Using remote sensing data and crop modelling to improve ...data.daff.gov.au/data/warehouse/9aac/9aacm/rsfcmd9... · Using remote sensing data and crop modelling to improve crop production

Using remote sensing data and crop

modelling to improve production

forecasting

A scoping study Sonja Nikolova, Sarah Bruce, Lucy Randall, Guy Barrett, Kim Ritman

and Margaret Nicholson

Research by the Australian Bureau of Agricultural

and Resource Economics and Sciences

Technical Report 12.3 September 2012

ii

© Commonwealth of Australia 2012 Ownership of intellectual property rights Unless otherwise noted, copyright (and any other intellectual property rights, if any) in this publication is owned by the Commonwealth of Australia (referred to as the Commonwealth). Creative Commons licence All material in this publication is licensed under a Creative Commons Attribution 3.0 Australia Licence, save for content supplied by third parties, logos and the Commonwealth Coat of Arms.

Creative Commons Attribution 3.0 Australia Licence is a standard form licence agreement that allows you to copy, distribute, transmit and adapt this publication provided you attribute the work. A summary of the licence terms is available from creativecommons.org/licenses/by/3.0/au/deed.en. The full licence terms are available from creativecommons.org/licenses/by/3.0/au/legalcode. This publication (and any material sourced from it) should be attributed as: Nikolova, S, Bruce, S, Randall, L, Barrett, G,

Ritman, K, & Nicholson, M 2012, Using remote sensing data and crop modelling to improve crop production forecasting: a scoping study, ABARES, technical report 12.3, Canberra, September. CC BY 3.0. Cataloguing data This publication (and any material sourced from it) should be attributed as: Nikolova, S, Bruce, S, Randall, L, Barrett, G,

Ritman, K, & Nicholson, M 2012, Using remote sensing data and crop modelling to improve crop production forecasting: a scoping study, ABARES, ABARES, technical report 12.3, Canberra, September. ISSN 189-3128 ISBN 978-1-74323-044-2 ABARES project 43369 Internet

Using remote sensing data and crop modelling to improve crop production forecasting: a scoping study is available at daff.gov.au/abares/publications Australian Bureau of Agricultural and Resource Economics and Sciences (ABARES) Postal address GPO Box 1563 Canberra ACT 2601 Switchboard +61 2 6272 2010| Facsimile +61 2 6272 2001 Email [email protected] Web daff.gov.au/abares Inquiries regarding the licence and any use of this document should be sent to [email protected]

The Australian Government acting through the Department of Agriculture, Fisheries and Forestry represented by the

Australian Bureau of Agricultural and Resource Economics and Sciences, has exercised due care and skill in the preparation

and compilation of the information and data in this publication. Notwithstanding, the Department of Agriculture, Fisheries

and Forestry, ABARES, its employees and advisers disclaim all liability, including liability for negligence, for any loss,

damage, injury, expense or cost incurred by any person as a result of accessing, using or relying upon any of the information

or data in this publication to the maximum extent permitted by law.

Acknowledgements

The authors thank the Department of Agriculture, Fisheries and Forestry’s Trade and Market Access Division for funding the

Integrated regional crop forecasting workshop, held in Canberra from 13 to 14 February 2012. Thanks to the participants of

the workshop for their invaluable contributions (see Appendix for a list of participants) and to Nathaniel Newlands,

João Soares, Alex Held and Suk-Young Hong for commenting on this paper.

http://creativecommons.org/licenses/by/3.0/au/deed.en

http://creativecommons.org/licenses/by/3.0/au/legalcode

http://daff.gov.au/abares/publications

mailto:[email protected]

http://daff.gov.au/abares

mailto:[email protected]

iii

Contents

Summary ............................................................................................................................................................ 1

1 Introduction .......................................................................................................................................... 2

2 Agricultural activities in Australia ................................................................................................ 3

Agricultural census and surveys and farm survey ................................................................. 4

3 Current approaches to crop yield forecasting in Australia ................................................. 5

National commodity forecasts ....................................................................................................... 5

State commodity forecasts .............................................................................................................. 6

Regional-scale crop yield modelling ............................................................................................ 7

4 Summary of approaches to crop yield forecasting in other countries ........................... 8

5 Proposal for an integrated crop monitoring and forecasting system .......................... 10

System requirements ...................................................................................................................... 11

Limitations and issues .................................................................................................................... 14

Appendix Integrated regional crop forecasting workshop ......................................................... 16

Glossary ........................................................................................................................................................... 18

References ...................................................................................................................................................... 20

Figures

Figure 1 Crop forecasting performance in Australia, 1985–2006 ............................................... 6

Figure 2 Prototype national crop monitoring and production forecasting system ........... 11

Maps

Map 1 Location and extent of Australian agriculture ....................................................................... 3

Using remote sensing data and crop modelling to improve crop production forecasting ABARES

1

Summary This scoping study discusses the development of an integrated forecasting system that will

achieve near real-time crop area estimates and combine them with near real-time yield

predictions to produce more accurate production estimates.

The study incorporates specialist knowledge and experience from Australia and other nations in

the use and integration of space observations in crop production estimation and other

agriculture applications. Although the prototype of this system will be applicable to the

Australian environment, it could form a template for other countries.

Crop forecasting aims to monitor crop performance and produce crop yield and production

forecasts. These forecasts provide policymakers with information on food security, trade and

market access, and the impacts of drought and other extreme climate events.

