RICA: A rice crop calendar for Asia based on MODIS multi-year data

 Bhogendra Mishra, Lorenzo Busetto, Mirco Boschetti, Alice Laborte, and Andrew Nelson   |  

Crop calendars report the timing of key agricultural activities and crop development. At their most detailed, crop calendars can support the planning of many farm activities, from land preparation, planting, crop, soil, and water management through the growing season, harvesting, storage, and fallow. In this study, we present a method to automatically generate a spatial rice crop calendar covering the main rice producing countries of Asia based on a long duration time series of remote sensing data to generate RICA: a RIce Calendar for Asia.

Crop calendars report the timing of key agricultural activities and crop development. At their most detailed, crop calendars can support the planning of many farm activities, from land preparation, planting, crop, soil, and water management through the growing season, harvesting, storage and fallow. Most crop calendars, however, contain information only on the timing of crop planting and harvesting for major crops, and report these dates as planting and harvesting windows for large geographic areas. This is particularly true for crop calendars that go beyond national scale (from multiple countries up to continental and global coverage).

Crop calendars have been characterised as essential or critical components of agricultural monitoring systems in a review and survey of such systems.The same review outlines their importance for specific applications such as monitoring crop condition, estimating crop area and harvested production, agricultural policy, emergency planning in the event of disasters. Crop calendars are also important layers of information for the United Nations Sustainable Development Goal (SDG) target 12.3 – to halve per capita global food waste at the retail and consumer levels and reduce food losses along production and supply chains, including post-harvest losses by 2030.

Crop calendars are constructed using one or more of the following approaches: (i) census-based calendars, compiled from multiple sources such as ground observations, expert opinion and/or survey data; or (ii) model-based calendars determined by climate data, crop characteristics and land surface phenology from remote sensing data.

Census-based crop calendars often rely on heterogeneous sources of data which are compiled into a single dataset. By relying on disparate data sources there may be issues of varying data quality and uncertainty that is rarely, if ever, documented within the crop calendar database. They often report at national level only or by large agro-ecoregions, because obtaining information at higher levels of spatial detail is time consuming. Also, calendars from different sources rarely refer to the same period.

As a result, they can vary in the level of spatial and temporal detail, are challenging to update on a regular basis, and are often ‘one shot’ efforts, despite known changes in planting and harvesting dates over time in response to changes in farming practices, markets, and climates. Model-based calendars often use climatic data such as temperature and rainfall, assumptions about crop varietal characteristics and management options, and Earth observation (EO) data to derive spatially and temporally consistent representations of planting and harvesting dates.

They may also use census-based calendars as an input. Their use of a wide range of input data, and their large geographic coverage (often global) limits effective validation with independent data, other than comparisons against other existing calendars. This lack of validation data as one of the biggest challenges, especially for large scale EO-based approaches (an example of model-based calendars) that are now established components of crop monitoring systems.

As one of the world’s most important staple crops, rice features in all major crop monitoring systems and is included in various crop calendar databases. EO-based rice crop calendars have been derived for major rice growing regions within countries: e.g., the Mekong delta in Vietnam, and northeast China ; whole countries, e.g., dry season rice cropping patterns in Bangladesh, and Japan; and for larger regions within Asia e.g., South Asia, and South and Southeast Asia.

While the national level EO-based studies were able to obtain validation data, the studies with larger geographic coverage were unvalidated, which is a persistent problem with large scale EO-based analysis. Few of the EO-based studies covered multiple years, whereas crop calendars should represent the typical cropping sequences rather than annual snapshots where the cropping calendar may be influenced by unusual weather patterns, or changes in labour availability that can delay planting or harvesting activities. None of the studies covered all of Asia, and we could not find any EO-based study of rice crop calendars for China.

Rice is the only crop to have its own detailed, global calendar called RiceAtlas. RiceAtlas is census-based, and reports planting and harvesting windows at subnational level for all major rice producing countries and many others, and otherwise reports at national level for small or minor rice producing countries. RiceAtlas reports the start, peak and end of planting and harvesting for each rice growing season in each spatial (administrative) unit. The average planting and harvesting window (i.e., the difference between start and end of planting/harvesting) in RiceAtlas is 42 days for the main season, 38 days in the second, and 30 days in the third season.

