Infrastructure Engineering - Research Publications

Permanent URI for this collection

Search Results

Now showing 1 - 6 of 6
  • Item
    Thumbnail Image
    In-season crop classification using optical remote sensing with random forest over irrigated agricultural fields in Australia
    Gao, Z ; Guo, D ; Ryu, D ; Western, A ( 2021-03-03)
    <p>Timely classification of crop types is critical for agronomic planning in water use and crop production. However, crop type mapping is typically undertaken only after the cropping season, which precludes its uses in later-season water use planning and yield estimation. This study aims 1) to understand how the accuracy of crop type classification changes within cropping season and 2) to suggest the earliest time that it is possible to achieve reliable crop classification. We focused on three main summer crops (corn/maize, cotton and rice) in the Coleambally Irrigation Area (CIA), a major irrigation district in south-eastern Australia consisting of over 4000 fields, for the period of 2013 to 2019. The summer irrigation season in the CIA is from mid-August to mid-May and most farms use surface irrigation to support the growth of summer crops. We developed models that combine satellite data and farmer-reported information for in-season crop type classification. Monthly-averaged Landsat spectral bands were used as input to Random Forest algorithm. We developed multiple models trained with data initially available at the start of the cropping season, then later using all the antecedent images up to different stages within the season. We evaluated the model performance and uncertainty using a two-fold cross validation by randomly choosing training vs. validation periods. Results show that the classification accuracy increases rapidly during the first three months followed by a marginal improvement afterwards. Crops can be classified with a User’s accuracy above 70% based on the first 2-3 months after the start of the season. Cotton and rice have higher in-season accuracy than corn/maize. The resulting crop maps can be used to support activities such as later-season system scale irrigation decision-making or yield estimation at a regional scale.</p><p>Keywords: Landsat 8 OLI, in-season, multi-year, crop type, Random Forest</p>
  • Item
    Thumbnail Image
    Using short-term ensemble weather forecast to evaluate outcomes of irrigation
    Guo, D ; Western, A ; Wang, Q ; Ryu, D ; Moller, P ; Aughton, D ( 2021-03-03)
    <p>Irrigation water is an expensive and limited resource. Previous studies show that irrigation scheduling can boost efficiency by 20-60%, while improving water productivity by at least 10%. In practice, scheduling decisions are often needed several days prior to an irrigation event, so a key aspect of irrigation scheduling is the accurate prediction of crop water use and soil water status ahead of time. This prediction relies on several key inputs such as soil water, weather and crop conditions. Since each input can be subject to its own uncertainty, it is important to understand how these uncertainties impact soil water prediction and subsequent irrigation scheduling decisions.</p><p>This study aims to evaluate the outcomes of alternative irrigation scheduling decisions under uncertainty, with a focus on the uncertainties arising from short-term weather forecast. To achieve this, we performed a model-based study to simulate crop root-zone soil water content, in which we comprehensively explored different combinations of ensemble short-term rainfall forecast and alternative decisions of irrigation scheduling. This modelling produced an ensemble of soil water contents to enable quantification of risks of over- and under-irrigation; these ensemble estimates were summarized to inform optimal timing of next irrigation event to minimize both the risks of stressing crop and wasting water. With inclusion of other sources of uncertainty (e.g. soil water observation, crop factor), this approach shows good potential to be extended to a comprehensive framework to support practical irrigation decision-making for farmers.</p>
  • Item
    Thumbnail Image
    Improving the representation of cropland sites in the Community Land Model (CLM) version 5.0
    Boas, T ; Bogena, H ; Grünwald, T ; Heinesch, B ; Ryu, D ; Schmidt, M ; Vereecken, H ; Western, A ; Hendricks-Franssen, H-J ( 2021-03-04)
    <p>The incorporation of a comprehensive crop module in land surface models offers the possibility to study the effect of agricultural land use and land management changes on the terrestrial water, energy and biogeochemical cycles. It may help to improve the simulation of biogeophysical and biogeochemical processes on regional and global scales in the framework of climate and land use change. In this study, the performance of the crop module of the Community Land Model version 5 (CLM5) was evaluated at point scale with site specific field data focussing on the simulation of seasonal and inter-annual variations in crop growth, planting and harvesting cycles, and crop yields as well as water, energy and carbon fluxes. In order to better represent agricultural sites, the model was modified by (1) implementing the winter wheat subroutines after Lu et al. (2017) in CLM5; (2) implementing plant specific parameters for sugar beet, potatoes and winter wheat, thereby adding the two crop functional types (CFT) for sugar beet and potatoes to the list of actively managed crops in CLM5; (3) introducing a cover cropping