Currently, in Australia, most yield and production estimates are calculated from climate-driven

crop models and the majority are limited to the main cereal crops. They are combined with

estimates of area planted and production estimates derived primarily from local expert

knowledge. Limitations in these forecasts necessitate the use of more spatially explicit

observations of Earth’s biophysical properties. Operational remote sensing systems are being

used in several countries to monitor crop growth, climate variables and forest cover, and to map

impacts on crop production. Linking these data to traditional crop forecasting systems could

improve the accuracy of forecasts by enhancing the spatial and temporal resolution.

This report outlines the structure and requirements of a prototype of an integrated crop

forecasting system. The system will deliver and link remotely sensed and contextual data,

historical trends and model outputs at spatial and temporal resolutions appropriate for

agricultural industries. The challenges and limitations of such a system are identified and

discussed.


2

1 Introduction This scoping study discusses the integration of remote sensing data and biophysical crop growth

modelling to improve land cover monitoring and crop forecasting in Australia and globally. It

encompasses the developmental methodology presented in the KREI report (Randall et al. 2010)

and the outcomes and recommendations of the Integrated regional crop forecasting (IRCF)

workshop, held from 13 to 14 February 2012, in Canberra.

This study discusses the development of a mid to long-term crop forecasting system for

Australia, based on integrating satellite Earth observations and crop production models. It

incorporates specialist knowledge and experience from Australia and other nations in the use of

space observations in agriculture. The prototype of this system will be applicable to the

Australian environment, but could form a template for other countries.

A number of activities that use near real-time satellite data are currently carried out in Australia

and globally to assess land cover and forecast crop yield and production. These operational

geomonitoring systems, which include land surveys and crop projections, provide data on land

cover, crop condition and growth. Other environmental and climate observation systems with

associated continuous data streams from space allow monitoring of land surface and forest

cover. In addition, space data on climate variables, such as ocean temperatures, precipitation

and soil moisture, provide invaluable information for assessing the impact of climate on

agriculture, including drought monitoring (Climate R3 2011).

The linking of these data to traditional crop forecasting systems has the potential to improve the

spatial and temporal resolution of crop forecasting and provide lines of evidence to support

traditional surveys. A crop forecasting system that integrated remotely sensed information into

crop area and production forecasts would:

provide a nationwide objective crop yield and production forecast, based on objective data and observations.

enable Australia to forecast crop yields and condition more frequently and at any time.

The IRCF workshop assessed the technical feasibility of developing a crop forecasting system

that integrates remotely sensed and modelled biophysical data to predict crop production. The

workshop identified issues in current Australian and international activities, and opportunities

for further development and international cooperation. Future activities will be aligned with the

intergovernmental GEO–GLAM initiative aimed at enhancing global agricultural production

through the use of Earth observations from space.

Increasing the volume and application of remote sensing data in agriculture monitoring is

consistent with DAFF’s role in the Earth Observation from Space (EOS) program. EOS data is a

cost-effective solution that enables DAFF to routinely collect and use agricultural data.

Partnerships and information sharing with other countries are vital to this process, as Australia

depends on other agencies for supply and processing of satellite data.


3

2 Agricultural activities in Australia Australia has a landmass of 7659 million sq km and stretches approximately 4000 km west to

east and 3700 km north to south. In general agriculture in Australia is located on flatter, wetter

areas within 500 km of the coast. The main areas of agricultural activity include the Murray–

Darling Basin, the Western Australian wheat–sheep belt and coastal catchments of the eastern

seaboard including the Great Barrier Reef catchments in Queensland (Map 1).

Map 1 Location and extent of Australian agriculture

Source: ABARES

Australia is one of the largest agricultural export nations. According to ABARES (2011),

agriculture generated $36 billion in exports in 2010–11. Of this, total crops accounted for

$17.7 billion. Agricultural industries accounted for around 2.4 per cent of Australia’s GDP.

The sector employed 33 per cent of the Australians in rural and regional areas (DAFF 2010).

Agriculture utilised nearly 60 per cent of the continent or 7 687 million hectares in 2005–06

(ABS 2008). Of this figure, grazing of native pastures represented 78 per cent and grazing of

modified or sown pastures up to 15.9 per cent, cropping, 5.9 per cent, and horticulture,

0.1 per cent.

The dominant crops in Australia by area are wheat and barley. The major wheat and barley

growing regions are in the southern states. Grain sorghum, cotton and sugarcane are primarily

grown in Queensland.


4

Clearing of native vegetation for grazing has occurred since European settlement in 1788,

reaching a peak in the 1970s. State government legislation has largely eliminated the clearing of

remnant native vegetation, although substantial regrowth is still cleared regularly. The two

broad types of pasture are native and modified (or sown) pastures. Native pastures occur in

woodlands and native grasslands, and may include areas that are sown with exotic grasses such

as buffel grass, which is highly productive and can withstand heavy grazing.

Livestock grazing of native pasture is the dominant land use in Australia by area. Sheep grazing

mainly occurs in the south and beef grazing is concentrated in the north of the country.

Sown or modified pastures often occur on cultivated land as part of a crop–pasture rotation, and

in dairy or other high quality grazing systems. Modified pastures can include native pastures

with tree and shrub layers removed and planted pastures as part of a crop–pasture rotation.

Permanent sown pastures tend to be limited to dairy cattle and high value sheep and cattle

breeds, and are frequently irrigated.

Horticultural crops include perennial crops, such as fruit and vines, as well as seasonal crops,

such as vegetables. The dominant crops by area include grapes (168 790 hectares), apples

(13 640 hectares), bananas (11 168 hectares) and potatoes (35 000 hectares).

Agricultural census and surveys and farm survey

Conducted every five years, the Australian Bureau of Statistics (ABS) Agricultural Census is the

prime source of statistics on agricultural commodities. The census is sent to the 170 000 farms

with an estimated value of agricultural operations of greater than $5000.

