In this paper we desribe a novel study which examines the adequacy of historical Australian temperature data for monitoring daily variability. Specifically, we focus on the pre-1957 period, an era for which the vast majority of daily observations are effectively unavailable to users, as they are not digitised, existing only on archived paper manuscripts. Using a station analoguing technique, we generate hypothetical datasets which replicate the pre-I957 digital network, together with a range of improved networks which could be gained through varying digitisation efforts (each at a different cost). These data are used in conjunction with the Barnes Successive Correction and Statistical Interpolation analysis methods, together with statistics on the decorrelation length-scales and observational errors of temperature, to estimate how well Australian daily temperature fields are known prior to 1957, and how this knowledge might improve through the addition of extra digital data. We show that the currently available pre-1957 digital data provides a very limited understanding of the daily temperature field. This reflects the fact that the decorrelation length-scales for the daily field are relatively modest, being about half those of monthly temperature, but more particularly the overwhelming sparsity of the available digital network. The examination of a range of improved data networks reveals the scope for a major improvement in the level of understanding, and in large areas the best network, in which all outstanding manuscripts are digitised, allows for a level of accuracy which approaches the theoretical limit. In the process we design simple cost-benefit curves, which describe how field analysis errors vary as a function of the number of stations, and hence with the financial cost of digitising. These reveal a highly non-linear relationship with, for example an outlay of approximately 2 million (165 extra stations) providing about a 40% reduction in the root-mean-square errors of analysis, while the expenditure of an additional 2-3 million (339 extra stations in total) provides fora further error reduction of around 5% for less. These complexities underscore the care which needs to be taken when designing observing networks, and the difficulties in the a priori estimation of what constitutes an adequate network for the past or future.
Jones et al. (Sun,) studied this question.