NRM – Module 8

 

Lecture notes

 

 

Approaches and methods for regional production estimation:

an (historical) overview

 

 

 

 

 

 

 

 

 

 

 


CONTENTS

 

Introduction. 3

Methods and approaches. 4

Crop growth simulation in retrospect5

Up-scaling crop growth simulation. 7

Temporal up-scaling. 7

Using regression analysis. 7

Using seasonal weather forecasting. 8

Spatial up-scaling using a GIS/RS. 9

Using a geographic information system.. 9

Using remote sensing. 10

Direct Vi-Yield Models. 10

Indirect Yield Models. 10

Rainfall11

Insolation. 11

Land Surface Temperature. 11

Soil Moisture. 12

Leaf Area Index. 12

Crop Phenology. 13

Linking Crop Simulation Models to RS inputs and GIS. 13

Extreme Events /Disasters. 14

References. 16

 

 

 

 

 


Introduction

Since the 1960s, growing demand for food and fibber products has been met through substantial in-creases in both area and per hectare yield (FAO, 1992).  Agricultural planners face huge challenges, since nearly all land suitable for agricultural use has already been allocated.  Moreover, the total area is expected to decrease in size due to depletion of land resources beyond their recovery.  In addition, agricultural production especially in underdeveloped countries is variable due to inconstant availability of natural resources such as water.  Since it is hard to keep food availability up with current and expected population increase and per capita consumption level, food security management is becoming increasingly important.

 

To cater for these problems many attempts have been made to evaluate and finally model land use and environmental processes that determine the productivity of land.  Although there remains doubt whether processes of such complexity can be understood and modeled, progress has undeniably been made over the last decade. Before the ‘90s, the predictive strength of such models was limited since comprehensive evaluations of varying climatic-soil-crop conditions were lacking and processes were incompletely understood.  Now they are better understood and can be partly modeled. 

 

If confidence in simulation modeling increases, i.e. if conditions can be adequately modeled, crop growth simulation holds promise for early-warning applications.  A reliable forecast system would reduce risks in potentially 'bad' years and maximize returns in potentially ‘good years’.  Such a framework for monitoring and forecasting yield would be useful to a broad range of users.  Possible beneficiaries are grain producers and traders; particularly those involved in import and export, governments, financiers, farmers or farming co-operatives and production input suppliers, milling companies, extension services, and state planners, policy- and decision-makers.  Decision-makers in South Africa already use maize yield forecasts to fix the maize price in March for bad seasons or farmers are advised to change their crops to more water-stress tolerant ones.  In cases of ‘good’ seasons, milling activities are planned accordingly as well as arrangements for transport and silo storage and production loans.  This has resulted in decreased risks and overall greater profits (de Jager et al., 1998).

 


Methods and approaches

Different approaches to quantifying regional production levels classified according to the information they use are (Dehghan, 1998):

 

(a)        Air (or space) - based information.

 

Remotely sensed vegetation indices (LAI, NDVI, etc.), and satellite estimates on evapotranspiration, and indirectly crop water stress have the potential to provide real-time information over vast regions. Although such remote sensing techniques exist, they do not permit yet, for various reasons the quantitative prediction, and assessments of crop yield.  The procedures still remain challenging and up-to-date no country dares to base its policies on solely RS-based estimates independent from other data sources.

 

(b)        Ground-based information

 

There are many different methods successfully used in this approach which can be classified as:

 

i.          Forecasts based on pre-harvest crop reports.

Infiltration, systematically, flows from villages to provincial level, reporting crop conditions.  Many examples could be given here, to name just a few: crop forecasting in India, paddy yield forecasting in Indonesia (Singh & Pariyar, 1994) and crop forecasting in Germany (Stadler, 1994).

 

ii.          Time trend analysis

This is the most commonly used approach, and crop-weather time trend analysis models are the most successful methods used in most of the continents. The trend extrapolations adjusted for weather conditions of European Statistical Office and various methods applied in the former USSR and FAO's Early Warning System GIEWS and etc. are based on statistical regressions between weather or agro-meteorological conditions and crop yield (Vossen and Rijks 1995).

 

iii.         Crop growth simulation models

This is the most advanced method introduced for yield forecasting in the last decade.  A yield gap is calculated by regression analysis between simulation results as potential yield and actual yield observations from the field.  These models are either used to modify time trend yields, experience of CGMS (ibid.) or field observable land quality indicators (above ground mass or LAI) are collected during the growing season to relate the simulation results to actual land-use systems (Driessen, 1997), or simulation results and actual yields are directly regressed to build forecasting model. For example, Vossen and Rijks (1995) proposed to combine a linear time trend with the results of crop growth simulation to explain the annual variation of yield per hectare.  The Crop Growth Monitoring System (CGMS) at the Joint Research Center of the European Union is based on this premise.  Statistical analysis is used to select the most robust predictor of yield for different stages in the growing season.  To accomplish this, four indicators (potential and water-limited yield biomass and storage organ) of modeled yields are regressed against historical yields, and the most significant one is used to forecast yield.  These model results reflect the compound effect of soil-weather conditions throughout the growing season on crop growth.  To account for the influence of increasing farmers’ skill and use of technology a fifth predictor, the so-called ‘time trend’, is used.  This trend of rising yields can often be observed in official yield records (Hooijer and van der Wall, 1994).

 


Crop growth simulation in retrospect

There are many models of crop growth, ranging from the fairly simple to the extremely complex.  Unfortunately, there is a scarcity of reviews of the subject in recent years. The review of Loomis and Williams (1969), and the collection of papers from the International Biological Programme (IBP/PP) technical meeting in Trebon, Czechoslovakia, in 1969, provided a comprehensive view of the discipline at that time.  A brief review by Hesketh and Jones (1976) covers the modeling of cotton.  Thornley (1976) covered the subject of modeling and reviewed some models in his book.  Hildreth (1976) covered several of the better-known crop models.

