1. Introduction

The increasing availability and access to satellite earth observation (EO) data at higher spatial, temporal and spectral resolutions, and UAVs and smart phone cameras to supplement or complement satellite data, as well simultaneous advances in using this data in crop modelling and machine learning and AI based analytics, has great potential in a wide range of agricultural applications. Potential applications include  crop monitoring for drought management, on-farm crop  production management, crop surveys for yield prediction,  crop insurance among others. Effectively leveraging these technological advances in crop management across scales (farm to village  to regional and national for crop insurance payouts at village level, regional drought management or or other contingency planning, food security policies, etc.) requires understanding their capabilities and limitations in two contexts: (i) the context of how they help to capture crop health and yield at local farm/village scale,  and (ii) how to organize and process the growing variety and massive volume (several petabytes) of EO data for easy access by users for application  from local to  region to national scales  in a systematic, reliable, consistent and traceable way. (as the data far exceeds the memory, storage and processing capabilities of traditional personal computers).

In this review, we identify and assess the state-of-art of advances in: (i)satellite remote sensing technologies  for crop  monitoring, (ii) earth observation and big data analytics platforms to store and process massive EO data, and (iii)  global crop monitoring platforms.

2. Satellite remote sensing data sources and access

Over the last five decades, government and private space agencies have launched numerous earth observation (EO) enabling progressively increasing access to data at higher spatial, temporal and spectral resolutions (Table 1).  Among the most important are the Landsat missions since 1970s, and MODIS missions since 1980s, that provided continuous time series data over the past five decades which is largely responsible for the development of remote sensing science that enabled new applications in agricultural and economic development, including agricultural insurance.

Table 1: Indicative list of selected satellites/sensors and gridded data products of relevant indicators available in public and private domains for
crop monitoring and insurance

