
Bulletin
wall bulletinMenu
- Overview
- Country analysis
- Afghanistan
- Angola
- Argentina
- Australia
- Bangladesh
- Belarus
- Brazil
- Canada
- Germany
- Egypt
- Ethiopia
- France
- United Kingdom
- Hungary
- Indonesia
- India
- Iran
- Italy
- Kazakhstan
- Cambodia
- Sri Lanka
- Morocco
- Mexico
- Myanmar
- Mongolia
- Mozambique
- Nigeria
- Pakistan
- Philippines
- Poland
- Romania
- Russia
- Thailand
- Turkey
- Ukraine
- United States
- Uzbekistan
- Viet Nam
- South Africa
- Zambia
- Algeria
- Kyrgyzstan
- Lebanon
- Mauritius
- Madagascar
- Kenya
- Syria
Authors: air_panqc | Edit: lirui
Annex B.Quick reference to CropWatch indicators,spatial units and methodologies
The following sections give a brief overview of CropWatch indicators and spatial units, along with a description of the CropWatch production estimation methodology. For more information about CropWatch methodologies, visit CropWatch onlineat cloud.cropwatch.com.cn.
Agroecological zones for 49 key countries
Overview
232 agroecological zones for the 49 key countries across the globe
Description
49 key agricultural countries are divided into 232 agro-ecological zones based on cropping systems, climatic zones, and topographic conditions. Each country is considered separately. A limited number of regions (e.g., region 001, region 027, and region 127) are not relevant for the crops currently monitored by CropWatch but are included to allow for more complete coverage of the 49 key countries. Some regions are more relevant for rangeland and livestock monitoring, which is also essential for food security.
| ||
CropWatch indicators
The CropWatch indicators are designed to assess the condition of crops and the environment in which they grow and develop; the indicators—RAIN (for rainfall), TEMP (temperature), and RADPAR (photosynthetically active radiation, PAR)—are not identical to the weather variables, but instead are value-added indicators computed only over crop growing areas (thus for example excluding deserts and rangelands) and spatially weighted according to the agricultural production potential, with marginal areas receiving less weight than productive ones. The indicators are expressed using the usual physical units (e.g., mm for rainfall) and were thoroughly tested for their coherence over space and time. CWSU are the CropWatch Spatial Units, including MRUs, MPZ, and countries (including first-level administrative districts in select large countries). For all indicators, high values indicate "good" or "positive."
INDICATOR | ||||
BIOMSS | ||||
Biomass accumulation potential | ||||
Crop/ satellite | Grams dry matter/m2, pixel or CWSU | An estimate of biomass that could potentially be accumulated over the reference period given the prevailing rainfall and temperature conditions. | Biomass is presented as maps by pixels, maps showing average pixels values over CropWatch spatial units (CWSU), or tables giving average values for the CWSU. Values are compared to the average value for the recent fifteen years, with departures expressed in percentage. | |
CALF | ||||
Cropped arable land and cropped arable land fraction | ||||
Crop/ | [0,1] number, pixel or CWSU average | The area of cropped arable land as fraction of total (cropped and uncropped) arable land. Whether a pixel is cropped or not is decided based on NDVI twice a month. (For each four-month reporting period, each pixel thus has 8 cropped/ uncropped values). | The value shown in tables is the maximum value of the 8 values available for each pixel; maps show an area as cropped if at least one of the 8 observations is categorized as "cropped." Uncropped means that no crops were detected over the whole reporting period. Values are compared to the average value for the last five years, with departures expressed in percentage. | |
CROPPING INTENSITY | ||||
Cropping intensity Index | ||||
Crop/ | 0, 1, 2, or 3; Number of crops growing over a year for each pixel | Cropping intensity index describes the extent to which arable land is used over a year. It is the ratio of the total crop area of all planting seasons in a year to the total area of arable land. | Cropping intensity is presented as maps by pixels or spatial average pixels values for MPZs, 47 countries, and 7 regions for China. Values are compared to the average of the previous five years, with departures expressed in percentage. | |
NDVI | ||||
Normalized Difference Vegetation Index | ||||
Crop/ Satellite | [0.12-0.90] number, pixel or CWSU average | An estimate of the density of living green biomass. | NDVI is shown as average profiles over time at the national level (cropland only) in crop condition development graphs, compared with previous year and recent five-year average, and as spatial patterns compared to the average showing the time profiles, where they occur, and the percentage of pixels concerned by each profile. | |
RADPAR | ||||