 

To organize the existing models for a logical discussion, the classification of Hildreth (1976) was adopted: a) plant function models, b) crop growth and yield models, and c) crop development and

yield models.  If physical principles are simulated throughout, the degree of complexity generally increases from a) to c).  The plant function models concern instantaneous processes.  The crop growth and yield models integrate the plant function models at the plant or plant community scale over short time periods.  The development models extend the growth models over the growing season, including simulation of physiological events such as flowering and maturity.

There do exist crop growth and crop development models that start with empirical observations and that are more simple than the integrated process type.

 

Simulation objectives of the plant function models include radiation interception, photosynthesis rate, respiration rate, translocation or carbohydrate partitioning, leaf energy balance, transpiration rate, and root water uptake.  An extensive review of light interception models was given by Lemeur and Blad (1974).  Since then, analyses have been made by Sinclair and Lemon (1974), Anderson and Miller (1974), Norman and Jarvis (1974, 1975), Mann et al. (1977), Kimes et

al. (1980), Denholm (1981a, 1981b), Oker-Blom and Kellomaki (1982), and Sinclair and Knoerr (1982).  Models of photosynthesis have been reported by Chartier (1970), Lommen et al. (1971), Van Bavel (1975), Tenhunen et al. (1976a, 1976b), Enoch and Sacks (1978), and Thornley

et al. (1981), with a review by Thornley (1976). Respiration has been modeled by McCree (1970, 1974), Penning de Vries (1972, 1974, 1975), Penning de Vries et al. (1974), Thornley (1976, 1977), Gay (1981), Thornley et al. (1981), and others.  For a review and analysis, see Gay (1981).  Leaf energy balance was simulated by Gates (1968) and Van Bavel et al. (1973), among others.  Models of crop evapotranspiration abound.  Thornthwaite (1948), Penman (1948),

Blaney and Criddle (1962), Jensen and Haise (1963), Monteith (1965a), and Van Bavel (1966) are examples.  A comprehensive comparison of water use models was made for a volume edited by Jensen (1973).  Much work has been done on root water uptake, with Lascano (1982) giving an extensive, recent review.

 

In the third category, there exist two major classifications of models, based on the method of simulation.  The first is the multiple regression or other statistical method of empirical modeling of crop yield, development, or status, based on environmental variables, usually temperature and rainfall.  These will not be further discussed.  The second integrates the plant functions in some manner, summing the individual effects to result in growth or yield.

 

A historical perspective of the evolution of crop simulation models is useful.  One of the earlier works was by Monsi and Saeki (1953) and Kasanaga and Monsi (1954), who studied the importance of light in dry matter production, and introduced the idea of dividing the canopy into layers.  Monteith (1965b), De Wit (1965) and Duncan et al. (1967) reported models of photosynthesis that were based on interception of radiation only, and not on temperature or CO2

concentration.  Stewart and Lemon (1969) used the light interception models of Duncan et al. (1967) and De Wit (1965) in the Soil-Plant-Atmosphere Model (SPAM), which considered the microclimate in each layer of a canopy.

 

The work cited above was known at the time of the IBP/PP technical meeting in Trebon, Czechoslovakia.  In that meeting, several papers of note were given.  De Wit et al. (1970) described the Elementary Crop Simulator ELCROS.  Ross (1970), Anderson (1970), and Kuroiwa

(1970) reviewed light interception and photosynthesis models. Acock et al. (1970) discussed spatial variability of light in the canopy. Tooming (1970) discussed net photosynthesis and plant adaptation. Monsi and Murata (1970) discussed dry matter distribution in crops. Denmead (1970) and Uchijima (1970) both discussed simulations of transfer processes within canopies.  Also, the paper of McCree (1970), listed earlier, was given.

 

After the Trebon meeting, the work started in the Netherlands by De Wit (1965) and De Wit et al. (1970) continued.  Goudriaan and Waggoner (1972) described an early version of a model updated and described fully by Goudriaan (1977), and tested by Stigter et al. (1977).  The model BACROS, for Basic Crop Simulator, evolved (De Wit et al., 1978).  A third model simulated field water use and crop yield (Feddes et al., 1978).

 

In the United States, Chen et al. (1969) started work that evolved into the Nebraska corn model (Splinter, 1973; Splinter, 1974; Childs et al., 1977).  A group in Arizona developed a cotton model

(Stapleton and Meyers, 1971; Stapleton et al., 1973).  In Ohio, Curry (1971) and Curry and Chen (1971) described a dynamic model of plant growth, and Curry et al. (1975) described the soybean growth model SOYMOD I, which was used by Meyer et al. (1981) to simulate reproductive processes and senescence.  At Purdue University in Indiana, Miles et al. (1973) and Holt et.al. (1975) developed a model of alfalfa, SIMED.  In Texas, Arkin et al. (1976) and Maas and Arkin

(1978) described a simulation model of grain sorghum, SORGF.

 

Other models developed in the 1970's include the CORNMOD model of Baker and Horrocks (1974), the model of Phragmites communis reported by Ondok and Gloser (1978a, 1978b), the barley model of Kallis and Tooming (1974), the shortgrass prairie model of Conner et al. (1974),

the wheat model of Milthorpe and Moorby (1974), the tobacco model of Wann et al. (1978), and corn model of Russo and Knapp (1976).

 

The descendants of the Duncan et al. (1967) model will now be examined.  Three partial tests of their model were reported (Loomis et al., 1968; Williams et al., 1968; and Loomis and Williams, 1969), and an independent test was reported by Keener (1972) and Keener and McCree (1975).  Duncan (1971) studied crop architecture and its influence on canopy photosynthesis using the 1967 model, and included a CO2 transport routine to study the vertical profiles of CO2 within a canopy (Duncan and Barfield, 1970, 1971).  Duncan also collaborated with the group at Mississippi State University to create SIMCOT and related models of cotton (Hesketh et al., 1971, 1972; Baker et al., 1972).  The SIMCOT model series was further documented by Jones et al. (1974), and McKinion et al. (1975), who included the nitrogen balance of the cotton crop. Duncan's coauthors in the 1967 paper developed a model of sugar beet growth, SUBGOL (Fick et al., 1975; Loomis and Ng, 1977; Hunt and Loomis, 1979).

