Last modified on Oct 09, 2023 16:13

Contributors: Allard de Wit (WAGENINGEN ENVIRONMENTAL RESEARCH), Ulan Turdukulov (WAGENINGEN ENVIRONMENTAL RESEARCH)

Issued by: WEnR

Issued Date: 07/10/2020

Ref: D422Lot1.WEnR.2.3.2

Official reference number service contract: 2017/C3S_422_Lot1_WEnR/SC2

History of Modifications

Acronyms

1. Scope of the document

This ATBD describes the input requirements and algorithms underlying the implementation of the Earth observation based crop productivity indicators (further to be called "AgDMP"). This product is similar to the Dry Matter Productivity (DMP) product that is already available from the Copernicus Global Land Service (https://land.copernicus.eu/global/products/DMP). The current product has been made more agriculturally relevant by:

1. Spatially selecting those areas (pixels) that likely belong to a particular crop
2. Including more agronomic knowledge into the algorithm including crop management (calendars) and crop specific parameters (phenology and light use efficiency)

This product thus combines satellite observed fAPAR (Fraction of Absorbed Photosynthetic Active Radiation) from the Copernicus Global Land service, with meteorological data from C3S and a simple crop simulation model to provide information on crop development, total accumulated crop biomass and crop yield.

2. Algorithm description

2.1. Introduction

The algorithm that was developed for the AgDMP product represents a hybrid algorithm that takes the satellite derived fraction of absorbed PAR (FAPAR) from the Copernicus Global Land Service, with elements from the WOFOST and LINTUL crop simulation models (De Wit et al. 2019). Our assumption is that the agriculturally tuned AgDMP product provides a better representation of crop productivity compared to the standard DMP product. This assumption has yet to be validated which will be carried out in the future through comparison with regional crop yield statistics.

Technically, the hybrid model has some interesting aspects. The simulation of the crop canopy (and thus light interception) is the most complicated part of simulation models like WOFOST/LINTUL. In the hybrid model we replace the simulation of the canopy by directly taking a satellite derived estimate of the canopy light interception (FAPAR).

Next, the conversion of intercepted light to biomass is done in the standard DMP product by a standardized (v1) or biome-specific (v2) light use efficiency which is hardly representative for the light use efficiency of many crops. Therefore, the assimilation routine from the LINTUL4 model has been used to convert intercepted light to dry matter.
Finally, the standard FAPAR product has no concept of cropping seasons or crop phenology which regulates dry matter partitioning and yield formation. Therefore, we take the phenological model from WOFOST and use that to define the length of the growing season in terms of growing degree days (optimized on the local crop calendar).

2.2. Input data used

2.2.1. Overview

The AgDMP algorithm requires inputs from a large number of sources includes meteorological data, satellite data, crop calendars, crop parameters and crop masks (Table 1). Below a detailed description of all inputs will be provided.

Table 1: Overview of the input data required for the AgDMP product

2.2.2. Daily meteorological data

Daily meteorological data are required because the crop simulation model runs at a fixed time-step of one day. For generating the crop indicators product for the period 2000-2018, we used meteorological variables from ERA-Interim at 0.25x0.25 degree which we obtained through the MARSOP4 project from the European Commission. Each 0.25x0.25 degree cell was linked to its nearest 0.1 degree cell. For the operational product, the system uses the AgERA5 dataset at 0.1x0.1 as the database schema and all other datasets have already been prepared at 0.1x0.1 degree.

The current AgDMP product only uses the weather variables related to temperature and radiation (Table 2) because no water balance is currently included. However, the current data provider modules assume those variables to be present. Moreover, if the system needs to be extended with a water balance, these variables will be required anyway.

Table 2: Daily variables provided by the MARSOP4 ERA-Interim and AgERA5 datasets. Variables which are used by the current version of AgDMP are marked in italics.

2.2.3. Copernicus GLS Fraction Absorbed PAR

The AgDMP product uses the FAPAR from the Copernicus Global Land Service to estimate the amount of radiation intercepted by the crop. For the period 2000 up till June 2020 the 1km products from the VGT, Proba-V and Sentinel3/OLCI sensors were used. Since June 2020 the 333m product is used because the 1km product has been discontinued. The 333m product is first aggregated to 1 km before the processing is continued. More information on the FAPAR product can be found at: https://land.copernicus.eu/global/products/fapar

2.2.4. Crop calendars

Two types of crop calendar data sets have been used in setting up the AgDMP product:

1. The crop calendars provided by the Center for Sustainability and the Global Environment (SAGE) of the University of Wisconsin-Madison described by Sacks et al. (2010) and available from http://nelson.wisc.edu/sage/data-and-models/crop-calendar-dataset/index.php.
2. The crop calendars from the Global Agro-ecological Zonation (GAEZ) as provided by the FAO (http://www.fao.org/nr/gaez/en/) and described by Fischer et al. (2021).

