Nitrogen (N) fertilization is an important and flexible management option for increasing the growth, wood yield and profitability of many types of forest plantations, including temperate eucalypts in Australia. Process-based models are increasingly being used to simulate plantation growth. However, few have been tested on fertilizer experiments with a range of responses across a broad range of environmental conditions. The APSIM Eucalyptus model was tested for its ability to simulate observed growth responses to N fertilization by plantations of Eucalyptus globulus and E. nitens. Twenty-four experiments in temperate Australia with treatments of 0 or 400 kg ha-1 N were established in 1-to 3-year-old plantations (early treatments), with treatments reapplied after three years (late treatments), and monitoring continued for another 2 years. Twelve experiments that were well replicated and randomized provided an adequate prediction of percentage responses to fertilizer (-8-51 % observed;- 13-39 % predicted; R2 = 0.71). Most of the response (89-100 %) was attributed to the first application of fertilizer. For plantations receiving early and late fertilization, average predicted water stress accounted for 71 % of variability in mean annual increment of stem volume. Modelling scenarios suggested that early fertilization was essential for maximizing growth at most sites. Field experimentation should continue to improve knowledge of N-responsiveness in changing climates, and to evaluate the model's suitability for simulating those responses.

Abstract Post planting seedling management activities are important factors that determinegrowth and survivalof tree species. A study was conducted to evaluate the effect of management regimes on growth and survival ofCordia africana, Croton macrostachyus, Vachelliaabyssinica, and Grevillea robustatree species. The management regimes (treatments)were with and without application of fertilizer, irrigation and weeding. Management regimeswere arranged in split-split-split plot design replicated three times with tree species as main plot factor and other treatments randomized totheir assigned sub plots. Plant height, root collar diameter and survivaldata were collectedfor statistical analysis. The nonparametrictestKaplan-Meierwas used to estimatesurvival probabilities. Besides, multivariate analysis and Tukey HSD for post hoc testswereemployed for assessing management effects on growth parameters of the tree species. Accordingly, time and management regimes significantly affected growth and survival of tree species. Seedlings with one and four months after planting had better survival and growth rates than twenty months after planting. Apart from main effects fertilizer and weeding, management regimes revealed no interaction effects.Vachelliaabyssinica and Cordia africanahadhigher survival rates, whileVachelliaabyssinica revealed least in growth rate. Height and diameter of Cordia africana and Croton macrostachyushad significant and strong correlations with fertilizerand so does weeding withGrevillea robusta.Generally, management regimes improved growth and survival of the tree species. However, effect of irrigation and fertilizer rates on growth and survival of agroforestry tree speciesrequires further systematic study for packaging management recommendation.

Forest plantations can access water from some unconfined aquifers that also contain nitrate at concentrations that could support hydroponic culture, but the separate effects of such additional water and nitrogen availability on tree growth have not hitherto been quantified. We demonstrate these effects using simulation modelling at two contrasting sites supporting Eucalyptus globulus Labill. or Pinus radiata D.Don plantations. The APSIM Eucalyptus and Pinus models simulated plantation growth within 2% of observed growth where the water table was at 4 m depth for eucalypts (height 28 m, MAI 32 m3 ha−1 year−1) and at 23 m for pines (height 37 m, MAI 20 m3 ha−1 year−1). In simulations without an aquifer, observed growth could only be matched using unrealistically high surface soil nitrogen (N) supply, suggesting this is an unlikely mechanism. Simulated aquifer N concentrations, evapotranspiration, and net N mineralization and leaching (emergent properties of modelling) were similar to measured values. These results strengthen the plausibility that aquifer N uptake by plantations could be contributing to tree growth. This hypothesis warrants further research that quantifies these processes at multiple sites. Simulations included growth of herbaceous and tree weed species, and pasture, which demonstrated the utility of the process-based APSIM modelling framework for dynamically simulating carbon, water and N of plantations and other mixed-species systems.