No.	Satellite/Sensor	Launch date	Ongoing / end date	Spatial resolution	Temperature sensor spatial resolution	Temporal resolution days	Category	Imagery Products available for crop monitoring *	Agency	Public/ private
1.	AVHRR	08/24/1981	ongoing	1.1 km		1.0,0.5	Biophysical	Vegetation Indices	NOAA	public
2.	MODIS -Terra	12/18/1999	ongoing	250m,500m,1km	1km	1.0	Biophysical	Temperature, LST	NASA	public
3.	MODIS-Terra	12/18/1999	ongoing	250m,500m,1km	1km	1.0	Biophysical	Vegetation Indices	NASA	public
4.	VIIRS	11/28/2011	ongoing	500m		1.0	Biophysical	Vegetation Indices	NASA/NOAA	public
5.	VIIRS_T	11/28/2011	ongoing	750m		1.0	Weather	Temperature, LST	NASA/NOAA	public
6.	ECOSTRESS	06/29/2018	ongoing	30m, 60m	60m	1-7 days	Biophysical	Plant Temperature, Evaporative Stress Index	NASA	public
7.	Landsat-1 MSS	07/23/1972	01/06/1978	30m, 60m		16	Biophysical	Vegetation Indices	NASA	public
8.	Landsat-2 MSS	01/22/1975	02/25/1982	30m, 60m		16	Biophysical	Vegetation Indices	NASA	public
9.	Landsat-3 MSS	03/05/1978	03/31/1983	30m, 60m		16	Biophysical	Vegetation Indices	NASA	public
10.	Landsat-4 MSS	07/16/1982	12/14/1993	30m, 60m		16	Biophysical	Vegetation Indices	NASA	public
11.	Landsat-5 TM	03/01/1984	01/15/2013	30m, 60m	120m	16	Biophysical	Vegetation Indices	NASA	public
12.	Landsat-7 ETM+	04/15/1999	ongoing	30m, 60m	60m	16	Biophysical	Vegetation Indices	NASA	public
13.	Landsat-8 OLI/TIRS	02/11/2013	ongoing	30m, 60m	100m	16	Biophysical	Vegetation Indices	NASA	public
14.	ASTER	Dec 1999	ongoing	15m, 30m, 90m	15m	1	Biophysical	Elevation, Surface reflectance, Emissivity, LST	NASA, METI(Japan)	public
15.	Sentinel-1A SAR	04/03/2014	ongoing	10m			Biophysical	Soil Moisture, ET	ESA	public
16.	Sentinel-1B SAR	04/25/2016	ongoing	10m			Biophysical	Soil Moisture, ET	ESA	public
17.	Sentinel-2A MSI	06/23/2015	ongoing	10m,20m,60m		10 (5)	Biophysical	Vegetation Indices, FAPAR, LAI	ESA	public
18.	Sentinel-2B MSI	03/07/2017	ongoing	10m,20m,60m		10 (5)	Biophysical	Vegetation Indices	ESA	public
19.	Sentinel-3 SLSTR	02/16/2016	ongoing	1km			Weather	Temperature, LST	ESA	public
20.	Sentinel-3A	02/16/2016	ongoing	300m			Biophysical	Vegetation Indices	ESA	public
21.	Sentinel-3B	04/25/2018	ongoing	300m			Biophysical	Vegetation Indices	ESA	public
22.	Sentinel-3B	04/25/2018	ongoing	1km			Weather	Temperature, LST	ESA	public
23.	Sentinel-5P TROPOMI	10/13/2017	ongoing	7km	17		Biophysical	Air quality, Ozone, GHGs	ESA	public
24.	SPOT-1	02/22/1986	12/31/1990	20m			Biophysical	Vegetation Indices	CNES	public
25.	SPOT-2	01/22/1990	07/31/2009	20m			Biophysical	Vegetation Indices	CNES	public
26.	SPOT-3	09/26/1993	11/14/1997	20m			Biophysical	Vegetation Indices	CNES	public
27.	SPOT-4	03/24/1998	07/31/2013	20m			Biophysical	Vegetation Indices	CNES	public
28.	SPOT-5	05/04/2002	03/31/2015	10m,20m			Biophysical	Vegetation Indices	CNES	public
29.	SPOT-6	09/11/2012	ongoing	6m			Biophysical	Vegetation Indices	CNES	public
30.	SPOT-7	06/30/2014	ongoing	6m			Biophysical	Vegetation Indices	CNES	public
31.	Pleiades 1A HiRI	12/17/2011	ongoing	2.8 m		1	Biophysical	Vegetation Indices	CNES	public
32.	Pleiades 1B HiRI	12/02/2012	ongoing	2.8m		1	Biophysical	Vegetation Indices	CNES	public
33.	RADARSAT-1	11/04/1995	12/14/2007	10 -100m		7	Biophysical	Vegetation structure, Soil Moisture, ET	CSA	Public
34.	RADARSAT-2	12/14/2007	ongoing	3-100m		7	Biophysical	Vegetation structure, soil moisture, ET	CSA	public
35.	RADARSAT-Constellation	06/12/2019	ongoing	3-100		4		Vegetation structure, Soil Moisture, ET	CSA	public
36.	ALOS 1,2,3-AVNIR	2006, 2014, 2019	ongoing	10m		5	Biophysical	Vegetation indices	JAXA (Japan)	public
