Green Memory, Dry Season:
A Five-Year Sentinel-2 NDVI Diagnosis of Machakos and Kulalu

Executive Summary

Two land areas speak through five years of orbital light. One tells a story of modest but measurable improvement; the other tells a quieter, harder story of persistent stress.

This report presents a forensic interpretation of five years of Sentinel-2 L2A NDVI time-series data from two Kenyan land areas: Machakos (Dataset 1, 396 observations, approximately 2,090 pixels) and Kulalu (Dataset 2, 402 observations, approximately 1,080 pixels), spanning May 2021 to May 2026. The analysis was commissioned by the Government of Kenya to support evidence-based land, water, and vegetation planning at landscape scale.

The primary findings are as follows:

Machakos displays a detectable upward vegetation trend from 2021 to 2026. Its cloud-filtered mean NDVI rose from approximately 0.298 in 2021 to 0.476 in 2026 (partial year). The land responds visibly to wet-season rainfall, peaking in January, April–May, and December. Strong seasonal rhythm suggests active agropastoral land use with intact rainfall response. However, dry-season periods (August–October) remain notably stressed, with mean NDVI dropping to approximately 0.24–0.25, consistent with dryland conditions in semi-arid Kenya.

Kulalu is consistently less vegetated than Machakos across all five years, with a cloud-filtered mean NDVI of 0.224 — falling within the low to moderate vegetation range. Its best months reach only 0.30–0.34 on average, and its worst months (September–October) fall below 0.15. The p10 percentile — the weakest tenth of the land — regularly approaches or drops below zero, indicating persistent bare, waterlogged, degraded, or shadow-affected surfaces. The p90 percentile improved through 2023 then declined in 2024–2025, suggesting that even the best-performing vegetation patches may be losing resilience. No sustained upward trend is evident.

Both datasets are burdened by heavy cloud contamination (median cloud cover >50% across all observations), which requires strict filtering and reduces the number of analytically usable observations. After applying a 30% cloud threshold, 127 usable dates remain for Machakos and 115 for Kulalu — sufficient for trend analysis, but not for precise phenological reconstruction.

These data, interpreted with appropriate humility and verified by ground teams, suggest that Machakos is a candidate for vegetation protection and agroforestry intensification, while Kulalu requires prioritised soil-water rehabilitation, bare-surface restoration, and sustained monitoring before more ambitious interventions are warranted.

What the Two Datasets Contain

Both datasets were exported from the Sentinel Hub EO Browser or equivalent platform, using the Sentinel-2 L2A (Level-2A, atmospherically corrected) product with NDVI as the derived index. Each observation corresponds to one satellite overpass within approximately a 5-day temporal cadence, though cloud cover frequently prevents usable observations.

The filenames and pixel counts reveal the following:

• Dataset 1 — Machakos: 396 total observations; constant sample count of 2,090 pixels per observation (fixed study-area extent); no-data count consistently 999, suggesting a defined pixel mask.
• Dataset 2 — Kulalu: 402 total observations; constant sample count of 1,080 pixels per observation — a smaller area of interest, roughly half the spatial extent of Machakos.

Both datasets include: date, minimum NDVI, maximum NDVI, mean NDVI, standard deviation, sample count, no-data count, median NDVI, p10 NDVI, p90 NDVI, and cloud coverage percentage. These columns, together, allow multi-layered interpretation of vegetation condition, spatial variability, and temporal dynamics.

Data Quality and Cloud Reliability Assessment

Cloud cover is the first and most consequential filter in any satellite vegetation analysis. These datasets are substantially affected by cloud contamination, and this must be acknowledged before any ecological interpretation is offered.

Cloud Coverage Distribution

Both datasets have median cloud coverage exceeding 50%. This is characteristic of equatorial East African environments, where the Inter-Tropical Convergence Zone (ITCZ) drives prolonged wet seasons with persistent cloud decks — precisely when vegetation is greenest and most interesting to observe, but also most obscured.

Recommended Cloud Filter

For this analysis, a 30% cloud coverage threshold was applied — meaning observations with more than 30% of the study-area pixels classified as cloud were excluded from statistical interpretation. This threshold was chosen as a balance between:

• Retaining enough usable observations across all seasons and years (≥115 observations per site);
• Excluding observations where cloud contamination significantly depresses or artificially elevates NDVI values;
• Maintaining comparability between the two sites.