Declining soil fertility and climatic extremes are among major problems for agricultural production in most dryland agro-ecologies of sub-Saharan Africa. In response, the agroforestry technology intercropping of Gliricidia (Gliricidia sepium (Jacq.)) and Maize (Zea mays L.) was developed to complement conventional soil fertility management technologies. However, diversified information on the profitability of Gliricidia-Maize intercropping system in dryland areas is scanty. Using data from the Gliricidia and maize models of the Next Generation version of the Agriculture Production Systems sIMulator (APSIM), this study estimates the profitability of the Gliricidia-Maize system relative to an unfertilized sole maize system. Results show significant heterogeneity in profitability indicators both in absolute and relative economic terms. Aggregated over a 20-year cycle, Gliricidia-Maize intercropping exhibited a higher Net Present Value (NPV = Tsh 19,238,798.43) and Benefit Cost Ratio (BCR = 4.27) than the unfertilized sole maize system. The NPV and BCR of the latter were Tsh 10,934,669.90 and 3.59, respectively. Moreover, the returns to labour per person day in the Gliricidia-Maize system was 1.5 times those of the unfertilized sole maize system. Sensitivity analysis revealed that the profitability of the Gliricidia-Maize system is more negatively affected by the decrease in output prices than the increase in input prices. A 30% decrease in the former leads to a decrease in NPV and BCR by 38% and 30%, respectively. Despite the higher initial costs of the agroforestry establishment, the 30% increase in input prices affects more disproportionally unfertilized sole maize than the Gliricidia-Maize system in absolute economic terms, i.e., 11.1% versus 8.8% decrease in NPV. In relative economic terms, an equal magnitude of change in input prices exerts the same effect on the unfertilized sole maize and the Gliricidia-maize systems. This result implies that the monetary benefits accrued after the first year of agroforestry establishment offset the initial investment costs. The Gliricidia-Maize intercropping technology therefore is profitable with time, and it can contribute to increased household income and food security. Helping farmers to overcome initial investment costs and manage agroforestry technologies well to generate additional benefits is critical for the successful scaling of the Gliricidia-Maize intercropping technology in dryland areas of Dodoma, Tanzania.

Agroforestry is one nature-based solution that holds significant potential for improving the sustainability and resilience of agricultural systems. Quantifying these benefits is challenging in agroforestry systems, largely due to landscape complexity and the diversity of management approaches. Digital tools designed for agroforestry typically focus on timber and crop production, and not the broader range of benefits usually considered in assessments of ecosystem services and natural capital. The objectives of this review were to identify and evaluate digital tools that quantify natural capital benefits across eight themes applicable to agroforestry systems: timber production and carbon sequestration, agricultural production, microclimate, air quality, water management, biodiversity, pollination, and amenity. We identified and evaluated 63 tools, 9 of which were assessed in further detail using Australia as a case study. No single tool was best suited to quantify benefits across each theme, suggesting that multiple tools or models could be combined to address capability gaps. We find that model complexity, incorporation of spatial processes, accessibility, regional applicability, development speed and interoperability present significant challenges for the tools that were evaluated. We recommend that these challenges be considered as opportunities to develop new, and build upon existing, tools to enhance decision support in agroforestry systems.

The use of process-based modelling of wood production in forest plantation has increased in recent decades amongst researchers and forest companies. Although, such models are used by several plantation researchers and managers, improved options and sensitivity to soil characteristics, genotype, and management options are desirable. A new generation of forest productivity modelling needs to extend previous capabilities and incorporate modern software engineering technologies. Our objective was to develop and evaluate an Agricultural Production Systems sIMulator (APSIM) Next Generation model for simulating the growth of Eucalyptus grandis and hybrids with or of E. globulus and E. urophyylla. The model simulates stem, canopy and root development, resource capture and use (light, water, N), and C and N allocation as mediated by climate, soil, genotype physiological characteristics and management. Tree dimensions (stem diameter, height, volume) are calculated as empirical functions of above-ground biomass. Datasets used for model calibration or independent evaluation were from diverse conditions in Australia (5 sites) and Brazil (13 sites), and at several of these sites there were treatments for fertilizer, irrigation or genotype. For the calibration and evaluation datasets, model performance was very good for above-ground biomass (Nash Sutcliffe Efficiency, NSE = 0.96 and 0.84 respectively). Notwithstanding this general performance, and as an example, local calibration improved performance in one of the independent test datasets, suggesting that applications of the model for specific sites or clones may benefit from parameterization to local conditions. Simulation of management for weed cover, N fertilizer and genotype are also demonstrated. As the model performed well and has high flexibility, it warrants consideration by forest plantation managers and researchers for knowledge synthesis and operational productivity predictions of Eucalyptus and other plantation genotypes.