 

Meanwhile, SPAM (Stewart and Lemon, 1969), itself partially based on the 1967 model and also De Wit (1965), was generating excitement in modeling.  The initial journal article (Lemon et al., 1971) showed many researchers the potential for modeling the microclimate within crops.  Lemon et al. (1973) studied evapotranspiration with SPAM. Shawcroft et al. (1974) described SPAM and a sensitivity analysis. Van Bavel (1974) used part of SPAM and the leaf action model of Van

Bavel et al. (1973) to create CANLAM and study the behavior of sunflowers with respect to soil water potential.  This combination of SPAM and the leaf action model was used in an optimization study of water use efficiency (Ahmed, 1974; Ahmed et al., 1976).  The CO2 assimilation equations of Van Bavel (1975) were incorporated into their model, then named CANLAM2.  This was used in simulations of the efficiency of field CO2 enrichment by Takami (1974) and Takami and Van Bavel (1975), and in investigations of the effect of respiration on crop production by McCree and Van Bavel (1977).  Takami and Kumashiro (1982) used CANLAM2 to study the effect of canopy architecture on rice photosynthesis.  In unrelated work, Sinclair et al. (1977) compared the original SPAM, a simplified version of SPAM, and a "big leaf" model similar to that of Monteith (1965b).

Up-scaling crop growth simulation

Crop growth simulation requires time and space discrete information. As a result, the outputs are only valid for one location and one period. For decision-making purposes this is often inadequate, since it is not of interest to know what the current status of a particular land use systems is but to know the final outcome for multiple land use systems for a whole region. Preferably well in advance of harvesting so that for example in times of drought grain can still be purchased at reasonable (world market) price-levels. It is for this reason we examine the various techniques for up-scaling crop growth simulation in applications for food security.

Temporal up-scaling

Results from crop growth simulation reflect the compound effect of soil-weather conditions throughout the growing season on crop growth. However, these simulations cannot directly be considered as the final yields, because actual yields are in many cases considerably lower than the potential or water-limited yields due to sub-optimal cultivation practices. 

Using regression analysis

To explain the annual variation of actual or obtainable yield per hectare Vossen and Rijks (1995) proposed to combine a linear time trend with the results of crop growth simulation. The Crop Growth Monitoring System (CGMS) used at the Joint Research Center of the European Union is based on this premise.  Under this approach statistical analysis is used to select the most robust predictor of yield for different stages in the growing season. To accomplish this, four indicators of modeled yields are regressed against historical yields, and the most significant one is used for forecasting.  To account for the influence of increasing farmers’ skill and increasing use of technology on yield, a fifth indicator, the so-called ‘time trend’, is tested as well. Literature suggests that a simple linear model to describe this trend is sufficient in most cases (Swanson and Nyankori, 1979 cited by Hooijer and van der Wal, 1994).  A smooth trend of any type over a large number of years assumes a continuity that might be unrealistic.  For that reason, Hooijer and van der Wal (1994) suggest to base this indicator only on data from the recent past.  Its length should nevertheless be long enough to give a sufficient number of degrees of freedom in a regression analysis.  In practice, the length of the time series used for the statistical model validation has been set to k = 9 years (if the total length n of the available series is <k, then k=n).  In Figure 25 the time trend for Mashonaland West is given based on yield statistics provided by AGRITEX, provincial office Chinhoyi.

Figure 1: Time trend observed in maize yield statistics 1990-1998.

 

The time trend and model outputs are simultaneously taken into account and validated based on the hypothesis that weather conditions can suppress or modify the expression of a time trend.

 

Exercise:

 

Using seasonal weather forecasting

Under the latest approach to yield forecasting, a sub-research started that primarily focused on preparing adequate surrogate weather data series for crop growth modeling (Hammer and Nicholls, 1996).  These methods vary in complexity but have in common that they forecast a season’s climatic patterns with the implicit assumption that the variability of future climate will be similar to that of the past. At best such an analysis of historical climatic conditions can provide an envelope in which season forecasts can be fitted; the direction and extend of the variance cannot be predicted. Hence, modeled yields may prove inaccurate.

 

Scientists aware of this drawback attempted to improve the ‘robustness’ of weather forecasting itself.  Stone et al. (1996) who showed that southern oscillation index (SOI) phase system provides an accurate predictor of rainfall in certain regions of the world.  The SOI method considers 'phases' of the SOI; that is, the method uses both change and value of the SOI to derive cumulative rainfall probability distributions for any location.  Because every month for every year can be placed into a particular analogue of months, corresponding historical records can be placed together to take out daily rainfall, evaporation, temperature, and radiation. These data can be used as input for a crop simulation model.  Even though data demanding, El Niño Southern Oscillation (ENSO) based weather forecasts remain useful for preliminary estimation of regional production since they are available well in advance of the actual production conditions and with reasonable accuracy. Thus, timely information on expected weather conditions of a coming season remain useful, specifically if predictions were to have any value for farm management. Temporal and spatial characterization of ENSO for rainfall estimation is made possible by the Agro-meteorological Rainfall Pattern Analysis and Forecast model (ARAF) as developed by Venus (Venus, 2000). Based on this model he proved that roughly 75% of the rainfall variability and a corresponding 80% of maize yield variability could be explained for Mashonaland West Province, Zimbabwe. Further justification of this technique can be found in Meinke and Hammer (1998) who demonstrated that highly significant differences in peanut yields in Australia exist among seasons grouped according to the SOI phases of Stone.  De Jager et al. based their weather forecast component of a calibrated CERES—maize model on this same principle, with the intention to forecast the extent and severity of drought in maize in the Free State Province of South Africa one month before the growing season started.  The accuracy of this type of forecasting system is yet uncertain but the high correlation value (r²=0.86) for simulated versus actual yields is an encouraging sign (de Jager et al. 1998).  Hodges et al. (1987) cited by de Jager et al. (1998) selected appropriate analogue historical weather data series, depending upon the 90-day weather outlook (below, above or normal).  Randomized weather data series generation (e.g. the climate model, Weathergen) is a possibility, as is the use of the daily rainfall data series generator of Zucchini and Adamson (1984) as cited by de Jager et al. (1998).  Lourens and de Jager (1997) forecast weather data within a growing season with historical data series that had delivered lower quartile, median and upper quartile seasonal rainfall (de Jager et al., 1998).  Fouché (1992) cited by de Jager et al. (1998) constructed seasonal rainfall scenarios of composite monthly rainfall data from historical meteorological records, assuming that each month received median monthly rainfall.  De Jager and Singels (1990) used combinations of daily sunshine, maximum, and minimum temperature and daily rainfall data selected randomly from historical data series (de Jager et al., 1998).  McKeon (1996) sited by de Jager et al. (1998) simulated forage yields by completing the season with 5-10 analogue years of weather data from which he determined the mean and coefficients of variation. 