Observations with Negative NDVI

Negative minimum NDVI values (ranging from approximately −0.08 to −0.52 in individual observations) appear across both datasets even in cloud-filtered records. These negative values indicate pixels contaminated by cloud shadow, standing water, construction material, or deep shade — not healthy land. Their presence in the minimum column is expected and does not invalidate the analysis, but confirms that some pixels within both study areas are consistently stressed or non-vegetated. The mean and median are more robust than the minimum for ecological interpretation.

NDVI Interpretation Framework

NDVI (Normalized Difference Vegetation Index) is calculated from Sentinel-2 red and near-infrared bands as: (NIR − Red) / (NIR + Red). It ranges from −1 to +1. Values are dimensionless. The ecological interpretation depends critically on land cover type, region, season, and soil background.

The following interpretive guide has been adapted for semi-arid dryland East Africa, where vegetation responses are more muted than humid equatorial forests, and where NDVI values above 0.5 during wet seasons represent genuinely healthy and productive land:

Throughout this report, the following statistical metrics are used as described:

• Mean NDVI — spatial average of all pixels; sensitive to outliers
• Median NDVI — central tendency; more robust in heterogeneous landscapes
• p10 NDVI — condition of the weakest 10% of the land surface (stressed, bare, waterlogged zones)
• p90 NDVI — condition of the strongest 10% (best vegetation refugia)
• Standard deviation (stdev) — spatial variability; high stdev = fragmented, uneven landscape
• Min NDVI — single worst pixel; often cloud shadow or bare rock; not a reliable landscape indicator
• Max NDVI — single best pixel; useful for identifying vegetation refugia but not representative

Dataset 1: Machakos — Detailed Analysis

Machakos speaks in a language of recovery. The land breathes with the rains, rests in the dry season, and — over five years — appears to be accumulating slightly more green with each passing cycle.

Overall NDVI Statistics (cloud <30%, 127 observations)

A mean NDVI of 0.382 places Machakos solidly in the low to moderate vegetation to moderate healthy vegetation transition zone — approximately consistent with a semi-arid landscape with functioning seasonal grassland, recovering bush, and dryland cultivation. The narrow gap between mean (0.382) and median (0.367) suggests a relatively symmetrical distribution without extreme outliers distorting the average. This is a coherent, internally consistent signal.

Annual Breakdown

The progression from 2021 (mean 0.298) through 2024 (mean 0.433) and into early 2026 (mean 0.476) represents a clear positive trajectory. The p90 value — the best vegetation patches — climbed from 0.360 in 2021 to 0.605 in 2026, indicating that even the strongest parts of the landscape are becoming more productive. The rising standard deviation (from 0.058 to 0.128) suggests increasing spatial complexity: some areas may be intensifying under cultivation or tree cover, while others remain persistently stressed.

Seasonal Pattern

Machakos exhibits a clear bimodal rainfall-driven seasonal pulse:

• January (NDVI 0.535): Post-short-rains (October–December) peak greenness
• April–May (NDVI 0.543–0.546): Long-rains (March–May) peak — typically the strongest agricultural season in Machakos County
• June (NDVI 0.484): Gradual decline as long rains end
• July–October (NDVI 0.24–0.30): Dry-season stress — vegetation retreats to refugia along drainage lines and under trees
• December (NDVI 0.513): Short-rains response — rapid greening before January peak

This pattern is highly consistent with the known rainfall climatology of Machakos County, where the long rains (Masika) run March to May and short rains (Vuli) run October to December. The land's ability to respond rapidly to rainfall — reaching 0.54 in April–May from a dry-season low of 0.24 in September — indicates functional rainfall–vegetation coupling, suggesting the land retains some capacity for infiltration and rooting-depth moisture storage.

Best and Worst Observed Periods

The single highest-quality observation was recorded on 22 April 2024 (mean NDVI 0.659, cloud 0.7%) — an exceptional wet-season capture showing the land at near-peak greenness. The weakest reliable observations occurred in August–October 2022 (mean NDVI 0.157–0.168), which may reflect a particularly poor dry season in that year or represent the structural baseline of Machakos under prolonged rainfall absence.

Dataset 2: Kulalu — Detailed Analysis

Kulalu speaks more quietly. Its vegetation is sparse, its responses to rainfall are muted, and its dry season is more severe. The land is not dead — but it is under strain that satellite signals alone cannot fully explain.