In non-census years the ABS conducts a survey of around 30 000 farms, stratified by industry,

size of operation and region. This survey collects a similar data set to the Census and estimates

are produced at national, state and regional levels.

ABS surveys focus on aggregated production information. ABARES conducts a much smaller

annual farm survey to collect detailed financial information for around 2000 farms from the

broad acre cropping, grazing and dairy industries. These industries account for around 70 per

cent of all farms (Lubulwa et al. 2010).

The ABARES survey is also stratified by industry, size and survey region, and produces estimates

at the regional, state and national levels. The Australian grain industry is covered by 12 of the

survey regions. In order to integrate this survey information with other spatial data a range of

data capture methods have been tried. At a minimum, a single point location is recorded for each

farm. However, polygon data capture has also been used for specific projects (Davidson et al.

2006).


5

3 Current approaches to crop yield forecasting in Australia

Crop forecasting aims to monitor crop behaviour and produce crop yield and production

forecasts. These forecasts provide policymakers with information on food security, trade and

market access, and the impacts of drought and other extreme climate events. A report to the Asia

and Pacific Commission on Agricultural Statistics (FAO 2001) found the following shortcomings

are generic to crop forecasting:

forecasts are largely subjective

forecasts are largely limited to a few key commodities

forecasts are generally available only at a national scale

estimates may conflict where more than one national agency is involved.

Most crop forecasting systems in Australia focus on the nation’s most important agricultural

commodities—the major cereal crops. Crop forecasts have been developed to assist market

participants in handling and marketing crop commodities and help governments assess the

impact of drought. Modelling systems for grazing lands, including AussieGRASS, GRASP (McKeon

et al. 1990) and GrassGro (Moore et al. 1997), are used by government and industry to help

manage the impacts of climate variability, such as droughts. These systems will not be discussed

in detail in this study.

Most yield estimates are calculated from climate-driven crop models. Production forecasts are

then derived by combining estimates of yield with estimates of area planted. The information is

drawn primarily from local expert knowledge. Approaches used differ across Australia and

between the states as detailed in this chapter.

National commodity forecasts

ABARES produces the national crop production forecast as point forecasts in February, June,

September and December. The point forecasts are composite products developed from a variety

of sources available at the time the forecasts are made, and supported by discussion with

industry experts and professional judgement.

Inputs include regional yield forecasts based on a simple stress index model provided by the

Queensland Alliance for Agriculture and Food Innovation (QAAFI). Information on the seasonal

climate outlook is provided by the Australian Government Bureau of Meteorology and estimates

of planted area are sourced from bulk handlers, traders, agronomists and industry bodies.

Unforseen changes to input factors, including seasonal conditions over the forecast period,

present a risk in forecasting production. As a result, actual production can sometimes differ from

the initial point forecasts. Figure 1 illustrates the percent difference between yield predictions

and final yields for three stages of the growing season in Australia, from 1985–2006 (Walcott

2012). Highest variation in the early stage was experienced in the south-eastern states (up to 20

per cent, while South Australia showed the least difference (less than 10 per cent). Most

successful mid-term forecasts were made for South Australia (2.5 per cent).


6

Figure 1 Crop forecasting performance in Australia, 1985–2006

Source: ABARES

State commodity forecasts

QAAFI uses seasonal crop outlooks that integrate a shire-based stress index model with seasonal

climate forecasts. A similar product has been developed by the Department of Agriculture and

Food Western Australia (DAFWA). These models are run for each month of the growing season.

QAAFI runs seasonal crop outlooks for wheat and grain sorghum for local government areas in

Queensland and north-eastern New South Wales. These outlooks are updated each month

during wheat and grain sorghum growing seasons. Both outlooks use a simple agroclimatic

stress index model, either Oz-Wheat or sorghum stress index model (SSIM), which is sensitive to

water deficit or excess during the growing season. The model is integrated with actual climate

data, up to the forecasting date, and projected climate data based on the Southern oscillation

index (SOI) phase system after the forecasting date. Modelling of the yields from 1930–2009 has

shown that the variability of the yields in eastern Australia are more variable than in the south-

western Australian wheat belt, mainly because of the impact of El Niño–Southern Oscillation

(ENSO) (D Stephens 2012, IRCF workshop).

DAFWA produces a seasonal crop outlook for wheat for local government areas in Western

Australia, which is updated each month from April to December. This outlook combines the

simple stress index model (STIN) with actual climate data up to the forecasting date and

projected seasonal climate data based on the average for the last 30 years.

Simple agroclimatic models are preferred for regional crop forecasts. Stress index models

(Oz-Wheat, SSIM and STIN) use a weekly simple dynamic tipping bucket water balance model,

climate and crop-specific parameters integrated to produce a stress index (Nix & Fitzpatrick

1969). The model input parameters are selected based on the best fit when calibrated against

actual shire (local government area) yields from the ABS. Stress index model outputs are

generated at point scale and then aggregated to create a local government area-scale index. The


7

local government area-scale stress index value is transformed to yield/unit area through a

simple regression model for each local government area.

A comparative study by Hammer and colleagues (1996) found that simpler empirical models

and agroclimatic stress index approaches had better predictive ability with fewer input

requirements than some of the more complex simulation approaches, such as APSIM.

These crop models estimate up to 80 per cent of the variation in yield. Estimates of regional

yield contend with the use of average inputs for the regional scale that encompasses

heterogeneity of those inputs, inaccuracy of climate forecasts and the inability to model pests,

disease and erosion.