Poor agricultural productivity has led to food shortages for smallholder farmers in Ethiopia. Agroforestry may improve food security by increasing soil fertility, crop production, and livelihoods. Agroforestry simulation models can be useful for predicting the effects of tree management on crop growth when designing modifications to these systems. The Agricultural Production Systems sIMulator (APSIM) agroforestry tree-proxy model was used to simulate the response of maize yield to N fertilizer applications and tree pruning practices in the parkland agroforestry system in the Central Rift Valley, Ethiopia. The model was parameterized and tested using data collected from an experiment conducted under trees and in crop-only plots during the 2015 and 2016 growing seasons. The treatments contained three levels of tree pruning (100% pruned, 50% pruned, and unpruned) as the main plots, and N fertilizers were applied to maize at two rates (9 or 78 kg N ha−1) as sub-plots. Maize yield predictions across two years in response to tree pruning and N applications under tree canopies were satisfactorily simulated (NSE = 0.72, RSR = 0.51, R2 = 0.8). Virtual experiments for different rates of N, pruning levels, sowing dates, and cultivars suggest that maize yield could be improved by applying fertilizers (particularly on crop-only plots) and by at least 50% pruning of trees. Optimal maize yield can be obtained at a higher rate of fertilization under trees than away from them due to better water relations, and there is scope for improving the sowing date and cultivar. Across a 34-year range of recent climate, small increases in yields due to optimum N-fertilizing and pruning were probably limited by nutrient limitations other than N, but the highest yields were consistently in the 2–4 m zone under trees. These virtual experiments helped to form hypotheses regarding fertilizers, pruning, and the effects of trees on soil that warrant further field evaluation.

Smallholder maize growers are experiencing significant yield gaps due to sub-optimal agricultural practices. Adequate agricultural inputs, particularly nutrient amendments and best management practices, are essential to reverse this trend. There is a need to understand the cause of variations in maize yield, provide reliable early estimates of yields, and make necessary recommendations for fertilizer applications. Maize yield prediction and estimates of yield gaps using objective and spatial analytical tools could provide accurate and objective information that underpin decision support. A study was conducted in Rwanda at Nyakiliba sector and Gashora sector located in Birunga and Central Bugesera agro-ecological zones, with the objectives of (1) determining factors influencing maize yield, (2) predicting maize yield (using the Normalized Difference Vegetation Index (NDVI) approach), and (3) assessing the maize yield gaps and the impact on food security. Maize grain yield was significantly higher at Nyakiliba (1.74 t ha −1 ) than at Gashora (0.6 t ha −1 ). NDVI values correlated positively with maize grain yield at both sites ( R 2 = 0.50 to 0.65) and soil fertility indicators ( R 2 = 0.55 to 0.70). Maize yield was highest at 40 kg P ha −1 and response to N fertilizer was adequately simulated at Nyakiliba ( R 2 = 0.85, maximum yield 3.3 t ha −1 ). Yield gap was 4.6 t ha −1 in Nyakiliba and 5.1 t ha −1 in Gashora. Soil variables were more important determinants of social class than family size. Knowledge that low nutrient inputs are a major cause of yield gaps in Rwanda should prioritize increasing the rate of fertilizer use in these agricultural systems.

Agroforestry parklands are a common land-use in Ethiopia and many parts of the tropics. These systems play an important role in climate change mitigation and adaptation, through carbon (C) sequestration. However, C sequestration in both tree biomass and soil has not been extensively studied for parklands of the Central Rift Valley (CRV), Ethiopia. Therefore, here we sampled a small number of F. albida trees and soil from the Adulala watershed, CRV, to provide a preliminary estimate of the C sequestration potential of these systems. Mean above-ground total dry biomass of trees was estimated at 844 kg tree(-1). Tree density was 5.80 ha(-1), which corresponded to 2.45 t C ha(-1) in above-ground biomass and 0.76 t C ha(-1) below-ground; and 118 t C ha(-1) in soil (0-80 cm depth) under trees, compared to 84 t C ha(-1) in the soil of crop-only areas. We speculate that if tree density was increased to 100 trees ha(-1), the rate of soil C sequestration could be estimated as 0.48 t C ha(-1) year(-1) for 42 years. Faidherbia albida tree density is sparse in the study area, but could be increased by encouraging farmers to protect planted seedlings or natural regeneration.