Spatial up-scaling using a GIS/RS

In estimating regional production levels, classical approaches such as crop growth simulation developed amongst others by the “de Wit school”, Wageningen University, The Netherlands, heavily rely on ground observations, crop parameters, and meteorological observation from synoptic weather stations. For example, WOFOST is a member of this family of models yields fairly accurate results, and was further developed by several scientists, notably van Keulen and Wolf, 1986; Van Diepen et.al., 1988; Van Keulen and van Diepen, 1990 (Supit et.al., 1994). Unfortunately, for operational applications this model was of little use. The European Union for example requires regional explicit estimates of crop production levels for its subsidy control measures for all EU-member states, but unfortunately WOFOST 6.0 is a point-based model. To be useful, spatial up-scaling is required.

Using a geographic information system

Point observations of soil, weather and crop parameters can be interpolated into a regular grid, a technique also referred to as “objective analysis”. This technique was adopted by the Joint Research Center of the European Union (EU) in order to make the simulations of WOFOST regional. The resulting system, also referred to as Crop Growth Monitoring System (CGMS) is currently used for quantified crop growth simulation for all EU member states. A simulation unit was defined as a combination of a crop number (which determines the crop parameters), a grid number (which determines the weather data) and a combination of the soil parameters, Soil Physical Group (SPG) and Rooting Depth (RD). The SPG together with the RD determined the necessary soil parameters for a simulation run.  This implies that all soils that are suitable for a crop have the same SPG/RD combinations and consequently will have the same yield if lying within the same grid. A unique Soil Mapping Unit (SMU) covers each EMU.

Figure 2, Example of an Elementary Mapping Unit

Using remote sensing

Crop yield assessments in third world countries can be a difficult task owing amongst other reasons to the scarcity of such data. Satellite-based remote sensing (RS) provides a vantage point in space to view large areas repeatedly in different regions of electromagnetic spectrum, much beyond the capabilities of human eye. By analyzing the radiation, reflected or emitted by objects, it is possible to detect, identify and classify various objects/ phenomena.

Direct Vi-Yield Models

Various Vegetation Indices (VI) have been used in RS literature by combining red and near-infrared and sometimes additional spectral channels, in ratio or other algebraic expressions, as a means of expressing crop vigour, LAI or biomass. The direct VI-yield empirical approach is based on results from a large number of ground studies where high correlation between VI at specific stage and final grain yield has been observed. The scientific rationale for direct linear VI-yield relation is provided by Spectral Component Analysis (SCA) approach of Wiegand et al (1986), which relates asymptotic increase in both VI and yield to LAI, resulting in the following:

 

LAI/VI X YIELD/LAI = YIELD/VI

 

This equation relates the two identities through LAI/VI and also indicates a direct VI –yield model.

 

A number of studies have been carried out in India that have related VI to district-level crop yields. The experience on use of such models for yield forecasting indicates that additional information in form of weather improves performance and a number of RS data normalization steps and zonation of districts is essential (Dadhwal and Ray, 2000). The direct approach has been applied to field scale and extended to multi-date data and a recent review on wheat yield modeling provides an overview of the Indian experience (Dadhwal et al., 2003).

Indirect Yield Models

The role of RS in agrometeorology could first be explained in terms of parameters that can be obtained from RS data. This could then be extended to description and modeling of processes, such as crop growth, energy and mass balance over crop canopies and lastly to real-life applications such as agromet advisory, precision agriculture, crop forecasting, cropping systems, agroclimatic zonation, sustainability and vulnerability to global change. The parameters of interest to agrometeorology and amenable to remote sensing-based assessment of three major categories are (a) Basic Weather Parameters: Insolation, rainfall, PAR, land surface temperature, etc., (b) Energy and water balance components: Albedo, net radiation, soil moisture, actual evapotranspiration, etc., and (c) Crop Growth: leaf area index, phenology, etc. It may also be mentioned that RS is now used with other technologies such as geographic information system (GIS). The RS-derived parameters also are an ideal way to merge models runs and observations to provide continuous spatial and temporal fields. The models that have now been coupled to RS inputs cover a wide range of SVAT (soil vegetation atmosphere transport), LSM (land surface models), and CSM (Crop Simulation Models). Sensors onboard satellites in geostationary orbit, (ca. 36,000 km above earth, e.g., INSAT, METEOSAT, GOES) provide high repetivity (typically 30 minutes for INSAT VHRR) but at poor spatial resolution, and are used for meteorology applications. Satellites in polar orbit, which orbit earth in an inclined north-south plane at 500-900 km above earth, are generally preferred for earth observations. Sensors onboard these satellites provide global coverage, a fixed repetivity (1-35 days) and capable of very high spatial resolution (currently best civilian resolution is under 1m). Satellite meteorology began with the launch of TIROS-1 in 1960 by USA and it gained momentum with the launch of the NOAA-Advanced Very High Resolution Radiometer (AVHRR) having visible and thermal (10.5-12.5 m) channels. The main uses of the AVHRR has been in analysis of cloud motion vectors (CMV), outgoing longwave radiation (OLR), quantitative precipitation estimates (QPE), sea surface temperature (SST) and cyclone tracking. These applications focus on meteorology, climatology or as inputs to numerical weather prediction models, but do not directly find use in agrometeorology. However, unintended applications of the AVHRR sensor data, such as land surface temperature proved to be also useful for agrometeorology.