Both products have their limitations when describing cropping patterns across the globe. The SAGE calendars are based on regional crop calendars collected by the FAO and others with limited spatial variability in dates of sowing and harvest. Therefore, they often do not represent gradients in sowing and harvest dates very well. Also, the spatial distribution of crops does not always reflect the true distribution (e.g. soybean areas missing in N-China) or is based on somewhat arbitrary criteria (such as the distinction between spring and winter-wheat). The advantage of the SAGE calendars is that they do properly reflect farmer choices in management, such as the choice for double cropping instead of a single long growing season.

The GAEZ calendars are based on a climatic assessment of the growing potential of crops. The latter includes an assessment of the start and length of the growing period based on climatic data. As such the GAEZ calendars are far better capable of describing gradients in the sowing/harvest dates as a result of gradients in climate. Also, the spatial delineation is based on a theoretical crop distribution that is defined by the climatic opportunity to grow a crop on a particular location, irrespective of whether that crop is actually cultivated there. Therefore the GAEZ covers all current growing areas of all crops and provides a start and length of the growing season for those. A drawback of the GAEZ is that cropping patterns that are a result of farmer decisions rather than climate are not properly reflected in the GAEZ. For example, the GAEZ does not provide any information on double cropping which is particularly relevant for rice cultivation in Asia. For the agDMP product, the GAEZ calendars for "intermediate input, rainfed cropping" were implemented.

For the implementation of the AgDMP product, the GAEZ crop calendars were used for soybean, maize, spring-wheat and winter-wheat. The SAGE crop calendar was used for the first and second season rice.

2.2.5. Crop parameters

The crop parameters used by the CSSF model are taken from the WOFOST and LINTUL crop parameter datasets (http://wageningenur.nl/wofost). Table 3 lists the crop parameters currently used by the simulation model. All parameter values are crop-specific and therefore the actual values are not listed here. A distinction is made between scalar (type S) parameters and tabular parameters (type T) where the value of the parameter depends on another state, usually development stage or temperature.

Most of the parameters listed in Table 3 are globally constant for a given crop, except for the TSUM1 and TSUM2 parameters. Those two parameters define the total length of the growth cycle as the number of growing degree days required to reach maturity. The TSUM1 and TSUM2 parameters have been optimized for each grid where a crop is cultivated. The optimization was carried out by running the CSSF model from sowing date up till the harvest date (as given by the crop calendar' for all years 2005 up till 2015. Next, the average total TSUM overall all years was calculated and split into a separate TSUM1/TSUM2 using a fixed fraction.

Table 3: Crop parameters used by the CSSF crop simulation model.

2.3. Crop simulation model

The crop simulation model used for generating the AgDMP product is related to the LINTUL crop simulation model (Light INTensity UtiLisation) which describes the daily increase in crop biomass through a simple light use efficiency approach (Shibu et al. 2010). LINTUL has been applied for many different crops including cereals, potato and grasslands. However, the algorithm has been adapted to allow for a hybrid model that combines the satellite inputs with a daily time-step crop simulation model.
The hybrid model used to generate the AgDMP product is called CSSF (Crop Simulation by Satellite- derived Fapar) and consists of the following modules:

• A phenology module that computes crop phenological development for sowing or emergence up till crop maturity. The approach used here is a simple growing degree day model which accumulates degrees Celsius above a base temperature until the number of growing degree days has been reached that is required for maturity. This phenology module has been taken directly from the LINTUL model.
• An assimilation module that computes the net daily amount of assimilates as a function of intercepted light, radiation use efficiency and reduction factors for daily average and minimum temperatures. The amount of intercepted light is directly computed from the satellite FAPAR product and the incoming global radiation. Maintenance respiration is not computed separately but is assumed to be part of the light use efficiency.
• A partitioning module that determines the partitioning of assimilates to the different plant organs based on the plant phenological stage
• An overarching module that computes the dry matter increase of the different plant organs using plant organ specific conversion efficiencies to convert assimilates into dry matter.