Increasing pressure on river catchment water resources has heightened the need for real-time monitoring data for water management agencies and users. This study investigates the capability to develop a technical basis for water co-management at the catchment scale. A web application (dashboard) was created that displayed real-time data for weather, soil water, river flow and water quality at high spatial and temporal resolutions, including streamflow and rainfall forecasts. A method to predict stream water temperature was developed and tested using an interactive table for visualisation of complex data. The dashboard was used to provide information and alerting to water users in an agricultural catchment, through the Ringarooma Water Users Group (RWUG) in NE Tasmania, Australia. The dashboard was updated several times per day using telemetry and data platform services, and accessed by RWUG members mainly during periods of low flow when water extraction restrictions can be applied. Irrigators gained an improved understanding of streamflow behaviour in response to weather variability, and water extractions and releases, and developed abilities to voluntarily manage the water resource as three sub-groups. The Department of Primary Industries, Parks, Water and Environment (DPIPWE) collaborated with the RWUG in this adaptive management approach, as enabled by the Ringarooma River Catchment Water Management Plan, and later informed a new Ministerial Policy for 'Water Resource Management During Extreme Dry Conditions'. Group management was assisted by using alerts sent through the existing DPIPWE text messaging service. Results showed that stakeholder understanding of technical information and scenario implications using the sensor-based dashboard increased capacity of irrigators to make decisions that improve production and social outcomes, while considering environmental values. This study demonstrates that the information provided through the dashboard enabled a high level of community and inter-agency co-operation, demonstrating potential to apply this management approach in other catchments.

A number of agroforestry models have been developed to simulate growth outcomes based on the interactions between components of agroforestry systems. A major component of this interaction is the impact of shade from trees on crop growth and yield. Capability in the agricultural production systems simulator (APSIM) model to simulate the impacts of shading on crop performance could be particularly useful, as the model is already widely used to simulate agricultural crop production. To quantify and simulate the impacts of shading on maize performance without trees, a field experiment was conducted at Melkassa Agricultural Research Centre, Ethiopia. The treatments contained three levels of shading intensity that reduced incident radiation by 0 (control), 50 and 75% using shade cloth. Data from a similar field experiment at Machakos Research Station, Kenya, with 0, 25 and 50% shading were also used for simulation. APSIM adequately simulated maize grain yield (r(2)=0.97) and total above-ground biomass (r(2)=0.95) in the control and in the 50% treatments at Melkassa, and likewise in the control (r(2)=0.99), 25% (r(2)=0.90) and 50% (r(2)=0.98) treatments at Machakos. Similarly, APSIM effectively predicted Leaf Area Index attained at the flowering (r(2)=0.90) and maturity (r(2)=0.94) stages. However, APSIM under-estimated maize biomass and yield at 75% shading. In conclusion, the model can be reliably employed to simulate maize productivity in agroforestry systems with up to 50% shading, but caution is required at higher levels of shading.

Hydrological indicators have been used for interpreting stream flow patterns in large catchments. It would be useful to evaluate such indicators for use also in headwater catchments close to the source of on-farm management impacts. Reforestation of cleared farmland is encouraged internationally for water quality and biodiversity improvements. One practice option is to establish fast-growing plantations in streamside management zones (SMZs), but impacts on stream flows are not well understood. A eucalypt plantation was established in a steep headwater catchment in temperate Australia and stream flows compared 2 years pre- and 4 years post-establishment with an adjacent unplanted catchment. The flow ratio of SMZ to control was never significantly different from one (no SMZ effect) despite a significant decreasing trend during the six-years of monitoring. The decrease in flow ratio was attributable mainly to a reduction in some higher flows resulting from increased surface roughness and interception of overland flow. Nonparametric analyses also indicated a reduction of some low flows due to plantation establishment. Overall, changes in flow patterns appeared minor, but the value placed on such changes will depend on social and environmental contexts. Because of the rarity of such data and the importance of understanding water flows as affected by plantations established on farmland, similar studies should be conducted elsewhere. The range of flow indicators tested in this case were useful and should be considered for other similar applications. In addition, a simulation capability needs to be developed to enable the effects of SMZ plantations on stream flows to be predicted for a range of climate, landscape and management contexts and to contribute further to policy and practice discussions.

Faidherbia albida is an important tree species in the parkland agroforestry system of the Rift Valley region, central and south-eastern Ethiopia. Positive effects of F. albida on crop production are widely recognised. However, the effects of tree pruning, zone and fertiliser interactions on crop growth have not been addressed in earlier studies. A field experiment containing three levels of tree pruning (100% pruned, 50% pruned, and unpruned) as main plots, and application of recommended rates of N and P fertilisers as sub-plots, was conducted during the 2015 and 2016 growing seasons. Maize grain yield and biomass, light intensity, and soil nutrients and moisture were measured at different positions from each F. albida tree trunk (0–2, 2–4 and 4–6 m) and in crop-only plots. Biomass and yield of maize were significantly greater under tree canopies compared to crop-only plots in both the 2015 and 2016 growing seasons, regardless of pruning levels. Fertilisation significantly increased yields under tree canopies compared to crop-only plots in both years. Light intensity increased with distance from trees and with greater pruning levels. Soil carbon and nutrient concentrations and moisture content decreased with increasing distance from tree and with soil depth. These results suggest that maize production and profitability could be maintained or improved through only partial pruning of F. albida rather than pollarding, and by preferentially applying fertilisers in normal and wet years. Recommendations need to be evaluated in a total system context including other rotational crops, fuel, livestock and socio-economic factors.

