Search

• We demonstrate flaws in the CMRSET implementation in Bretreger et al. (2020). • Actual evapotranspiration biases were because they used an incorrect Landsat band. • The calculation of GVMI should have used Landsat SWIR1 band instead of SWIR2. • CMRSET's suitability for irrigation monitoring is demonstrated via new experiments. • CMRSET evapotranspiration using the correct SWIR1 band did not show large biases. In a recent paper published in this journal, Bretreger et al. (2020) estimate irrigation water use from satellite remotely sensed estimates of actual evapotranspiration in five irrigated districts of the Murray-Darling Basin (southeast Australia). They used three models that scale crop reference evapotranspiration with vegetation indices acquired by recent Landsat satellites: (i) IrriSAT, (ii) Kamble and (iii) CMRSET. In their paper, irrigation water use computed with CMRSET generally overestimated observed irrigation water use, sometimes fivefold. Based on these results, Bretreger et al. (2020) discouraged the use of CMRSET for irrigation monitoring. In this comment, we reproduce the experiments in Bretreger et al. (2020), and demonstrate that their overestimation was because they used an incorrect Landsat band in their implementation of CMRSET. CMRSET was originally calibrated using both MODIS-derived Enhanced Vegetation Index (EVI) and Global Vegetation Moisture Index (GVMI). To calculate GVMI, a shortwave infrared (SWIR) band with a wavelength of ∼1.6 μm was used, which for MODIS is named SWIR2, and for Landsat is named SWIR1 (i.e., different sensors have different bands and a different number of bands and SWIR1 in MODIS has a wavelength of ∼1.2 μm). In their Landsat CMRSET implementation, Bretreger et al. (2020) computed GVMI using the Landsat SWIR2 band, which has a wavelength of ∼2.1 μm (see their Table 3). We show that CMRSET implemented with the Landsat SWIR1 band (i.e., with a wavelength of ∼1.6 μm, so being the correct Landsat band to calculate GVMI) yields similar results (both for temporal patterns and magnitude) when compared to the other two remote sensing actual evapotranspiration models in Bretreger et al. (2020). For the irrigation districts, these similar results meant that the mean water year (i.e., July to next June from 2010 to 2017) actual evapotranspiration mean absolute relative difference was 15.3% (with a 5.4%–26.5% range) and the water year irrigation water use mean absolute relative difference was 4.1% (with a 9.8%–18.9% range). Conversely, we show that a Landsat CMRSET implementation using the incorrect Landsat band for the calculation of GVMI (i.e., with a wavelength of ∼2.1 μm) led to a large overestimation agreeing with the CMRSET results reported by Bretreger et al. (2020), both for water year actual evapotranspiration (mean absolute relative difference of 64.6%, with a 51.1%–83.6% range) and irrigation water use (mean absolute relative difference of 44.1%, with a 31.4%–65.1% range). The use of the incorrect Landsat band means that the Bretreger et al. (2020) recommendation specific to their Landsat CMRSET implementation is invalid. A demonstration of Bretreger et al.'s (2020) incorrect Landsat CMRSET implementation and the correct one can be accessed at: https://jorgepena.users.earthengine.app/view/comment-on-bretreger-2020-joh-paper. Despite Bretreger et al.'s (2020) shortcomings in their implementation of the Landsat CMRSET actual evapotranspiration model, timely monitoring of irrigation water use via satellite remote sensing is required at Landsat resolutions (i.e., 30 m) or higher due to the heterogeneous nature of irrigation in most agricultural landscapes, and we applaud them for pursuing this line of research. [ABSTRACT FROM AUTHOR]

• Remote sensing provides a way to monitor irrigation water use at the catchment scale. • Accurate irrigation simulations are observed over a range of catchment sizes and crop types. • Better results are seen in wetter years (water allocations above 40%). • Long term negative drought indices (SPIs), indicating dry years, shows a decrease in accuracy. Irrigated agriculture has been identified as using approximately 72% of water globally. Australia, like many places in the world, is subject to water sharing plans that cross government boarders and are subject to a mixture of management policies. There is a pressing need to develop a method to monitor irrigation water use to aid in water resource assessments and monitoring. This paper aims to test a method previously developed which monitors irrigation water use using remotely sensed observations over the catchment scale, without the need for in-situ observations, ground data or in-depth knowledge of crops and their planting dates. Using conservative assumptions about agricultural land management practice, irrigation is calculated as I r r = A E T - P. The method uses three vegetation indices derived from Landsat images to calculate crop coefficients (K c) based on multiple published relationships. These are combined through the FAO56 methodology using gridded rainfall and two reference evapotranspiration ( E T o) products to find actual evapotranspiration as A E T = E T o × K c , providing six E T o - K c combinations. Results indicate this study method can effectively assess irrigation water use over a range of catchment sizes from ~6000 to ~600,000 ha, although issues arise when regions have a designated low allocation volume for that season (less than40%). Comparisons with the Standardised Precipitation Index (SPI) and Evaporative Stress Index (ESI) show that the proposed method is robust to the rapid onset and short-term droughts, However, its performance was poor during the long term droughts with low water allocation years. The study results during these years has been predominately attributed to water stress in certain crops being undetected, agricultural managers skipping annual crop commodities as well as stock and domestic water use making up larger portions of total water use. This is a limitation of this approach, although when only comparing results in years with greater than 40% allocations, the results improved significantly showing it can monitor water use effectively. When adequate water is available, this approach is able to accurately predict irrigation water use for the sites examined. [ABSTRACT FROM AUTHOR]

• Event mean concentrations of water quality properties using stratified sampling. • The outlined method provides an easily implemented probabilistic approach. • Flexible for multiple water quality properties. • Estimates of key hydrograph components can be determined for each event. Short rainfall events contribute to large portions of annual sediment and nutrient exports. Most water quality sampling schemes rely on regularly spaced temporal sampling and increasingly monitoring schemes are including a form of event-based sampling. A typical approach is to sample each event using equal intervals in time using an automatic sampler. The use of this form of sampling is systematic in nature and requires model-based statistics to be analysed correctly. Probabilistic based sampling methods allow for easier and more defendable statistical inference as the assumptions are not based on a model, rather they are based on the sample design. Several probabilistic methods have been developed, however these methods commonly require additional hardware to implement. In this paper we present a method using a stratified random sampling procedure for automatic samplers which does not require any additional hardware. Our approach is to divide the mean event hydrograph into strata based on key features such as the rising and falling limbs. Random sampling is applied within each strata. A problem of this approach is that the length of the event and strata must be defined before each event. We therefore outline how the samples can be post-stratified after each event based on the key hydrological components of each event. The sampling scheme is outlined using continuously sampled electrical conductivity and turbidity data of three events from a creek in south eastern Australia. Limited to 24 samples per event, estimated event mean CIs were within the observed event means for all three events. This method provides a flexible low-cost sampling scheme providing unbiased estimates of key event hydrological components which can be easily adapted by catchment management authorities. [ABSTRACT FROM AUTHOR]