CropWatch indicator for Photosynthetically Active Radiation (PAR), based on pixel based PAR | ||||
Weather/Satellite | W/m2, CWSU | The spatial average (for a CWSU) of PAR accumulation over agricultural pixels, weighted by the production potential. | RADPAR is shown as the percent departure of the RADPAR value for the reporting period compared to the recent fifteen-year average, per CWSU. For the MPZs, regular PAR is shown as typical time profiles over the spatial unit, with a map showing where the profiles occur and the percentage of pixels concerned by each profile. | |
RAIN | ||||
CropWatch indicator for rainfall, based on pixel-based rainfall | ||||
Weather/ satellite | Liters/m2, CWSU | The spatial average (for a CWSU) of rainfall accumulation over agricultural pixels, weighted by the production potential. | RAIN is shown as the percent departure of the RAIN value for the reporting period, compared to the recent fifteen-year average, per CWSU. For the MPZs, regular rainfall is shown as typical time profiles over the spatial unit, with a map showing where the profiles occur and the percentage of pixels concerned by each profile. | |
TEMP | ||||
CropWatch indicator for air temperature, based on pixel-based temperature | ||||
Weather/ satellite | °C, CWSU | The spatial average (for a CWSU) of the temperature time average over agricultural pixels, weighted by the production potential. | TEMP is shown as the departure of the average TEMP value (in degrees Centigrade) over the reporting period compared with the average of the recent fifteen years, per CWSU. For the MPZs, regular temperature is illustrated as typical time profiles over the spatial unit, with a map showing where the profiles occur and the percentage of pixels concerned by each profile. | |
VCIx | ||||
Maximum vegetation condition index | ||||
Crop/ | Number, pixel to CWSU | Vegetation condition of the current season compared with historical data. Values usually are [0, 1], where 0 is "NDVI as bad as the worst recent year" and 1 is "NDVI as good as the best recent year." Values can exceed the range if the current year is the best or the worst. | VCIx is based on NDVI and two VCI values are computed every month. VCIx is the highest VCI value recorded for every pixel over the reporting period. A low value of VCIx means that no VCI value was high over the reporting period. A high value means that at least one VCI value was high. VCI is shown as pixel-based maps and as average value by CWSU. | |
VHI | ||||
Vegetation health index | ||||
Crop/ | Number, pixel to CWSU | The average of VCI and the temperature condition index (TCI), with TCI defined like VCI but for temperature. VHI is based on the assumption that "high temperature is bad" (due to moisture stress), but ignores the fact that low temperature may be equally "bad" (crops develop and grow slowly, or even suffer from frost). | Low VHI values indicate unusually poor crop condition, but high values, when due to low temperature, may be difficult to interpret. VHI is shown as typical time profiles over Major Production Zones (MPZ), where they occur, and the percentage of pixels concerned by each profile. | |
VHIn | ||||
Minimum Vegetation health index | ||||
Crop/ | Number, pixel to CWSU | VHIn is the lowest VHI value for every pixel over the reporting period. Values usually are [0, 100]. Normally, values lower than 35 indicate poor crop condition. | Low VHIn values indicate the occurrence of water stress in the monitoring period, often combined with lower than average rainfall. The spatial/time resolution of CropWatch VHIn is 16km/week for MPZs and 1km/dekad for China. | |
CPI | ||||
Crop Production Index | ||||
Crop/ | Number, pixel to CWSU | The average crop production situation for the same period in the past five years was used as a benchmark to make an overall estimate of the current season's agricultural production situation. | Based on the VCIx, CALF, land productivity and area of irrigated and rainfed cropland in the current monitoring period and the same period in the past five years for the spatial unit, a mathematical model proposed by CropWatch is used to calculate the index expressed as a normalized value. A value of 1.0 represents the basic normal crop production situation in the current period for the spatial unit, and the higher the value, the better the crop production situation in the current period. Conversely, the lower the value, the worse the crop production situation for the spatial unit in the current period. | |
Note: Type is either "Weather" or "Crop”; source specifies if the indicator is obtained from ground data, satellite readings, or a combination; units: in the case of ratios, no unit is used; scale is either pixels or large scale CropWatch spatial units (CWSU). Many indicators are computed for pixels but represented in the CropWatch bulletin at the CWSU scale.
