The objective of this study was to improve the current wheat yield and production forecasting system for Western Australia on a LGA basis. PLS regression models including temporal NDVI data from AVHRR and/or MODIS, CR, and/or SI, calculated with the STIN, were developed. Census and survey wheat yield data from the Australian Bureau of Statistics were combined with questionnaire data to construct a full time-series for the years 1991–2005. The accuracy of fortnightly in-season forecasts was evaluated with a leave-year-out procedure from Week 32 onwards. The best model had a mean relative prediction error per LGA (RE) of 10% for yield and 15% for production, compared with RE of 13% for yield and 18% for production for the model based on SI only. For yield there was a decrease in RMSE from below 0.5 t/ha to below 0.3 t/ha in all years. The best multivariate model also had the added feature of being more robust than the model based on SI only, especially in drought years. In-season forecasts were accurate (RE of 10–12% and 15–18% for yield and production, respectively) from Week 34 onwards. Models including AVHRR and MODIS NDVI had comparable errors, providing means for predictions based on MODIS. It is concluded that the multivariate model is a major improvement over the current DAFWA wheat yield forecasting system, providing for accurate in-season wheat yield and production forecasts from the end of August onwards.
Current methods to measure aboveground biomass (AGB) do not deliver adequate results in relation to the extent and spatial variability that characterise rangelands. An optimised protocol for the assessment of AGB is presented that enables calibration and validation of remote-sensing imagery or plant growth models at suitable scales. The protocol combines a limited number of destructive samples with non-destructive measurements including normalised difference vegetation index (NDVI), canopy height and visual scores of AGB. A total of 19 sites were sampled four times during two growing seasons. Fresh and dry matter weights of dead and green components of AGB were recorded. Similarity of responses allowed grouping into Open plains sites dominated by annual grasses, Bunch grass sites dominated by perennial grasses and Spinifex (Triodia spp.) sites. Relationships between non-destructive measurements and AGB were evaluated with a simple linear regression per vegetation type. Multiple regression models were first used to identify outliers and then cross-validated using a 'Leave-One-Out' and 'Leave-Site-Out' (LSO) approach on datasets including and excluding the identified outliers. Combining all non-destructive measurements into one single regression model per vegetation type provided strong relationships for all seasons for total and green AGB (adjusted R 2 values of 0.65-0.90) for datasets excluding outliers. The model provided accurate assessments of total AGB in heterogeneous environments for Bunch grass and Spinifex sites (LSO-Q 2 values of 0.70-0.88), whereas assessment of green AGB was accurate for all vegetation types (LSO-Q 2 values of 0.62-0.84). The protocol described can be applied at a range of scales while considerably reducing sampling time.
Remote sensing for the assessment and mapping of total standing biomass relies on accurate ground data for calibration and validation. The spatial heterogeneity of rangelands pose challenges in sampling methodologies, demanding a large number of replicate measurements that are expensive and labour demanding when working on the scale of pastoral stations. In this paper we present a ground truthing protocol that can be used for biomass estimation in heterogeneous rangeland environments, important for the development of assessments based on remote sensing or growth modelling. The protocol is based on a combination of visual estimates, crop circle NDVI, and disk-plate meter height recordings. Relationships between these indirect measurements and biomass were specific for either season or vegetation type. A combination of these measurements in a multivariate regression provided an accurate alternative, while strongly reducing the number of cuts required.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.