Regional-scale crop yield modelling

Over the past decade, increased availability of regional yield data, improvements in seasonal

forecasting, regional downscaling and modelling technology have made it possible to extend the

scale of crop modelling from point to regional. Furthermore, improved coordination among

relevant Australian EOS data users, international space agencies and satellite data providers

make the integration of remotely sensed data of vegetation greenness, crop types and/or climate

with crop yield models technically more feasible.

To generate a seasonal crop yield forecast, models are run using actual climate data up to the

forecasting date; projected climate data is used for the rest of the season. The projected climate

data can come from statistical systems with median climate and analogue years, such as ENSO,

SOI, SAM or IOD phases, as predictors. Australia has two operational seasonal climate

forecasting systems based on the statistical relationship between climate and historical

analogues and predictors (usually using the SOI phase). The seasonal outlook for rainfall is

provided as probabilities for three-month rainfall being over a given threshold. However, this

approach is less accurate during the critical winter cropping season and does not provide intra-

seasonal information on a specific amount of rainfall over an area of interest. In addition, the

linearity of these systems does not allow nonlinear processes to be accounted for, thus affecting

the accuracy of estimates.

Seasonal climate forecasting can be improved by adopting outputs from dynamical global

circulation models (GCMs) which can predict many aspects of future climate and associated

uncertainty, without relying on history. GCM output can be readily adapted for a wide range of

applications, potentially improving accuracy and reducing lead time of seasonal predictions in

Australia. The Australian Community Climate and Earth System Simulation (ACCESS) project is a

joint venture between CSIRO, BoM and five Australian universities. The project will provide the

framework for dynamical prediction across all time scales. To improve seasonal forecasting the

project team will investigate processes such as the Madden-Julian Oscillation—the dominant

mode of intra-seasonal variability—and look at better ways to assimilate ocean and climate data

into coupled dynamical models.

Hansen and colleagues (2004) studied the use of GCM-predicted seasonal rainfall using a wheat

simulation model to forecast regional and state yields in Queensland. The results show an

improved yield forecast accuracy during the pre-planting period when the SOI phase seems to be

less predictable. This encouraging result may substantially increase the role of these forecasts in

handling and marketing the Australian grain crop production.


8

4 Summary of approaches to crop yield forecasting in other countries

Most crop forecasting systems in Australia and overseas focus on the main agricultural

commodities—the major cereal crops. Approaches to yield forecasting are determined from

either climate-driven models or remotely sensed sources, but rarely in an integrated fashion.

The information on approaches to crop forecasting was collected from APEC representatives at

the Integrated regional crop forecasting workshop, held in Canberra from 13 to 14 February

2012.

Republic of Korea

Rural Development Administration (RDA) combines climate and MODIS–NDVI imagery at

national level to predict rice yields and estimate rice nitrogen and protein. According to

Statistics Korea, differences between predicted and statistical data for milled rice yields over the

period 2002–2010 were approximately 130 kg/hectare. NDVI-derived time series of rice leaf

nitrogen content explained 90 per cent of variation (R2=0.90). NDVI-based estimates of grain

protein content in rice canopies indicated different predictive power, depending on rice

varieties.

Sequential radar backscattering coefficients data from crop canopies are also used to estimate

rice and soybean growth and yields, with the L-band data performing better for both rice and

soybean. RDA is expanding the yield estimation methodologies to other grain crops, including

wheat and corn. However, there are still limitations for operational use of remote sensing

technology in crop growth and yield forecasting, mainly because of the difficulties with

acquisition of satellite data in a monsoon climate zone. The launch of radar satellites by Korea

Aerospace Research Institute in 2012 is expected to provide better data for crop monitoring.

Canada

Statistics Canada maintains the Crop Condition Assessment Program (CCAP) for drought

monitoring and wheat yield forecasting, with wheat yield forecasts varying up to ±9 per cent.

The Canadian crop yield forecaster prototype combines remote sensing and agroclimatic data to

produce in-season forecast updates. It is designed to provide near real-time information on

major crop yield and growth status for crops such as spring wheat. CCAP uses the outputs of the

Bayesian approach (a statistical regression model) that incorporates annual wheat yields and

measurements of crop water stress (SI) and growing degree days. Statistics Canada report a gain

of 10 per cent accuracy with the use of NDVI, reaching 89 per cent accuracy in the crop forecast.

China

The China CropWatch system uses remote sensing data include crop monitoring (acreage, yield,

drought and soil moisture) and crop condition assessment (crop phenophase, biomass and

nitrogen status). The system includes forecasts of seasonal and annual crop yields and

production based on the spatial cropping index derived from NDVI time series. These

applications suggest that remote sensing technology is more than capable of predicting crop

yields without the use of crop models.

http://www.rda.go.kr/foreign/eng/

http://www26.statcan.ca/ccap-peec/start-debut-eng.jsp

http://atlas.agr.gc.ca/ccyf

http://www.cropwatch.com.cn/en/index.html


9

Japan

In April 2012 Japan launched an online GIS automatic satellite data processing system JAMSS to

monitor agriculture. The system uses high time resolution and medium spatial resolution

satellite data from the Disaster Monitoring Constellation (DMC) satellite network and MODIS.

The high spatial resolution observations (2.5–32 metres) will enable farmers to watch their

fields online.

Japan Meteorological Agency (JMA) uses the Automated Meteorological Data Acquisition System

(AMeDAS) to predict the quality of rice before harvest, focusing on a low protein variety. JMA

found a high correlation between the rice protein and NDVI time series. The comparison with

measurements shows good agreement. JMA uses MODIS data and a regression model to forecast

winter wheat yields with a predictive power of R2=0.64.

