National Drought Early Warning System

The development of a national drought early-warning system for Australia commenced with a Queensland prototype in 1991. The need for such a system was clear. A major land degradation episode in northeastern Queensland in the mid-1980s provided clear evidence that failure to reduce stock numbers during drought can damage the land and pasture resource. A survey across northern Australia in 1991 indicated widespread deterioration in pasture and land condition (Tothill and Gillies, 1992). An improved capacity for seasonal forecasting based on the ENSO phenomenon provided hope that the impact of future droughts could be reduced if appropriate action was taken in response to early warnings of drought. With the collaboration of several state agencies and funding support from Land and Water Australia, the Queensland Department of Natural Resources and Mines has developed a national modeling framework (figure 29.4) that is being used to monitor and forecast drought across Australia (AussieGRASS; Carter et al., 2000; http://insite.nrm.qld.gov.au/resourcenet/rsc/agrass).

Spatial Modeling Framework

The AussieGRASS spatial modeling framework (figure 29.4) allows agricultural simulation models to be run at a continental scale on a 5-km grid. The framework runs on a supercomputer and calculates daily outputs simultaneously across the continent. The framework is capable of efficiently running any daily time-step biological simulation model, provided the model is recoded to simultaneously operate across all pixels. A grazing system model, GRASP, (GRASs Production) is currently incorporated in the modeling framework to operationally monitor drought across the nation.

Pasture Model

The GRASP model was developed as a generic plant growth model and has been used to simulate growth of native pastures, sown pastures, and crops. The soil water budget is simulated using four layers (0-10 cm, 10-50 cm, 50-100 cm, and a deeper layer available only to trees). Daily calculations

Figure 29.4 The AussieGRASS modeling framework for drought monitoring and alerts.

of runoff, drainage, soil evaporation, and transpiration are based on inputs of rainfall and pan evaporation. A daily plant-growth index is calculated from separate indices representing plant growth response to water stress (ratio of actual to potential transpiration), air temperature, vapor pressure deficit, solar radiation, and nitrogen availability. At low pasture cover, plant growth is calculated as a function of the plant growth index, plant density, and potential regrowth rate. As green cover increases, plant growth is calculated from a combination of temperature response, transpiration-efficiency, radiation-efficiency, and nitrogen limitations. For native pasture simulations, trees compete for water and nitrogen. Pasture biomass is calculated as the net result of pasture growth, detachment, and intake by grazing animals.

The GRASP model is calibrated against field data to obtain the main soil and plant parameters (McKeon et al., 1990). GRASP has been evaluated at a small plot and paddock scale for tropical pastures in northern Australia (over 100 locations, <29° S latitude; figure 29.1; Day et al., 1997) and temperate pastures in southern Australia (16 locations, >29° S latitude; figure 29.1; Tupper et al., 2001). Output from the spatial implementation of this model has been evaluated through broad-scale field surveys of pasture biomass (e.g., Carter et al., 2000; Hall et al., 2001) and, as described later in this chapter, against historical drought records.

Spatial Implementation of the Models

Rainfall and climate inputs are a generic requirement of all agronomic models. As the spatial framework runs daily time-step models, rainfall and climate data are input on individual 5-km grids (surfaces) for each day and climate element (Jeffrey et al., 2001). The surfaces are automatically created by interpolating records from individual recording stations, which are downloaded each day from the national climate database maintained by the NCC. Although rainfall is interpolated on a daily basis, the monthly rainfall totals are considered more accurate because they have been manually checked at BoM. Hence the daily rainfall surfaces are corrected such that they sum to equal the monthly surface as generated from the manually checked monthly rainfall totals. As a final check, the monthly rainfall surface is visually compared to the equivalent map on BoM's Web site. The framework also includes historical rainfall data from 1890 onward and daily climate data (temperature, humidity, and solar radiation) from 1957 onward (from 1970 for evaporation). Daily climate averages are being used for the 1890-1956 period until archival data from 1890 are added to the database.

