• unlimited access with print and download
    $ 37 00
  • read full document, no print or download, expires after 72 hours
    $ 4 99
More info
Unlimited access including download and printing, plus availability for reading and annotating in your in your Udini library.
  • Access to this article in your Udini library for 72 hours from purchase.
  • The article will not be available for download or print.
  • Upgrade to the full version of this document at a reduced price.
  • Your trial access payment is credited when purchasing the full version.
Buy
Continue searching

Using remote sensing and grid-based meteorological datasets for regional soybean crop yield prediction and crop monitoring

Dissertation
Author: Preeti Mali
Abstract:
Regional crop yield estimations using crop models is a national priority due to its contributions to crop security assessment and food pricing policies. Many of these crop yield assessments are performed using time-consuming, intensive field surveys. This research was initiated to test the applicability of remote sensing and grid-based meteorological model data for providing improved and efficient predictive capabilities for crop bio-productivity. The soybean prediction model (Sinclair model) used in this research, requires daily data inputs to simulate yield which are temperature, precipitation, solar radiation, day length initialization of certain soil moisture parameters for each model run. The traditional meteorological datasets were compared with simulated South American Land Data Assimilation System (SALDAS) meteorological datasets for Sinclair model runs and for initializing soil moisture inputs. Considering the fact that grid-based meteorological data has the resolution of 1/8 th of a degree, the estimations demonstrated a reasonable accuracy level and showed promise for increase in efficiency for regional level yield predictions. The research tested daily composited Normalized Difference Vegetation Index (NDVI) from Moderate Resolution Imaging Spectroradiometer (MODIS) sensor (both AQUA and TERRA platform) and simulated Visible/Infrared Imager Radiometer Suite (VIIRS) sensor product (a new sensor planned to be launched in the near future) for crop growth and development based on phenological events. The AQUA and TERRA fusion based daily MODIS NDVI was utilized to develop a planting date estimation method. The results have shown that daily MODIS composited NDVI values have the capability for enhanced monitoring of soybean crop growth and development. The method was able to predict planting date within ±3.4 days. A geoprocessing framework for extracting data from the grid data sources was developed. Overall, this study was able to demonstrate the utility of MODIS and VIIRS NDVI datasets and SALDAS meteorological data for providing effective inputs to crop yield models and the ability to provide an effective remote sensing-based regional crop monitoring. The utilization of these datasets helps in eliminating the ground-based data collection, which improves cost and time efficiency and also provides capability for regional crop monitoring.

v TABLE OF CONTENTS Page DEDICATION....................................................................................................................ii

ACKNOWLEDGEMENTS...............................................................................................iii

LIST OF TABLES...........................................................................................................viii

LIST OF FIGURES.............................................................................................................x CHAPTER I.INTRODUCTION................................................................................................1

1.1

Research Introduction..................................................................................1

1.2

Literature Review.........................................................................................4

1.2.1

Crop Yield Modeling.......................................................................4

1.2.2

NDVI and Crop Productivity Monitoring........................................5

1.2.3

Crop Yield Models and Remote Sensing.........................................7

1.2.3.1

Regression Based Emperical Method...............................7

1.2.3.2

Semi-Empirical Method....................................................9

1.2.3.3

Mechanistic or Agro-meteorological Method.................10

1.2.3.4

Methods useful for regional predictions.........................12

1.2.4

Sinclair Crop Model.......................................................................13

1.3

Statement of Problem.................................................................................15

1.4

Research Objectives...................................................................................17

1.5

References..................................................................................................19

II.ANALYSIS OF SALDAS METEOROLOGICAL FORCINGS AND SALDAS SIMULATED SOIL MOISTURE FOR SOYBEAN YIELD ESTIMATION MODELING....................................................21

2.1

Abstract......................................................................................................21

2.2

Introduction................................................................................................22

2.2.1

Meteorological datasets in crop yield modeling............................24

2.2.1.1

SALDAS forcing’s as input to Sinclair model...............28 2.2.1.2

SALDAS for intializing soil moisture parameters..........29

vi 2.3

Study Area.................................................................................................31