37.	ALOS 1,2,3-PALSAR	2006, 2014,2019	ongoing	10m		3	Biophysical	Soil moisture
38.	ALOS 1,2,3-PRISM	2006,2014,2019	ongoing	2.5m; ALOS 3, 0.8m)		5	Biophysical	Elevation
39.	Resourcesat-1 LISS-3	10/17/2003	ongoing	23.5m			Biophysical	Vegetation Indices	ISRO	public
40.	Resourcesat-1 LISS-4	10/18/2003	ongoing	5.8m			Biophysical	Vegetation Indices	ISRO	public
41.	Resourcesat-1 AWIFS	10/19/2003	ongoing	56m			Biophysical	Vegetation Indices	ISRO	public
42.	Resourcesat-2 LISS-3	4/20/2011	ongoing	23.5m			Biophysical	Vegetation Indices	ISRO	public
43.	Resourcesat-2 LISS-4	4/21/2011	ongoing	5.8m			Biophysical	Vegetation Indices	ISRO	public
44.	Resourcesat-2 AWIFS	4/22/2011	ongoing	56m			Biophysical	Vegetation Indices	ISRO	public
45.	OCO-2	07/02/2014	ongoing	3km	16		Biophysical	Solar induced fluorescence	NASA	public
46.	OCO-3	05/04/2019	ongoing	3km			Biophysical	Solar induced fluorescence	NASA	public
47.	CBERS 1	10/14/1999	8/1/2003	20m, 80m	160m	3, 26	Biophysical	Vegetation indices	CNSA / INPE	public
48.	CBERS 2	10/21/2003	12/31/2009	20m, 80m,260m	160m	3,5,5	Biophysical	Vegetation indices	CNSA / INPE	public
49.	CBERS 3	12/1/2013	12/1/2013	10m, 20m, 40m	80m	3,5,26	Biophysical	Vegetation indices	CNSA / INPE	public
50.	CBERS 4	12/1/2014	ongoing	10m, 20m, 40m, 64m	80	3,5,26	Biophysical	Vegetation indices	CNSA / INPE	public
51.	CBERS 4A	12/20/2019	ongoing	8m, 16m, 55m		5,31	Biophysical	Vegetation indices	CNSA / INPE	public
52.	GOSAT	2009	ongoing	10.5 km		3	Biophysical	Green house gases (GHG)	JAXA	public
				0.5, 1.5 km		3	Biophysical	Clouds and aerosols
53.	AMSRE
54.	SMAP	01/31/2015	ongoing	36km		2 to 3 days	Biophysical	Soil Moisture	NASA
55.	SMOS	11/02/2009	ongoing	40km		2 days	Biophysical	Soil Moisture	ESA	public
56.	TRMM	11/27/997	2014	5km		1 day	weather	Precipitation	JAXA/NASA	public
57.	GMI (GPM)	2/27/2014	ongoing	11km			Weather	Precipitation	NASA/JAXA	public
58.	GOES-1-17	10/16/1975	ongoing	1km		1day	Weather	Soil Moisture, ET	NOAA/NASA	public
59.	Meteosat-1-7	11/23/1977	1984	1km		30 min	Weather	Precipitation, Soil Moisture, ET	EUMETSAT/E	public
60.	Meteosat-8-11	08/28/2002	ongoing	1km		5, 15 min	Weather	Precipitation, Soil Moisture, ET	EUMETSAT/E	public
61.	CHIRPS (Gridded Data products)	01/01/1981	present	5km		1.0	Weather	Precipitation	UCSB	public
62.	CHIRTS(Gridded Data products)	01/01/1983	present	5km		1.0	Weather	Temperature	UCSB	public
63.	GRIDMET (Gridded Data products)	1979	present	4km				Rainfall, Relative Humidity, Tmin, Tmax, wind speed, vapour pressure deficit, reference evapotranspiration	UC Merced	Publ;ic
64.
65.	VanderSat	06/01/2002	ongoing	100m			Biophysical	Soil Moisture, ET	Vandersat	private
66.	IKONOS	1999	2015	4m		3	Biophysical		Digital Globe	private
67.	WorldView-1	09/18/2007	ongoing	0.5m-2m		1.7	Biophysical	Vegetation Indices	Maxar (Digital Globe)	private
68.	WorldView-2	10/08/2009	ongoing	0.5m-2m		1.1	Biophysical	Vegetation Indices	Maxar	private
69.	WorldView-3	08/23/2014	ongoing	0.3m-1.24m			Biophysical	Vegetation Indices	Maxar	private
70.	WorldView-4	11/11/2016	ongoing	0.3m-1.24m			Biophysical	Vegetation Indices	Maxar	private
71.	RapidEye -Planet	08/29/2008	03/31/2020	5m			Biophysical	Vegetation Indices	Planet	private
72.	PlanetScope	06/22/2016	ongoing	3m,5m			Biophysical	Vegetation Indices	Planet	private
73.	SkySat (Planet)	11/21/2013	ongoing	0.72m			Biophysical	Vegetation Indices	Planet	private
74.	aWhere	01/01/2006	ongoing	9km			Weather	Precipitation, Temperature	aWhere	private
75.	ICEYE-X1 to X10;  SAR	01/01/2018	ongoing	0.25 to 5m, 5 to 20m		hourly  to daily	Weather	Soil Moisture, flood mapping	ICEYE	private

