What one row of the cropland index holds
On this platform, NDVI for East Africa comes as a single table. It gives a cropland vegetation index for each administrative unit in Kenya, Tanzania, Uganda and Rwanda, with one value per unit per week. The value is unitless: a ratio from -1 to 1, where higher means greener. The cropland NDVI by county dataset page lists coverage as 47 Kenyan counties, 31 Tanzanian regions, 135 Ugandan districts and 30 Rwandan districts. That is 47 plus 31 plus 135 plus 30, or 243 units. Every row has the unit, the week, the index averaged across cropland pixels inside that unit, and a comparison with that unit's own seasonal normal. So Isiolo is judged against Isiolo in that week of the year, not against Vihiga.
Two satellite products feed it. A 10 m weekly product supplies the fine series, and a 300 m dekadal product supplies the coarse one. Both have eight years of history. Eight years can tell you what a normal late October looks like in a county. It says nothing about a decade earlier. Treat the normal as a short memory, not a climatology.
The table does not hold three things. It has no yield, because greenness is a reflectance ratio and a harvest is tonnes. It has no forecast, because the platform serves observations and arithmetic on observations. It also has no value for a week when cloud blocked the view. That cell is empty, not a guess.
A value of 0.38 is a land cover first and a crop second
A regional climate centre's NDVI explainer gives the scale most readers want on the wall. Shrub and grassland run 0.2 to 0.3. Temperate and tropical rainforest run 0.6 to 0.8. Values near zero or below mean bare rock, soil, water, ice or cloud. On that scale, Kenya's latest median of 0.38 sits between grassland and forest. That is roughly where mixed cropland across semi-arid and sub-humid counties should sit. Uganda's 0.59 is brushing the bottom of the forest band.
Trouble starts with that comparison. From orbit, a tea estate, a maize field at tasselling and a banana grove can all look like a 0.6. An absolute value tells you how much green leaf is reflecting light. It does not tell you which crop, how far through the season it is, or whether that much leaf is good news. In the other direction, the FAO country earth-observation pages note that for their relative indices, a high value in a sparsely vegetated setting such as savannah means above-average performance for that setting, not lush vegetation. Isiolo's 0.13 is low on the regional scale. It may be ordinary for Isiolo cropland in that week. Read the seasonal comparison column first.
The mask changes the meaning too. A cropland vegetation forecasting notebook from a continental earth-observation service builds its series from a 20 m optical satellite product. It masks that series with a crop mask. The eastern version covers Ethiopia, Kenya, Tanzania, Rwanda and Burundi; a continental version covers the whole continent. Cropland NDVI is only comparable across units cut with the same mask version. A method note that omits the mask has left out the fact that decides whether two medians belong on one axis.
Four countries in the week to 1 November 2026
For the week to 1 November 2026, platform satellite data carries a value for every one of the 243 units. Kenya's 47 counties have a median of 0.38, ranging from 0.13 in Isiolo to 0.64 in Vihiga. Tanzania's 31 regions have a median of 0.23, ranging from 0.17 in Shinyanga to 0.61 in Kaskazini Pemba. Uganda's 135 districts have a median of 0.59, ranging from 0.18 in Moroto to 0.70 in Masindi. Rwanda's 30 districts have a median of 0.42, ranging from 0.27 in Kayonza to 0.61 in Musanze.
Some things in those lines need a second look. Tanzania's median of 0.23 sits at the grassland end of the regional scale. Its floor of 0.17 in Shinyanga is close to that median. That pattern says most Tanzanian regions read low that week. One island region, Kaskazini Pemba at 0.61, pulled up the top of the range alone. Uganda has the widest spread. Moroto at 0.18 and Masindi at 0.70 are the same country in the same week. The gap between them, 0.52, is larger than the gap between the Kenyan and Ugandan medians, 0.21. Rwanda's 30 districts form the tightest cluster, from 0.27 to 0.61.