Overall NDVI Statistics (cloud <30%, 115 observations)

A mean NDVI of 0.224 places Kulalu within the bare to low vegetation range — consistent with degraded dryland, overgrazed rangeland, dry acacia bushland, or sparsely cultivated small-scale farms. The mean (0.224) and median (0.211) are closely aligned, suggesting relatively uniform low greenness rather than a landscape with productive patches lifting the average. The p10 mean of 0.081 — approaching bare soil — is a particularly diagnostic signal: at least 10% of the Kulalu study area is persistently near-bare at any given usable observation.

Annual Breakdown

Unlike Machakos, Kulalu does not demonstrate a clear upward trend. The data suggests an oscillating pattern: modest improvement in 2022–2023 followed by retreat in 2024–2025, with 2026 showing early-year wet-season recovery. The p90 trajectory is particularly concerning: it rose from 0.262 (2021) to 0.410 (2023) — then fell back to 0.329 by 2025. This suggests that even the most productive vegetation patches in Kulalu are not holding their gains between rainfall seasons. Possible explanations include grazing pressure, soil crusting preventing infiltration, or absence of tree cover to buffer the dry season.

Seasonal Pattern

• January (NDVI 0.340): Best month — post-short-rains response
• April (NDVI 0.282): Long-rains response — notably weaker than Machakos's 0.543
• July–November (NDVI 0.147–0.185): Prolonged dry-season stress — the land's weakest period
• December (NDVI 0.304): Short-rains onset — beginning of recovery cycle

Kulalu's seasonal amplitude is narrower than Machakos's. The difference between wet-season peak (~0.34) and dry-season trough (~0.15) is approximately 0.19 NDVI units, compared to Machakos's amplitude of approximately 0.31 units. This reduced amplitude suggests lower soil moisture storage capacity, reduced vegetation biomass to respond to rainfall, or greater evapotranspiration stress. The land appears to gain less from each wet season and lose more during each dry season.

The p10 Signal: A Persistent Warning

The p10 percentile — representing the weakest 10% of the Kulalu study area — averages only 0.081 across five years. In September–October, it drops to 0.017–0.027, within a whisker of zero. In several individual observations, it turns negative. This is a robust, consistent signal that a significant fraction of the Kulalu study area is persistently bare, crusted, waterlogged, or eroded. No single good rainfall season has been able to restore these zones to healthy vegetation cover within the five-year window.

Comparative Diagnosis: Machakos vs. Kulalu

Two landscapes. One satellite. The comparison is clear in its broad strokes — and humbling in its details.

Visual Comparison: Mean NDVI by Year

Which Area Is More Ecologically Resilient?

By every metric examined, Machakos demonstrates greater vegetation resilience: higher baseline NDVI, stronger wet-season response, improving long-term trend, and strengthening vegetation refugia. This suggests the land in the Machakos study area retains more functional ecosystem capacity — likely some combination of better soil water retention, vegetation structure, land management, or more favourable topography within the study polygon.

Kulalu's higher standard deviation — despite lower mean NDVI — indicates a more fragmented landscape where productive patches coexist with severely degraded zones. This internal heterogeneity is itself a diagnostic signal: the landscape may contain dry riverbeds, bare compacted areas, and small productive patches in close proximity, with no continuous vegetation matrix connecting them.

Land, Water, Soil, and Vegetation Meaning

NDVI is not a water sensor. It is not a soil probe. It is a mirror held up to plant chlorophyll. But plant chlorophyll speaks the truth of what lies beneath: what the soil holds, what the roots find, what the rainfall has given or withheld.

Machakos: Land and Water Inferences

• Soil moisture: The strong wet-season NDVI response suggests the soil in the Machakos study area retains adequate moisture for at least 4–6 weeks post-rainfall. This is consistent with moderate clay content, possible mulch or litter cover, or some agroforestry structure retaining soil humidity.
• Infiltration: The speed of vegetation response after seasonal rains (detectable within one 5-day observation cycle) suggests reasonable infiltration rather than dominant runoff. However, dry-season bare soil suggests surface crusting may develop between rains.
• Erosion risk: Dry-season mean NDVI of 0.24 still leaves significant bare soil exposure. Sloped terrain in Machakos is known to be erosion-prone during the onset of long rains. Ground inspection of soil compaction and rill formation is recommended.
• Cropping potential: April–May NDVI values of 0.54–0.55 are consistent with actively cropped maize or legumes during a good rainfall season. The Machakos area likely supports rainfed smallholder cultivation with moderate to good yields in above-average rainfall years.
• Tree establishment: Rising p90 values reaching 0.605–0.720 in 2024–2026 are consistent with growing tree canopy. Agroforestry systems (e.g., Grevillea robusta, Calliandra, moringa) may already be developing within the study area.