Note that current version of CSSF does not take water/limitation into account as no water balance is included. However, we assume that water stress will be reflected in the satellite observed FAPAR because of a decrease in leaf area due to drought stress.

2.3.1. Phenology

Having an accurate simulation of the phenological development of a crop is of utmost importance to have an accurate estimation of the crop life cycle, canopy development, growth and yield.

Therefore, phenological development must be simulated using a robust approach taking into account differences in phenological response between crop types. For simulating phenological development we use a growing degree days approach with crop-specific estimates for the cardinal temperatures for phenological development.

The daily development rate is calculated as the daily effective temperature divided by the temperature sum needed to complete each development stage. Daily effective temperature is derived from the daily mean temperature (T_avg) corrected for a base temperature (T_base) below which phenological development halts and a cutoff temperature (T_ctf) above which phenological development does not increase anymore:

Where DVR is the development rate [day-1], T_eff the daily effective temperature [C] and TSUM_i [C day] the temperature required to complete stage i. The development stage can then be expressed as the daily accumulation of the development rate. The temperature sums for development stages are expressed as a TSUM1 being the sum of effective temperature between emergence and flowering, and TSUM2 as the sum of effective temperature from flowering to maturity. The impact of day length on crop development (photoperiodicity) is currently not taken into account. Estimates of the base and cut-off temperatures are available for many crops from the LINTUL and WOFOST models and the general literature.

2.3.2. Assimilation

The net assimilation rate [GASS in kg dry matter ha-1 day-1] is computed as the amount of PAR (estimated as 48% of the incoming global radiation) multiplied by the fraction of intercepted light by the crop canopy, the crop-specific light-use efficiency and three additional factors that describe the impact of atmospheric CO2 concentration, daytime temperature and daily minimum temperature on assimilation:

Here fAPAR is the satellite observed fraction of absorbed PAR, which is found by linear interpolation using the satellite derived faPAR at dekadal time steps. Figure 1 provides an example of the satellite derived fAPAR for wheat as the aggregated value for all wheat pixels in a 0.1x0.1 grid in in the Netherlands.

Light use efficiency (LUE) at each time-step is found by linearly interpolating the value through the light use efficiency table and taking the DVS from the phenology module as abscissa value (Figure 2).

The factor describing the impact of ambient CO2 concentration is based on a linear relationship between relative LUE and CO2 concentration. The relative impact is 1.0 (no impact) at 360ppm CO2, currently a fixed CO2 concentration of 400 ppm is assumed.

The last two factors determine the impact of temperature on CO2 assimilation which is needed because different plant species have a different response of assimilation to temperature.

The 𝑓_Tday represents the impact of the day time temperature on the assimilation. Day time temperature represents the average temperature during the period of daylight and during which assimilation takes place. It is computed as:

The daytime temperature is than used to derive a reduction factor by linear interpolation through a response function (Figure 4). The figure clearly demonstrates the difference in photosynthesis response function between wheat and maize.

Finally, 𝑓_Tmin represents the impact of minimum temperature on assimilation. The background of this reduction is that during night-time the assimilates, produced during daytime, are transformed into structural biomass. This process is hampered by low temperature. If these low temperatures prevail for several days, the assimilates accumulate in the plant and the assimilation rate diminishes and ultimately halts. The actual reduction factor based on the function as shown in Figure 5 is not computed on daily minimum temperature but a 7-day running average of daily minimum temperature is used instead.

Figure 1: Fraction of absorbed PAR [-] derived for wheat growing an area in The Netherlands in 2017. Wheat is usually harvested at the beginning of August coinciding with the depression in FaPAR values.

Figure 2: The light use efficiency for wheat as function of crop development stages. The drop in light use efficiency is due to a decrease in leaf nitrogen concentration towards the end of the growth cycle.

Figure 3: Relative impact of ambient CO2 concentration on light use efficiency for wheat and maize.

Figure 4: Impact of daytime temperature on daily assimilation rate for wheat (C3 photosynthesis) and maize (c4 photosynthesis).

Figure 5: Impact of low temperature on daily assimilation rate.

2.3.3. Partitioning

Partitioning is not regarded as a separate process but is directly attached to the development stage of the crop. Static partitioning tables describe the fraction of assimilates that will be sent to the different plant organs depending on the development stage of the crop. The mechanism is simple and fairly robust but it strongly depends on having a correct description of the crop phenological development and cropping calendar. Depending on their development and physiology , the partitioning pattern of a crop can be quite different. An example is provided for wheat (Figure 6) and soybean (Figure 7) with soybean showing a much more gradual increase in the partitioning to beans compared to the partitioning to grains in wheat.

