Tree crop age estimation in oil palm and rubber plantations

Canopy height from TanDEM-X and ICESat-2, combined with Landsat planting-date detection, lets commodity traders and development banks map the age class of oil palm and rubber stands with enough precision to drive replanting schedules and carbon accounts.

Sensors

TanDEM-X (TerraSAR-X add-on for Digital Elevation Measurements): X-band interferometric SAR pair producing canopy height models at 12 m posting (global DEM) or down to approximately 6 m in high-resolution spotlight mode. Phase coherence between the two satellites encodes vertical structure; over closed-canopy plantations the interferometric phase centre sits within the upper canopy, enabling height retrieval with published RMSE of roughly 2 to 4 m over tropical tree crops in independent validation studies.
ICESat-2 ATLAS photon-counting lidar: Six-beam 532 nm lidar with approximately 17 m footprint diameter and 70 cm along-track sampling. Provides highly accurate canopy top heights (published uncertainty around 1 m) along sparse transects at 91-day repeat. Not wall-to-wall coverage; most useful for calibrating and validating area-wide SAR-derived height models rather than mapping entire estates independently.
Landsat archive (Landsat 5 TM through Landsat 9 OLI-2): 30 m multispectral imagery from 1984 onwards at 16-day revisit (single satellite). NDVI and spectral change detection across the full archive allows retrospective identification of clearing and replanting events, pinning planting dates to within one or two growing seasons. The archive depth is the critical asset here; no other free optical record goes back as far.
Sentinel-1 C-band SAR: 10 m, 6-day revisit (two-satellite constellation over the tropics). Backscatter and coherence time series complement TanDEM-X height estimates by tracking canopy development through cloud cover. C-band penetrates less deeply than L-band into the canopy, but the dense temporal stack is useful for detecting the abrupt backscatter drop that follows clear-felling and the gradual rise through early regrowth.
Sentinel-2 MSI: 10 m visible and NIR, 20 m red-edge and SWIR, 5-day revisit at the equator. Red-edge bands (705 nm, 740 nm, 783 nm) are sensitive to chlorophyll content and canopy closure, both of which correlate with stand age in the first decade of growth. Used to fill temporal gaps and cross-check spectral trajectories derived from Landsat.

Why canopy height is a clock

Oil palm height increases at roughly 0.4 to 0.5 metres per year through the first decade after planting, then slows sharply as the frond base accumulates and the trunk growth rate declines. Published remote-sensing studies from Malaysia and Indonesia have demonstrated that this trajectory is consistent enough across Borneo, Sumatra and the Malay Peninsula to assign stands to five-year age classes with reasonable confidence using height alone. Rubber (Hevea brasiliensis) follows a broadly similar pattern, with canopy closure occurring within three to four years and height plateauing in mature stands at around 25 to 30 m.

The practical implication is that height, measured once with sufficient accuracy, encodes age information that would otherwise require ground surveys or company records that are often incomplete, disputed or commercially sensitive. A height model that is accurate to two or three metres translates to an age uncertainty of roughly four to six years in the linear growth phase, which is acceptable for replanting planning and carbon accounting even if it is too coarse for precise yield modelling.

What TanDEM-X and ICESat-2 each contribute

TanDEM-X provides the area-wide canopy height model. The satellite pair operates in bistatic mode, meaning both spacecraft acquire simultaneously, which eliminates temporal decorrelation and makes the interferogram stable even over dynamic vegetation. Over oil palm, the X-band phase centre sits at roughly 70 to 90 percent of canopy height, so a correction factor derived from field measurements or ICESat-2 transects is needed to recover true canopy top. Without that correction, raw TanDEM-X heights will underestimate by two to five metres in mature stands.

ICESat-2 ATLAS supplies the calibration. Its photon-counting approach produces canopy top heights along its six beam tracks with sub-metre precision, and where those tracks cross a plantation block the comparison between ICESat-2 and TanDEM-X directly quantifies the phase-centre offset. The limitation is coverage: ICESat-2 tracks are spaced roughly 3 km apart at the equator on a given 91-day cycle, so calibration relies on the assumption that the offset is relatively stable across canopy types within a region. That assumption holds reasonably well within a single plantation species but should be validated separately for oil palm and rubber.

Anchoring height to time: the Landsat archive method

Height tells you how tall a stand is today. The Landsat archive tells you when it was planted. The combination converts height into age with far less ambiguity than either source alone.

The detection method works backwards through the archive. A dense time series of NDVI or NBR (Normalised Burn Ratio) values from 1984 to the present shows a characteristic pattern at each cleared-and-replanted pixel: a sharp drop to bare soil or low-biomass values, followed by a monotonic rise over the subsequent years. Automated change-detection algorithms, including variants of the Continuous Change Detection and Classification (CCDC) method developed at Boston University, can identify the break-point date to within a few months in areas with low cloud frequency. Over the humid tropics, persistent cloud cover is the main constraint. Landsat's 16-day revisit means that in a region with 80 percent cloud cover, a clear observation may arrive only once every two to three months, and the break-point date estimate degrades accordingly. Sentinel-2 improves this with its shorter revisit, but its archive only extends to 2015.

The output is a pixel-level planting-year map. Combined with the current-year height model, any residual ambiguity between a young tall stand and an old short stand (caused by soil or management differences) can be partially resolved.