Kulalu: Land and Water Inferences

• Soil moisture: Weak wet-season NDVI response (peak ~0.34 in January) suggests the soil in Kulalu does not retain adequate moisture between rainfall events. Sandy or degraded soils with low organic matter are consistent with this signal. Alternatively, shallow soil over impermeable hardpan (duripan or calcrete) may prevent deep percolation.
• Bare soil exposure: p10 NDVI of 0.08 on average — and near-zero in dry months — is unambiguous evidence that a substantial fraction of the Kulalu area is persistently bare, compacted, or covered with non-photosynthesising material. This is consistent with heavy grazing pressure, soil crusting, or the presence of rocky or lateritic outcrops.
• Runoff and erosion: Bare soil under intense East African rainfall events is vulnerable to flash surface runoff, rill erosion, and gully formation. The low vegetation cover and weak infiltration signal suggests Kulalu may be losing more rainfall as runoff than Machakos — a critical inefficiency in a water-scarce environment.
• Waterlogging: Negative p10 values in some observations may indicate seasonally waterlogged depressions or dark shadows from rocky scarps. These zones — if confirmed as seasonal water bodies — could represent water-harvesting opportunities rather than pure liabilities.
• Crop suitability: Mean NDVI of 0.224 at the landscape scale, combined with a dry-season trough below 0.15, suggests that conventional rainfed cropping in Kulalu faces significant climate risk. Drought-tolerant crops (sorghum, cowpea, mung bean) and integrated water-harvesting systems would be more appropriate than maize without supplemental irrigation.

Practical Action Plan for Kenyan Land and Water Managers

Phase 1: Ground-Truthing (Months 1–3)

No satellite-based plan should be implemented without ground verification. The following field activities are recommended as the first priority for both study areas:

• Soil texture and structure assessment at 10–20 sample points per site — clay content, organic matter, compaction depth
• Double-ring infiltration tests to measure how quickly water enters the soil under current conditions
• Slope mapping using GPS or available DEM data (SRTM 30m or Copernicus 10m) to identify erosion-vulnerable slopes and natural drainage lines
• Vegetation transects — walk the land and record what actually grows at the locations where NDVI is high versus low
• Community memory mapping — interview elders and farmers about what the land looked like 20, 30, and 50 years ago; this is irreplaceable historical data that satellites cannot provide

Phase 2: Priority Interventions for Kulalu (Months 3–18)

Given Kulalu's evidence of persistent degradation, the following interventions are prioritised:

• Contour bunds and half-moon earthworks on degraded bare slopes to capture runoff and force infiltration
• Zai pit planting — traditional West African technique proven in East African drylands — to concentrate water and nutrients around individual tree or crop seedlings in bare zones
• Grass strip planting (e.g., Napier grass, Rhodes grass, lemon grass) along contour lines to slow runoff and build soil organic matter
• Moringa establishment in water-harvesting basins — moringa tolerates poor soils, produces rapidly, and provides food, fodder, and soil improvement simultaneously
• Pond siting in confirmed low-lying waterlogged zones — tilapia aquaculture may be viable if seasonal water retention can be confirmed over 60+ days
• Bare-soil rehabilitation using compost, manure application, and direct seeding of native grass species to rebuild soil biological activity before crops are introduced

Phase 3: Intensification and Protection for Machakos (Months 6–24)

• Agroforestry expansion — the rising p90 signal in Machakos is consistent with some existing tree cover; this should be supported, protected, and deliberately extended using Grevillea, Calliandra, Acacia senegal, and indigenous species such as Melia volkensii
• Seasonal planting calendars aligned with the observed NDVI-rainfall rhythm: long-rains planting (March) for maize, beans, cowpea; short-rains planting (October) for drought-tolerant sorghum and vegetables
• Swales and water retention earthworks on dry-season bare slopes to extend soil moisture availability through the dry period (July–October)
• Kitchen garden establishment at household scale in conjunction with aloe vera, indigenous vegetables (terere/amaranth, managu), and fruit trees (mango, avocado, papaya) to improve nutrition and income during dry-season NDVI troughs
• Community-based NDVI monitoring — train local land stewards to use Sentinel Hub browser or equivalent to check their own NDVI values seasonally, creating a feedback loop between satellite science and farm management decisions

Government Policy and Planning Implications

A satellite does not govern. But those who govern must learn to read what the satellite shows — and then act with the humility, urgency, and intergenerational commitment that the land demands.