subroutine that allows multiple crop types on the same column within one year. The latter modification allows the simulation of cropping during winter months before usual cash crop planting begins in spring, which is an agricultural management technique with a long history that is regaining popularity to reduce erosion and improve soil health and carbon storage and is commonly used in the regions evaluated in this study. We compared simulation results with field data and found that both the new crop specific parameterization, as well as the winter wheat subroutines, led to a significant simulation improvement in terms of energy fluxes (RMSE reduction for latent and sensible heat by up to 57 % and 59 %, respectively), leaf area index (LAI), net ecosystem exchange and crop yield (up to 87 % improvement in winter wheat yield prediction) compared with default model results. The cover cropping subroutine yielded a substantial improvement in representation of field conditions after harvest of the main cash crop (winter season) in terms of LAI magnitudes and seasonal cycle of LAI, and latent heat flux (reduction of winter time RMSE for latent heat flux by 42 %). Our modifications significantly improved model simulations and should therefore be applied in future studies with CLM5 to improve regional yield predictions and to better understand large-scale impacts of agricultural management on carbon, water and energy fluxes.</p>
  • Item
    Thumbnail Image
    On the relationship between the variability of catchment hydroclimate and physiography, and the uncertainty of runoff generation hypotheses
    Khatami, S ; Fowler, K ; Peel, M ; Peterson, TP ; Western, A ; Kalantari, Z ( 2021-03-04)
    <p>Question #20 of the UPH aspires to disentangle and reduce model prediction uncertainty. One feasible approach is to first formulate the relationship between variability (of real-world hydrological processes and catchment characteristics) and uncertainty (of model components and variables), which links the UPH theme of “modelling methods” to “time variability and change” and “space variability and scaling”. Building on this premise, we explored the relationship between runoff generation hypotheses, derived from a large ensemble of catchment model simulations, and catchment characteristics (physiographic, climatic, and streamflow response characteristics) across a large sample of 221 Australian catchments. Using ensembles of 10<sup>6 </sup>runs of SIMHYD model for each catchment, runoff generation hypotheses were formulated based on the interaction of 3 runoff generating fluxes of SIMHYD, namely intensity-based, wetness-based, and slow responses. The hypotheses were derived from model runs with acceptable performance and sufficient parameter sampling. For model performance acceptability, we benchmarked Kling-Gupta Efficiency (KGE) skill score against the calendar day average observed flow, a catchment-specific and more informative benchmark than the conventional observed flow mean. The relative parameter sampling sufficiency was also defined based on the comparative efficacy of two common model parameterisation routines of Latin Hypercube Sampling and Shuffled Complex Evolution for each catchment. Across 186 catchments with acceptable catchment models, we examined the association of uncertain runoff generation hypotheses (i.e. ensemble of modeled runoff fluxes) with 22 catchment attributes. We used the Flux Mapping method (https://doi.org/10.1029/2018WR023750) to characterise the uncertainty of runoff generation hypotheses, and a range of daily and annual summary statistics to characterise catchment attributes. Among the metrics used, Spearman rank correlation coefficient (R<sub>s</sub>) was the most informative metric to capture the functional connectivity of catchment attributes with the internal dynamics of model runoff fluxes, compared to linear Pearson correlation and distance correlation coefficients. We found that streamflow characteristics generally have the most important influence on runoff generation hypotheses, followed by climate and then physiographic attributes. Particularly, daily flow coefficient of variability (Qcv) and skewness (Q Skewness), followed by the same summary statistics of precipitation (Pcv and P Skewness), were most important. These four attributes are strongly correlated with one another, and represent the dynamics of the rainfall-runoff signal within a catchment system. A higher Pcv denotes a higher day-to-day variability in rainfall on the catchment, responded by a higher Qcv flow response. A higher variability in rainfall propagates through the catchment model and translates into a higher degree of equifinality in model runoff fluxes, which implies larger uncertainties of runoff generation hypotheses at catchment scale, and hence a greater challenge for reliable/realistic simulation and prediction of streamflow.</p>
  • Item
    No Preview Available
    Testing uncertainty in a model of stream bank erosion
    Jha, S ; Western, AW ; Rutherfurd, ID ; Grayson, RB ( 2020-01-01)
