Contributors: Allard de Wit (WAGENINGEN ENVIRONMENTAL RESEARCH)
Issued by: WEnR
Issued Date: October 2021
Ref: D422Lot1.WEnR.2.3.2
Official reference number service contract: 2017/C3S_422_Lot1_WEnR/SC2
History of Modifications
Acronyms
Scope of the document
This document serves as the Product User Guide for the crop productivity indicators based on observations developed as part of the C3S Global Agriculture contract for the Sectoral Information Systems (SIS) in Copernicus Climate Change Service. More information about the project can be found at https://climate.copernicus.eu/global-agriculture-project
Executive summary
The EO-based crop productivity indicators provide insight in the productivity and yield of the four main crops (wheat, maize, soybean, rice) and main production regions at a global scale over the period 2000-2018. The algorithm combines earth observation data of plant light interception with a simple crop model that converts intercepted light into crop biomass and uses a phenological model to determine the cropping season length.
The product is provided at a 0.1x0.1 degree resolution and can be used to analyse the effect of climate variability on crop yields at regional scale. The product is not suitable for field scale analysis and is therefore only provide at the 0.1 degree aggregated level. The data product contains three main variables:
- The total above-ground production in kg/ha (acronym: TAGP). This variable provides an estimate of the total above-ground productivity of the plant in terms of dry matter content.
- The harvestable plant product in kg/ha (acronym: TWSO). This variable provides an estimate of the yield of the crop (grains, beans) in terms of dry matter content.
- The crop development stage (acronym: DVS). This variables provides insight into the phenological development of the crop and is expressed as 0.0 at emergence, 1.0 at anthesis (flowering) and 2.0 at physiological maturity.
The product is provided as individual netCDF files which contain the results at a single dekadal (10-daily) time step.
The current version of the productivity indicators has several limitations:
- The basic satellite input data has a spatial resolution of 1 km. This means that only agricultural areas with cropping patterns that contain a limited number of dominant crops are included. For example, most of the agricultural areas in Africa are excluded because the cropping patterns in areas with smallholder farming cannot be resolved at a 1km spatial resolution.
- The simulation model does not directly include crop water limitations in the simulation. Currently, water limitations are expressed through a decrease in crop light interception as observed by the satellite (basically the leaves of the crop are dying). This implies that long term drought effects are reflected in the product but short term droughts will not be properly accounted for. In a next version of the product a crop water balance may be integrated that accounts for this kind of effects.
- The earth observation based data on plant light interception are aggregated based on a crop mask at 1km level. This crop dominance mask is a static product, although it is generally known that cropping areas change or expand.
Also, the crop mask has a limited accuracy. These problems are most likely to occur in regions with complex terrain with landscape mosaics of mixed agriculture and natural vegetation. In such cases, for example, satellite pixels with evergreen forest may have been misclassified as rice leading to unrealistic simulation results.
The product is based on well-known principles of crop growth which can be found in many other crop simulation models. A formal validation of the product against regional reported crop yields has been carried out. The results have been documented and at the time of writing are currently under review for publication in a peer-reviewed journal (Climate Services).
Figure 1: Development stages for maize at 30 June 2018. Crops in the southern hemisphere have already reached maturity (green), while crops in the Northern hemisphere are various stages of development.
Product description
Description for version 1.0
Introduction
One of the products available in the Copernicus Global Land Service is the Dry Matter Productivity product (DMP1). DMP aims to estimate the vegetation growth rate and the cumulated DMP values over the growing season should yield to the total vegetation dry matter production. The DMP product employs a simple light use efficiency model to estimate CO2 assimilation and the corresponding growth of dry matter. Although the Copernicus DMP product does a fair job in estimating regional vegetation productivity, it has some severe limitations for estimating agricultural productivity:
- The light use efficiency (LUE) parameter of the DMP algorithm is taken from a study focusing on broadleaf forests and is therefore not optimized for agricultural A review of the LUE efficiency parameters of the LINTUL crop simulation model shows that parameter values range from 2.7 to 3.6 kg DM/GJ for C3 crops and from 4.5 to 5.0 kg DM/GJ for C4 crops. This differ considerable from the value of 2.54 used in the DMP algorithm (v1).