The Kenyan government is encouraged to treat this NDVI time-series not as a one-off study but as the beginning of a living environmental intelligence system. Sentinel-2 data is freely available through the Copernicus programme. With minimal capacity building, county-level officials and trained community monitors can update this analysis every six months, building an evidence base for adaptive management that no single report can provide.

For the Young Farmers of Edinburgh — and the Young Custodians of Kenyan Land

Tables Reference

Conceptual Diagrams

┌─────────────────────────────────────────────────────────┐
│           SENTINEL-2 SATELLITE (786 km altitude)        │
│         5-day revisit · 10m resolution · L2A product    │
└───────────────────────────┬─────────────────────────────┘
                            │  Red + NIR bands → NDVI
                            ▼
┌─────────────────────────────────────────────────────────┐
│                  CLOUD FILTERING                        │
│         Remove observations with cloud > 30%            │
│         Machakos: 396 → 127 usable observations         │
│         Kulalu: 402 → 115 usable observations           │
└───────────────────────────┬─────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────┐
│              STATISTICAL INTERPRETATION                 │
│   mean · median · p10 · p90 · stdev · min · max        │
│   ↓ each tells a different ecological story ↓           │
└───────────────────────────┬─────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────┐
│            ECOLOGICAL TRANSLATION                       │
│  NDVI 0.0–0.2 → bare soil / degraded land              │
│  NDVI 0.2–0.4 → dry pasture / stressed crop            │
│  NDVI 0.4–0.6 → active crop / healthy shrub            │
│  NDVI 0.6–0.8 → dense canopy / irrigated land          │
└───────────────────────────┬─────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────┐
│              GROUND VERIFICATION (essential!)           │
│  Walk the land · Soil tests · Community knowledge       │
│  Infiltration tests · Elder memory · Crop records       │
└───────────────────────────┬─────────────────────────────┘
                            │
                            ▼
┌─────────────────────────────────────────────────────────┐
│              LAND MANAGEMENT ACTION                     │
│  Contour bunds · Agroforestry · Water harvesting        │
│  Crop calendars · Restoration planting · Policy         │
└─────────────────────────────────────────────────────────┘

NDVI
0.65 ┤                           ╔═══╗
     │                         ╔╝   ╚╗      (Machakos wet-season peak)
0.55 ┤            ╔════╗      ╔╝     ╚╗
     │           ╔╝    ╚╗   ╔╝         ╚╗
0.45 ┤         ╔╝        ╚═╝             ╚═╗
     │        ╔╝                             ╚╗   ← MACHAKOS
0.35 ┤       ╔╝                               ╚═╗
     │──────╝                                     ╚════  (dry season)
0.25 ┤                                                  
     │   ╔══╗               ╔══╗           ╔══╗
0.20 ┤  ╔╝  ╚╗             ╔╝  ╚╗         ╔╝  ╚╗   ← KULALU
     │ ╔╝    ╚╗           ╔╝    ╚╗       ╔╝    ╚╗
0.15 ┤╔╝      ╚═══════════╝      ╚═══════╝      ╚══  (prolonged bare stress)
     └─────────────────────────────────────────────────────
       Jan  Mar  May  Jul  Sep  Nov  Jan  Mar  May  Jul
       ← Long rains →    ← Dry →  ← Short rains →  ← Dry →

  MACHAKOS: Strong bimodal pulse; large seasonal amplitude (~0.31)
  KULALU:   Weak bimodal pulse; narrow amplitude (~0.19)
  GAP:      Machakos typically 0.15–0.27 NDVI units greener than Kulalu

