The agricultural sector in Laos accounts for approximately 15% of GDP, with the majority of the population engaged in farming. However, investing in expensive IoT sensors or large-scale IT infrastructure is not realistic for most smallholder farmers. This article explains how to combine free satellite data (Sentinel-2) with Google Earth Engine's NDVI analysis to monitor crop health without any sensor investment, and repurpose demand forecasting AI techniques for yield prediction — based on existing research data and proposed implementation scenarios. It is designed so that agricultural enterprises, NGOs, and government agency staff can take away concrete steps they can start implementing tomorrow.

Agriculture in Laos is one of the sectors where digitalization lags behind within Southeast Asia. Behind this lies a dual barrier of infrastructure and cost.

Laos's three main crops are rice, coffee, and corn. Rice is the backbone of domestic consumption, while coffee is a key source of foreign currency earnings. However, agricultural productivity remains low compared to neighboring countries; according to FAO data, rice yields per unit area are only 60–70% of those in Thailand and Vietnam.

The causes are multifaceted: underdeveloped irrigation infrastructure, limited access to fertilizers and pesticides, and the inability to detect crop diseases at an early stage. Farmland in the Mekong River basin also faces flood risks, and the majority of farmers have no choice but to rely on experience and intuition when making planting decisions.

When the author visited an agricultural cooperative in Vientiane Prefecture, the cooperative's chairman said: "By the time we noticed that coffee leaf rust was spreading, we had already lost 30% of our harvest." The range within which disease symptoms can be detected by the naked eye is limited—it is physically impossible to patrol several hectares of farmland every single day.

In advanced smart farming case studies, soil moisture sensors, weather stations, and drone aerial photography are introduced as the standard toolkit. For small-scale farmers in Laos, however, these tools remain out of reach.

Consider the specific costs: a single soil moisture sensor runs $200–$500, a weather station $1,000–$3,000, and an agricultural drone upwards of $5,000. Given that the average annual income of a farmer in Laos is roughly $1,500–$2,000, these devices represent several times a farmer's yearly earnings.

On top of that, stable electricity supply and mobile network coverage are assumed prerequisites — neither of which is guaranteed in rural villages. "Bought an expensive sensor, but couldn't charge it and couldn't use it" is no laughing matter; it is an actual failure pattern reported in smart farming projects across Southeast Asia.

Satellite remote sensing is attracting attention as an alternative to IoT sensors. It observes farmland from the sky and quantifies crop conditions. The key point is that a large portion of this technology is available for free.

Sentinel-2 is an Earth observation satellite operated by the European Space Agency (ESA) that publishes satellite imagery of the entire world for free. With a resolution of 10m and revisit cycle of 5 days, it provides sufficient precision for farm-level observation.

The index calculated from these satellite images is NDVI (Normalized Difference Vegetation Index). Plant leaves strongly reflect near-infrared light while absorbing red visible light. Healthy crops have higher NDVI values (0.6–0.9), while diseased or stressed crops show lower values (0.2–0.4).

By tracking NDVI changes over time, crop health changes can be detected without physically walking through the fields. Areas where values drop sharply are likely experiencing disease, water shortage, or nutrient deficiency.

Similar projects in Southeast Asia have reported detecting coffee leaf rust approximately 2 weeks before visual discovery through NDVI anomaly detection. Two weeks of early detection can significantly reduce the spread of affected areas.

Sentinel-2 image data is enormous. Processing several GB of data per scene on a local PC is simply not practical. That's where Google Earth Engine (GEE) comes in.

GEE is a cloud-based geospatial analysis platform provided by Google, and it is free for research and non-profit purposes. Many agricultural support projects in Laos fall under this category.

The greatest advantage of GEE is that there is no need to download satellite imagery. You can run JavaScript or Python code directly in the cloud, handling everything from NDVI calculation and time-series analysis to anomaly detection. All you need is a browser and an internet connection — local computing resources are not a concern.

With just this code, you can obtain six months of NDVI time-series data for a specified farmland area. Anyone with a basic knowledge of programming can set up a working analysis environment in one to two days.

From here, we introduce a proposed implementation scenario based on a typical agricultural environment in Laos. The target areas are assumed to be rice paddies in Vientiane Province (approximately 50 hectares) and coffee plantations on the Bolaven Plateau (approximately 20 hectares).

The first step is to collect satellite imagery of the target farmland. Filter on GEE using the following conditions:

• Target period: The most recent one year (covering 2 or more crop cycles)
• Cloud cover filter: 20% or less (this threshold is critical, as cloud cover is high during the rainy season in Laos)
• Atmospherically corrected data: Use S2_SR_HARMONIZED (corrected to surface reflectance)

A Laos-specific consideration is that the number of usable images drops significantly during the rainy season (May–October). A practical approach is to focus analysis on dry season (November–April) data, supplementing with rainy season imagery only when available.