In Asia, the window is narrower for the main and second seasons at 35 days. RiceAtlas is not an EO-based approach and suffers from the limitations of census-based methods in that it is not easily updated, it is compiled from data sources with varying spatial and data attribute quality, and it is limited to spatial units that do not correspond to the spatial patterns of rice crop management practices. That said, RiceAtlas is the most spatially detailed, global crop calendar for any crop and remains the standard against which any EO-based rice crop calendar for all of Asia should be compared.

One such EO-based approach is PhenoRice. PhenoRice uses hypertemporal remote sensing imagery (spectral indices and bands from MODIS timeseries) to estimate the start of season (SoS), peak of season (PoS) and end of season (EoS) dates for all rice seasons detected within the timeseries. PhenoRice generates high spatial resolution maps (250 m in the case where MODIS data are used as input) of rice SoS, PoS and EoS. PhenoRice has been applied and tested in different rice growing environments, producing robust results. This robust performance across diverse sites suggests that PhenoRice can be applied to larger areas.

The global, and relatively spatially detailed information in RiceAtlas coupled with the potential to apply PhenoRice at scale provides a unique opportunity to develop, test and validate an EO-based approach to map rice SoS and EoS consistently over large geographic areas. The unique characteristics of rice systems–high cropping intensity with up to three crops per year on the same plot of land, wide diversity in crop management practices, and its widespread cultivation in the cloudy monsoon season, suggest that this is one of the most challenging staple crops for such an approach.

Here we demonstrate an EO-based approach to provide crop calendar information that is comparable to the best available rice crop calendar (RiceAtlas). By using an EO-based approach, we ensure that the method is automated, replicable, and repeatable. Such approaches are particularly valuable in the context of tracking changes in cropping cycles over time, due to climate change or due to policies that affect diversification and land consolidation.

We use East, South, and Southeast Asia for this comparison since these regions account for 85% of the global rice area and 90% of global rice production. We use the PhenoRice algorithm and MODIS data for 2003–2016 to detect the number of rice seasons per calendar year and estimate SoS and EoS per pixel for each detected season. We further process the PhenoRice pixel-level outputs to estimate the number of seasons and the peak detections of SoS and EoS dates for each season per RiceAtlas administrative unit to generate RICA: a RIce Calendar for Asia.

Our aim is to test whether an automated EO-based method can derive a rice crop calendar for Asia that is comparable to RiceAtlas, but which is repeatable, replicable, and consistent. We assess the accuracy of the EO-based RICA dataset by comparing our SoS and EoS estimates to the peak crop establishment and harvesting dates for rice seasons reported in RiceAtlas. In addition to a continent-wide accuracy assessment, we explore the case of the Philippines as an example country where RiceAtlas is known to have particularly accurate and detailed rice season information.

In this study we present a method to automatically generate a spatial rice crop calendar covering the main rice producing countries of Asia based on a long duration timeseries of remote sensing data. The PhenoRice algorithm was used to generate pixel-level SoS and EoS estimates for rice for 2003–2016 which were then pooled and analysed to determine the number, SoS and EoS of rice seasons per administrative unit to create the RICA.

We compared RICA to the state-of-the-art RiceAtlas crop calendar across Asia and made country comparisons for Japan and the Philippines where RiceAtlas data are known to be of particularly high quality. Ad hoc comparisons against more locally tuned EO-based rice crop calendars suggested that the RICA performed comparatively well.

This comparison of EO-based and survey-based calendars showed that the EO-based RICA could estimate the number, SoS and EoS of rice seasons sufficiently well and could accommodate the wide range of SoS/EoS dates, cropping intensities, climates, and rice crop management practices across Asia. The RICA dataset consists of SoS and EoS dates per rice season per administrative unit and is a first crop-specific example of using EO information to generate a crop calendar across such a large geographic area using multi year data.

Read the study:
Mishra B, Busetto L, Boschetti M, Laborte A, and Nelson A. (2021) RICA: A rice crop calendar for Asia based on MODIS multi year data. International Journal of Applied Earth Observation and Geoinformation, Volume 103,102471.

Leave A Response