Vigorous Eucalyptus plantations produce 105 to 106 km ha−1 of fine roots that probably increase carbon (C) and nitrogen (N) cycling in rhizosphere soil. However, the quantitative importance of rhizosphere priming is still unknown for most ecosystems, including these plantations. Therefore, the objective of this work was to propose and evaluate a mechanistic model for the prediction of rhizosphere C and N cycling in Eucalyptus plantations. The potential importance of the priming effect was estimated for a typical Eucalyptus plantation in Brazil. The process-based model (ForPRAN – Forest Plantation Rhizosphere Available Nitrogen) predicts the change in rhizosphere C and N cycling resulting from root growth and consists of two modules: (1) fine-root growth and (2) C and N rhizosphere cycling. The model describes a series of soil biological processes: root growth, rhizodeposition, microbial uptake, enzymatic synthesis, depolymerization of soil organic matter, microbial respiration, N mineralization, N immobilization, microbial death, microbial emigration and immigration, and soil organic matter (SOM) formation. Model performance was quantitatively and qualitatively satisfactory when compared to observed data in the literature. Input variables with the most influence on rhizosphere N mineralization were (in order of decreasing importance) root diameter > rhizosphere thickness > soil temperature > clay concentration. The priming effect in a typical Eucalyptus plantation producing 42 m3 ha−1 yr−1 of shoot biomass, with assumed losses of 40 % of total N mineralized, was estimated to be 24.6 % of plantation N demand (shoot + roots + litter). The rhizosphere cycling model should be considered for adaptation to other forestry and agricultural production models where the inclusion of such processes offers the potential for improved model performance.

Thermal stratification refers to difference of temperature across water column and can act as a proxy of algal bloom. Algal bloom is a problem in Lake Trevallyn in Launceston, Tasmania. Administrators are interested in finding the causes of algal bloom and in prediction of such events in Lake Trevallyn. The results presented in this paper are the findings from a study to predict thermal stratification in Lake Trevallyn using a machine learning based approach.

A number of agroforestry models have been developed to simulate growth outcomes based on the interactions between components of agroforestry systems. A major component of this interaction is the impact of shade from trees on crop growth and yield. Capability in the agricultural production systems simulator (APSIM) model to simulate the impacts of shading on crop performance could be particularly useful, as the model is already widely used to simulate agricultural crop production. To quantify and simulate the impacts of shading on maize performance without trees, a field experiment was conducted at Melkassa Agricultural Research Centre, Ethiopia. The treatments contained three levels of shading intensity that reduced incident radiation by 0 (control), 50 and 75% using shade cloth. Data from a similar field experiment at Machakos Research Station, Kenya, with 0, 25 and 50% shading were also used for simulation. APSIM adequately simulated maize grain yield (r = 0.97) and total above-ground biomass (r = 0.95) in the control and in the 50% treatments at Melkassa, and likewise in the control (r = 0.99), 25% (r = 0.90) and 50% (r = 0.98) treatments at Machakos. Similarly, APSIM effectively predicted Leaf Area Index attained at the flowering (r = 0.90) and maturity (r = 0.94) stages. However, APSIM under-estimated maize biomass and yield at 75% shading. In conclusion, the model can be reliably employed to simulate maize productivity in agroforestry systems with up to 50% shading, but caution is required at higher levels of shading.