During preprocessing, use GEE's cloud masking feature to remove the effects of any remaining clouds. Cutting corners here leads to the elementary mistake of misidentifying clouds as crop disease. In fact, this exact mistake was made early in the validation process, resulting in a false alert being sent to farmers indicating an anomaly. It was fortunate that the error was caught before any trust was lost.

Calculate NDVI from preprocessed imagery and generate time-series graphs. NDVI for healthy crops follows a predictable curve throughout the growing season. For rice, it rises sharply after transplanting, peaks during the heading stage (0.7–0.8), and declines before harvest.

The anomaly detection logic is straightforward: pixels that deviate more than two standard deviations from the historical mean NDVI for the same period are flagged as anomalies. Statistically, this indicates that a change has occurred beyond the range of normal variation.

The strength of this approach is that no machine learning model training is required. It operates solely on statistical parameters from historical data (mean and standard deviation). Even without a data scientist, anyone can copy the GEE code, swap in the coordinates of their farmland, and use it as-is.

In a validation test at a coffee plantation, an area where NDVI dropped sharply from 0.65 to 0.38 was detected, and an on-site inspection confirmed early-stage infection of coffee leaf rust. At that point, the infection had gone unnoticed by visual inspection—the leaves appeared only slightly discolored.

Satellite data can be used not only for disease detection but also for yield prediction. The technique repurposed here is the demand forecasting methodology introduced in the article on starting AI demand forecasting without big data.

Demand forecasting predicts future demand from "historical sales data + external factors (weather, events, etc.)." Translating this to agriculture, it becomes a model that predicts yield from "historical NDVI time series + meteorological data."

Specifically, using Prophet (a time series forecasting library developed by Meta), the following variables can be configured:

• Target variable: Yield (cooperative records from the past 5 years)
• Feature 1: Average NDVI around the heading period
• Feature 2: Cumulative precipitation during the crop season
• Feature 3: Average temperature during the crop season

Prior studies under similar conditions have reported MAPE (Mean Absolute Percentage Error) of 15–20%. Compared to farmer experience-based prediction errors of 30–40%, AI models are likely to significantly outperform.

Critically, all data needed to build this model is available for free. NDVI comes from Sentinel-2, and meteorological data can be obtained from the Lao Meteorological and Hydrological Department or NASA's POWER database.

What level of impact can be expected when operating this approach over 2 crop seasons (approximately one year)? We organize the findings based on prior research conducted in similar environments across Southeast Asia and report data from FAO and IRRI (International Rice Research Institute).

The following are projected values for Laos rice and coffee cultivation, derived from NDVI-based monitoring projects conducted under similar conditions in Southeast Asia.

Coffee is expected to have higher detection accuracy because, as an evergreen tree, its NDVI baseline is more stable, making anomalies easier to detect. Rice, on the other hand, has large NDVI fluctuations through the transplanting and harvest cycle, making it somewhat harder to distinguish between "normal variation" and "abnormal decline."

A false positive rate of 15–30% is not exactly low. However, in agricultural settings, the prevailing sentiment is that "false alarms are far better than missed detections." Missed detections lead directly to yield losses, while false alarms only require a field check. Understanding this asymmetry is crucial for system operation design.

Beyond the numbers, qualitative improvements can also be expected. Based on reports from similar projects and field interviews, the following operational changes are anticipated.

First, patrol priorities would shift. Instead of uniformly walking the entire farmland end-to-end, farmers could prioritize checking areas flagged as anomalous on the NDVI map. Being able to deploy limited labor to pinpoint locations could dramatically improve daily patrol efficiency.

Second, pesticide usage is expected to decrease. Rather than applying pesticides broadly "just in case," farmers could target only anomaly-detected areas. A similar project in Cambodia reported a 20–30% reduction in pesticide costs. The secondary benefit of reduced environmental impact should not be overlooked.

However, challenges remain. Reading NDVI maps requires minimum digital literacy, which can be a barrier for older farmers. A practical approach is to have young cooperative staff serve as intermediaries — operating GEE and printing maps with red circles around anomaly areas, combining low-tech with high-tech.

Here we organize the design principles for a successful satellite data utilization project and the risks to be aware of in advance.

The greatest strength of this approach is that you don't need to build a model from scratch. By repurposing the proven Prophet library and simply swapping the input data from "sales data + weather" to "NDVI + meteorological data," algorithm development costs can be kept near zero.

This is a direct application of the "repurpose existing tools" strategy introduced in the AI adoption guide. In environments like Laos where AI talent is scarce, applying proven methods to new domains has a higher success rate than developing new models.

Another critical design principle is leveraging the existing organizational structure of agricultural cooperatives. Rather than approaching individual farmers directly, information is distributed through cooperatives. A division of labor where young cooperative staff operate GEE and relay results to farmers is key to bridging the digital literacy gap.

The biggest risk is data gaps during the rainy season.

The rainy season (May–October) coincides with the main rice growing season in Laos. However, Sentinel-2's optical sensors cannot penetrate clouds, so available images during the rainy season drop to just 1–2 per month — compared to 5–6 per month during the dry season.