    Sediment and nutrient loads in Australian rivers are a significant management concern. The National Land and Water Audit (2002) identified bank erosion as a major source of sediment, particularly in southern Australian systems. This paper tests a method of incorporating uncertainty into and the up-scaling of a cross-section scale stream bank erosion model. The cross-section scale model is based on an understanding of fluvial erosion and mass failure processes in which fluvial erosion is estimated using an excess shear stress approach while mass failure is estimated using a limit equilibrium analysis at the cross-section scale. Figure 1 shows a schematic of the model. A Monte-Carlo framework is used to propagate input uncertainty to output uncertainty in the model and to scale up to the reach scale. Widely available databases are used to estimate variables for the two model components. A range of spatial information (GIS layers) is used to describe spatial variations in general properties such as soil type and catchment area. These are considered to be relatively well known (compared with cross-section geometry, geotechnical properties of the bank materials, riparian tree density, and hydrologic variables), although spatially coarse. A variety of empirical models and assumptions are used to transform the spatial information into model parameters, which are considered to be relatively poorly known. Two major challenges, which are related, involve incorporating the effects of natural variability along a river reach and estimating the uncertainty in the model inputs and the effect that this has on uncertainty in the model prediction. A Monte Carlo framework is used to achieve this. This involves developing a series of statistical models to predict the erosion model inputs and their (co)variability. A hierarchical approach is used to develop these input models. An attempt is first made to construct a statistical model that predicts each model parameter from available spatial information using multiple regressions. Uncertainty in these parameters is incorporated using the regression error statistics. Where cross-correlations were found to be important, these were incorporated in the generation models. Where it was not possible to develop empirical relationships with available spatial data sets, a suitable parametric distribution is fitted for those input variables for which some data is available. Where no data were available for fitting a distribution, a distribution was assumed with a shape and parameters based on heuristic consideration of the relevant processes. Once both the erosion model and the various input models were established, the Monte Carlo technique was applied. This involves generating sets of the input variables of the model from the respective stochastic input models and the running the erosion model. This allows the probability distribution for the model output to be estimated for a location in the stream network. The model is tested using historical records of plan form change from a 40km reach of the Goulburn River downstream of Eildon Dam in Victoria, Australia. The results obtained from the model are promising; with bank erosion rates being predicted within a factor of two without calibration. A series of sensitivity analyses (detail sensitivity analysis, scenario analysis, and advance sensitivity analysis) were conducted to identify key variables for predicting bank erosion rates using this particular bank erosion model. This suggested that bank angle, bank material physical characteristics, stream bed slope, and the high-flow flow regime (bankfull duration) control the behaviour of the model for loam bank materials.
  • Item
    Thumbnail Image
    A bayesian hierarchical model to predict spatio-temporal variability in river water quality at 102 catchments
    Guo, D ; Lintern, A ; Webb, A ; Ryu, D ; Bende-Michl, U ; Liu, S ; Western, A (Copernicus GmbH, 2020)
    Our current capacity to model stream water quality is limited particularly at large spatial scales across multiple catchments. To address this, we developed a Bayesian hierarchical statistical model to simulate the spatio-temporal variability in stream water quality across the state of Victoria, Australia. The model was developed using monthly water quality monitoring data over 21 years, across 102 catchments, which span over 130,000 km2. The modelling focused on six key water quality constituents: total suspended solids (TSS), total phosphorus (TP), filterable reactive phosphorus (FRP), total Kjeldahl nitrogen (TKN), nitrate-nitrite (NOx), and electrical conductivity (EC). The model structure was informed by knowledge of the key factors driving water quality variation, which had been identified in two preceding studies using the same dataset. Apart from FRP, which is hardly explainable (19.9%), the model explains 38.2% (NOx) to 88.6% (EC) of total spatio-temporal variability in water quality. Across constituents, the model generally captures over half of the observed spatial variability; temporal variability remains largely unexplained across all catchments, while long-term trends are well captured. The model is best used to predict proportional changes in water quality in a Box-Cox transformed scale, but can have substantial bias if used to predict absolute values for high concentrations. This model can assist catchment management by (1) identifying hot-spots and hot moments for waterway pollution; (2) predicting effects of catchment changes on water quality e.g. urbanization or forestation; and (3) identifying and explaining major water quality trends and changes. Further model improvements should focus on: (1) alternative statistical model structures to improve fitting for truncated data, for constituents where a large amount of data below the detection-limit; and (2) better representation of non-conservative constituents (e.g. FRP) by accounting for important biogeochemical processes.