Agroforestry systems, containing mixtures of trees and crops, are often promoted because the net effect of interactions between woody and herbaceous components is thought to be positive if evaluated over the long term. From a modelling perspective, agroforestry has received much less attention than monocultures. However, for the potential of agroforestry to impact food security in Africa to be fully evaluated, models are required that accurately predict crop yields in the presence of trees. The positive effects of the fertiliser tree gliricidia (Gliricidia sepium) on maize (Zea mays) are well documented and use of this tree-crop combination to increase crop production is expanding in several African countries. Simulation of gliricidia-maize interactions can complement field trials by predicting crop response across a broader range of contexts than can be achieved by experimentation alone. We tested a model developed within the APSIM framework. APSIM models are widely used for one dimensional (1D), process-based simulation of crops such as maize and wheat in monoculture. The Next Generation version of APSIM was used here to test a 2D agroforestry model where maize growth and yield varied spatially in response to interactions with gliricidia. The simulations were done using data for gliricidia-maize interactions over two years (short-term) in Kenya and 11years (long-term) in Malawi, with differing proportions of trees and crops and contrasting management. Predictions were compared with observations for maize grain yield, and soil water content. Simulations in Kenya were in agreement with observed yields reflecting lower observed maize germination in rows close to gliricidia. Soil water content was also adequately simulated, except for a tendency for slower simulated drying of the soil profile each season. Simulated maize yields in Malawi were also in agreement with observations. Trends in soil carbon over a decade were similar to those measured, but could not be statistically evaluated. These results show that the agroforestry model in APSIM Next Generation adequately represented tree-crop interactions in these two contrasting agro-ecological conditions and agroforestry practices. Further testing of the model is warranted to explore tree-crop interactions under a wider range of environmental conditions.

Our aim was to quantify the effects of forest plantation and management (clear cut or 30% partial harvest) in relation to pasture, on catchment discharge in southeast Rio Grande do Sul state, Brazil. A paired-catchment approach was implemented in two regions (Eldorado do Sul and Sao Gabriel municipalities) where discharge was measured for 4 years at three catchments in each region, two of which were predominantly eucalypt plantation (mainly Eucalyptus saligna, rotation of approximately 7-9 years) with native forest and grass in streamside zones. The third catchment was covered with grazed pasture. Weather, soils, canopy interception, groundwater level, tree growth, and leaf area index were also measured. The 3-PG process-based forest productivity model was adapted to predict spatial daily plantation and pasture water balance including precipitation interception, soil evaporation, transpiration, soil moisture, drainage, discharge, and monthly plantation growth. The TOPMODEL framework was used to simulate water pools and fluxes in the catchments. Discharge was higher under pasture than pre-harvesting plantation and increased for 1-2 years after complete plantation harvest; this change was less pronounced in the catchments under partial harvest. The ratio of discharge to precipitation before harvesting varied from 7% to 13% in the eucalypt catchments and 28% to 29% under pasture. The ratio increases to 23-24% after total harvest, and to 17% after partial harvesting. The ratio under pasture also increases during this period (to 32-44%) owing to increased precipitation. The baseflow, in relation to total discharge, varied from 28% to 62% under Eucalyptus and from 38% to 43% in the pasture catchments. Hence, eucalypt plantations in these regions can be expected to influence discharge regimes when compared with pasture land use, and modelling suggests that partial harvesting would moderate the magnitude of discharge variation compared with a full catchment plantation harvesting. The model efficiency coefficient (Nash-Sutcliffe model efficiency coefficient) varied from 0.665 to 0.799 for the total period of the study. Simulation of alternative harvesting scenarios suggested that at least 20% of the catchment planted area must be harvested to increase discharge. This model could be a useful practical tool in various plantation forestry contexts around the world. Copyright (C) 2016 John Wiley & Sons, Ltd.

Accurate models for one day ahead prediction of stream flow are crucial for water management catchment scale for agriculture. This is particularly important for a country such as Australia where weather conditions can be harsh and varying. The Support Vector Regression (SVR) and the Vector Auto Regression (VAR) are standard methods used for time series prediction [12, 14]. However, since they use fixed-sized time windows, these models cannot capture long-term dependencies that are often present in stream flow time series. Recurrent Neural Networks (RNNs) do not have this weakness, yet they have not been extensively used on this type of time series. In this work, we tested various types of RNN architectures, including the recently introduced clock-work RNN (CW-RNN) [10], on two different stream flow datasets in Tasmania. We compared their accuracy with that of the SVR and VAR methods on the task of one day ahead prediction of the stream flow. In our experiments, the CW-RNNs outperformed the SVR and VAR methods across both datasets. In particular, when evaluated on the largest test set, which contained approximately 4 years of daily records, the normalised Root Mean Squared Error (nRMSE) and the Nash-Sutcliffe Efficiency (NSE) of the best CW-RNN architecture were equal to 0.166 and 0.956, respectively. This is a significant improvement over the best SVR model, which had nRMSE = 0.202 and NSE = 0.936. Our results suggest that RNNs are well suited for the task of one day ahead prediction of stream flow.