United States

The National Agricultural Statistics Service of the United States Department of Agriculture

maintains CropScape—Cropland data layer. CropScape (USDA 2012) uses Landsat and MODIS

satellite imagery to estimate crop types, condition and in-season acreage. Their approach also

includes on-ground observations, human and automated computer models for identification,

deriving forecasts from models and satellite data and verification and validation using ground

truth. CropScape produces monthly crop forecasts based on growers’ reports and surveys. The

system uses average climate conditions derived from historical data.

http://nassgeodata.gmu.edu/CropScape/


10

5 Proposal for an integrated crop monitoring and forecasting system

The main aim of developing an integrated crop forecasting (ICF) system is to achieve spatially

explicit, near real-time crop area estimates that will then be combined with near real-time yield

predictions to produce more accurate production estimates.

The proposed system will help decision-makers identify and assess potential risks and

vulnerability from climate variability and extremes. In addition, quantifying crop growth and

yield will assist governments to meet the demand for food security.

We propose a conceptual approach to developing a prototype of a national integrated

forecasting system for agricultural industries that will deliver and link:

remotely sensed and contextual data

historical trends and model outputs at industry-appropriate spatial and temporal scales.

The United States and China have developed similar systems that use remote sensing data,

integrated and updated with field observations.

The prototype system will initially be developed for crop yields forecasting in Australia, with the

potential to be standardised for application in other countries.

The ICF prototype (Figure 2) links initiatives already implemented in Australia at regional scale

and will be based on recommendations made by APEC representatives at the IRCF Workshop in

Canberra in 2012.

ABARES presented a paper on the main components of the ICF prototype at the combined Group

on Earth Observations and Global Agricultural Monitoring (GEO–GLAM) meeting in Geneva in

September 2011.


11

Figure 2 Prototype national crop monitoring and production forecasting system

Source: ABARES

System requirements

Figure 2 describes a prototype for a crop forecasting system, with the inputs, outputs and

outcomes required to report on crop area and production.

Crop type and crop area estimate

Crop mask

Purpose

The crop mask will constrain the crop forecasting modelling to locations where cropping is

likely to take place.

Work underway

Land use is mapped by state agencies and modelled using Australian Bureau of Statistics

Agricultural Census data. The data are compiled using moderate resolution imagery, such as

Landsat TM and SPOT 5. The modelled output currently uses 1 km resolution NOAA AVHRR data

(Randall et al. 2010).

Australian agencies, such as CSIRO and Geoscience Australia, are working closely with space

agencies under the GEO-GLAM and the Global Forest Observations Initiative Committee on Earth

Observation Satellites’ Satellite Data Coordination Group. These groups aim to secure better

access and continuity of relevant satellite data sets for use in operational land cover and crop

monitoring. Once the data is available on a regular basis, a mask can be made of land uses such

as rain-fed and irrigated modified pastures and cropping. USDA and NASA have developed a

dynamic continuous global crop mask using MODIS data (Pittman et al. 2010) for maize, rice,


12

soybeans and wheat. This data could also be used to map crop types in a spatially explicit

fashion.

Further work to be done

Regular updates of the crop mask will ensure it reflects the current conditions. This should be

achieved by detailed analysis of satellite data requirements for crop mapping, alternative data

sources and associated processing approaches.

With new GMO/hybrid wheat varieties and increased interest in crop diversification, a land-use

change component should be integrated to capture impacts on land use of government policy-

driven incentives or other global drivers, such as biofuel. These activities will require inclusion

of data on land capability, crop suitability and soil fertility not readily available in Australia.

Validation

The mask will be checked against current crop forecasting sources for accuracy.

Crop condition (green-up)

Purpose

Accurate forecasting of crop condition requires an assured supply of medium resolution satellite

data with sufficient quality and frequency to accurately determine the date of green-up, as an

indicator of a growing crop.

Work underway

State agencies, such as DAFWA and QAAFI, use coarse-resolution (for example, MODIS, AVHRR

1km) time-series satellite imagery to detect when a crop starts to grow. The satellite imagery is

used to build up a crop profile and help identify different crops. This can be assisted by using

agricultural survey or Census data, to establish which crops normally grow in the area, and crop

calendars to identify growing season and land management practices.

Further work required

Improvements in crop-type identification from remote sensing will greatly improve the fidelity

of the resulting map products. The DAFWA and QAAFI methods should be tested across other

cropping regions of Australia. Contingency satellite data sources need to be identified and access

secured, as well as sources of other space-derived data, such as soil moisture, plant moisture

index and temperature.

To add robustness of the forecasts to possible extremes or shocks from changing inputs,

information on price volatility of relevant agricultural crop or commodities should be integrated.

This activity will make it possible to track the response of producers to different price volatility

levels within international markets. It will also be possible to incorporate the effect of

producer’s decision to grow last year’s high-yielded crops.

Validation

Crop types and growth stage derived from remotely sensed data will be checked against local

experts for accuracy.


13

Integrated crop modelling

Purpose

Research suggests that significant improvements in crop simulation models can be achieved by

integrating additional datasets and variables derived from remote sensing to scale-up crop yield

models. Further improvements could be made using new land cover dynamic datasets being

produced operationally in programs such as TERN AusCover (Auscover 2012) and TERN Soils,

combined with gridded climate variables from the Bureau of Meteorology.

Work underway

There are two experimental approaches to integrating remote sensing (RS) data into crop

models. The direct integration of RS data explores the correlation between vegetation greenness

indices (for example, NDVI) and crop yields, based on the absorption of photosynthetically

active radiation by the canopy. Some field-scale studies show that NDVI data can be integrated

into crop growth simulations to calibrate or adjust parameters during the simulation period

(run-time calibration) (Jongschaap 2006). Results show that run-time calibration of mechanistic

simulation models may improve the accuracy of predicted yields.

