Geoscience Reference
In-Depth Information
ability in soybean yields (Batchelor et al., 2002). Although a large amount of vari-
ability is explained by these models, prediction errors are still substantial and are a
source of uncertainty for resource managers. The models are also of limited use in
off-site applications because of the level of detail of spatial variables needed for
prediction. Recent attempts to use remotely sensed data have shown promise (Basso
et al., 2001; Batchelor et al., 2002), and there is the possibility of linking models,
for example water balance models based on Digital Elevation Models linked to
Normalised Difference Vegetation Index, to highlight areas where water availability
is a major limitation on yields (Batchelor et al., 2002). In general, predictive crop
models have a reasonable level of yield prediction, but are either too vague and
general, or are more accurate but require substantial data to validate for relatively
small spatial areas.
Increasing evidence of the environmental impacts of intensive farming techniques
have led to three farming scenarios with different attitudes to prediction and man-
agement of variability: (i) increased investment in precision agriculture to reduce
spatial and temporal variability as much as possible; (ii) use of precision agriculture
to reduce variability and improve yields in selected locations, while other areas (e.g.,
buffer zones) are left 'natural' and so allow some level of natural variability; and
(iii) acceptance of natural variability and changing in farming practices to green/
organic farming, wherein crop yields are exchanged for variability (and the provi-
sion of ecosystem services, which society will pay increased prices for). This is an
example of a move back towards variability and away from single-resource predic-
tion and management.
Predictability in intensively managed ecosystems:
marine fi sheries
Marine fi sheries are less intensively managed than arable fi elds, as it is not possible
to exert the same level of control on the oceans that it is on land. Nevertheless, fi sh
populations around the world are carefully managed to obtain maximum sustain-
able yield (MSY) and consequently accurate predictions of fi sh populations and their
variability are crucial to the fi shing industry (Gaichas, 2008). Fisheries management
usually focuses on a single species, and fi sh stocks of several species have previously
crashed in various locations due to over fi shing (e.g., Roughgarden and Smith,
1996). Accurate predictions of population dynamics are essential because this will
determine what the MSY is for a given fi sh population at a given time point (often
calculated on an annual basis) to avoid population decline or collapse. Predictive
models have generally been based on observations of variability in previous catches,
sometimes supported by population monitoring. However, the focus on maximum
sustainable yield poses risks, as such predictive models are not perfect, and taking
perceived MSY does not allow for non-linearity and stochasticity in population
dynamics (see Gaichas, 2008 for a detailed critique of MSY).
Despite long-standing management practice of ignoring non-linear dynamics and
many ecosystem interactions (e.g., Hilborn and Walters, 1992), there is now a
greater focus on developing non-linear models for fi sh population prediction. For
example, the use of linear univariate time-series models (see, e.g., Czerwinski et al.,
2007) were common predictive methods, but now non-linear models, for example
artifi cial neural networks, are gaining in popularity. In order to evaluate the
effectiveness of different predictive modelling approaches, Czerwinski et al. (2007)
