Soil water modelling in hill country, New Zealand : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Earth Science at Massey University, Palmerston North, New Zealand
Loading...
Date
2019
DOI
Open Access Location
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Massey University
Rights
The Author
Abstract
As the importance of environmental sustainability and increasing market demands have expressed pressure on New Zealand’s hill country farming systems, the more effective use of available resources and additional inputs has become crucial. Pastoral hill country farms are critical components of New Zealand’s economy, and precision agriculture solutions have been increasingly utilised to improve the sectors’ financial stability and resilience, and to satisfy the elevated expectations in yield. Profitability is dependent on pasture productivity that is highly influenced by the availability of nutrients as well as the amount of soil moisture (θv, m³ m⁻³). However, high variability of soil and landscape factors that control productivity is the primary concept describing these diverse landscapes. Hence, a study was conducted on a 2600 ha dominantly beef and sheep farm in the southern east coast of the North Island of New Zealand representing typical hill country settings. Some of the specific concerns of this research were the examination of the role of accurate, calibrated θv measurements via a wireless sensor network (WSN) (1) and the spatiotemporal variability of θv (2). Furthermore, the study investigates the potential of remote sensing for the mapping of near surface θv in sloping lands (3) and the characterisation of pasture yield patterns induced by the topography (4). These primary points were addressed to better understand the complexity occurring behind the environmental factors governing pasture yield and to potentially achieve improvement in pasture growth simulations. Systematic θv measurements have been used increasingly to inform decisions regarding fertiliser applications, feed supply and stock management in non-irrigated farming systems. To assist near real time θv and soil temperature (Ts) monitoring, 400 mm capacitance-based AquaCheck (AquaCheck, South Africa) probes (four θv and four Ts sensors per probe) were installed at 20 locations (hereinafter microsites) in predominantly silt loam soils. The spatially distributed probes were arranged into a WSN to capture data from various topographical positions. The application of manufacturer-provided calibration formula resulted in a mean root mean square error (RMSE) of 0.106 m³ m⁻³, a mean bias error of -0.099 m³ m⁻³ (indicating underestimation), and a coefficient of determination (R²) of 0.58 when correlated to directly measured reference (θv values. A single custom formula, relevant to the local soils resulted in an improved RMSE of 0.039 m³ m⁻³, while microsite-specific calibrations achieved an RMSE of 0.029 m3 m−3 and R² of 0.77. The application of a sensor-specific calibration resulted in an RMSE of 0.019 m³ m⁻³ with R² = 0.9. Sensor performance and accuracy errors were observed to vary as a function of soil wetness, bulk density (ρb, gcm⁻³) clay and total organic carbon (TOC) content. These effects were significant (P value < 0.001) but eliminated by the sensor-specific custom calibration. Sensor specifically calibrated θv was utilised the examine the effect of highly variable terrain attributes such as aspect, slope angle and soil physical properties on the θv patterns, stability and distribution both spatiotemporally and along the soil profile. Non-normal θv distribution was observed in the study period. The statistical analysis confirmed that the temporal stability of θv was higher in the deeper sections in both dry and wet seasons, while the spatial variability of θv increased with decreasing mean θv, although the greatest was in the rewetting stages. The degree of temporal persistence of the θv patterns varied with soil wetness conditions and seasons. Based on the temporal stability assessment, a representative location was selected based on a north-facing and open slope with silt loam soils. The θv distribution patterns were influenced by the topographic attributes showing that north-facing steep and moderately steep slopes were characterised with the highest variation, while east- and west-facing slopes showed similar trends.
Due to the significant variability, near surface θv mapping at a spatial resolution that would be useful for describing within farm heterogeneity has been challenging for researchers. The near surface θv modelling performance of a Random Forest (RF) ensemble learning method and the synergetic use of various remote sensing data with terrain attributes were investigated at 20x 20 m pixel size. The RF model was trained using a two-year reference dataset containing Sentinel-1 SAR backscatter data (i), normalized difference vegetation index (NDVI derived from Sentinel-2, Landsat 7 and Landsat 8 images) (ii), a number of landscape parameters (iii) and in situ near surface θv values obtained by the WSN (iv) as ground truth. The RF algorithm captured a significant amount of the complex relationships and the model predicted θv with a mean RMSE of 0.047 m³ m⁻³ and adjusted R² of 0.76 at the point scale as given by the repeated cross validation. The fine-tuned RF regressor was trained using 15 microsites and a series of near surface θv maps was developed. The maps were validated using the five left out microsites resulting in 0.049 m³ m⁻³ RMSE and 0.76 adjusted R² indicating good agreement between modelled and observed θv values. The general annual trend of θv was closely reflected in the developed maps.
The role of near surface and root zone θv, Ts, climatic variables and topographical attributes on the spatiotemporal pattern of pasture productivity was investigated at 13 selected microsites at which pasture herbage accumulation was monitored by the moveable exclusion cage method in 2016-2018. Considerable differences were found in the stored soil water response to significant rainfall events and climatic variables influencing pasture production. On the created multitable dataset, a multiple factor analysis was executed. As a result of this analysis, the role of various environmental parameters was defined highlighting the role of slope angle as the most significant determinant of pasture growth. The effect of landscape position was found to be more significant than aspect, which showed a seasonal dependence. Additionally, the contribution of terrain attributes was not consistent during the study period and changed from year to year. Ts and θv at a soil depth of 100 mm demonstrated the strongest governing effect on pasture production among the monitored parameters. In conclusion, the outcomes of this study imply that an extended and improved version of the proposed methods have the potential to be a basis of more accurate water balance simulations in complex landscapes at the regional scale. The presented quantification and isolation of the influencing topographic factors on pasture production may assist in hill country intensification by adding value to the generation of regulatory nutrient management plans. Ultimately, these advancements will enable the better characterisation of the dynamic hill country pastoral systems, which will lead towards helping hill country sheep and beef farmers to grow more pasture and increase returns while reducing the degrading effects of fertiliser applications on the environment.
Description
Listed in 2020 Dean's List of Exceptional Theses
Keywords
Soil moisture, New Zealand, Mathematical models, Pastures, Hill farming, Dean's List of Exceptional Theses