 

The following is an introduction to the vast and growing field of use of RS in agrometeorology with emphasis on Indian experiences.

Rainfall

Rainfall over land has been estimated using thermal data from INSAT VHRR using cold cloud duration algorithm by India Meteorology Department. These are on monthly scale and over 2.5x2.5 degree grid cells and Roy Bhowmick and Sud (2003) have recently evaluated its errors. These have application in climatology and are not useful for agro-meteorological applications.

 

Globally, the rainfall retrieval from space has become a reality with the Tropical Rainfall Measuring Mission (TRMM), a joint mission between NASA of USA and NASDA of Japan. It carries TRMM Microwave Imager (TMI), the precipitation radar (PR) and the Visible and Infrared Radiometer System (VIRS) (Kummerow et al., 1998). Coincident measurements TMI and PR are complementary. The frequency dependence of electromagnetic properties of cloud and precipitation particles allows for the design of multi-channel passive microwave radiometers which can "sound" to different depths in a precipitating cloud, but the height assignment of cloud properties is not very specific. Active microwave sensors (radars) provide specific height information based upon the time delay of the precipitation-backscattered return power. The VIRS on TRMM adds cloud-top temperatures and structures to complement the description of the two microwave sensors. Daily rain rate products are being produced and improved space-based precipitation missions are planned. There have not been many studies on use of passive microwave data for rainfall over land in India.

Insolation

A number of algorithms have been developed for insolation retrieval, such as, (a) empirical regression between visible clear radiance (VCR) from geostationary satellites (Brakke and Kanemasu, 1983) and ground measurements, or with single acquisition of NOAA-AVHRR, (b) using multiple acquisitions per day to derive atmosphere turbidity and cloudiness and simple parameterisation to compute insolation, e.g. Modified Heliosat method (Rigollier and Wald, 2000), day-night algorithm (Rosema, 1993), and (c) physical models of varying complexity applicable to multiple acquisitions per day geostationary (Tanahashi et al., 2001) sensor.

 

Recently, Bhattacharya et al (2001b) have compared day-night, Heliosat and VCR approaches on METEOSAT data and found Heliosat algorithm to outperform other two approaches. Kimothi et al (2003) retrieved and validated daily and five-day average daily total insolation using METEOSAT data over Indian landmass in clear and cloudy skies. The accuracy of retrieval was 88% for pooled datasets over three winter months.

Land Surface Temperature

Land surface temperature (LST) is retrieved from thermal channels (10-12 m) and requires information on emissivity, correction for atmospheric water vapour and view angle for accurate retrieval. Based on sensor characteristics, three major categories of algorithms have been developed, (a) Mono-window method, (b) Split-window methods, and (c) Multi-angle method. Technique (a) gives least accuracy but can be applied when only single thermal channel, such as in INSAT VHRR, is available. Split-window was developed to use two-channel of NOAA-AVHRR and is able to compensate for atmospheric water vapour but requires an emissivity retrieval approach to work on land and biases 2-3K can be obtained. The multi-angle method can work with advanced sensors such as ASTER and allow separation of component temperatures, such as foliage and soil.

 

A number of studies on estimation of LST over agricultural regions have been carried out in India. Gupta et al (1995) have retrieved land surface temperature (LST) and surface albedo over north-west India using NOAA AVHRR LAC images and developed relationship between LST and NDVI. RaviKumar et al (1999) compared NOAA/AVHRR derived LST with screen air temperatures and observed LST to be within 5 degrees of air temperatures. Using NOAA AVHRR LAC data Bhattacharya et al (2000, 2001a) retrieved land (soil-vegetation complex) surface temperature (LST) over Land Surface Processes Experiment (LASPEX) sites in Gujarat using NDVI-emissivity relations developed by Casselles et al (1997). The LST were midway between air temperature (AT) and soil temperature (ST) during winter (January) and closer to ST in summer (April).

Soil Moisture

Soil moisture is highly variable in space and time, and thus it is difficult to collect reliable information over a large area using conventional methods of point measurements. Generally thermal and microwave sensors are used for surface soil moisture estimation. Thermal IR techniques include (i) thermal inertia (van de Griend et al., 1985) (ii) temperature-vegetation triangular space (Gillies and Carlson, 1997) (iii) Directional radiometry (Francois, 2003) and (iv) morning rise in land surface temperature (Wetzel and Woodward, 1987). Microwave radiations are capable of penetrating clouds and vegetation cover and hence microwave sensors offer great potential in soil moisture determination over cropped areas. The measurement of soil water content by means of microwave sensors relies on the large difference between the dielectric properties of dry soil and liquid water. Microwave RS techniques have been shown to provide soil moisture estimates in the upper layer of soils up to 10cm. Both active and passive sensors have been used for soil moisture estimation. The major drawback with the passive sensors is their poor spatial resolution.

 

With the launch of the AQUA satellite in June 2002 and ADEOS II in December 2002, data from the Advanced Microwave Scanning Radiometer (AMSR) are becoming available at two equatorial overpass times per day. The C-band of the AMSR (6.9 GHz) has better sensitivity than the 19.4-GHz channels of the Special Sensor Microwave Imager (SSM/I) for retrieving soil moisture. In the past, soil moisture has been measured from space using SSM/I with the 19-, 37-, and 85-GHz channels. This sensitivity is highest at lower frequencies (L-band: 1.4 GHz and C-band: 6.9 GHz) and decreases as the frequency of observation increases due to increased contribution from the atmosphere and vegetation.

 

Thermal inertia approach based on mid-morning rise in LST from 0830 to 1130 hours as obtained from INSAT VHRR Infra red channel was investigated by Pathak et al (1993). The change in LST in September (1991) was found to be negatively correlated with area averaged rainfall over Gujarat. Bhattacharya and Dadhwal (2003b) used NOAA AVHRR thermal data and vegetation index triangular space to characterize surface soil moisture availability at 8km spatial resolution over Central-Northwest India. Rao et al (2001) have related microwave brightness temperatures at 6 and 10 GHz NIMBUS-SMMR data of 1984 and 1987 to API (apparent precipitation index) at 150 km grid cell. This field-level estimates could be done using SAR (Synthetic Aperture Radar) which has high spatial resolution (5-50 m). A number of studies have been carried out in India.

