Grow Asia

Dataset

Units: meters
Resolution: 30m downsampled to 90m

Introduction

The two most commonly used global elevation data sources are SRTM (provided by NASA) and ALOS DSM (provided by JAXA). ALOS DSM is generally better in cloud-free regions and areas where there is no dense forest coverage.
ALOS DSM is a digital surface model (DSM) which means it measures height of the earth’s surface including features such as buildings, trees, etc whereas SRTM is a digital elevation model (DEM) which means it measures just height of the earth’s surface.
ALOS DSM is built from AW3D (5m resolution, NTT DATA and Remote Sensing Technology Center of Japan), the world’s most precise global 3D map. It is developed using stereoscopy, a method of using imagery taken from different angles to produce a 3D model. The images are collected by ALOS satellite which measure elevation with PRISM. SRTM lacked global coverage (~80% coverage) but void-fills using other elevation data and interpolation.

Evaluation

Santillana and Makinano-Santilla‘s paper compared AW3D30 with SRTM at 5,121 ground-based check-points in Japan and around the world. AW3D30 had a lower root mean squared elevation error (4.40m vs 7.50m) and a lower standard deviation (4.38m vs 7.43m). Another study in the Philippines also came to similar conclusions (AW3D30 RSME 5.68m, SRTM RSME 8.28m, AW3D30 standard deviation 3.66m, SRTM standard deviation 4.57m). They also found that AW3D30 had the lowest mean and RMSE errors across five different land covers - brushland, built-up, cultivated, dense vegetation and grassland.

Summary statistics

Average: 1095.37 / Standard deviation: 788.27 / Min: -34.00 / 25 percentile: 396.00 / 50 percentile: 1057.00 / 75 percentile: 1643.00 / Max: 4091.00

Historical availability

Data recorded in 2006 to 2011

Dataset

Units: degrees
Resolution: 30m downsampled to 90m

Introduction

Slope (30m) is calculated from the ALOS elevation data using the four neighbor method (calculates 2 vectors along x and y axis neighbors to form a plane, calculates the normal to the plane at the point, and then takes the degree difference between the normal from 90 degrees).

Summary statistics

Average: 18.84 / Standard deviation: 12.16 / Min: 0.00 / 25 percentile: 8.95 / 50 percentile: 19.06 / 75 percentile: 27.78 / Max: 83.69

Dataset

Data: 1 if have water, 0 if not water
Resolution: 30m downsampled to 90m

Introduction

JRC Global Surface Water is compiled by the European Commission's Joint Research Centre and is the output of "High-resolution mapping of global surface water and it's long term changes'”, a Nature paper which uses Landsat classified tiles (64,254 samples) to build a model to detect water bodies.

Evaluation

The same Nature paper collected 40,124 control points around the world across 32 years and found the system had <1% of false water detections and missed <5% of water surfaces.

Summary statistics Summary statistics

4% of land mass is water body

Historical availability

We use 2017 to 2019 data.

Dataset

Units: Celsius
Resolution: 1km, applied Gaussian smoothing to interpolate values and get 90m resolution

Introduction

MOD11A1 is provided by NASA and is temperature data derived from MOD11_L2, retrieved using the split-window algorithm and under clear-sky conditions.

Evaluation

MOD11A1 V5 has previously been shown to be accurate within ± 2K across 42 sites, which represent a wide range of land surface temperature across seasons and years, except in six bare soil sites. MOD11A1 naturally performs poorly in bare arid areas since bands 31 and 32 (algorithm inputs) are more variable in these types of regions. “New refinements and validation of the collection-6 MODIS land-surface temperature/emissivity product” tested the v6 update on five bare soil sites in north Africa, with validation errors within ± 0.6K in 10 of 12 validatation data sets (0.8K and 1.9K for the other two). Standard deviation is similarly low at <0.7K for all 12 data sets. Similarly, “Evaluation of Collection-6 MODIS Land Surface Temperature Product Using Multi-Year Ground Measurements in an Arid Area of Northwest China” evaluates v5 and v6 data on various sand desert land types and concludes that the RMSE of V6 is smaller compared to the RMSE of V5.

Summary statistics

Temperature monthly average (these are statistics of the average monthly temperature for each pixel) Average: 23.78 / 50 percentile: 24.00

Historical availability

2000 onwards, we use 2016 2017 and 2018 data.

Dataset

Units: millimeters
Resolution: 5km, applied Gaussian smoothing to interpolate values and get 90m resolution

Introduction

Precipitation datasets are largely divided into gauge-based (accurate at single points), satellite-related (provides broader coverage using microwave to conduct radiative transfer techniques and infrared to measure temperature of cloud tops, but inaccurate in areas with complex topologies), methods merging gauge and satellite-related methods, and reanalysis systems (methods to synthesize irregular observations). Reanalysis systems are generally inconclusive - in a global study of 30 precipitation datasets, reanalysis systems have been shown to have larger variance, whereas in another reanalysis systems performed better than satellite systems. Generally, precipitation datasets perform poorly in arid climates and satellite systems are challenging in areas with complex topography.
CHIRPS (Climate Hazards Group, UCSB) is a 35+ year quasi-global rainfall dataset. It uses a satellite-based, gauge corrected method by incorporating 0.05° resolution satellite imagery with ground station data.