- Currently, the DMP algorithm does not estimate the harvestable product from the vegetation productivity estimates. Combination of DMP with a simple phenological model would allow estimating the harvest index and therefore the harvestable
- The estimated vegetation growth rate is not limited by water availability. This water-limitation factor may be partially embedded in the fAPAR (fraction of absorbed photosynthetic active radiation) itself due to a reduced fAPAR as a result of a wilting crop However, this effect will lag with the true water stress effect and will fail to capture short term droughts.
- The DMP product in its current state has no relation to cropping cycles and can therefore not be meaningfully connected to statistical or reported crop yields. An aggregation is needed that connects with the locally prevailing cropping The latter may be estimated from vegetation cycles as represented by NDVI time-series.
Some of the issues noted above have been outlined by the authors of the DMP product as well.
Within the C3S project on global agriculture a new product on global crop productivity indicators has been implemented. This takes a similar approach as the original DMP product but several improvements have been implemented which make the product more relevant for agriculture:
- The satellite based fAPAR data have been made more crop-specific by combining them with a crop dominance mask taken from the USGS' Global Food Security Support Analysis Data (GFSAD) on global Crop Dominance2. This allows us to generate a 'pseudo' crop-specific mask for wheat, maize, rice and soybean and thus focus on areas with dominant cropping patterns.
- The satellite time series of fAPAR were aggregated using the pseudo crop- specific mask. Next they were integrated into a simple crop model that was calibrated on the local cropping season in order to have realistic estimates of crop
- The crop parameters were specific for each Particularly the light use efficiency differs considerably from the values used by the original DMP algorithm.
Product description
Biophysical description
As part of the C3S project on global agriculture, three different crop productivity indicators have been implemented (Table 1). The first product is the crop development stage (acronym: DVS) which represents the stage of phenological development of the crop. DVS is a dimensionless indicator which starts at 0 at crop emergence, reaches 1.0 at flowering and finally 2.0 at physiological maturity. Other crop stages (tillering, heading, etc.) are not explicitly defined but because the scale is numeric it is easy to compare the crop development stages between years in order to determine if the crop is developing normal, slower or faster compared to the long term average. The phenological development also determines the grain filling stage and a correctly simulated phenology is very import to obtain a good estimate of the crop yield.
The second indicator is the crop total above-ground production (acronym: TAGP) which represents the total mass of the above-ground plant parts (stems, leaves and grains/pods) in terms of dry matter (kg/ha). Under normal growing conditions, the total above-ground production is mainly a function of the light use efficiency of the plant and the length of its cycle.
Finally, the third indicator is the crop yield (acronym: TWSO) in terms of grain/pod dry mass in kg/ha. It represents the part of the total above-ground production that ends up in the reproductive organs. The crop yield divided by the total above-ground production is known as the harvest index. The harvest index tends to vary from 0.6 for crops that mainly produce starch or sugar (potato) to 0.3 for crops with high protein/lipid content. Most cereal crops have a harvest index between 0.4 and 0.5.
See the Algorithm Theoretical Baseline Document (ATBD) for a more elaborate overview of the algorithms behind the crop productivity indicators.
Table 1: Overview of the crop productivity indicators generated in C3S Global Agriculture
Acronym | Description | Units |
DVS | Crop phenological stage: 0 at emergence, 1.0 at flowering, 2.0 at crop maturity | - |
TAGP | Total above-ground production (dry matter) | Kg/ha |
TWSO | Total weight storage organs (dry matter), e.g. grains or pods | Kg/ha |
Input data required to generate the product
Several input data sources are required to generate the crop productivity indicators. Amount those data sources are static data like the crop calendars, crop mask and crop
model parameters. The latter data sets are relatively static and will only require occasional updates and recalibration.
Other data sources are dynamic and need real-time updates in order to generate the productivity indicators. The real-time data requirements are the following:
- Global daily meteorological data including temperature, global radiation, precipitation, wind speed and humidity at 0.1 degree global resolution. The ERA-Interim reanalysis dataset has been used for the products spanning the campaign years 2000 up till 2018. For the campaign years since 2019, weather inputs have been derived from the new AgERA5 dataset (https://doi.org/10.24381/cds.6c68c9bb).