CropWatch spatial units(CWSU)
CropWatch analyses are applied to four kinds of CropWatch spatial units (CWSU): Countries, China, Major Production Zones (MPZ), and global crop Monitoring and Reporting Units (MRU). The tables below summarize the key aspects of each spatial unit and show their relation to each other. For more details about these spatial units and their boundaries, see the CropWatch bulletin online resources.
SPATIAL LUNITS | |
CHINA | |
Overview | Description |
Seven monitoring regions | The seven regions in China are agro-economic/agro-ecological regions that together cover the bulk of national maize, rice, wheat, and soybean production. Provinces that are entirely or partially included in one of the monitoring regions are indicated in color on the map below. |
Countries (and first-level administrative districts, e.g., states and provinces) | ||
Overview | Description | |
“Forty eight plus one” countries to represent main producers/exporters and other key countries. | CropWatch monitored 49 countries together represent more than 80% of the production of maize, rice, wheat and soybean, as well as 80% of exports. Some countries were included in the list based on criteria of proximity to China (Uzbekistan, Cambodia), regional importance, or global geopolitical relevance (e.g., four of five most populous countries in Africa). The total number of countries monitored is “48 + 1,” referring to 48 and China itself. For the nine largest countries— United States, Brazil, Argentina, Russia, Kazakhstan, India, China, and Australia, maps and analyses may also present results for the first-level administrative subdivision. The CropWatch agroclimatic indicators are computed for all countries and included in the analyses when abnormal conditions occur. Background information about the countries’ agriculture and trade is available on the CropWatch Website, cloud.cropwatch.com.cn. | |
| ||
Major Production Zones (MPZ) | ||
Overview | Description | |
Six globally important areas of agricultural production | The six MPZs include West Africa, South America, North America, South and Southeast Asia, Western Europe and Central Europe to Western Russia. The MPZs are not necessarily the main production zones for the four crops (maize, rice, soybean, wheat) currently monitored by CropWatch, but they are globally or regionally important areas of agricultural production. The seven zones were identified based mainly on production statistics and distribution of the combined cultivation area of maize, rice, wheat and soybean. | |
Global Monitoring and Reporting Unit (MRU) | |
Overview | Description |
105agro-ecological/agro-economic units across the world | MRUs are reasonably homogeneous agro-ecological/agro-economic units spanning the globe, selected to capture major variations in worldwide farming and crops patterns while at the same time providing a manageable (limited) number of spatial units to be used as the basis for the analysis of environmental factors affecting crops. Unit numbers and names are shown in the figure below. A limited number of units are not relevant for the crops currently monitored by CropWatch but are included to allow for more complete coverage of global production. Additional information about the MRUs is provided online under cloud.cropwatch.com.cn . |
Crop area estimation
In a global context, cultivated land exhibits significant differences in plot sizes, topography, and other influencing factors. The complexity of crop planting structures arises from diverse planting patterns seen in countries such as China, India, and many others across Asia and Africa. These regions often feature fragmented farmlands, common intercropping practices, and intricate crop planting arrangements. Despite the availability of high-resolution (10m) remote sensing data, mixed pixels are prevalent. Traditional crop area monitoring, reliant on image classification, faces challenges in accurately identifying crop types and assessing crop area due to these complexities, leading to high uncertainties. To overcome these challenges, the project team has pioneered a revolutionary approach by integrating Crop Planting Proportion Remote Sensing Monitoring (CPTP) with crop classification proportion sampling surveys. This innovative technique enables the operational monitoring of global crop sowing progress and planting areas during the growing season. The methodology achieves a remarkable accuracy exceeding 96%, both domestically and internationally. Specifically, identifying specific crop types in areas with complex planting structures proves challenging and uncertain. However, determining whether farmland is planted with crops is comparatively easier and more accurate. Leveraging the advantage of remote sensing in distinguishing between vegetation and non-vegetation, the project has introduced a universal method and technology for identifying cultivated land conditions. This approach allows precise dynamic monitoring of crop planting proportions just one month after crop emergence, achieving accuracy rates exceeding 98% in various regions worldwide. Crop planting structures are obtained through the GVG via field sampling. This innovative approach not only addresses the intricacies of crop planting structures in fragmented farmlands but also ensures highly accurate and timely monitoring of global crop planting progress.