2.4

Methodology..............................................................................................32

2.4.1

Data Collection..............................................................................32

2.4.1.1

The South American Land Data Assimilation System (SALDAS).............................................32

2.4.1.2

Selection of Soybean Fields............................................33

2.4.1.3

Ground Meteorological Data..........................................34

2.4.1.4

Soil Moisture Field Sampling.........................................35

2.4.1.5

Yield, Cultivars and Planting Date.................................35

2.4.2

Data Preprocessing.........................................................................35

2.4.3

Use of SALDAS forcings meteorological input............................36

2.4.4

Use of SALDAS soil moisture as initializing inputs.....................37

2.5

Results and Discussion..............................................................................38

2.5.1

Yield results from ground meteorological inputs..........................38

2.5.2

Yield results from all SALDAS forcings.......................................39

2.5.3

Yield results from SALDAS precipitation.....................................40

2.5.4

Yield results from SALDAS minimum temperature.....................41

2.5.5

Yield results from SALDAS maximum temperature.....................42

2.5.6

Yield results from SALDAS solar radiation..................................43

2.5.7

SALDAS soil moisture as initializing inputs.................................44

2.5.7.1

Use of SALDAS moisture values for initializing Sinclair model....................................................45

2.6

Conclusions................................................................................................51

2.7

References..................................................................................................53

III.THE UTILITY OF MODIS AND SIMULATED VIIRS IMAGERIES FOR MONITORING CROP PRODUCTIVITY...................................55

3.1

Abstract......................................................................................................55

3.2

Background and Introduction....................................................................56

3.3

Methodology..............................................................................................61

3.3.1

Study Area.....................................................................................61

3.3.2

Data Sources..................................................................................62

3.3.2.1

Moderate Resolution Imaging Spectroradiometer (MODIS)............................................................62

3.3.2.2

Visible Infrared Imaging Radiometer Suite (VIIRS)......63

3.3.3

Preprocessing.................................................................................64

3.3.3.1

Field Boundary Delineation and Verification.................64

3.3.3.2

Data Subsetting, Reprojection and Format Change........65

3.3.3.3

VIIRS Simulation............................................................65

3.3.3.4

NDVI Calculation...........................................................66

3.3.3.5

Zonal Value Extraction...................................................66

3.3.3.6

Large Data Size and Volume Handling with Batch Processing..........................................................67

3.4

Analysis and Results..................................................................................71

3.4.1

Comparison between MODIS and simulated VIIRS.....................71

vii 3.4.1.1

Comparison using scatter plot.........................................71

3.4.1.2

Validation of VIIRS simulation......................................76

3.4.1.3

Validation of time-series curves based on daily NDVI with respect to Soybean Phenology for MODIS and simulated VIIRS......................78

3.4.1.4

NDVI time-series depicting soybean growth..................79

3.4.1.5

Cross Validation Approach.............................................83

3.5

Results and Discussions.............................................................................93

3.6

Conclusions................................................................................................95

3.7

References..................................................................................................97

IV.IMPROVEMENT IN PLANTING DATE ESTIMATION THROUGH THE USE OF NDVI DERIVED FROM SATELLITE IMAGERY.............................................................................................99

4.1

Abstract......................................................................................................99

4.2

Background and Introduction..................................................................100

4.3

Methodology............................................................................................105

4.3.1

Data Collection and Preprocessing..............................................105

4.3.1.1

Field Boundary Verification.........................................106

4.3.1.2

Field Based Datasets Collected.....................................107

4.3.2

Analysis and Results....................................................................107

4.3.2.1

NDVI Calculation and Fusion Based Compositing......107

4.3.2.2

Difference between AQUA and TERRA MODIS mean NDVI values...........................................110

4.3.2.3

Cross Platform Fusion and Composite Creation...........110

4.3.2.4

Zonal Processing...........................................................111

4.3.2.5

Planting Date Estimation Process.................................112

4.4

Results and Discussion............................................................................113