- Satellite observed fAPAR data provided by the Copernicus Global Land Service at 1 km spatial resolution and at dekadal (10-daily) time steps. For the campaign years 2000-2020, the 1 km product has been used provided by a series of satellites starting with SPOT-VEGETATION (2000-2013), PROBA-V (2014-2020) and Sentinel2/OLCI (2020-current). Since June 2020, the 1km FAPAR product is not available anymore. Therefore, the 333m product is used and aggregated to
File contents and data structure
The product is provided as individual netCDF files which contain a single dekadal time step of a particular variable (either TAGP, TWSO or DVS). The netcdf files are organized by variable, crop, campaign year, season and date. A campaign year is defined as the year when the crop harvest takes place. Since cropping calendars are different between the northern and southern hemisphere and also differ across regions, the total length of one campaign year can be more than a calendar year. For example, the campaign year 2010 for soybeans starts in Oct/Nov 2009 in Argentina and ends in November 2010 with the last soybeans of the 2010 campaigns being harvest in Northern China. At the same time, sowing has already started for the 2011 campaign in Argentina again.
Technically, the time-series of indicators are available as netcdf files up till the end of the campaign. This means that after the cropping season has ended for a given region, the indicator values are copied up till the end of the campaign year (Figure 2). This approach makes it easier to spatially aggregate results for a given campaign year over regions with varying sowing and maturity dates.
Finally, crops may have two growing seasons within one campaign year. This only applies for rice, which can have two growing seasons, particularly in South-East Asia. In the latter case there is a season id which is either 1 or 2.
Figure 2: Simulated total above-ground production for soybean for a grid in La Salle county, Illinois
Product target requirements
Crop productivity indicators only become useful when a sufficiently long archive is available in order to understand how current conditions compare against the historical record. For example, is current growing season warmer or wetter than normal and what is the expected impact on the crops. Therefore, the product requirements are that both a long historical archive is available as well near real-time product generation for monitoring current conditions.
Product Gap analysis
For many users in the agricultural community, assessments of crop development and associated climate anomalies for the running cropping season are highly relevant. This is especially true for the agro-policy and the agro-business communities, as early indications of production anomalies are of paramount importance for tax/subsidies, price volatility, logistics and production shortfall.
The current version of this dataset largely fulfils the target requirements. The length of the historical archive is limited by the availability of the satellite archive which is available since 2000. Although from a climatological perspective this is relatively short it can be regard as sufficiently long for agricultural purposes. Currently, the real-time products are generated with a delay of ~8 days on real-time. This delay is caused mainly by the availability of AgERA5 which comes available with a delay of 7 days on real-time. The Copernicus FAPAR satellite products have a delay on real-time which varies between 3 to 5 days. Closing the gap on real-time is only possible by integrating more recent weather data although for regional analysis an 8 day delay will often be sufficient.
Data usage information
Practical usage considerations during use of products
It should be taken into account that the model setup has been done using data with a global perspective. This means that, while being globally consistent, considerable deviations may occur when crop productivity indicators are compared with local data. A clear example of this type of problems is given by the crop calendars that were used to generate the product. The two sources of data for global crop calendars (Sacks et al. and GAEZ) both have their limitations (see ATBD section 3.2.4). The former is based on the regional crop calendars and often fails to describe gradients in sowing dates, while the latter neglects farmer decisions on cropping cycle. As a result, simulation results may deviate from local cropping patterns because of a difference in the prescribed sowing date and season length.
It is therefore not recommended to use the product for decision making at a very local level, but look more at regional aggregate values. Further, given the lack of calibration with local data, it is likely that the interannual and spatial variability displayed by the indicators is more useful than the absolute value.
A formal validation of the product against regional reported crop yields has been carried out. The results have been documented and are currently under review for publication in a peer-reviewed journal (Climate Services).
Known Limitations of product
The current version of the EO productivity indicators has several limitations:
- The basic satellite input data has a spatial resolution of 1 km. This means that only agricultural areas with cropping patterns that contain a limited number of dominant crops are included. For example, most of the agricultural areas in Africa are excluded because the cropping patterns in areas with smallholder farming cannot be resolved at a 1km spatial
- The simulation model does not directly include crop water limitations in the Currently, water limitations are expressed through a decrease in crop light interception as observed by the satellite (basically the leaves of the crop are dying). This implies that long term drought effects are reflected in the product but short term droughts will not be properly accounted for. In a next version of the product a crop water balance may be integrated that accounts for this kind of effects.
- The earth observation based data on plant light interception are aggregated based on a crop mask at 1km level. This crop dominance mask is a static product, although it is generally known that cropping areas change or expand. Also, the crop mask has a limited accuracy. These problems are most likely to occur in regions with complex terrain with landscape mosaics of mixed agriculture and natural vegetation. In such cases, for example, satellite pixels with evergreen forest may have been misclassified as rice leading to unrealistic simulation