START: Site has consistently low NDVI (below 0.25)?
       │
       ├─ YES ─→ Is p10 near zero or negative?
       │            │
       │            ├─ YES ─→ Bare/compacted/eroded land
       │            │          → Ground survey → Zai pits, half-moons,
       │            │            compost, pioneer grasses FIRST
       │            │
       │            └─ NO ──→ Some vegetation but stressed
       │                       → Soil test → Add organic matter
       │                         Consider agroforestry
       │
       └─ NO ──→ NDVI above 0.25; seasonal response present?
                   │
                   ├─ YES ─→ Is there an upward trend (year-on-year)?
                   │            │
                   │            ├─ YES ─→ Land recovering
                   │            │          → Protect, expand, agroforest
                   │            │
                   │            └─ NO ──→ Land stable but not improving
                   │                       → Investigate management gaps
                   │                         Water harvesting? Grazing control?
                   │
                   └─ NO ──→ Seasonal response absent or very weak
                              → Investigate soil structure / hardpan
                                Infiltration test essential
                                Consider subsoil ripping or deep compost trenches

                    ★ SATELLITE (786 km up)
                    │  "I can see colour. I can measure green."
                    │
                    ▼
           ┌──────────────────────────────┐
           │   NDVI = "Green Heartbeat"   │
           │                              │
           │  p90 = the strongest patch   │  ← "The healthiest corner of the field"
           │  mean = the whole field avg  │  ← "How is everyone doing on average?"
           │  p10 = the weakest patch     │  ← "Who is struggling most?"
           │  stdev = how uneven is it?   │  ← "Is the field all the same, or wild?"
           └───────────────┬──────────────┘
                           │
              What goes UP after rain?    What goes DOWN in dry season?
                           │                           │
              ┌────────────┴──────────────┐           ┌┴─────────────────────────┐
              │ Good land: responds fast  │           │ Stressed land: recovers  │
              │ Stores water in soil      │           │ slowly or not at all     │
              │ Has roots and organic     │           │ Bare, compacted, eroded  │
              │ matter to hold moisture   │           │ Low organic matter        │
              └────────────┬──────────────┘           └┬─────────────────────────┘
                           │                           │
                           ▼                           ▼
              PROTECT, AGROFOREST              RESTORE FIRST:
              WATER HARVEST, EXPAND            Half-moons, Zai pits,
              SEASONAL PLANTING                Pioneer grasses, Moringa
                           │                           │
                           └───────────────┬───────────┘
                                           │
                                    ★ YOUR FEET ON THE GROUND
                                    "Now go check. Walk it. Smell it. Feel it."

Limitations and Ground-Truthing Needs

Science that does not know its own limits is not science. It is confidence masquerading as knowledge. These limitations are not apologies — they are invitations for the next phase of investigation.

What NDVI Cannot Tell Us

• Soil fertility: A green signal means chlorophyll, not fertility. Nitrogen-deficient but well-watered grass can be green. Deeply fertile soil with drought-stressed crops can look brown. NDVI cannot measure soil nutrients.
• Exact rainfall amounts: NDVI responds to vegetation, not directly to rainfall. We infer rainfall timing from vegetation response, but cannot calculate how many millimetres fell.
• Land ownership and tenure: The satellite sees vegetation, not boundaries, rights, or history. Who owns these lands, how they are managed, and what conflicts may exist around them are entirely outside the orbital frame.
• Crop type: We cannot distinguish maize from cowpea from millet from fallow grass at this scale and pixel size. Ground surveys are essential for crop mapping.
• Cause of degradation: Whether Kulalu's low NDVI reflects overgrazing, drought, fire history, soil type, slope position, historical clearing, or some combination of these cannot be determined from NDVI alone.
• Sub-pixel heterogeneity: Each 10m² pixel averages everything within it — a healthy tree canopy and a bare path beside it may average to a moderate NDVI value that represents neither accurately.
• Cloud gaps: The 68–71% of observations excluded by the cloud filter represent an unknown period of vegetation dynamics. We may be systematically missing peak greenness events (which coincide with cloud cover) in our analysis.

Essential Ground-Truthing Steps

• Stratified random field visits to high-NDVI, mid-NDVI, and low-NDVI zones within each study area
• Soil sampling (0–20cm and 20–50cm) for texture, pH, organic carbon, available nutrients
• Community participatory mapping to overlay social, historical, and management contexts
• Hydrological survey: slope gradients, drainage network, any existing water harvesting structures
• Comparison with any available rainfall station data from Machakos and Garissa/Tana River Districts
• Cross-reference with Kenya Land Use Survey maps and county land-use plans