|dc.identifier.citation||Palmer, D. J. (2008). Development of national extent terrain attributes (tanz), soil water balance surfaces (swatbal), and environmental surfaces, and their application for spatial modelling of pinus radiata productivity across new zealand (Thesis, Doctor of Philosophy (PhD)). The University of Waikato, Hamilton, New Zealand. Retrieved from https://hdl.handle.net/10289/4403||en
|dc.description.abstract||The most widely distributed and commercially important forestry crop in New Zealand is Pinus radiata D. Don. Until recently foresters have focussed on maintaining plantation management systems that are highly productive, while remaining sustainable. However, the new era of reduced carbon emissions and carbon trading means forestry systems are now viewed as potential sinks for the sequestration of carbon. Never before has the need to quantify the productive capacity of New Zealand's plantation forests at the national extent been so great. Furthermore, regions of relatively low productivity may become increasingly desirable because these sites require lower capital outlay.
In this research, a series of spatial surfaces potentially useful in the modelling and mapping of forest productivity across the national extent of New Zealand have been developed. Modelled surfaces include 15 primary and four secondary terrain attributes; 13 shortwave radiation surfaces, topographically adjusted (one annual and 12 monthly surfaces); and 39 soil water balance model surfaces (one annual and 12 monthly surfaces for fraction of available root zone water storage, available root zone water storage, and drainage).
Terrain attributes were developed using a 25-m floating point DEM and are unique and currently the best comprehensive surfaces for the following reasons.
(1) Terrain attributes comprehensively encompass the entire country compared with previous piecemeal and site-specific surfaces.
(2) Terrain attributes were modelled using a macro-catchment concept that divides the New Zealand landscape into large, naturally draining catchments to avoid the modelling problems associated with edge effects at catchment boundaries.
(3) Upslope contributing areas were calculated by switching between an FD8 algorithm that modelled flow divergence in upland regions above defined stream channels and a D8 algorithm used in low-lying areas where modelling of flow convergence is appropriate.
(4) Where appropriate, terrain attributes were corrected for undesirable spurious sinks inherent in the 25-m floating point DEM, while retaining naturally occurring sinks in karst environments, depressional lakes and wetlands. This correction provided a continuous surface that modelled flow either to a sink or continuously across the surface until reaching the sea.
The soil water balance model, SWatBal, is a dynamic spatial model that can be updated over time as new and improved data become available. SWatbal calculates the fraction of available root-zone water content, available root-zone water content, and drainage for the P. radiata species at a 100-m resolution throughout New Zealand. SWatBal was applied in this study to derive monthly mean soil water balance values, but the model can easily be adjusted to calculate any spatial extent or period. A further advance of SWatBal is the development of reasoned and allocated virtual (RAV) rainfall data. RAV consist of 365 rainfall surfaces representing the normal rainfall distribution on a monthly basis. The advantage RAV data have over monthly mean rainfall is that rainfall distribution of an actual month is used, making the data realistic, rather than assuming constant rainfall across each day for a month.
A shortwave-radiation model was developed for New Zealand at a 25-m cell-size resolution utilising a national extent DEM and a latitude surface. This shortwave radiation model encompassed slope and aspect adequately while simultaneously accounting for the influence of terrain shading. As a model it has simplicity, flexibility, and minimal computation time and storage requirements.
A partial least squares (PLS) regression technique was used to develop the surfaces of (i) stem volume mean annual increment at age thirty years for a defined reference regime of 300 stems ha-1 (300 Index), and (ii) mean top height at age twenty (Site Index) using TANZ, SWatBal and other developed and existing New Zealand spatial datasets. Together, (i) and (ii) provided the basis for a spatial model of P. radiata productivity. Initially, the 300 Index and Site Index values were calculated for 1698 permanent sampling plot (PSP) locations. For cross validation purposes, 552 PSP sites were withheld from all modelling procedures. PLS regression was used to model and predict 300 Index and Site Index values using previously developed and some existing datasets including climate, landuse, terrain, and their environmental surfaces. Best models explained 58% and 67 % of the variance for 300 Index and Site Index, respectively. The PLS models were also used to develop quantitative productivity maps across the national extent of New Zealand. In addition, a regression kriging (RK) technique was used, where ordinary kriging (OK) of the PLS model residuals was undertaken to improve model outcomes by summing the PLS and OK surfaces. Cross validation showed that prediction precision increased for both the 300 Index and Site Index RK models. However, only Site Index predictions were considered less biased using the RK technique. Findings from the commonly used and relatively straight forward spatial interpolation technique, inverse distance weighting (IDW), were compared with those derived using the more complex RK, OK, and PLS techniques. Cross validation showed that all techniques performed better than their respective data means. OK, RK, and IDW techniques were similar in prediction precision with the IDW prediction precision best for the 300 Index, and RK best for the Site Index. However, OK predictions showed reduced prediction bias. Having stated that RK, OK, and IDW interpolation techniques provided overall better predictions than PLS, it is emphasised that cross validation locations only occur within currently forested landscapes. Beyond these forested regions PLS regression has an inordinate advantage over OK and IDW prediction techniques by utilising local environmental and landform information. Additionally, there is the potential of prediction improvement through the coupling of the PLS model with its kriged regression residuals. Indeed, the main purpose of producing the 300 Index and Site Index maps was to provide empirically based predictions of regions currently without forests as much as regions with forests through spatial interpolation of existing national extent observed PSP data. Possible drivers of P. radiata productivity across 14 broad LENZ-derived environmental regimes were also assessed. It was found that generally air temperature and water balance variables were the predominate drivers.||en_NZ