4.5

Conclusions..............................................................................................118

4.6

References................................................................................................120

V.FINAL SUMMARY AND RECOMMENDATIONS......................................123

viii LIST OF TABLES 2.1 Satellite based meteorological data sources...........................................................26

2.2 Integrated Data Sources.........................................................................................27

2.3 Soybean fields used for the study for the year 2006/2007 in Argentina................34

2.4 Comparison between Sinclair simulated base yields using with ground meteorological data and the simulated yield using SALDAS meteorological data....................................................................................40

2.5 Comparison between Sinclair simulated base yields using with ground meteorological data and the simulated yield in which the ground meteorological input is replaced with only the SALDAS precipitation...............................................................................................41

2.6 Comparison between Sinclair simulated base yields using with ground meteorological data and the simulated yield in which the ground meteorological input is replaced with only the SALDAS minimum temperature................................................................................42

2.7 Comparison between Sinclair simulated base yields using with ground meteorological data and the simulated yield in which the ground meteorological input is replaced with only the SALDAS maximum temperature...............................................................................43

2.8 Comparison between Sinclair simulated base yields using with ground meteorological data and the simulated yield in which the ground meteorological input is replaced with only the SALDAS solar radiation.....................................................................................................44

2.9 Comparison between SALDAS soil moisture values and field observed soil moisture...............................................................................................45

2.10 Yield from six sets of simulations using different ESW and DEEP values results obtained...............................................................................49

ix 2.11 Percentage differences of the simulated yields with ESW and DEEP from SALDAS with base yields (simulated yield results using all experts provided initializing variables and ground meteorological datasets as inputs)......................................................................................50 2.12 Root mean square error (RMSE) calculated between yield from six sets of simulations using different ESW and DEEP values and base yield............................................................................................................50 3.1 Comparison between spatial, temporal and spectral resolution of MODIS, VIIRS and AVHRR....................................................................59

3.2 Soybean fields for the year 2006/2007 in Argentina.............................................69

3.3 Statistical differences between VIIRS images simulated from MODIS and AWiFS.................................................................................................77

3.4 Correlation coefficient value between image to image comparison between MOD02 (atmospherically uncorrected level 1 image) vs. MOD09 (atmospherically corrected level 2 image)...................................78

3.5 “Temporal Metrics” (Reed et al. 1994) and the corresponding phenological stages....................................................................................83

3.6 Values of various phenological stages for the soybean fields and the corresponding representative NDVI value from Terra MODIS and LAI values simulated from Sinclair Model.........................................86

3.7 Values of various phenological stages for the soybean fields and the corresponding representative NDVI value from Terra MODIS and LAI values simulated from Sinclair Model.........................................87

4.1 Estimated planting dates of early planted (October to early November) soybeans...................................................................................................117

x LIST OF FIGURES 1.1 Abridged schematic flowchart depicting inputs and program flow of major modules for Sinclair soybean model...............................................14

2.1 AWIFS imagery showing study sites selected for soybean farms and also for installing seven automatic weather stations (except Monte Buey)..........................................................................................................31

2.2 Gridded SALDAS meteorological forcings and soil moisture data.......................33

2.3 An example of fields selected for the study: Marcos Juarez fields over AWIFS imagery.........................................................................................33

2.4 Geo-processing methods utilized to extract and process SALDAS grid datasets for providing “Sinclair model ready” inputs................................36

3.1 The AOI of our interest is shown by the red box inside the larger magenta colored box. The magenta colored box is the MODIS tile H12V12 used for this experiment........................................................63

3.2 Simulated VIIRS NDVI (400 m) from MODIS data (250 m) using Application Research Toolbox (ART).......................................................66

3.3 Flowchart of the geo-processing methodology used for the analysis....................68

3.4 Flowchart of the pre-processing steps performed before analysis.........................70

3.5 TERRA MODIS NDVI and simulated VIIRS NDVI scatter plots for 14 fields in the area for 2006-2007 soybean growing season.........................72

