) we are developing prototype processing pathways as a template for operational systems –underpinned by the research summarised in this paper – to better account for water losses in Australia’s rural landscapes.

www.csiro.au/WIRADA-Science-Sympsosium-Proceedings

198 McVicar TR, Van Niel TG et al.

ACKNOWLEDGEMENTS

Thanks to Dr Zhu and Dr Chen (Beijing Normal University, China) for providing access to the ESTARFM code; Drs Tivi Theiveyanathan, Masoud Edraki and Carl Daamen (Water Resources Assessment Section, Bureau of Meteorology) for numerous fruitful discussions; and Dr Leo Lymburner (Geoscience Australia) and members of his team for making the Landsat‑5 TM Lower Gwydir Catchment database available.

REFERENCES

Australian Government Department of the Environment and Water Resources (2007) Directory of Important Wetlands in Australia (DIWA) spatial database. Canberra. Available online at <http://www.environment.gov.au/metadataexplorer/full_metadata.jsp?docId=%7B3F179472‑DE1F‑4C6C‑B7CA‑535DF2896656%7D&loggedIn=false>.

Bureau of Rural Sciences (2009) Catchment scale land use mapping for Australia (April 2009 dataset), Canberra. Available online at <http://adl.brs.gov.au/mapserv/landuse/index.cfm?fa=app.landUseInformation>.

CSIRO (2008) Water availability in the Murray–Darling Basin. A report to the Australian Government from the CSIRO Murray‑Darling Basin Sustainable Yields Project. CSIRO, Australia. Available online at <http://www.csiro.au/files/files/po0n.pdf >.

Donohue, R.J., McVicar, T.R., Roderick, M.L. (2009) Climate‑related trends in Australian vegetation cover as inferred from satellite observations, 1981–2006. Global Change Biology 15(4), 1025–1039.

Donohue, R.J., Roderick, M.L., McVicar, T.R. (2007) On the importance of including vegetation dynamics in Budyko’s hydrological model. Hydrology and Earth System Sciences 11(2), 983–995.

Donohue, R.J., Roderick, M.L., McVicar, T.R. (2011) Assessing the differences in sensitivities of runoff to changes in climatic conditions across a large basin. Journal of Hydrology 406(3–4), 234–244, doi:10.1016/j.jhydrol.2011.07.003.

Emelyanova, I.V., Van Niel, T.G., McVicar, T.R., Li, L.T., Van Dijk, A.I.J.M. (2011a) Assessing the accuracy of blending Landsat‑MODIS reflectances in two heterogeneous landscapes. Remote Sensing of Environment. In prep.

Emelyanova, I.V., Van Niel, T.G., McVicar, T.R., Li, L.T., Van Dijk, A.I.J.M. (2011b) Estimating actual evapotranspiration for hydrologically active areas using blended Landsat‑MODIS time‑series imagery: Implications of spatial, spectral and temporal sampling. Journal of Hydrology. In prep.

Geoscience Australia (2006) GeoData Topo 250K Series 3 Topographic Data. Canberra. Available online at <http://www.ga.gov.au/oracle/agsocat/geocat_brief.php?catno=64164>.

Guerschman, J.P. Van Dijk, A.I.J.M., Mattersdorf, G., Beringer, J., Hutley, L.B., Leuning, R., Pipunic, R.C., Sherman, B.S. (2009) Scaling of potential evapotranspiration with MODIS data reproduces flux observations and catchment water balance observations across Australia. Journal of Hydrology 369(1 –2), 107–119.

Jones, D.A., Wang, W., Fawcett, R. (2009) High‑quality spatial climate data‑sets for Australia. Australian Meteorological and Oceanographic Journal 58(4), 233–248.

Kalma, J.D., McVicar, T.R., McCabe, M.F. (2008) Estimating land surface evaporation: a review of methods using remotely sensed surface temperature data. Surveys in Geophysics 29(4–5), 421–469.