In regions characterized by simple planting structures and large plots, complexities arising from variations in crop growth and phenological stages across extensive areas lead to locally low accuracy in detailed crop classification mapping. Additionally, the absence of remote sensing data covering the entire growing season presents a significant challenge for precisely identifying crop types during the current season. This situation creates a fundamental contradiction between the high dependence on samples and the difficulty in acquiring sample data. To address these challenges, the report introduces an automated extraction of training samples and structured single-class classification technology. This is achieved by identifying key phenological stages and features specific to different crop types. The method integrates deep learning models with temporal and spatial features and incorporates shape attention mechanisms. This approach facilitates intelligent extraction of plot boundaries and crop classification at the plot level, eliminating salt-and-pepper noise in classification. This innovative approach achieves accurate identification of major crops such as wheat and rice without sample support, with classification accuracy exceeding 90%. This method not only resolves the complexities associated with varying crop growth stages but also ensures precise and reliable crop classification, even in the absence of extensive sample data.
To tackle challenges in sensitive areas for ground data acquisition, a method incorporating encrypted temporal feature reconstruction, phenological correction, and time-series matching technology was developed. This innovative approach resulted in the creation of a crop classification model with the capability for annual migration application. The accuracy of cross-year migration application reached 92%, markedly reducing the reliance on ground-based sample data for fine classification of crops in international regions. This methodology has operationalized large-scale crop classification at the regional level, providing reliable and consistent results across different years.
For estimating the planting area of major producing countries where ground observations cannot be conducted effectively and crop growth characteristics cannot be obtained, the report introduces the concept of Cropland Planting Fraction (CALF) for calculation, as shown in the following formula:
In this formula, a and b are two coefficients obtained through linear regression of the time series cropland planting fraction (CALF) from 2002 to 2022 and the area statistical data from FAOSTAT or those released by various countries from 2002 to 2022. The cropland planting fraction for each country is calculated through the CropWatch system. The formula calculates the area change based on the planting area values for the current year and the previous year.
Crop yield estimation
The factors influencing crop yields exhibit dynamic variations across different regions worldwide. Meteorological conditions, soil moisture, pest and disease pressures, nutrient supply, and management practices diverge significantly from one region to another. Current models lack the precision required for accurately estimating crop yields on a global scale.
To address the complexities arising from diverse and dynamic factors impacting crop yields across vast regions, the research team of this report has developed four distinct remote sensing models for crop yield estimation. These models encompass the biomass-harvest index model, remote sensing index model, coupled remote sensing and crop model, and data-driven model.
The crop yield model that integrates biomass and harvest index quantifies the influence of diurnal temperature variation, maximum temperature, minimum temperature, and water stress on crop photosynthetic efficiency, thereby improving yield monitoring accuracy. This model is particularly effective for predicting crop yields in rain-fed agricultural regions.
In the biomass productivity model, crop biomass can be represented as the cumulative value of absorbed and converted photosynthetically active radiation (APAR) during the growing season, as shown in the following equation:
Here, APAR, the absorbed photosynthetically active radiation, can be obtained by multiplying photosynthetically active radiation (PAR) with the fraction of absorbed photosynthetically active radiation (FPAR). PAR represents the portion of total solar radiation reaching the Earth's surface that can drive vegetation photosynthesis. ε denotes the efficiency with which vegetation converts absorbed APAR through photosynthesis into organic carbon. Generally, vegetation's efficiency in utilizing light varies with environmental conditions during the growing season and is primarily influenced by temperature and water stress.