3.6 AQUA MODIS NDVI and simulated VIIRS NDVI scatter plots for 14 fields in the area for 2006-2007 soybean growing season.........................74

3.7 NDVI curves depicting soybean growth characteristics for the fourteen soybean test fields in Argentina from AQUA MODIS (a), VIIRS simulated from AQUA MODIS (b), TERRA MODIS (c), VIIRS simulated from TERRA MODIS (d)..........................................................80

xi 3.8 Charts showing the NDVI curves for various selected soybean fields in Argentina from TERRA MODIS and VIIRS simulated from TERRA MODIS and specific phenological stages for each field..............88

4.1 Time chart of various phenological stages for the soybean fields as calculated by the Sinclair crop model......................................................103

4.2 Comparison between various zenith angle constraints for compositing process. The zenith angle of 48º provided a good cutoff point................109

4.3 Flowchart of the geo-processing methodology used for the analysis..................112

4.4 Planting date estimation using Vegetation Index and Temporal map algebra (Mali et al., 2006)........................................................................113

4.5 Graph showing soybean phenology and growing degree days for soybean planted over soybean field in Rio Segundo.............................................114

4.6 Graph showing soybean phenology and growing degree days for soybean planted over wheat field in MonteBuey...................................................115

1 CHAPTER I INTRODUCTION 1.1 Research Introduction Crop models have been used for predicting crop yield before harvest. The benefits of such predictions have potential effects from local to regional to global. Such predictions warn decision makers about potential reductions in crop yields and allow timely import and export decisions. These pre-harvest crop yield estimations also help in regional and global crop pricing and trade policies. Thus, reliable yield prediction methods are highly important for national and global food security.The Production Estimation and Crop Assessment Division (PECAD) of United States Department of Agriculture (USDA)/ Foreign Agricultural Service (FAS) provide global crop yield forecasts for major food grain and oil seed crops. These estimates require a tremendous amount of ground data collection network. The ability of remote sensing and meteorological grid datasets to provide information on crop growth and environmental conditions that affect crop growth is a huge benefit for agencies such as USDA/FAS PECAD for regional yield predictions. With the benefits and limitations of remote sensing considered, this research has been developed with the hypothesis that utilization of remote sensing and spatial technologies can greatly benefit in regional scale crop yield

2 estimation modeling. A detailed literature review has showed that most remote sensing based regional yield prediction models use coarse resolution imageries such as the National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) and Moderate Resolution Imaging Spectro-radiometer (MODIS). The USDA/FAS PECAD currently utilizes (MODIS) Normalized Difference Vegetation Index (NDVI) for assessing the crop growth conditions. However, a future sensor Visible/Infrared Radiometer Suite (VIIRS) is expected to replace MODIS in the near future. This research compares MODIS and VIIRS to assess whether VIIRS sensor product is applicable to replace MODIS for regional yield prediction and crop productivity monitoring. The types of models used for crop predictions are highly varied. Different methods of crop prediction using remote sensing and spatial datasets for crop prediction have been researched. However, most of these methods either use highly empirical methods, or methods that use parameters that cannot be utilized for regional level predictions (Rasmussen, 1998; Dabrawoska et al., 2002; Bastiaanssen and Ali, 2003; Lobell et al., 2003). Doraiswamy et al.(2005) applied remote sensing to a semi- mechanistic crop model for regional yield assessment and found that with the use of the correct crop model, the information from remote sensing observations can be effectively integrated into crop modeling methodologies. This research uses Sinclair model for soybean yield prediction for Argentina. Sinclair model has been used operationally by USDA/FAS PECAD to provide estimation on soybean production (Reynolds, 2001). This model has been described as “semi- mechanistic” and is considered as “a compromise between completely empirical approaches and extremely detailed mechanistic approaches” (Speath et al., 1987, p. 298);