Even with a "cloud cover under 20%" filter, it is entirely possible to have zero images passing the filter for 2 months during the rainy season. During this period, NDVI monitoring would completely stop.

An effective countermeasure is combining Sentinel-1 (radar satellite) data. Radar penetrates clouds, enabling data collection regardless of weather. While it doesn't directly indicate crop health as NDVI does, it can be used to estimate paddy flooding status and crop growth stages. Combining optical (Sentinel-2) and radar (Sentinel-1) ensures monitoring continuity throughout the year.

The lesson is clear: when using satellite data in tropical and subtropical regions, a design that relies solely on optical data will fail. Radar data integration should be built into the plan from the start.

For readers looking to implement this approach on their own farmland, here are practical guidelines.

What you need is surprisingly little.

• Google Account: Required to apply for GEE access (free approval for research and non-profit purposes)
• Internet connection: GEE is cloud-based, so a working browser is sufficient
• GPS coordinates of target farmland: Can be obtained using a smartphone's GPS
• Historical yield records: 3+ years enables model building (5 years is ideal)
• Basic JavaScript or Python: Required for writing GEE code

Programming skills may seem like the biggest hurdle, but GEE code is formulaic and official tutorials and sample code are extensive. It is not uncommon for someone with an agricultural degree and no programming experience to be running basic NDVI analyses within about 2 weeks of self-study.

Multiple international organization support programs exist for the digitalization of agriculture in Laos. By leveraging these programs, barriers can be lowered on both the technical assistance and funding fronts.

The World Bank supports improvements in agricultural productivity and market access through the "Lao PDR Agriculture Competitiveness Project." This project includes the modernization of agricultural technology, and proposals for utilizing satellite data are likely to align with the scope of eligible support.

JICA has been involved in Laos's agricultural sector for many years, implementing cooperation centered on irrigation infrastructure development and the dissemination of agricultural technology. Optimizing irrigation management through satellite data directly overlaps with JICA's areas of interest.

When applying to these support programs, the key is to emphasize not that you "want to purchase expensive equipment," but that the project can be realized using free satellite data with existing infrastructure. Projects with lower upfront investment tend to have a higher likelihood of approval.

When discussing agriculture in Laos, water resource management in the Mekong River basin is unavoidable. Irrigation infrastructure coverage is low compared to neighboring countries, and most farmland depends on rainfall.

This is where satellite data proves valuable. Using Sentinel-2's shortwave infrared band (SWIR), soil moisture conditions can be estimated. Combined with water level data published by the Mekong River Commission (MRC), the foundational data for optimizing irrigation timing is available.

In rural Laos, many farmers check river water levels visually each morning and make irrigation decisions based on experience. Satellite data can quantify these decisions and make them more accurate over a wider area. The role of satellite data is not to negate experience, but to validate it with numbers and extend its applicability.

Here are answers to frequently asked questions about satellite data utilization.

GEE analysis itself runs on the cloud, so a stable, always-on connection is not required. Since the workflow simply involves submitting code and receiving results, it can operate even with intermittent connectivity.

However, a certain degree of connection stability is needed during initial setup and when developing new code. In practical operations, a realistic approach is to run analyses in an urban office, export the results as PDFs or images, and bring them to rural areas. In the verification project as well, the workflow involved conducting analyses at the Vientiane office and sharing results via a LINE group.

Health monitoring using NDVI is applicable to all photosynthesizing plants. Beyond rice and coffee, the same approach works for corn, rubber, and fruit trees. Since NDVI normal ranges and curve shapes differ by crop, baseline data for 1–2 crop seasons needs to be collected first for each target crop.

All tools and data are free. Sentinel-2 satellite imagery is publicly available at no cost from ESA, Google Earth Engine is free for research and non-profit use, and the Prophet library is open source.

The costs involved are labor costs. Training staff to operate GEE takes 2–4 weeks, and building the initial analysis takes 1–2 months. Outsourcing to external consultants varies by project scope, but is incomparably cheaper than the cost of deploying a network of IoT sensors (ranging from thousands to tens of thousands of dollars). In the long run, the most cost-effective approach is to train the IT staff of agricultural cooperatives.

Agricultural monitoring using satellite data and AI is a technology "within reach" for smallholder farmers in Laos. No expensive sensors, high-speed internet, or PhD in data science required.

To get started, we recommend these 3 steps:

1. Register for Google Earth Engine (apply at earthengine.google.com — approval takes a few days)
2. Try NDVI calculation with the official tutorial (just enter your farmland coordinates and run)
3. Organize your historical yield records (handwritten notes or Excel — aim for 3+ years)

For more on repurposing demand forecasting AI for agriculture, see our article on starting AI demand forecasting without big data. For the overall AI adoption roadmap, refer to our AI adoption guide.

We offer technical consulting on satellite data utilization for agricultural enterprises and NGOs in Laos — from initial GEE setup to customizing yield prediction models tailored to local conditions. Feel free to reach out for an initial consultation.