The research, published in Theoretical and Applied Climatology, describes the algorithms underpinning the Cambodia Agrometeorological Service, known as CAS, a web-based system that delivers real-time monitoring and seasonal forecasts of rainy-season onset, cessation, and duration. The work was carried out by researchers from Cambodia’s Ministry of Water Resources and Meteorology and the Food and Agriculture Organization of the United Nations, with support from the Green Climate Fund through the PEARL Project, an initiative aimed at building ecologically sound agriculture and resilient livelihoods in the basin. Until now, Cambodia had no operational system capable of translating raw rainfall data and global forecast models into actionable seasonal guidance for farmers.

To pin down the rhythm of the rainy season, the team turned to more than four decades of daily rainfall observations, drawing on the Climate Hazards Group InfraRed Precipitation with Station data dataset, a satellite-based rainfall record blended with ground observations that covers the period from 1981 to 2024. Rather than relying on a single threshold, such as the first day of measurable rain, the researchers applied the cumulative precipitation anomaly method, a technique originally developed for the Amazon basin that tracks the accumulated departure of daily rainfall from its climatological mean. Onset is declared when the cumulative anomaly curve begins a sustained climb, signaling that rains have become persistent rather than episodic, while cessation is identified when the curve begins a sustained decline, marking the reliable end of the wet season.

The results reveal a basin-wide climatology with a mean onset date of 7 May and a mean cessation date of 22 October, yielding an average rainy-season length of roughly 168 days. But those averages conceal a striking degree of variability. The timing of onset swings by more than 70 days from one year to the next, meaning that in some years the effective growing season begins in early April while in others farmers may wait until mid-June for dependable rains. Cessation and season length likewise show pronounced interannual and spatial variability across the basin, underscoring why fixed calendar-based planting recommendations so often fail Cambodian farmers and why a dynamic, data-driven service is needed.

Characterizing the past is only half the challenge; predicting the future is what makes such a service genuinely useful. The team evaluated seasonal hindcasts from the European Centre for Medium-Range Weather Forecasts, comparing retrospective model predictions against observed rainy-season metrics over an independent validation period. The raw model output showed systematic biases: onset was predicted 8.65 days too early on average, cessation 1.75 days too late, and total season length 10.40 days too long. These are meaningful errors in an agricultural context, where a week of misjudged timing can determine whether seedlings survive their first dry spell or perish before the rains stabilize.

To sharpen the forecasts, the researchers applied a split-sample bias correction, calibrating the model output against observations in one portion of the record and testing it on another. The correction reduced the onset bias to 6.76 days and the season-length bias to 3.73 days, a substantial improvement for the metrics that matter most to planting decisions. Cessation bias, however, shifted to 3.03 days in the opposite direction, and improvements in absolute error proved dependent on which metric was examined. The authors are candid about these limitations, noting that at its current stage CAS should be interpreted primarily as basin-scale guidance rather than field-level precision, with forecast skill depending on lead time and the character of the event being predicted.

Encouragingly, the corrected hindcasts performed well in the years that matter most. When the team examined extreme seasons, the years with unusually early or late onsets and exceptionally long or short rainy periods, the bias-corrected forecasts reproduced the broad evolution of rainfall anomalies with reasonable fidelity. For a farmer deciding whether to invest in seed, labor, and fertilizer ahead of a season, capturing the sign and rough magnitude of an upcoming anomaly is often the difference between a useful forecast and a useless one. The analysis of extreme years suggests CAS can deliver exactly that kind of heads-up guidance at seasonal lead times.

Beyond forecast evaluation, the study digs into the physics of why some seasons start early and end late. Dynamical diagnostics revealed that years with early onset and late cessation are linked to enhanced low-level convergence between airflows originating over the Gulf of Thailand and the South China Sea, the two great moisture reservoirs that feed the Indochinese monsoon. In those years, the atmosphere carries more precipitable water and exhibits stronger vertical ascent, the upward motion that cools humid air and triggers deep convection. This mechanistic link is more than academic curiosity; it provides a physically grounded diagnostic framework that forecasters can monitor in real time, watching the interplay of regional moisture sources to anticipate whether the coming season will run long or short.

The implications extend well beyond Cambodia’s borders. Across the tropics, from the Sahel to South Asia to the Brazilian Amazon, the timing of the rainy season is a first-order control on food security, and studies across multiple continents have shown that aligning sowing dates with observed onset can measurably improve yields and water productivity. Yet many developing countries lack the institutional infrastructure to convert global forecast products into local decision support. The CAS framework, built on freely available satellite rainfall data, publicly accessible seasonal hindcasts from the Copernicus Climate Change Service, and a transparent, replicable detection algorithm, offers a template that other agricultural regions with similar data constraints and predictive challenges can adapt at relatively low cost.

For the farmers of the Northern Tonle Sap Basin, where the flood pulse of the Tonle Sap lake and the timing of monsoon rains shape livelihoods that have persisted for centuries, the arrival of an automated, locally tuned forecasting service marks a quiet but consequential shift. The system will not eliminate the uncertainty inherent in seasonal prediction, and its developers are careful to frame it as probabilistic guidance rather than a crystal ball. But by quantifying exactly how variable the rainy season is, demonstrating how far global models can be pushed with careful bias correction, and anchoring the whole enterprise in the underlying monsoon dynamics, the research validates the core algorithms of CAS and establishes a credible scientific basis for a new generation of climate services tailored to the people who need them most.

Subject of Research: Rainy-season onset, cessation, and duration prediction for agrometeorological services in Cambodia's Northern Tonle Sap Basin

Article Title: Characterizing rainy-season onset, cessation, and duration across Cambodia’s Northern Tonle Sap Basin for operational agrometeorological services

Article References: Chinn, R., Ngeang, L., Roth, S., Sok, S., Chea, S., & Libanda, B. (2026). Characterizing rainy-season onset, cessation, and duration across Cambodia’s Northern Tonle Sap Basin for operational agrometeorological services. Theoretical and Applied Climatology, 157(10), Article 650. https://doi.org/10.1007/s00704-026-06586-0

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06586-0

Keywords: Cambodia, monsoon, rainy season onset, agrometeorology, seasonal forecasting, Tonle Sap Basin, ECMWF, bias correction, CHIRPS, climate services, rice agriculture, Green Climate Fund