3 therefore, suitable for adapting to a geoprocessing environment. The model uses daily inputs of temperature (daily minimum and maximum), precipitation, solar radiation, day length and planting date. The input variables are obtained from meteorological stations. Besides these variables, the model also requires initialization of certain parameters during each model run, of which planting date and soil moisture content are very sensitive to the yield estimates. Remote sensing is used in a number of crop prediction models. These range from simple regression-based models to very complicated models based on a number of inputs. Although remote sensing is beneficial for local uses such as precision agriculture, remote sensing is increasingly being used in regional predictions due to the ability to efficiently provide spatially based results for larger areas. Most regional level yield prediction methods consider using MODIS and AVHRR due to their wide swath width. Since crop yield models are usually developed from field-based experiments, regional prediction models based on remote sensing are usually adapted from crop models developed from field-level experimentations. Therefore, in order to obtain as much accuracy as possible in predicting yield, the spatially-based input variables should be able to represent field level conditions as much as possible.Different types of models for field-level crop yield predictions for various crop types are available. The adaptation of crop models to regional-level predictions of yield are lacking validated mechanisms for their application. The major challenge in such adaptation lies in the area of scaling or substituting model inputs to obtain representative estimates that extend capabilities to a regional or national level. In scaling models to regional-level analysis, field-level conditions such as row spacing, amount of fertilizer per field, and other field-level details cannot be used.

4 Difficulties arise when field-specific input variables to the models are replaced by information extracted from satellite image based observations. The probability that the inaccuracy of the model output would increase cannot be neglected because on one hand a field level model is being used for regional level estimates, and on the other hand,the input parameters are estimated from remote sensing. Even then, it can be quite beneficial for using such crop models for regional predictions using remote sensing based inputs. In fact researchers agree that remote-sensing technologies can help to reduce the costs, time, and money to effectively predict crop yield (Reynolds et al., 2000; Wiegand et al., 1991). Therefore, the research evaluates the use of remote sensing (current and future sensors) and grid-based meteorological datasets to reduce the need for detailed time consuming field data and for providing improvement in efficiency to monitor and model crop bio- productivity. 1.2 Literature Review 1.2.1 Crop Yield Modeling Crop growth modeling was initiated and developed by C.T. de Wit. The origin of crop modeling can be traced into the publication in 1965 by C.T. de Wit on modeling photosynthesis as a function of leaf canopies (Boumat et al., 1996). The crop models that follow the modeling philosophy of C. T. de Wit are considered to belong to the “School of de Wit”. De Wit and Penning (1982) proposed a basic classification of modeling system that consisted of four production situations: a) potential production b) water limited production c) nitrogen limited production and d) nutrient limited production

5 (Bouman et al., 1996). De Wit (1965) demonstrated that canopy photosynthesis is the sum of photosynthesis of all the individual leaves. De Wit (1982) introduced growth rate calculation as a function of time, a dynamic system of crop modeling was introduced. 1.2.2 NDVI and Crop Productivity Monitoring The utilization of remote sensing in crop yield estimation and crop growth monitoring can be traced back to the development of vegetation indices that are based upon the plant spectral characteristics. Vegetation indices utilize the properties of the chlorophyll reflectance in the red and near-infrared region of the electromagnetic spectrum (Myneni et al., 2005). Among the many vegetation indices, NDVI is the most researched and widely used index.The development of the NDVI is shared by Tucker (1979) and Deering (1978). Tucker (1977) found that leaf water content was best estimated in the region of 0.4-0.5, 0.63-0.69 and 0.74 to 0.8 µm in the electromagnetic spectrum. He concluded that these resulted due to the strong chlorophyll absorption on the 0.4-0.5 µm, 0.63-0.69 µm and high reflectance of vegetation in the 0.74 to 0.8 µm regions.Tucker (1979) studied in-situ spectral reflectance of grass for 80% green biomass, 50% green biomass and dead biomass and compared various vegetation indices which included NIR/R, Visible/IR ratios. He concluded that for vegetation studies, the IR/Red ratio was most useful. Tucker (1979) also pointed out that some means of normalization for different irradiation conditions would be useful for studying the green leaf biomass of crops. Tucker (1979) found that the normalized difference transformation was effective in compensating for the variation in irradiation conditions. Tucker (1979) also found that the percentage of crop cover was closely related to vegetation indices. As