The other approach, involving a combination of NDVI and outputs from simulation models,

indicated an improvement in wheat yield and production forecasts. For example, Schut and

colleagues (2009) found that predictive models using remote sensing data are superior to

existing wheat yield forecasting systems in Australia, providing more accurate in-season

forecasts from the end of August onward. Results of other studies (Becker-Reshef et al. 2010;

Boken & Shaykewich 2002; Mkhabela et al. 2011; Ren et al. 2009) confirm that the predictive

power of integrated crop models is superior to that of simulation models that do not use

additional variables derived from remote sensing.

The Canadian crop yield forecaster prototype (Agriculture and Agri-Food Canada 2012) combines

RS and agroclimatic data to produce in-season forecast updates. It is designed to provide near

real-time information on major crop yield, for example, spring wheat, and growth status. The

forecaster uses the outputs of the Bayesian approach (a statistical regression model) that

incorporates annual wheat yields and measurements of a crop water stress index (SI) and

growing degree days. The agency reports a gain of 10 per cent in accuracy with the use of NDVI,

with an accuracy of 89 per cent in the crop forecast.

Further work required

Yield analysis and model parameterisations should be extended nationally and to all major

crops. Testing of alternative vegetation greenness measures should be investigated because

NDVI tends to saturate once crops reach full canopy cover.

Strategic place-based monitoring and modelling that pools information from a number of pilot

experimental test sites in different regions should be developed to provide data that can be used

to both train models and to validate integrated forecasting via a pilot-site/place-based training

component.

External factors, such as drought, floods and fire, can influence the reliability of crop forecasts. In most cases the location and spatial extent are mapped or have spatial information, including:

droughts, wind, storms, heat and frost—data provided by the Bureau of Meteorology

http://data.auscover.org.au/



14

fire, floods and natural hazards—data, such as the Sentinel bushfire website provided by Geoscience Australia (2012)

plague locusts—Locust bulletins provided by the Department of Agriculture, Fisheries and Forestry (DAFF 2012)

dust storms—DustWatch reports provided by the New South Wales Office of Environment and Heritage (OEH 2012)

pasture growth—collaborative projects Aussie GRASS, maintained by the Queensland Government (2012), and Pastures from space (2012), maintained by the Western Australian Government and CSIRO.

The impact of external factors is not always estimated. Research is required into crop area

estimation, and the link between yield and the extent, timing and severity of external factors.

Crop production forecast

Estimates of crop yields will be integrated with estimates of crop area to produce a biophysics-

based crop production forecast at high spatial and temporal resolution. At this stage,

professional judgement (heuristic approach) could be used to further refine and finalise the

forecast. The type and format of products, for example, maps and statistics, will depend on the

application and user requirements.

Outcomes

Producers and agricultural stakeholders have expressed an interest in integrated crop

forecasting system output for using mobile/wireless applications, as well as internet-based/web

portal technology. This would enable automated data assimilation, geospatial model and crop

forecast updating and rapid visualisation. It would also provide forecasts to the public/system

users via the internet and mobile devices within the mobile portal applications outcome

component.

Limitations and issues

The use of remotely sensed information has improved the predictive power of crop area and

yield models. However, as identified by IRCF workshop participants, integrated crop forecasting

systems have some technical and scientific limitations:

Inputs

Routine and consistent supply of satellite data—needs to be aligned to national and international coordination via GEO GLAM and the CEOS Satellite Data Coordination Group

Crop descriptors—phenology, crop distribution index

Better representation of soil properties—type, texture, biophysical and chemical limitations

Soil moisture—satellite measurements can be used

Climate input

o climatologies, including data on frost, land surface temperature and extreme climate

http://sentinel.ga.gov.au/acres/sentinel/index.shtml

http://www.daff.gov.au/animal-plant-health/locusts/current

http://www.environment.nsw.gov.au/dustwatch/dwreports.htm

http://www.longpaddock.qld.gov.au/rainfallandpasturegrowth/index.php?area=aus

http://www.pasturesfromspace.csiro.au/


15

o predictors, including SOI and other drivers; teleconnections

o outputs of dynamical regional climate models

Improvements in time and spatial resolution of satellite data—will improve the accuracy of predictions, but will incur higher costs

Masks—native vegetation

Data acquisition time—what is required

Impact of climate change and variability—how to incorporate into models

Cloud contamination—how to overcome this

Protein content modelling—what is required and how to incorporate in model.

Outputs

Prediction products—spatial scale

Timing of forecast—how quickly this needs to be made

Uncertainty—qualifying and quantifying uncertainties for operational real-world application.

Models

Main factors to consider in models—agroclimate variability versus technology/management

Scaling up/down—limitation of techniques

Model approach—complexity versus efficiency and accuracy

Approach in integrated crop modelling—climate data/crop models, satellite or combined approach.

Validation

Ground data—are we measuring the right things

Lack of ground-measured data—how to supplement this.


16

Appendix Integrated regional crop forecasting workshop DAFF hosted the workshop in Canberra from 13 to 14 February 2012. Participants from a range

of national and international organisations attended.