Leaf Area Index

The leaf area index (LAI) is one of the most important parameters characterizing a canopy and RS-based LAI estimation would greatly aid the application of LAI as input to models of photosynthesis, crop growth and yield simulation models, evapotranspiration, estimation of net primary productivity and vegetation/ biosphere functioning for regional applications. A number of techniques for space borne remote sensing data have been developed/tested, ranging from regression models to canopy reflectance model inversions with varying successes, which include (1) statistical models that relate LAI to band radiance (Badhwar et al., 1986) or develop LAI-vegetation index relation (Myneni et al., 1997), (2) biophysical models like Price (1993), and (3) inversion of canopy reflectance using numerical model or LUT based model (Knyazikhin et al., 1998). Myneni et al (1997) developed a simple approach for estimating global LAI from atmospherically corrected NDVI using NOAA-AVHRR data. One- or three-dimensional radiative transfer models were used to derive land cover-specific NDVI-LAI relations of the form

 

LAI = a x exp (b x NDVI + c)

 

where, coefficients a and c are determined by vegetation type and soil. With MODIS, onboard TERRA (launched in Dec. 1999), an eight-day composite ‘LAI product’, at a spatial resolution of 1km, which incorporates a canopy-radiation model and look-up-table based LAI retrieval algorithms (Knyazighin et al., 1998) is produced operationally. Pandya et al (2002) described results of a study to develop small area LAI maps using IRS-LISS-III data using field sampling and regression approach and using the generated maps to validate MODIS LAI product over Bhopal and Indore. Using NDVI of test fields, empirical models based on site-specific NDVI-LAI relation were developed and used to generate LAI maps for each acquisition and study site. The LAI images were aggregated to 1km spatial resolution and compared with MODIS LAI product and results indicated significant positive correlation between LAI derived from LISS-III data and MODIS data albeit with a positive bias, in the MODIS product.

 

Rastogi et al (2000) tested Price model on farmers fields during 1996-97 season in Karnal (Haryana, India) and 1997-98 in Delhi using IRS LISS-III data and estimated wheat attenuation coefficients. The root mean square error (RMSE) between RS estimates and ground measured LAI ranged between 0.78-0.87 when LAI was in the range of 1-4, while for higher LAI range (4-6), the RMSE varied from 1.25 to 1.5 in two sites. Such errors can severely reduce utility of a model using field-level LAI as input.

Crop Phenology

Crop phenology can be derived from temporal behaviour of vegetation indices (VI) such as NDVI. Coarse-resolution high repetivity data is preferred, since a maximum value compositing (MVC) algorithm can compensate for cloud and atmosphere caused limitations. While composited NDVI such as 10-day NOAA-AVHRR Pathfinder data clearly show annual cycle of crop growth and senescence, they could still be affected by various noise sources like persistent clouds, cloud shadows, surface anisotropy, aerosol and water vapour etc. Additional smoothing techniques (running averages or medians), compound smoothers, fourier methods or parametric smooth model functions are then applied. The phenology indicators extracted from RS data are (a) length of growing season, (b) date of peak VI (corresponds to ear-emergence/anthesis phase in grain crops), and (c) spectral emergence (indicator of date of sowing).

 

These RS-based phenology indicators have been used in regional crop monitoring applications such as (a) relating length of growing season to crop yields, (b) using spectral emergence date as input in crop simulation models, and (c) using temporal-spectral derived crop stages for accumulating phenology-based weather indices as opposed to fixed date-based approach. Recently multi-date WiFS data has been used for inventory of wheat and study relation between sowing date and pre-anthesis duration in wheat in Haryana and Punjab at district scale. A reduction in duration between spectral emergence and peak VI date as later is delayed, similar to field observations has been estimated (Rajak et al., 2002).

Linking Crop Simulation Models to RS inputs and GIS

The use of RS information through spectrally derived LAI to improve crop model accuracy, either as direct input to physiological crop model or as an independent check to model calculation for its re-initialization was proposed more than two decades ago by Wiegand et al., (1979). Only recent advances in LAI retrieval from space and availability of calibrated and validated CSM have led to pilot level regional demonstration of such a use. It is now realized that CSM and RS data can be combined in a number of ways (Delecolle et al., 1992, Moulin et al., 1998) (a) input, i.e., the direct use of RS-derived variable as driving variable in CSM; (b) forcing, i.e., the updating of a state variable of CSM (e.g., LAI) with RS parameter; (c) re-initialization, i.e., the adjustment of an initial condition in CSM to obtain agreement between simulation and RS derived parameter; (d) re-parameterization, i.e., the adjustment of CSM parameters to obtain agreement between simulation and RS-derived parameters; (e) corrective method, i.e., a relationship is developed between error in some intermediate variable as estimated from RS and error in final yield and applied to cases where yield forecast is needed.

 

The driving variables of CSM, i.e., weather inputs can be derived from RS data. This would overcome the limitation of sparse observation network and recently much progress has been made in deriving rainfall, surface temperatures, solar radiation and intercepted/absorbed PAR as discussed above. METEOSAT based decadal (10-day) rainfall using cold cloud duration has been used as input to CERES-Millet in Burkina Faso by Thornton et al (1997b) to forecast provincial millet yields halfway through crop duration to within 15% of their final values. The forcing strategy consists of updating at least one state variable in the model using remote sensing data. LAI has been the most commonly updated state variable, either on day of RS observation or daily LAI profile is generated using some simple parametric model for use. Mariappan et al (2003) have used this approach for rice with ORYZA and MODIS based LAI. In the re-initialization approach it is assumed that model is formally adequate but requires re-calibration. This is achieved by minimizing error between RS-derived state variable and its simulation by the model and again the state variable matched is LAI.