In the equation, ε* represents the maximum light use efficiency (gMJ-1), T1 and T2 indicate the inhibitory effect of environmental temperature on light use, and W represents the water stress coefficient, expressing the extent to which moisture influences vegetation's use of light.
The Harvest Index (HI), also known as the Coefficient of Economic Yield, represents the percentage of crop yield in relation to the above-ground biological yield. The biological significance of the Harvest Index mainly reflects the proportion of assimilated products allocated between grains and vegetative organs. Since the formation of the crop Harvest Index is a dynamic process, it can be simulated based on its formation mechanism to model the entire process. Before the fruiting stage, crops mainly grow organs such as roots, stems, and leaves, converting absorbed photosynthetic energy into stored carbohydrates in various organs. After the fruit starts growing, the crop's growth focus shifts towards the fruit, with absorbed photosynthetic energy primarily supplying the fruit's growth. Additionally, the carbohydrates stored earlier will also be partially transferred to the fruit. Based on this phenomenon during the fruiting process, modeling can be employed to simulate the allocation process of absorbed and already fixed photosynthetic energy within the crop. This modeling approach ultimately achieves the goal of estimating crop Harvest Index or yield.
The second type of model is the remote sensing index model. For rain-fed areas, this report primarily utilizes the Normalized Difference Vegetation Index (NDVI) as a regression parameter, analyzing its correlation with multi-year crop yields and establishing a statistical model for crop yield based on NDVI. However, in irrigated agriculture areas and high-density planting areas, existing remote sensing indices encounter signal saturation issues during the crop's peak growth period. To address this challenge, a coupled yield monitoring model integrating coherent polarization and red-edge index was developed. This model effectively resolved the optical remote sensing signal saturation problems in irrigated areas and high-density planting regions, reducing errors by 30%. Additionally, to account for the impact of water stress on rain-fed crop yields, a rain-fed crop yield prediction method considering multidimensional drought indices was proposed. This innovative approach enhances the reliability of the remote sensing index model.
The third type of model integrates crop models based on physiological and ecological mechanisms with remote sensing data. This approach typically employs data assimilation methods to incorporate soil and meteorological data into the models, aiming to accurately reflect the changes in environmental factors essential for crop growth and simulate crop yields. In this report, the advantages of ensemble filtering and variational assimilation algorithms are combined, taking into account the peak Leaf Area Index (LAI) during the flowering period of winter wheat. A novel assimilation method based on the double flow-dependent ensemble square root filtering for crop model remote sensing assimilation yield estimation is proposed. By coupling remote sensing with the Canopy Photosynthesis and Assimilation in agricultural Crops (CASA) model and different crop growth models like WOFOST or APSIM, and considering the computational efficiency of remote sensing models and the mechanistic advantages of crop growth models, the study achieves the prediction of crop yields throughout the entire growing season.
In recent years, machine learning methods have become prevalent in crop yield prediction due to the numerous factors influencing crop yields. However, the black box nature of machine learning models has faced criticism. Addressing this concern, the development team of this report proposed a crop yield prediction model that combines extreme gradient boosting with multidimensional feature engineering. They also integrated game theory and local interpretable model-agnostic explanations to enhance the model's interpretability. This innovative approach not only identified the optimal prediction window and the primary features influencing soybean yields but also enabled precise predictions of crop yields at both national and county scales.
Through the integration of multiple models and processes, a comprehensive global crop yield remote sensing monitoring system has been established. This advanced system tackles challenges like high uncertainty in single models and limited responsiveness to extreme events, providing a robust solution for monitoring global crop yields.
Crop production estimation
CropWatch estimates the current year's crop production based on the previous year's production. The calculation formula is as follows:
Where i represents the current year, and
represent the change rates of yield and area compared to the previous year, respectively.
For China, the total production of various crops () is estimated through the product of yield (
) and area (
), calculated using the following formula:
For other major crop producing countries, the variation in yield () is obtained by establishing a relationship between the NDVI of the current year and the NDVI of the previous year's time series. The calculation formula is as follows:
Here, and
represent the spatial averages of NDVI time series for the current year and the previous year, respectively, after masking the crops. Taking into account the phenology of different crops in various countries, yield variation is calculated based on peak or mean NDVI values from the time series curve.