A median is the right summary for this table, but it is a poor number to quote on its own. Half of Kenya's counties sat below 0.38 and half above it. That tells you nothing about where a particular county landed. Ranges help more, and even a range boils down 47, 31, 135 and 30 separate rows. If you are pricing a product by county, pull that county's row and its seasonal comparison. Dates need the same care. All four countries share the week to 1 November here, but the latest week does not have to line up across them. A query that assumes it does will one day compare October in one country with November in another and not notice.
Eight weeks of change, and where the greening was
Measured against the week of 6 September 2026, direction of travel depends on the country. Uganda's index rose by more than 0.02 in 109 of 135 districts and fell by more than 0.02 in 18. Rwanda rose in 20 of 30 and fell in 1. Kenya is mixed: up in 18 counties, down in 24. Tanzania leans the other way, up in 5 regions and down in 21. A 0.02 threshold keeps noise out of the count. A district that moved 0.01 either way lands in neither column. Add Kenya's two columns and 42 of 47 counties moved by more than 0.02 in one direction or the other. That is a lot of movement for a country whose median barely tells you anything.
All four of the largest greenings are Ugandan: Hoima at +0.33, Kasese at +0.31, Bunyangabu at +0.31 and Kiruhura at +0.29. A move of 0.33 on a scale that tops out at 1 is large. On the regional scale it is the distance from grassland to the edge of forest. Browning is Ugandan too, with one exception: Bundibugyo at -0.21, Uasin Gishu in Kenya at -0.21, Kotido at -0.18 and Karenga at -0.18. Two Ugandan districts top the greening list and two top the browning list, in the same country and the same eight weeks. A national count of risers hides a wide district distribution. And 109 of 135 says nothing about Kotido.
Uasin Gishu is the row a reader would not expect, and the one to handle with care. A fall of 0.21 over eight weeks can be a crop coming off the field, because harvested ground reads near zero like any bare soil, or it can be a crop under stress. The index alone cannot separate the two. Its seasonal comparison column can. A fall that matches the county's own normal for the time of year is a calendar. A fall that undershoots it is a signal. Nothing in the data block says which applies here, so nothing in this article will.
The literature settles what to lay beside the movers next. A peer-reviewed analysis of East African vegetation from 1982 to 2015 found NDVI correlated more closely with precipitation than with temperature. It also found the short October to December rains showed higher NDVI variability than the long rains. So the first join is the seasonal rainfall anomaly for the same units, not a yield table. The eight weeks from early September to early November sit precisely in the season that study flagged as the noisier one. Correlation with rain is not a cause for any single district. It is a reason to check rain before anything else.
One name, two instruments: absolute index and anomaly
Vegetation index covers two products that should never share a column. This platform's weekly value is an absolute index, a number between -1 and 1 for the week. A continental earth-observation service's monthly NDVI anomaly product is the other kind. It shows how far a month's vegetation sits from the long-term average, in standard deviations, at 30 m across all African countries, monthly from January 2017, with a baseline built from 1984 to 2020. Positive means greener than average, usually read as more rain. Negative means vegetation that may be stressed by drier conditions. An abrupt large negative can follow a fire. It is published as a low-latency product on the 5th day of the following month.
Set the two clocks side by side. A weekly absolute value arrives inside the season it describes. A monthly anomaly arrives after the month has closed, judged against a record reaching back to 1984. That makes the weekly series the provisional layer and the monthly anomaly the confirming one. A lender who sees a weekly browning in Uasin Gishu has a reason to look. A monthly anomaly that is negative in standard deviations is a reason to act.
Baselines are the catch. This platform's seasonal normal rests on eight years. That anomaly product's baseline spans 1984 to 2020. FAO's pages draw their maps of severe-drought frequency from a series running 1984 to 2025. They compute non-seasonal vegetation indicators dekadal and monthly, never weekly. Three instruments, three baselines, three cadences. A standard deviation from a record starting in 1984 and a departure from eight years of weekly readings are different numbers even when both are negative in the same county. A dashboard that charts them together without saying which is which has hidden how its number was made.