6 crop cover increased or decreased, the vegetation index values measured had a corresponding change. Thus, he concluded that due to the observed relationship between the vegetation indices and crop development, crop conditions could be monitored through spectral measurements. The early set of researches on relationship between the vegetation growth and the spectral bands of NDVI as well as the IR/Red ratios has led to a great deal of research such that NDVI has been used as proven index for monitoring vegetation condition.Wiegand and Richardson (1984) found that plants express their development, stress response, and yield capability through spectral observable canopy relating vegetation index to leaf area index, fractional observed photosynthetically active radiation (PAR) and economic yield (Richardson, 1990). Wiegand and Richardson (1990a) proposed a rationale in which the spectral observations i.e. vegetation indices were related to plant processes specifically leaf area, evapo-transpiration and yield. Wiegand and Richardson (1990b) tested the rationale proposed on relating vegetation indices to plant processes to cotton, wheat and corn. The vegetation indices used include NDVI, Perpendicular Vegetation Index (PVI), and red index and found that although limited, the vegetation indices do have relationships to crop growth and development and can be used to infer leaf area, evapo-transpiration and yield. Various researchers have since utilized NDVI to predict crop yield.

7 1.2.3 Crop Yield Models and Remote Sensing The review of different methods of crop yield predictions has shown that crop yield predictions using remote sensing and spatial technologies can be basically categorized into the following (Moulin et al., 1998): Regression based empirical method Semi-empirical based method (Monteith based model) Mechanistic or agro-meteorological based method 1.2.3.1 Regression Based Emperical Method Regression based crop yield models are developed on the basis of the relationship between crop yield to a variety of biophysical factors such as crop vigor, rainfall, temperature, and soil. Boken et al. (2002) used NOAA-AVHRR based composited NDVI for spring wheat model in Canadian Prairies, using a monthly model based on a cumulative moisture index. The main purpose of this research was to improve an operational wheat model using remote sensing information based on monthly weather data. The model uses monthly temperature and precipitation data, estimated daily crop water requirement to obtain the Cumulative Moisture Index (CMI), which provides the daily moisture data; these data are cumulated for the whole of the growing season (from sowing to harvest). The use of CMI is based on the theory that if soil moisture requirement has been met, optimum growth will be attained. The research found that the use of NDVI based variable in a regression model with CMI improves the prediction power of the model significantly. The coefficients of determination in a NDVI based model were 0.79, 0.96, 0.83, 0.95, and 0.39 in five districts as opposed to 0.13, 0.70,

8 0.70, 0.75, 0.50, and 0.00 in a regular monthly model. Boken et al. (2002) also compared different variables obtained from NDVI and crop growth with wheat yield, and found that average NDVI during the heading period correlated highly with the wheat yield. In other research, Rasmussen et al. (1998) used NOAA-AVHRR NDVI based model for predicting crop yield in Senegal, West Africa. The nine-day maximum value composited NDVI imagery of the years 1990 and 1991 were used. Besides NDVI, percent tree cover data were collected through low altitude systematic reconnaissance flights. Similarly, Tropical Livestock Unit (TLU) densities and percentage cultivated land data and population density data for grownup males were collected. These data were collected as point data and were interpolated using inverse distance weighted method with grid cell size of 500 m. On regression analysis of the various data parameters, it was found that grain yield and time weighted NDVI values (iNDVI) were highly correlated. The application of the model to stratified data with greater than 22% cultivated land improved the yield from r 2 = 0.62 to 0.73. The TLU density showed a significant relation with yield. The regression model, for cultivated area of percentage value greater than 22%, based on the iNDVI and TLU density,improved the yield prediction to r 2 = 0.88. Dabrawoska (2002) also used AVHRR NDVI based regression model for cereal yield estimation. In this research NOAA-AVHRR GAC (global area coverage) data with 4 km resolution was used to calculate NDVI and brightness temperature (BT). The NDVI and BT were further used to obtain VCI (vegetation condition index) and TCI (temperature condition index),respectively. Landsat data were also used to obtain agricultural distribution map to obtain pixels with agricultural land less than 50,50- 70 and 70-100%. High correlations of yield with VCI were noted in the weeks of 16, 22 of