Evaluation

In an evaluation of 8 datasets in Adige Basin, an area of complex topography, against 101 rain gauges and across multiple temporal (daily, monthly and annual) and spatial scales (grid and watershed), CHIRPS performed the best overall (bias < 0.05). All products do not perform well in low precipitation events (winter months).
In a 2018 study of 12 satellite-based systems compared against 72 gauges near Lake Titicaca, an area of contrasting emissivity, temperature and orography (topography), MSWEP v.2.1 and CHIRPS v.2 performed best in space-time accuracy and consistency. The authors conclude that “some SPPs are relatively stable at regional scale (CHIRPS v.2, MSWEP v.2.1) and could thus be used to study regional precipitation patterns with a relatively high degree of confidence”, with MSWEP v.2.1 being the best. Across all three assessment indicators (gauge precipitation estimate, streamflow and snow cover duration), systems with gauge data outperformed satellite-only systems.
In a study of three satellite-based systems over a hot desert climate in Egypt, CHIRPS was best at estimating rainfall amounts, IMERG was better at detecting the occurrence of rainfall than CHIRPS.
In a 2017 global study of 22 precipitation datasets on 76,086 gauges, MSWEP V2.0 performs best across the climate indicators. For tropical regions, CHIRPS V2.0 is the second best choice “if a daily temporal resolution suffices, and if the peak magnitude underestimation and spurious drizzle are less critical”.
In general, MSWEP and CHIRPS is on par, but MSWEP requires more manual integration and permission from the Princeton team for scientific purposes.

Summary statistics

Rainfall quarterly max (these are statistics of the quarter with the most rainfall for each pixel) Average: 788.01 / 50 percentile: 770.69
Rainfall annual average (these are statistics of the average annual rainfall for each pixel) Average: 2098.40 / 50 percentile: 1968.14

Historical availability

1981 to present, we use 2016 2017 and 2018 data.

Dataset

Units: millimeters
Resolution: 25km, applied Gaussian smoothing to interpolate values and get 90m resolution

Introduction

Both GLDAS (NASA, Noah land surface model) and SMAP (NASA) are global high resolution datasets providing surface and subsurface soil moisture data. GLDAS is a satellite and ground observations based system. Specifically it measures the backscattering of microwaves which interact with soil, wetter soil results in more backscattering. SMAP combines L-bands with C- and X-band microwaves.

Evaluation

The top most commonly used soil moisture products are SMAP and GLDAS.
Few papers directly evaluate GLDAS against SMAP. In a study by IBM over Malaysia area, GLDAS performs marginally better than SMAP. In another paper evaluating SMAP, ASCAT and AMSR2 against GLDAS, SMAP and GLDAS performed similarly in R-value (0.74 and 0.73), with GLDAS showing the lowest absolute bias at 0.03 and lowest ubRMSD values.
In a study in central Tibetan Plateau, a cold region, GLDAS models generally produced lower RMSE and BIAS compared to AMSR-E algorithms. The article also mentions GLDAS errors are high due and they systematically underestimate the moisture within the topsoil due to the absence of high amounts of soil organic carbon in the region.
In a study validating 5 passive satellites in Qinghai-Tibet Plateau and surrounding areas, SMAP had lowest uncertainty in comparison with in situ observations (large R and low RMSE and bias) and regional measurements, and offers the highest coverage. It performed well on in situ measurements in areas with sparse vegetation. The study also tested the models against three algorithm inputs (physical surface temperature, vegetation optical depth, and dielectric mixing model) and noted that retrieval algorithms and inputs were primary causes of error. Similarly, here SMAP had the lowest RMSE (0.039–0.063 m3 m−3) and bias (0.022–0.050 m3 m−3), and vegetation cover correction is also a major factor in the retrieval algorithm. Finally, in a study over the Tibetan Plateau, SMAP performs well in spatio and temporal tests. It performed poorly in areas of complex topography where stations were sparse.
In summary, we are indifferent to SMAP or GLDAS products, though recommend GLDAS as it has a longer history and there is still a dearth of studies validating SMAP against GLDAS.

Summary statistics

Soil moisture monthly average (these are statistics of the average soil moisture on a monthly basis for each pixel) Average: 28.10 / 50 percentile: 29.62

Historical availability

2000 onwards, we use 2016 2017 and 2018 data.