Honest limits: what the method cannot resolve

Age class accuracy degrades at both ends of the growth curve. In the first two years after planting, palms are too short for TanDEM-X to distinguish reliably from ground clutter, and the spectral signal in Landsat is easily confused with other low-biomass land covers. Beyond ten years in oil palm, height is no longer a reliable age proxy because the growth rate has flattened. A 15-year stand and a 22-year stand may differ by only two or three metres in canopy height, well within the measurement uncertainty of a single TanDEM-X acquisition.

Management practices introduce additional noise. High-input estates with intensive fertilisation grow faster than smallholder plots; drought-stressed stands may be stunted relative to their age. Neither TanDEM-X nor Landsat can observe these inputs directly. Age maps produced by this method are therefore best described as structural-age estimates, not calendar-age guarantees, and should be presented with a stated uncertainty band rather than a single-year figure.

Cloud cover over the humid tropics remains the most persistent operational constraint for the optical component. Buyers in Kalimantan or Aceh should expect planting-date uncertainty of one to three years in the cloudiest areas, even with multi-year Landsat compositing.

Who uses age maps and for what

Commodity traders and plantation companies use age-class maps to forecast production volumes. An estate approaching peak yield (roughly years 7 to 18 in oil palm) has a very different forward production profile from one with a large proportion of stands over 20 years old and approaching economically motivated replanting. Banks financing plantation bonds and development finance institutions assessing smallholder credit risk both need independent verification of the age distribution that does not rely on company-supplied records.

Carbon accounting is a growing application. Plantation age determines above-ground biomass stock, which feeds into REDD+ accounting and voluntary carbon market methodologies. An independently derived age map provides an auditable baseline that is harder to manipulate than self-reported data. Satellize's analytics work includes crop-estimation programmes for sovereign clients, including the Kingdom of Tonga, and the same combination of archive-based phenology and canopy-structure retrieval that underpins those programmes applies directly to plantation age mapping at estate or national scale.

Replanting planning is perhaps the most operationally immediate use. Oil palm replanting is capital-intensive and requires two to three years of zero production from the affected blocks. A spatial age map allows an estate manager to sequence replanting across blocks to maintain a continuous production profile rather than facing a production cliff when an entire estate planted in the same year reaches senescence simultaneously.

From data to a deliverable age map

A typical workflow begins with TanDEM-X acquisition over the area of interest, either from the global 12 m DEM already in archive or from a new tasked acquisition in spotlight mode for higher resolution. ICESat-2 transects crossing the estate are extracted from the publicly available ATL08 land and vegetation height product. The two are co-registered and a height-correction function is fitted. The corrected canopy height model is then intersected with the Landsat-derived planting-year map to produce an age-class raster, typically in five-year bins.

Delivery formats depend on the client's workflow: GeoTIFF rasters for GIS integration, polygon shapefiles aggregated to plantation block boundaries, or tabular summaries by block for direct import into estate management systems. Uncertainty layers should accompany every delivery. A map without uncertainty estimates is, in this application, less useful than it appears.

Typical figures

Spatial resolution (height model)	12 m (TanDEM-X global DEM); 6 m (TanDEM-X spotlight tasking)
Spatial resolution (planting-date layer)	30 m (Landsat); 10 m (Sentinel-2, from 2015)
Height accuracy (TanDEM-X, tropical forest)	RMSE approximately 2 to 4 m in published validation; phase-centre correction required for canopy top
Height accuracy (ICESat-2 ATLAS ATL08)	Published uncertainty approximately 1 m along-track; 17 m footprint diameter
Planting-date archive depth	1984 to present (Landsat 5 onwards); 2015 to present (Sentinel-2)
Planting-date temporal precision	Within one growing season in low-cloud areas; one to three years in persistently cloudy humid tropics
Age-class resolution	Five-year bins in the linear growth phase (years 2 to 10); coarser beyond canopy height plateau
Minimum detectable stand age	Approximately 2 years post-planting (height signal above ground clutter threshold)
Geographic coverage	Global tropics and subtropics; TanDEM-X global DEM covers 80°N to 80°S
Typical delivery format	GeoTIFF (age-class raster with uncertainty band), Shapefile (block-level polygons), CSV (block summary table)

Analytics Satellize can run

Canopy height model (CHM)	TanDEM-X bistatic InSAR phase-centre retrieval, corrected against ICESat-2 ATL08 transects using empirical offset function	GeoTIFF at 6 to 12 m resolution with per-pixel height uncertainty layer
Planting-year map	CCDC-class break-point detection on dense Landsat NDVI/NBR time series from 1984 to present, supplemented by Sentinel-2 from 2015	30 m raster of estimated planting year, with cloud-frequency-weighted confidence score per pixel
Age-class map	Intersection of CHM and planting-year map; age-class bins defined by species-specific growth curves from published Malaysian and Indonesian literature	GeoTIFF and block-level Shapefile with five-year age-class labels and uncertainty range
Replanting priority ranking	Spatial overlay of age-class map against estate block boundaries; blocks beyond economic yield threshold ranked by area and adjacency to processing infrastructure	Ranked block list with estimated replanting-year windows, delivered as CSV and PDF summary
Above-ground biomass stock estimate	Age-to-biomass allometric conversion using published species-specific equations for oil palm and Hevea; uncertainty propagated from age-class uncertainty	Block-level biomass table (tonnes per hectare, with range) for carbon accounting or REDD+ reporting
Change detection: recent clear-felling and replanting events	Sentinel-1 backscatter change detection and Sentinel-2 NDVI drop monitoring over rolling 12-month window	Alert layer of newly cleared or replanted blocks, updated quarterly, delivered as GeoJSON feed

Sources

Who does the work

We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.