9 crop growing period and TCI in the week 25 of the crop growing period. Therefore, Dabrawoska et al. (2002) developed a regression-based model using TCI at weeks 16 and 22 and VCI at week 25 of crop growing season. The prediction result showed only a mean average error of 4%. 1.2.3.2 Semi-Empirical Method Montieth-based models can be considered semi-empirical in nature (Moulin et al., 1998). Bastiaanssen and Ali (2003) used a Monteith-based model that uses accumulated above ground biomass to predict yield. The biomass is derived from APAR (absorbed photosynthetically active radiation) values, which are derived from NDVI-derived PAR and APAR/PAR fraction. PAR values are obtained from incoming solar radiation values measured from Gumble Stokes recorders at ground meteorological stations. The method used also required estimation of light use efficiency values. The estimation of light-use efficiency requires values of ‘impact soil moisture’, ‘heat effect factors’, and ‘evaporative fraction’ that require complex derivations and assumptions. The authors found that their method was successful in predicting wheat, rice, and sugarcane yields, with 22, 29 and 23% relative deviation from the observed yield values. However, this method was not successful in predicting cotton yield and the reason has been given as the inability of AVHRR in distinguishing cotton fields. Lobell et al. (2003) also used a Montieth-based model to predict crop yield using Landsat TM. But the derivation of variables of the model is different from the method used by Bastiaanssen and Ali (2003). The APAR/PAR fraction and light-use efficiency are calculated using a different procedure. PAR values are calculated in the field using a pyranometer. APAR values are calculated

10 using simple ratio (SR) and maximum and minimum possible APAR. The light-use efficiency is calculated from plot-based harvested biomass and APAR values. 1.2.3.3 Mechanistic or Agro-meteorological Method Mechanistic models usually contain a defined process using crop state variables and energy, carbon, nutrient fluxes at crop/soil/atmosphere interfaces (Moulin et al., 1998). One such mechanistic agro-meteorological model is FAO-based crop specific water balance model (CSWB) used by Reynolds et al. (2000). In this method, near real time satellite products such as NDVI, RFE (rainfall estimate) images are used. The NDVI is derived from NOAA-AVHRR and RFE images are obtained from stationary Meteosat- 5 satellite. Ground-based PET (potential evapo-transpiration data) from meteorological station was also used. This method incorporates remote sensing data with a ground-based model. The data are integrated in a GIS-based model called WINDISP3. This method also requires locally derived information such as yield reduction factor, maximum yield that differ spatially. The agrometeorological-based method has even been used by PECAD FAS to provide estimation on global agricultural production (Reynolds, 2001; NASA, 2003). PECAD’s method is based on an automated decision support system called Crop Condition Data Retrieval and Evaluation (CADRE). CADRE is an operational outgrowth of the Large Area Crop Inventory Experiment (LACIE) and Agriculture and Resources Inventory Surveys through Aerospace Remote Sensing (AgRISTARS). CADRE integrates remote sensing data, crop and soil models with weather information. It serves as an interface to different models and outputs data through GIS software, time-series plot and web interface displays. The agro-

11 meteorological data to CADRE is provided by Agricultural Meteorological Model (AGRMET) and World Meteorological Organization (WMO) network of weather stations. AGRMET provides precipitation, minimum and maximum temperature, snow depth, solar and long wave radiation and potential and actual evapo-transpiration (ET). CADRE computes its own ET from temperature inputs using the Penman-Monteith equation. The satellite-based data include AVHRR and SPOT vegetation data. CADRE also requires baseline data which are digital elevation model which comprises of, FAO generated Digital Soil Map of the World (DSMW) information, historical crop production database from FAS, average temperature, rainfall spatial data, and administrative boundaries. The CADRE crop model requires crop calendar models, crop stress models (CERES, AgRISTARS, Maas, URCROP, Sinclair) and a two-layer soil moisture model. The two-layer soil moisture model runs the crop calendar and crop stress models. The soil moisture model accounts for the total water gained or lost in the soil profile by recording the amount withdrawn by evapo-transpiration and replenished by precipitation. The crop calendar model is based on the growing degree-days algorithm that uses minimum, maximum and threshold temperatures defined by a particular crop report. The crop stress model developed by AgRIStars informs analyst on abnormal temperature or moisture stress that may affect yields. Thus, PECAD uses a highly operational crop yield prediction system that requires an extensive input of time series data, baseline data and crop information and models from various sources.