Delegate Position Organisation

Guy Barrett Scientist ABARES Department of Agriculture, Fisheries and Forestry

Dr Sarah Bruce (Chairman)

Manager, Mitigation and

Adaptation Sciences

ABARES Department of Agriculture, Fisheries and Forestry

Kelly Chow

Commodity Analyst


Fiona Crawford

Commodity Analyst


Beth Deards

Commodity Analyst


James Fell Commodity Analyst


Lee Georgeson

Scientist


Caroline Gunning-Trant Commodity Analyst


Dr Alex Held

Principal Research Scientist, Team Leader, Terrestrial Earth Observation

CSIRO Land and Water

Dr Zvi Hochman Research Scientist CSIRO Ecosystem Sciences

Kenton Lawson Economist, Productivity ABARES Department of Agriculture, Fisheries and Forestry

Dr Dath Mita

Senior Analyst/Technical Leader

International Production Assessment Division Office of Global Analysis USDA Foreign Agricultural Service

Paul Morris Executive Director ABARES Department of Agriculture, Fisheries and Forestry

Dr Nathaniel Newlands

Research Scientist

Environmental Program Agriculture and Agri-Food Canada Research Centre

Margaret Nicholson

Manager, Climate Impact Sciences


Dr Sonja Nikolova Scientist, Mitigation and

Adaptation Sciences



17

Dr Kei Oyoshi

Researcher

Earth Observation Research Centre, Japan Aerospace Exploration Agency

Dr Andries Potgeiter Research Fellow,

Crop monitoring

Queensland Alliance for Agriculture and Food

Innovation at University of Queensland

Dr Lucy Randall Senior Scientist ABARES Department of Agriculture, Fisheries and Forestry

Dr Kim Ritman Chief Scientist ABARES Department of Agriculture, Fisheries and Forestry

Dr Tom Schut Research Scientist, Crop monitoring

Curtin University

Simon Smalley Assistant Secretary Multilateral Trade Trade and Market Access Division Department of Agriculture, Fisheries and Forestry

Dr João Soares GEO Secretariat Group on Earth Observations

Dr David Stephens Climate and Modelling Science Department of Agriculture and Food Western Australia

Dr Suk-Young Hong Senior Researcher National Academy of Agricultural Science, Republic of Korea

Medhavy Thankappan Director, Science and Strategy

National Earth Observation Group Geoscience Australia

Dr James Walcott Senior Scientist ABARES Department of Agriculture, Fisheries and Forestry

Dr Koji Wakamori Chief Engineer, Space

Application Group Leader

Space Business Development Department,

Japan Manned Space Systems Corporation

Professor Wu Bingfang Chief Crop Forecaster Chinese Academy of Sciences


18

Glossary

ABARE Australian Bureau of Agricultural and Resource Economics

ABARES Australian Bureau of Agricultural and Resource Economics and Sciences

ABS Australian Bureau of Statistics

ACCESS Australian Community Climate and Earth System Simulator

AMeDAS Automated Meteorological Data Acquisition System

APEC Asia-Pacific Economic Cooperation

APRSAF Asia-Pacific Regional Space Agency Forum

APSIM Agricultural Production Systems Simulator

AVHRR Advanced very high resolution radiometer

DMC Disaster Monitoring Constellation remote sensing satellites

DAFWA Department of Agriculture and Food Western Australia

ENSO El Niño-Southern Oscillation

FAO Food and Agriculture Organization of the United Nations

GCM Global circulation model

GDP Gross domestic product

GEO–GLAM Group on Earth Observations Global Agricultural Monitoring

GIS Geographic information system

ICF Integrated crop forecasting

IOD Indian Ocean Dipole

LAI Leaf area index

LGA local government area

MODIS Moderate resolution imaging spectroradiometer

NASA National Aeronautics and Space Administration

NDVI Normalised difference vegetation index

QAAFI Queensland Alliance for Agriculture and Food Innovation

RDA Rural Development and Administration, Republic of Korea

RS Remote sensing

SAM Southern annular mode

SOI Southern oscillation index


19

SSIM Sorghum stress index model

STIN Stress index model

TERN Terrestrial Ecosystem Research Network

USDA United States Department of Agriculture


20

References ABARES 2011, Agricultural commodity statistics 2011, Australian Bureau of Agricultural and

Resource Economics and Sciences, Canberra, available at daff.gov.au/abares.

ABS 2008, ‘7125.0 Agricultural commodities: small area data, Australia 2005–06 (reissue)’,

Australian Bureau of Statistics, Canberra, available at abs.gov.au.

Agriculture and Agri-Food Canada 2012, Canadian crop yield forecasts, Government of Canada,

Ottawa, Ontario, available at atlas.agr.gc.ca/ccyf.

Auscover 2012, ‘Welcome to TERN/AusCover’, Marine and Atmospheric Research Division,

CSIRO, Canberra, available at data.auscover.org.au/.

Becker-Reshef, I, Vermote, E, Lindeman, M & Justice, C 2010, ‘A generalized regression-based

model for forecasting winter wheat yields in Kansas and Ukraine using MODIS data’, Remote

Sensing of Environment, vol. 114, issue 6, June, pp. 1312–23.

Boken, VK & Shaykewich, CF 2002, ‘Improving an operational wheat yield model using

phenological phase-based Normalized Difference Vegetation Index’, International Journal of

Remote Sensing, vol. 23, issue 20, pp. 4155–68.

China CropWatch 2012, ‘CropWatch monitoring map’, Institute of remote Sensing Application at

Chinese Academy of Sciences available at cropwatch.com.cn/en/index.html

Climate R3 2011, ‘Climate regional readiness review: background’, Commonwealth of Australia,

Canberra, available at space.gov.au/CLIMATER3/Pages/default.aspx.

DAFF 2012, ‘Current locust situation and news’, Department of Agriculture, Fisheries and

Forestry, Canberra, available at daff.gov.au/animal-plant-health/locusts/current.

DAFF 2010, Australia's Agriculture, Fisheries and Forestry at a glance 2012, Department of

Agriculture, Fisheries and Forestry, Canberra, available at

daff.gov.au/about/publications/glance2012.