^*Algorithms are available in respective toolkits of several satellites/sensors (AVHRR, MODIS, VIIRS, LANDSAT, SENTINEL ) for derived vegetation status indicators  of biophysical  importance in  representing important vegetation properties of crops  to assess  crop condition and yield estimation,  like  LAI (leaf area index),  FAPAR (Fraction of absorbed photosynthetically active radiation,  FVC (Fractional Vegetative Cover), and others.

Emerging Paradigm: Earth Observation Big Data and Analysis Cloud Platforms

Access to free, time series, global moderate to high resolution satellite remote sensing data has progressively increased over the last four decades (Table 1), starting with AVHRR (1981 to present) and MODIS (1999 to present), followed by publication by USGS of global Landsat (30m resolution) data Archive (1970 to present) in 2008/2009 under US open data initiative, and by the European Space Agency of Sentinel satellites high resolution data (10m, 20m, and microwave; 2015 onwards). A host of other public and private satellites also now provide access to high resolution (up to sub-meter) time series Earth Observation (EO) data, making the current EO data pool vastly different from a decade ago. These developments enabled scientists, businesses, and policy makers in various domains, including agriculture, to visualize the enormous potential of time series EO data in addressing a wide range of important environmental, economic and social problems, at local, regional, and global scales. But, the growing variety and volume (several petabytes) of EO data far exceeds the memory, storage and processing capabilities of traditional remote sensing data storage, distribution, and processing locally on personal computers. It is also not technically feasible to carry out mandatory pre-processing of long time series of raw EO data in the traditional paradigm. These factors have limited EO data use for application development in the traditional paradigm to only very small portions of available data.

The major hurdle in working with time series global scale EO data from diverse sources is in providing the proper connections between data, applications, and users. Overcoming this hurdle necessitates a paradigm shift in EO data retrieval, storage and analysis from local processing on personal computers towards: (i) adoption of next generation infrastructures based on cloud platforms and big data technologies for data storage and processing, and (ii) for automating the pre-processing stages of raw EO data into Analysis Ready Data (ARD). Analysis Ready Data are time-series stacks of satellite imagery that are ready for a user to analyze with minimal or no additional pre-processing of the imagery. They are a packaged product created after pre-processing raw EO data through standard multiple stages that include: searching and downloading data from various providers, image fusing, clipping the data to cover only the area of interest, correcting for geometry, sensors, radiometry, and atmosphere, identifying pixels shadowed by clouds or with poor quality data, and lining up the images pixel for pixel, by geospatially co-registering and resampling the data. Developing long term continuous ARD sets that are consistent (over time from same sensor, and across multiple sensors) is a work in progress that is evolving with new developments in both sensor technologies and algorithms.

The idea of ARD has also shifted the burden of pre-processing EO data from individual  users to data providers, and lowered technical barriers for users to fully utilize EO data. Usually ARD are provided as tiled interoperable, georegistered stacks of both Top of Atmosphere (TOA) reflectance and atmospherically corrected surface reflectance products, and with explicit quality assessment information and appropriate metadata for traceability. Users then work directly with data that are pre-processed and arranged in a coherent time series stack for their area of interest, instead of a bunch of randomly placed overlapping images. In the new paradigm, users also can significantly increase the scope and limits of their analysis by working with powerful cloud-computing platforms (instead of personal computers)  and advanced analyses with time series, ML, AI or other models, to address complex problems.