12 1.2.3.4 Methods useful for regional predictions Most of the methods reviewed have used NOAA-AVHRR-based NDVI images to estimate crop yield. The regression-based models may be used for regional yield estimations; however,it is highly empirical in nature. The three methods based on regression reviewed here (Boken, 2002; Rasmussen, 1998; Dabrawoska, 2002) are all empirical in nature and are locally based such that the models derived cannot be used for regional or global application. However, if a globally derived model can be obtained, regression-based yield models are the simplest of all models. According to Moulin et al. (Moulin et al., 1998), regression-based models based on vegetation index and yield are empirically derived; hence, not based on a theoretical and experimentally proved relation. Therefore, Moulin et al. (1998, p.1023) stated that, “more mechanistic and physiologically sound models are necessary to assimilate remote sensing data and to predict production of major crops”. However, even though Montieth-based models are physiologically sound and experimentally proven, their uses in regional based application remain questionable. The Montieth-based models reviewed required values such as light- use efficiency values which were calculated using highly complex relations in Bastiaansen and Ali (2003) and using field based values as in Lobell et al. (2003). Thus, the applicability of such models in regional yield predictions whose variables cannot be computed regionally or globally need to be further studied. Contrary to the empirical methods and Montieth-based method, the agro-meteorological-based crop yield prediction method seems to have a good scope in regional yield predictions using remote sensing. The variables in these methods are mostly obtained from either meteorological stations or remote sensing satellites; thus, they have global or regional applicability. One

Full document contains 142 pages
Abstract: Regional crop yield estimations using crop models is a national priority due to its contributions to crop security assessment and food pricing policies. Many of these crop yield assessments are performed using time-consuming, intensive field surveys. This research was initiated to test the applicability of remote sensing and grid-based meteorological model data for providing improved and efficient predictive capabilities for crop bio-productivity. The soybean prediction model (Sinclair model) used in this research, requires daily data inputs to simulate yield which are temperature, precipitation, solar radiation, day length initialization of certain soil moisture parameters for each model run. The traditional meteorological datasets were compared with simulated South American Land Data Assimilation System (SALDAS) meteorological datasets for Sinclair model runs and for initializing soil moisture inputs. Considering the fact that grid-based meteorological data has the resolution of 1/8 th of a degree, the estimations demonstrated a reasonable accuracy level and showed promise for increase in efficiency for regional level yield predictions. The research tested daily composited Normalized Difference Vegetation Index (NDVI) from Moderate Resolution Imaging Spectroradiometer (MODIS) sensor (both AQUA and TERRA platform) and simulated Visible/Infrared Imager Radiometer Suite (VIIRS) sensor product (a new sensor planned to be launched in the near future) for crop growth and development based on phenological events. The AQUA and TERRA fusion based daily MODIS NDVI was utilized to develop a planting date estimation method. The results have shown that daily MODIS composited NDVI values have the capability for enhanced monitoring of soybean crop growth and development. The method was able to predict planting date within ±3.4 days. A geoprocessing framework for extracting data from the grid data sources was developed. Overall, this study was able to demonstrate the utility of MODIS and VIIRS NDVI datasets and SALDAS meteorological data for providing effective inputs to crop yield models and the ability to provide an effective remote sensing-based regional crop monitoring. The utilization of these datasets helps in eliminating the ground-based data collection, which improves cost and time efficiency and also provides capability for regional crop monitoring.