Davidson, A, Beare, S, Gooday, P, Kokic, P, Lawson, K & Elliston, L 2006, Native vegetation: public

conservation on private land, ABARE research report 6.13, Australian Bureau of Agricultural and

Resource Economics, Canberra.

FAO 2001, Report of the Eighteenth Session of the Asia and Pacific Commission on Agricultural

Statistics, 6–10 November 2000, Bali, Indonesia, Regional Office for Asia and the Pacific, Food

and Agriculture Organization of the United Nations, Bangkok.

Geoscience Australia 2012, ‘Sentinel home’ [Sentinel national bushfire monitoring system],

Commonwealth of Australia, Canberra, available at

sentinel.ga.gov.au/acres/sentinel/index.shtml.

http://www.daff.gov.au/abares

http://www.abs.gov.au/


http://data.auscover.org.au/

http://www.cropwatch.com.cn/en/index.html

http://www.space.gov.au/CLIMATER3/Pages/default.aspx

http://www.daff.gov.au/animal-plant-health/locusts/current

http://www.daff.gov.au/about/publications/glance2012

http://sentinel.ga.gov.au/acres/sentinel/index.shtml


21

Hansen, JW, Potgieter, A & Tippett, MK 2004, ‘Using a general circulation model to forecast

regional wheat yields in northeast Australia’ Agricultural and Forest Meteorology, vol. 127,

pp. 77–92.

Hammer, GL, Stephens, D & Butler, D 1996, ‘Development of a national drought alert strategic

information system’, vol. 6a, Wheat modelling sub-project—Development of predictive models

of wheat production, in Report to the Land and Water Resources Research and Development

Corporation, Queensland Department of Natural Resources, Brisbane.

Jongschaap, REE 2006, Integrating crop growth simulation and remote sensing to improve

resource use efficiency in farming systems, PhD thesis, Wageningen University, The Netherlands.

Lubulwa, M, Martin, P, Shafron, W & Bowen, B 2010, ‘Statistical integration in designing

Australian farm surveys’, ABARE-BRS conference paper 10.13, Fifth international conference on

agriculture statistics, ISI Committee on Agricultural Statistics, 12–15 October 2010, Speke

Resort, Kampala, Uganda.

McKeon, GM, Day, KA, Howden, SM, Mott, JJ, Orr, DM, Scattini, WJ & Weston, EJ 1990,

‘Management for pastoral production in northern Australian savannas’, Journal of Biogeography,

vol. 17, pp. 355–372.

Mkhabela, MS, Bullock, P, Raj, S & Yang, Y 2011, ‘Crop yield forecasting on the Canadian prairies

using MODIS NDVI data’, Agricultural and Forest Meteorology, vol. 151, issue 3, pp. 385–393.

Moore, AD, Donnelly, JR & Freer, M 1997, ‘GrazPlan: decision support systems for Australian

grazing enterprises: III. Pasture growth and soil moisture submodels, and the GRASSGro DSS’,

Agricultural Systems, vol. 55, pp. 535–582.

Nix, HA & Fitzpatrick, EA 1969, ‘An index of crop water stress related to wheat and grain

sorghum yields’, Agricultural Meteorology, vol. 6, pp. 321–327.

OEH 2012, ‘DustWatch reports’, New South Wales Office of Environment and Heritage, Sydney,

available at environment.nsw.gov.au/dustwatch/dwreports.htm.

Pastures from Space 2012, Pastures from Space website, CSIRO Livestock Industries, the

Western Australian Department of Agriculture (DAWA), and the Western Australian Land

Information Authority (Landgate), Spearwood, Western Australia, available at

pasturesfromspace.csiro.au/index.asp

Pittman, K, Hansen, M, Becker-Reshef, I, Potapov, PV & Justice, C 2010, Estimating global

cropland extent with multi-year MODIS data, Remote Sensing, vol. 2, no. 7, pp. 1844–63.

Queensland Government 2012, The Long Paddock: Current rainfall and pasture growth for

Australia, Queensland Government, Brisbane, available at

longpaddock.qld.gov.au/rainfallandpasturegrowth/index.php?area=aus.

Randall, L, Bruce, S, Nikolova, S & Lawson, K 2010, ‘A review of the use of remote sensing for

forecasting crop production in Australia’, Australian Bureau of Agricultural and Resource

Economics and Sciences, Canberra, report for Korea Rural Economic Institute.

Ren, J, Yu, F, Du, Y, Qin, J & Chen, Z 2009, ‘Assimilation of field measured LAI into crop growth

model based on SCE-UA optimization algorithm’, paper presented at the, 2009 International

IEEE Geoscience and Remote Sensing Symposium, 17–12 July, Cape Town, South Africa.

http://www.environment.nsw.gov.au/dustwatch/dwreports.htm

http://www.pasturesfromspace.csiro.au/index.asp

http://www.longpaddock.qld.gov.au/rainfallandpasturegrowth/index.php?area=aus


22

Schut, AGT, Stephens, DJ, Stovold, RGH, Adams, M & Craig, RL 2009, Improved wheat yield and

production forecasting with a moisture stress index, AVHRR and MODIS data, Crop and Pasture

Science, vol. 60, no. 1, pp. 60–70.

USDA 2012, CropScape—Cropland Data Layer, United States Department of Agriculture,

Washington, DC, available at nassgeodata.gmu.edu/CropScape/.

Walcott, J 2012, Validation and external factors, paper presented at the Integrated regional crop

forecasting workshop, Canberra, 13–14 February.

http://nassgeodata.gmu.edu/CropScape/

Documents

Using remote sensing data and crop modelling to improve ...data.daff.gov.au/data/warehouse/9aac/9aacm/rsfcmd9... · Using remote sensing data and crop modelling to improve crop production