Operational EO big data platforms for agricultural applications

In recent years, several approaches to EO big data infrastructures for storage and processing on cloud platforms have evolved to enable and accelerate applications in different domains. Their common goals include helping users to achieve: (i)  easier access to EO data , (ii) easier use (storage and processing)  of EO data in the cloud, (iii) easier EO  big data analytics in the cloud, and (iv) better usability through tailored imagery web applications. Google was among the first to enable the shift towards using EO big data cloud platforms when it introduced the Google Earth Engine (GEE) in 2010, to enhance use of satellite imagery for large scale and time series applications. GEE set a benchmark in enabling universal access to its high power cloud computing resources for fast retrieval and processing of time series ARD from diverse sensors (nearly all sensors and gridded products in Table 1). In addition to ARD of multiple satellite data, the platform provides: (i) a large repository of other geospatial data, including environmental variables, weather, and climate forecasts, land cover, topography, and  socioeconomic data, and (ii) a portfolio statistical, ML and AI tools, for a wide array of applications (Gorelick et al, 2017). Its library comprises 600 + EO analysis ready datasets and 1000+ analytical tools. Each data source available on GEE has its own time series of EO/ARD data organized into a stack called Image Collection. Users can access Google Earth Engine with only an active internet connection and a Google account (250 GB free quota). With regular updates of EO data, tools, and features, the platform adapts to a wide range of user requirements and expertise. Users can also analyze their private data on the GEE platform with help of backend data and analytical tools. A summary of features of GEE and some other currently available EO data cloud platforms for potential use in crop monitoring and insurance is given in Table 2.

 

Table 2:  Cloud based platforms for EO data access and analysis

Platform	Satellite Datasets	Access	Special features
Google Earth Engine  (GEE) , 2010 (operated by Google)	Near real time and archived Global ARD Datasets with corresponding cloud masks; derived time series products of  major biophysical variables (NDVI, EVI, etc.);  Gridded weather  data products ; (nearly all sensors and gridded data sets of Table 1 are included on the platform)	Free and unlimited access to all public sensor time series data and  storage for research, education, and non-profit use ; Limited access to some private satellite datasets (Planet; MAXAR)	Users can leverage storage and computing resources of GEE platform;  allows scaling to large regional and global  analysis of time series data on the platform; open source code for extracting data and processing with a range of statistical and ML/AI algorithms; analysis results can be displayed on the fly in GEE browser and can also  be extracted to user systems for integration with other data;  users can contribute own datasets and algorithms and develop mobile apps with GEE  platform data and analysis tools; sufficiently user friendly for non-specialists in RS or  ML/AI tools; Closed source software, so cannot guarantee reproducibility as source code can change
Geospatial Big Data Platform GBDX (Maxar), 2018	ARD of MAXAR  (WorldView) data; and open data of  Sentinel and Landsat	Access based on purchase of subscription; Also offers a free Community Edition called GBDX Notebooks  that gives free access to open data (Landsat/ Sentinel) and some MAXAR data	Leverages Amazon Web Services (AWS) to deliver scalable storage and computing resources that can be used for geospatial analytics and AI machine learning applications; Does not allow export of derivatives or the open images except some limited extraction in the Community edition ( 6 GB instance and  20 GB of drive space).
Radiant Earth (Open source Platform of Radiant Earth Foundation, 2016)	Library of ARD of Sentinel-2 and Landsat data, ML algorithms and Training data sets	Free access to data, algorithms,  and training data sets for applications of ML/AI to support decisions in critical areas like agriculture, forests, disaster management, for sustainable  global development; Free access to available  crop land data to develop crop masks	Target audience Global Development Community (NGOs, Academics, entrepreneurs); Not designed to scale to large regional analyses; Focus on localized studies for ML applications; Repository of Training Data sets for ML algorithms; Allows maintaining own personal projects and bringing  in additional EO and secondary data; Allows sharing of data, training data sets, and algorithms with community.
Sentinel Hub Playground (Sinergise)	ARD of Sentinel, Landsat- MODIS and DEM	Flexible pricing from free to basic and enterprise level uses; Free access to explore and download satellite imagery for non-commercial/ research use; Paid access through specific protocols and API, data processing, mobile application data access, higher access limits	Between GBDX and GEE in function;  Limited variety of EO data; Closer to GEE in  terms of free access to data, analysis tools and direct display of results  in the browser; Allows customized analysis scripts but not sharing of scripts among users; Closed source code, so cannot guarantee Reproducibility; presents the best balance between the analyzed capacities. The drawback of the ODC solution is mainly the lack of support for reproducibility of science, which is not found in the others either. On the other hand, the other capacities evaluated are at least partially met.
SEPAL (System for Earth Observation Data Access, Processing and Analysis for Land Monitoring), 2018FAO –platform for forest and land monitoring	Open source ARD from GEE  and directly from other sources (including Sentinel, Pleiades, WorldView, etc)	Free access, largely  meant for developing countries with limited access to satellite data resources.	More focused on infrastructure management and  provision of tools for EO data analyses; So, . big data challenges are not directly addressed; Combination of GEE (for EO data)  and open source software ORFEO Toolbox (open source remote sensing data processing software for multiple sources),R and others for analysis; GEE is used for data retrieval and the Amazon Web Services Cloud (AWS) is used for data storage and  infrastructure for computing analyses.
Open Data Cube – ODC   First Developed  as  Australian Geoscience Data Cube (AGDC, 2017); Modified to allow diverse users, datasets and national or regional use options	Data cubes (time series stacks) of ARD of Landsat-5/7/8, Sentinel-1/2, MODIS ALOS-1/2, ASTER DEM and others	Available under Apache 2.0 license as a suite of applications These repositories include .	Generic framework composed of a series of data structures and tools for  organization and analysis of massive EO data sets; Open source code distributed through github repositories which include  web interface modules for data visualization, data statistics extraction tools as well as jupyter notebooks with examples of access and use of indexed data in ODC; Does not allow sharing of  applications and data Different National ODC implementations are operational in  Australia, Switzerland, Kenya,  UK, Taiwan, and many other countries