 

Sehgal et al (2003) have used a modified corrective approach for field-level wheat yield prediction in farmers’ fields during rabi 1998-99 in Alipur block (Delhi). The RS inputs as estimated LAI from LISS-III at end of January were linked to wheat simulation model WTGROWS for yield mapping and results were validated with yield observations on farmers’ fields. Biometric relation of grain yield and leaf area index (LAI) is derived from simulation model by running model for a combination of input resources, management practices and soil types occurring in the area. Then this biometric relationship is applied to all the crop fields of the study area for which the LAI is computed from remote sensing data. The comparison of predicted grain yield and observed yield for the 22 farmers’ fields showed high correlation coefficient of 0.8 and a root mean square error (RMSE) of 597 kg ha-1, which was 17 percent of the observed mean yield.

 

Sehgal et al (2001) have reported the development of a prototype Crop Growth Monitoring System (CGMS) for wheat using WTGROWS simulation model on a 5’X5’ grid in GIS environment for generating daily crop growth maps and predicting district-wise grain yield. The inputs used were RS based wheat distribution map, daily weather surfaces, soil properties map and crop management input databases in a GIS environment and analysis for wheat season of 1996-97 was carried out. The model predicted yields were within ±10% of reported yields in 12 out 16 districts. The RMSE of 335.4 kg ha-1, which is less than 10 percent of the State mean yield, was obtained. Sehgal et al (2002) demonstrated a technique for estimating date of sowing (DOS) using RS-derived spectral-temporal crop growth profiles and CGMS simulation capability and demonstrated that district-level yield prediction improves by using RS-CSM derived DOS as an input in CSM.

 

It is clear from the above studies that potential of integrating Crop simulation model, RS inputs and GIS has been well proven in a number of case studies. While techniques for geophysical and crop biophysical parameter retrieval are becoming available and producing products of required accuracy, the available crop simulation models need to be provided with GIS integration and iterative run options to benefit from this integration.

Extreme Events /Disasters

Crops are also susceptible to damage by episodic events, such as drought, flood and disease / pest outbreaks. Mapping areas affected by such events, quantifying damage class and monitoring recovery can be aided by RS data. In India, drought monitoring using MVC NDVI from NOAA AVHRR is being done regularly. Current research indicates that a combined use of optical and thermal data supports drought assessment as higher temperatures at a given level of vegetation index indicates water stress conditions. Mapping flood / cyclone caused inundated crop area and assessing damage to production has been done with combination of optical and microwave data. RS can also be used for detection and mapping of affected areas using RS data is feasible, if sufficiently large areas are affected. However, RS data per se cannot generally identify the cause of decline in canopy green biomass/LAI. Few case studies on large-scale infestation detection and mapping have been carred out. Amongst these is between-season greenness change detection for detecting and mapping incidence of yellow rust (Puccinia striiformis West) in West Punjab (Pakistan). RS data is also being used for forecasting and monitoring the desert locust (Schistocera gregaria Forsk.) infestation over large areas in Africa and Asia covering 55 countries and more than 16 million km2 of area. Potential locust breeding habitats are regularly monitored using IRS LISS and WiFS-based vegetation index in Rajasthan.

 


References

 

Badhwar, G.D., MacDonald, R.B., Mehta, N.C. (1986). Satellite-derived LAI and vegetation maps as input to global cycle models-a hierarchical approach. Int. Journal Remote Sensing, 7: 265-281.

 

Bhattacharya, B.K., Pandya, M.R., Dadhwal, V.K. (2001a). An approach to retrieve some land surface parameters using satellite images. Presented in Agromet seminar at GAU, Anand, Gujarat during 27-28th September, 2001

 

Bhattacharya B.K., Pandya M.R., Dadhwal V.K (2001b) A Preliminary Study to Retrieve Ground Insolation Using INSAT-2E and METEOSAT Images. Presented at ‘ISRS National Symposium’, 12-13 December, SAC, Ahmedabad.

 

Bhattacharya, B.K., Dadhwal, V.K. (2003a). Retrieval and validation of land surface temperature (LST) from NOAA AVHRR thermal images of Gujarat, India. Int. Journal of Remote Sensing. 24(6): 1197-1206

 

Bhattacharya, B.K. and Dadhwal, V.K. (2003b). Characterising Surface Soil Moisture Availability Using NOAA AVHRR Coarse Resolution Vegetation Index and Thermal Data. To be presented in National Seminar on Agrometeorology: Challenges and Perspectives in new millenium, at PAU, Ludhiana, October 29-31

 

Brakke, T.W., Kanemasu, E.T. (1981). Insolation estimation from satellite measurements of reflected radiation. Remote Sensing of Environment, 11: 157-167

 

Casselles, V., Coll, C., Valor, E. (1997). Land surface emissivity and Temperature determination in the whole HAPEX-Sahel area from AVHRR data. International Journal of remote sensing, 18:1009-1027.

 

Dadhwal, V.K., Ray S.S. (2000) Crop Assessment using remote sensing – Part II: Crop condition and yield assessment. Indian J. Agric. Economics 55(2, Suppl.): 55-67

 

Dadhwal, V.K., Singh, R.P., Dutta, S., Parihar, J.S. (2002). Remote Sensing Based Crop Inventory: A review of Indian Experience. Tropical Ecology 43(1): 107-122 25.

 

Dadhwal, V.K., Sehgal, V.K., Singh, R.P., Rajak, D.R. (2003). Wheat yield modeling using satellite remote sensing with weather data: Recent Indian experience. Mausam 54(1): 253-262.

 

Delecolle, R., Maas, S.J., Guerif, M., Baret, F. (1992). Remote sensing and crop production models: present trends, ISPRS J. Photogramm. Remote Sens., 47:145-161.