Cloud, and the rule that empties a cell
Cloud decides which units can carry a value in a given week. A NASA visualisation of NDVI anomaly sets out the rule: the sensor on that satellite cannot see the surface through cloud, so on cloudy days nothing is collected. The most recent prior day's value is carried forward for up to 10 days. An area with no clear view for ten or more days is marked as having no data. Two consequences follow. A map can look complete while a share of its pixels are last week's reading carried forward. And in a rainy season the cut-off bites exactly where the vegetation is responding to the rain the observer wants to see.
That visualisation covers only the maize, rice, soybean and wheat zones of the United States, Australia, Russia, Europe and southern Africa. East Africa is not on it. This platform's rule is simpler: cloud in the rainy seasons can leave a week without a value, and the week stays empty. For an analyst that is a join problem. A county series will have holes in the long rains and the short rains. Anything that fills a hole by interpolation has created a number nobody measured. Carry the gap through, and count filled weeks per unit before reporting any trend.
Water pixels are the second contamination. At the regional scale water reads near zero. So a unit whose boundary takes in lake or sea has pixels that would drag its cropland mean down unless the mask removed them. No coastal flag exists in the data block. Kaskazini Pemba at 0.61 shows an island region can still read high. But a reader has to assume the mask did its job rather than check it. Any method note that uses coastal or lakeside units needs that assumption written down.
Three questions this index cannot answer
Yield first. A study of productivity and water stress in tropical East Africa examined the 2010 to 2011 drought, when the October to December 2010 and March to May 2011 rains failed in succession. It found gross primary productivity about 32 percent below the long-term average, with chlorophyll fluorescence responding to soil moisture with a lag of 16 days. Productivity is closer to yield than greenness is, and even that study reports growth, not harvest. Vihiga's 0.64 says Vihiga's cropland was green that week. It does not say what Vihiga will harvest.
Cause second. A comparison of monitoring indicators over East Africa lays out the chain a vegetation index sits at the end of: rainfall deficit, then soil moisture deficit, then higher land surface temperature, then finally vegetation. It also found that evapotranspiration products, both the land-surface-model kind and SSEBop, were better indicators during extreme drought events, while NDVI gave better information during wetter than normal conditions. So the index reads a wet season well and a severe drought late, after damage has reached the leaf.
Forecast third. No forecast is served here. The forecasting notebook cited earlier is candid about why: it resamples to monthly or fortnightly steps and states that a longer forecast length means greater uncertainty. An eight-week change is the difference between two weeks, not a slope. Hoima's +0.33 says nothing about Hoima's next eight weeks. What the index can feed is a classification. FAO's Agricultural Stress Index counts the share of arable pixels with a Vegetation Health Index below 35 percent, a threshold from Kogan's 1995 work, and sorts drought into Extreme, Severe, Moderate and Mild per GAUL 2 region. That is a derivation the reader builds from the rows, not a column shipped.
Where the series is thin
History is short, and the long record shows why that matters. Over 1982 to 2015 the peer-reviewed East African analysis found mean annual NDVI ran a positive linear trend of 14.50 x 10^-4 before 1998 and a negative trend of -9.64 x 10^-4 after, with the break in 1998. Croplands changed insignificantly, by 6.9 x 10^-5 per year, while non-croplands changed by 5.16 x 10^-4 per year. Every eight-week comparison here sits inside a regime that was already running when the platform's history began. A cropland-only index hides that the land around it moves far faster. That published record stops in 2015 and the continental anomaly product starts in 2017; nothing in these sources stitches them together.
Field size is the second thinness. A land cover study of the Kibwezi watershed in Makueni County works on roughly 700 km² across two subcounties with bimodal rain averaging 600 mm a year. A county mean averages many such watersheds. A 300 m dekadal pixel averages whatever fields fall inside it. The 10 m weekly product resolves individual fields, which is why the two are served separately.
Coverage comes third, and vegetation wins here. FAO's pages note that vegetation indices are finer than precipitation products and yield valid zonal statistics for more sub-national units in certain countries. That holds for greenness by county before rainfall by county. Field-level weather and soil rows on the platform are different: they come from NuaSense's own sensor network, described on its sensor products page. Those rows answer only for the devices on the owning account and never feed the satellite index. A reader who needs a price beside a greenness row should take the county market table that the commodity data piece worked through, and join on unit and week.