 

High-resolution, high-frequency, consistent, and more detailed time series satellite data based crop monitoring is needed over extended periods for effective implementation of crop insurance schemes nationwide, including PMFBY. The data management and analysis challenges arising from the huge time series data volumes can be overcome only with new cloud computing infrastructures, technologies and data architectures, such as those listed in Table 2. Among these GEE is currently the most developed with access to more data and analysis resources.  However, since it is closed platform, it cannot guarantee reproducibility source code can change. Sentinel Hub has relatively limited EO data resources and also has similar drawback as GEE for reproducibility. SEPAL is more focused on infrastructure management and the provision of tools for the retrieval and analysis of EO data. It does not directly address big data challenges of EO data storage and processing. Radiant Earth is focussed more on small area studies and machine learning tools. ODC provides a toolkit to facilitate application development by the user. ODC,  SEPAL and Radian Earth are open source platforms and their code is available in open repositories. ODC is the only platform  that gives a user direct access to data and its infrastructure and data processing capabilities.  It  provides public documents of the platform governance process and of how to create or incorporate new features into the platform including new software tools. It allows high replicability, provides for high scalability for storage and processing and the maximum opportunity for data access  interoperability. ODC also uses distributed data storage to minimize data movement during processing so that processing occurs where data is stored. While GEE is the most useful ready to use platform or users with its multiple source ARD, library of processing tools, but the  transient nature  of its code and uncertainty of its availability for the long term raise questions about reliability for use in  nationally important long term public agricultural schemes. For such schemes ODC may be the preferred platform for national schemes despite considerable effort involved initially in building the platform,  because of its generic open source scalable framework, distributed data storage, data and storage scalability, and interoperability of data sources.

 

In India, most research and application development with Indian satellite data (Resourcesat series) has so far been in the traditional paradigm, that is, processing only a small portion of available data for relatively few periods on personal computers. Some early attempts at developing ARD time series stacks for public access, and building a national ODC have only recently been initiated by ISRO.