Apart from the rainfall and climate inputs, the GRASP model also requires inputs of tree density, stock numbers, and specific parameters for different pasture types. These inputs are incorporated in the modeling framework as separate surfaces. Although important for determining the absolute amount of pasture production or biomass, these factors do not vary greatly from year to year and are therefore not critical for drought assessment purposes. Drought assessment is concerned more with relative measures of pasture or crop production (i.e., how production in the current season compares with historical production levels). Stored soil moisture, rainfall, air temperature, humidity, and evaporation vary considerably from year to year and contribute more to the year-to-year variation in pasture and crop production.

Drought Monitoring

Outputs from the GRASP model have been stored as monthly surfaces from 1890 onward and are updated each month. Various model outputs are available for drought assessment such as soil moisture, pasture growth, and pasture biomass. Pasture growth, ranked as a percentile against historical levels, is probably the most appropriate single index of drought in grazing lands because it is (1) a direct measure of rainfall effectiveness; (2) relatively insensitive to management practices except in the long term; and (3) highly correlated with carrying capacity of livestock.

Percentile rainfall and pasture growth maps are output on a monthly, seasonal, annual, and biennial basis. However, longer term maps may be required to assess protracted droughts. Twelve-month pasture growth percentiles are used as the operational basis for drought monitoring. A 12-month period is appropriate for analyzing drought in grazing lands because it corrects for the strong seasonality of rainfall and pasture growth in many parts of the country. For drought assessment purposes, the model output is aggregated to a district (e.g., shire) level. A threshold of pasture growth less than 10th percentile is adopted for triggering drought and a threshold of pasture growth more than the 30th percentile for breaking drought. It could be argued that the 30th percentile threshold for breaking drought is too low and that it takes an above-average season to break drought. While this is a common perception, a risk-averse manager is likely to gear normal stocking rates to a level that would be safe at least 70% of the time (i.e., to a level commensurate with 30th percentile pasture growth).

Based on the above criteria, the spatial model has been used to construct an historical time-sequence of drought in Queensland on a shire-by-shire basis (figure 29.5). This modeled time-series is in close agreement with the record of official droughts from the Queensland drought scheme, described earlier in this chapter. The overall close agreement between the two time series provides independent validation both of the AussieGRASS modeling framework and the criteria for monitoring drought. The close agreement also clearly dismisses any overall suggestion that official droughts in Queensland were declared too often in terms of frequency and duration (e.g., Daly, 1994). The major difference between the two time series occurred in the late 1960s and late 1980s, when the model calculates a higher proportion of land stricken by drought than evidenced by official drought declarations (figure 29.5). The far-north of the state is a region of major

Percent of 80 " rural land 60 -in drought 40 ■

Percent of 80 " rural land 60 -in drought 40 ■

1965

1970 1975 1980

1985 1990

1995

Figure 29.5 Percentage of rural holdings officially drought declared in Queensland and the percentage of land calculated to be in drought (line) based on thresholds of simulated pasture growth.

2000

1965

1970 1975 1980

1985 1990

1995

Figure 29.5 Percentage of rural holdings officially drought declared in Queensland and the percentage of land calculated to be in drought (line) based on thresholds of simulated pasture growth.

discrepancy in both periods. In this region pastures are of poor quality due to high rainfall and low soil nutrient status. As a result, cattle numbers are generally low in relation to the amount of forage available, and animal condition is influenced more by the length of the growing season than by the amount of pasture grown per se. Thus, a model based on pasture growth alone may not adequately represent drought for this region.

In the late 1960s, model calculations also indicated more droughts in far-western Queensland than were officially declared. In this region pasture growth is highly variable from one year to the next (coefficient of variation > 100%), and extremely low pasture growth is more the norm than the exception. Stocking rates are, in general, extremely light, and some larger companies have the capacity to move livestock in and out of the region as seasonal conditions dictate. Hence, industry is more adapted to drought in this region than in regions with more reliable pasture growth.

Drought and Pasture Condition Alerts

The GRASP model is run forward in time to estimate the probability of future drought. The probability is dependent both on current conditions (e.g., soil moisture) and the likelihood of future rainfall. Current conditions are determined by operationally running the model up to the current month. The model is then run forward for several months using rainfall and climate for each past year for which historical rainfall surfaces are available. A subset of historical years are objectively chosen as being analogous to the current year in terms of the Southern Oscillation Index (SOI) or global sea-surface temperatures (SSTs). The likelihood of drought is based on the proportion of model runs based on these analogue years for which projected pasture growth for a 12-month period is less than the 10th percentile. When combined with calculations of pasture utilization by livestock and kangaroos, the drought alert can be modified to provide pasture condition alerts as first envisaged by Pressland and McKeon (1990). The alert is triggered when high grazing pressure (i.e., >30% utilization of pasture grown over a 12-month period) is likely to occur during the periods of low pasture growth (i.e., less than the 30th percentile). Such circumstances are likely to cause losses of perennial grasses and pasture cover.