Figure 6: Partitioning of daily assimilates between stems, grains and leaves as function of development stages.

Figure 7: Partitioning of daily assimilates between stems, beans and leaves as function of development stages.

2.3.4. Growth

The growth of the different plant organs is computed as the total assimilation rate multiplied by the fraction that is allocated to each plant organ and corrected for the conversion efficiency of assimilates into dry matter. The latter is required because conversion of carbohydrates into substances with high protein content is less efficiency compared to conversion to sugar or starch.

First, the total assimilates are partitioned into below-ground and above-ground production. Next, the total above-ground dry matter increase is partitioned into leaves, stems and storage organs (grains or beans).

2.4. Output data

The output of the CSSF model consists of three states:

1. Total above-ground production (TAGP) which represents the total above-ground dry matter accumulated in all the plant parts which includes the stems, leaves and storage organs (Figure 8).
2. Total weight storage organs (TWSO) which represents the dry matter of the harvestable product (grains or beans), see Figure 9.
3. The crop development stage (DVS) which represents the phenological development of the crop which starts at 0.0 at crop emergence, reaches 1.0 at flowering and reaches 2.0 at maturity (Figure 10).

All AgDMP products are provided as netCDF files providing the values of the three variables at particular dekad in the growing season. The dekadal values span the entire global cropping season for a given campaign year. A campaign year is defined as the year when the harvest of a particular crop takes place.

Since cropping calendars overlap from a global perspective, this has the implication that a netCDF file for a single campaign year actually consist of a time-series that is longer than one year. For example, the campaign year 2010 for soybean starts with the sowing of soybean in South America already in the beginning of September 2009, while the last soybeans in NE China only reach maturity at the end of September 2010.

Further, the maturity date cannot be predict beforehand as it depends on the number of growing degree days, therefore CSSF is always run until the 31st December of the campaign year. After reaching maturity, the values for the biomass and development stage do not change anymore and will be repeated up till December 31st. The additional advantage of this approach is that spatial aggregations of crop yields can be operated on all dekads without having to take the maturity date into account.

Figure 8: Simulated total above-ground production for soybean for a La Salle, Illinois.

Figure 9: Simulated total bean dry weight for soybean for a La Salle, Illinois.

Figure 10: Simulated crop development stage for soybean in La Salle, Illinois.

3. Product validation overview

The AgDMP product has been evaluated by comparing the predicted values with reported crop yields at regional level. In this study, the evaluation was carried out spatially and temporally for the United States of America, India, and China for the period 2000-2018, using reported yield statistics aggregated to the lowest available NUTS level of each region. For almost all the crops in the three countries, the skill error can be reduced by more than 25% for both the indicators TAGP and TWSO compared to a climatological trend, suggesting considerable performance in assessing interannual yield variability. The results have been documented in a paper which is currently under review for a peer-reviewed journal (Climate Services).

References

de Wit, Allard, Hendrik Boogaard, Davide Fumagalli, Sander Janssen, Rob Knapen, Daniel van Kraalingen, Iwan Supit, Raymond van der Wijngaart, and Kees van Diepen. "25 Years of the WOFOST Cropping Systems Model." Agricultural Systems 168 (January 2019): 154–167.

Fischer, G., Nachtergaele, F. O., van Velthuizen, H., Chiozza, F., Francheschini, G., Henry, M., ... & Tramberend, S. (2021). Global Agro-ecological Zones (GAEZ v4)-Model Documentation.

Sacks, William J., Delphine Deryng, Jonathan A. Foley, and Navin Ramankutty. "Crop Planting Dates: An Analysis of Global Patterns: Global Crop Planting Dates." Global Ecology and Biogeography (June 2010): no-no.

Shibu, M.E., P.A. Leffelaar, H. van Keulen, and P.K. Aggarwal. "LINTUL3, a Simulation Model for Nitrogen-Limited Situations: Application to Rice." European Journal of Agronomy 32, no. 4 (May 2010): 255–271.

Related articles

• Page:
  NEMO-ERSEM and POLCOMS-ERSEM: User Guides for Products (v1.1)
• Page:
  Sea level and ocean wave time series and indicators
• Page:
  Product user guide for sea level and ocean wave products - time series and indicators
• Page:
  ECDE indicators catalogue entry: Product User Guide
• Page:
  Climate variables for cities in Europe from 2008 to 2017 documentation