 

Gupta, R.K., Prasad, S., Sesha Sai, M.V.R., Nadham, T.S.V. (1995). Surface temperature estimation over agricultural region and its relationship with Normalized Difference Vegetation Index using NOAA-AVHRR data. In: Proceedings of National symposium on “Remote sensing of Environment with emphasis on green revolution”, PAU, Ludhiana, Nov 22-24,1995

 

Gillies, R.R., Carlson, T.N. (1997). A variation of the ‘triangle’ method for obtaining surface soil water content and energy fluxes from remote measurements of the normalized difference vegetation index (NDVI and surface radiant temperature. International Journal of Remote sensing, 18(1), 71-94

 

Kimothi, S., Bhattacharya, B.K., Semalty, P.D., Pandey, V.K., Dadhwal, V.K. (2003). Estimation of ground insolation using METEOSAT data over India. Current Science (communicated)

 

Knyazikhin, Y., Martonchik, J.V., Myneni,R.B., Diner, D.J., Running, R.W.(1998). Synergistic algorithm for estimating vegetation canopy leaf area index and fraction of absorbed photosynthetically active radiation from MODIS and MISR data. J. Geophys. Res.,103 : 32257-32275.

 

Kummerow, C., Barnes, W., Kozu, T, Shiue, J., Simpson, J. (1998) The Tropical Rainfall Measuring Mission (TRMM) Sensor Package. J Atmos. Ocean Technol.

 

Maracchi, G., Perarnaud, V., Kleschenko, A.D. (2000). Applications of geographical information systems and remote sensing in agrometeorology. Agric. Forest Meteorol., 103: 119-136.

 

Mariappan, V.E.N., Sehgal,V.K., Dadhwal,V.K. (2003). Linking MODIS LAI to ORYZA1 model in predicting rice yield in TamilNadu. National symposium on three decades of research in Agricultural Physics, IARI, New Delhi, 26-28 April.

 

Moulin, S., Bondeau, A., Delecolle, R. (1998). Combining agricultural crop models and satellite observations: from field to regional scales, Int. J. Remote Sens., 19(6): 1021 -1036.

 

Pandya, M.R., Chaurasia, S., Bhattacharya, B.K., Chaudhari, K.N., Singh, R.P., Sehgal, V.K., Vyas, S.P., Jaishanker, R., Nain, A.S., Sridhar V.N., Dadhwal, V.K. (2002). Leaf area index retrieval from IRS LISS-III data and validation of MODIS LAI product. Internat. Arch. Photogramm. Remote Sens. & Spatial Inf. Sci., 34 (7): 28-31.

 

Pathak, P.N., Rao, B.M., Potdar, M.B. (1993). Large-area soil moisture estimation using INSAT-VHRR infrared data. Indian J. Radio & Space Physics, 22: 156-164

 

Price J.C. (1993). Estimating leaf area index from satellite data. IEEE Trans. Geoscience Remote Sensing, 31 : 727-734.

 

Rao, B.M., Thapliyal, P.K., Pal, P.K., Manikiam, B., Dwivedi, A. (2001). Large scale soil moisture estimation using microwave radiometer data. J Agrometeorology

 

Rastogi, A., Kalra, N., Agarwal, P.K., Sharma, S.K., Harit, R.C., Navalgund, R.R., Dadhwal, V.K. (2000). Estimation of wheat leaf area index from satellite data using Price model. International J. Remote Sensing, 21(15): 2943- 2949

 

RaviKumar A., Ramana K.V., Rao, K.V., Rao, T.Ch. M., Venkatratnam, L. (1999). Land surface temperatures (LSTs) generation from NOAA/AVHRR thermal infrared channels. Vayu Mandal, 29(1-4): 390-395

 

Rosema, A. (1993). Using METEOSAT for operational Evapotranspiration and biomass monitoring in the Sahel region. Remote sensing of Environment, 46: 27-44.

 

Rajak, D.R., Oza, M.P., Bhagia, N., Dadhwal, V.K. (2002). Relating Wheat Spectral Profile Paratmeters to Yield and Phenology. Internat. Arch. Photogramm. Remote Sens. & Spatial Inf. Sci., 34 (7): 363-367.

 

RoyBhowmik, S.K., Sud, A.M. (2003) Seasonal and spatial characteristics of QPE errors over Indian region. Mausam, 54(3): 653–658.

 

Sehgal, V.K., Rajak, D.R., and Dadhwal, V.K. (2001). Issues in linking remote sensing inputs in a crop growth monitoring system: results of a case study, In ‘Proc. of the ISRS National Symposium’, Dec 11-13, 2001, Ahmedabad, India.

 

Sehgal, V.K., Rajak, D.R., Chaudhary, K.N., Dadhwal, V.K. (2002). Improved regional yield prediction by crop growth monitoring system using remote sensing derived crop phenology. Internat. Arch. Photogramm. Remote Sens. & Spatial Inf. Sci., 34 (7): 329-334.

 

Sehgal, V.K., Sastri C.V.S., Kalra N. and Dadhwal V.K. (2003) Farm-level yield mapping for precision crop management under Indian conditions using a simple approach of linking remote sensing information and crop simulation model. J. Indian Society of Remote Sensing (in press).

 

Tanahashi, S., Kawamura, H., Matsura, T., Yusa, H. (2001), A system to distribute satellite incident solar radiation in real-time. Remote Sens. Environ., 75 : 412-422.

 

Thornton, P.K., Bowen, W.T., Ravelo, A.C., Wilkens, P.W., Farmer, G., Brock, J., Brink, J.E. (1997). Estimating millet production for famine early warning: an application of crop simulation modelling using satellite and ground-based data in Burkina Faso, Agric. For. Meteorol, 83: 95 -112.

 

Van de Griend, A. A., Camillo, P.J., Gurney, R.J. (1985). Discrimination of soil physical parameters, thermal inertia and soil moisture from diurnal surface temperature fluctuations. Water Resources Research, 21, 997-1009

 

Wetzel, P.J. and Woodward, R.H. (1987). Soil moisture estimation using GOES-VISSR Infrared Data: A case study with a simple statistical method. J. Climate and Applied Meteorology, 26: 107-117

 

Wiegand, C.L., Richardsons, A.J., Kanemasu, E.T. (1979). Leaf area index estimates for wheat from LANDSAT and their implications for evapotranspiration and crop modeling, Agronomy J., 71: 336-342.

 

Wiegand, C.L., Richardson, A.J., Nixon, P.R. (1986). Spectral components analysis: A bridge between spectral observations and agrometeorological crop models, IEEE Trans. Geosci. Remote Sens., GE-24: 83-89.