Currently the method for selecting analogue years is based on phases of the Southern Oscillation (Stone et al., 1996). Alternative approaches are also available, such as the SST scheme developed by BoM (Drosdowsky, 2002). Seasonal climate forecasting is a rapidly evolving field, and the spatial modeling framework is flexible enough to incorporate new statistical forecast schemes or downscaled outputs from global circulation models.

Use of Satellite Imagery

Satellite imagery was initially envisaged as a means of improving the spatial resolution of the modeling framework and of providing an independent drought assessment. However, the role of satellite imagery has fallen short of initial expectations, particularly in grazing lands, due to (1) the unreliability of the signal, (2) the short historical record against which to rank current conditions, (3) an inability to project forward from the current situation to provide warnings, (4) the tree cover confounding the pasture signal, and (5) the difficulty in distinguishing bare ground from dry pasture. Current studies are addressing such difficulties using Landsat data (e.g., Taube, 1999) and normalized difference vegetation index (NDVI) data (Carter et al., 2000). NDVI data are also used to monitor wheat yields (e.g., Smith et al., 1995), particularly in western Australia, jointly through Agriculture Western Australia (http://www.agric.wa.gov.au) and Department of Land Administration (http://www.dola.wa.gov.au).

Incorporation of Crop Models in the Modeling Framework

Apart from the GRASP model, the Agricultural Production Systems Simulator (APSIM) model has been tested within the national modeling framework for calculating district crop yields. APSIM is a detailed modeling framework developed for farm-scale simulations of a range of crop and farming systems. However, Hammer et al. (1996) found that for modeling wheat at regional scales, simpler approaches tailored to that the scale (e.g., the stress index [STIN] model, Stephens, 1998) were more accurate, robust, and easier to implement. STIN incorporates a daily soil water balance and calculates an accumulated crop moisture-stress index (SI) from a nominal sowing date. SI is sensitive both to moisture deficits and moisture excesses through the growing season. Inputs for the model include sowing date, daily rainfall, and average daily climate data (maximum and minimum temperature and solar radiation). SI is transformed to the district wheat yields through regression relationships between SI and historical shire yield data from the Australian Bureau of Statistics.

Yield is generally considered the best measure of drought in cropping lands (e.g., Stephens, 1998), and calculated district wheat yields from STIN are used as the basis for a drought alert issued by the Queensland Department of Primary Industries for Queensland wheat-growing shires (http://www.dpi.qld.gov.au/climate). Although alerts are only issued in

Queensland, calculations are made for all wheat-growing districts in Australia. The alert is based on the likelihood of shire yields falling below the 10th percentile. Projected yields are calculated using rainfall from analogue years based on phases of the SOI as described previously for pastures. The calculations are made in near-real time using climate and rainfall from the spatial modeling framework. Similar calculations based on STIN are also made by Agriculture Western Australia (http://www.agric.wa.gov.au/ climate).

Operational Reports

An operational monthly report, "A Summary of Seasonal Conditions in Queensland," has been developed for grazing lands in northeastern Australia (Day and Paull, 2001; http://www.LongPaddock.qld.gov.au/About Us/Publications/ByAuthor/KenDay). This report combines a range of information including rainfall, pasture growth, remote sensing, and forecasts of rainfall and pasture growth into a four-page color leaflet. A similar product, "Regional Crop Outlook: Wheat," has now been developed for wheat in northeastern Australia (http://www.dpi.qld.gov.au/climate). Each of these products can be adapted to a national basis. Drought alerts for pastures and crops, as well as pasture condition alerts, have been incorporated into a more comprehensive prototype booklet (Day and Paull, 2001). A formal survey of the recipients of the seasonal conditions leaflet endorsed the usefulness of these reports. However, the production of these reports has been suspended due to the lack of demand and funding.

0 0

Post a comment