Dataset

Units: kg/kg
Resolution: 25km, applied Gaussian smoothing to interpolate values and get 90m resolution

Introduction

Both GLDAS (NASA, Noah land surface model) and SMAP (NASA) are global high resolution datasets providing surface and subsurface soil moisture data. GLDAS is a satellite and ground observations based system. Specifically it measures the backscattering of microwaves which interact with soil, wetter soil results in more backscattering. SMAP combines L-bands with C- and X-band microwaves.

Evaluation

The top most commonly used soil moisture products are SMAP and GLDAS.
Few papers directly evaluate GLDAS against SMAP. In a study by IBM over Malaysia area, GLDAS performs marginally better than SMAP. In another paper evaluating SMAP, ASCAT and AMSR2 against GLDAS, SMAP and GLDAS performed similarly in R-value (0.74 and 0.73), with GLDAS showing the lowest absolute bias at 0.03 and lowest ubRMSD values.
In a study in central Tibetan Plateau, a cold region, GLDAS models generally produced lower RMSE and BIAS compared to AMSR-E algorithms. The article also mentions GLDAS errors are high due and they systematically underestimate the moisture within the topsoil due to the absence of high amounts of soil organic carbon in the region.
In a study validating 5 passive satellites in Qinghai-Tibet Plateau and surrounding areas, SMAP had lowest uncertainty in comparison with in situ observations (large R and low RMSE and bias) and regional measurements, and offers the highest coverage. It performed well on in situ measurements in areas with sparse vegetation. The study also tested the models against three algorithm inputs (physical surface temperature, vegetation optical depth, and dielectric mixing model) and noted that retrieval algorithms and inputs were primary causes of error. Similarly, here SMAP had the lowest RMSE (0.039–0.063 m3 m−3) and bias (0.022–0.050 m3 m−3), and vegetation cover correction is also a major factor in the retrieval algorithm. Finally, in a study over the Tibetan Plateau, SMAP performs well in spatio and temporal tests. It performed poorly in areas of complex topography where stations were sparse.
In summary, we are indifferent to SMAP or GLDAS products, though recommend GLDAS as it has a longer history and there is still a dearth of studies validating SMAP against GLDAS.

Summary statistics

Specific humidity monthly average (these are statistics of the average specific humidity on a monthly basis for each pixel) Average: 0.01 / 50 percentile: 0.01

Historical availability

2000 onwards, we use 2016 2017 and 2018 data.

Dataset

Units: kWh/m2
Resolution: 250m, applied Gaussian smoothing to interpolate values and get 90m resolution

Introduction

Solargis, funded by The World Bank Group, provides Global Horizontal Irradiation (GHI) is the total irradiance from the sun on a horizontal surface. GHI is the sum of Direct Normal Irradiation (DNI) and diffuse horizontal irradiance (DHI). Both DNI and DHI are measured on a horizontal surface on Earth. DNI is defined as beam radiation (radiation coming from the sun disk) at that horizontal surface. DHI is radiation at that horizontal surface due to scattering effects from atmospheric components (eg particles, droplets, etc.).
GHI is constructed from the Solargis solar radiation model (which takes in data from geostationary satellites, meteorological models to calculate clear-sky irradiance, cloud index to get all-sky irradiance) and air temperature model (takes in meteorological parameters such as wind speed, direction, relative humidity, etc.)

Evaluation

Solargis evaluated their data in different geographies and situations and concluded “in most of the cases the expected deviation of the yearly values is expected in the range of ±4% to ±8% for GHI, and in the range of ±8% to ±15% for DNI. A higher deviation is expected in geographically complex conditions, and in regions which are not sufficiently covered by high quality meteorological measurements.”

Summary statistics

Average: 1483.73 / Standard deviation: 2189.29 / Min: -32768.00 / 25 percentile: 1464.00 / 50 percentile: 1644.00 / 75 percentile: 1794.00 / Max: 1980.00

Historical availability

2007-2015.

Dataset

Units: probability
Resolution: 90m

Introduction

We take waterbody, elevation and slope data to create a custom flood risk model.
For each tile within 1km from the river, we calculate:
- risk given the distance from the river, modeled as Normal(0, 500m)
- risk given the difference in elevation from river, modeled as Normal(0, 100m)
- risk given the slope from river, modeled as Normal(0, 20 degrees)
We then multiply the above risks to get the inundation risk.

Evaluation

None

Summary statistics

Average: 0.03 / Standard deviation: 0.14 / Min: 0.00 / 25 percentile: 0.00 / 50 percentile: 0.00 / 75 percentile: 0.00 / Max: 1.00

Dataset

Units: house or no house
Resolution: 90m

Introduction

This model has been used by Enveritas for rural areas in 12 countries and has 90% precision and recall.

Evaluation

None

Summary statistics

1% of land mass is houses.

Dataset

Units: population per 90m x 90m tile
Resolution: 90m

Introduction

HDX provides population and shapefiles at level 3 as updated in mid-2019. For each level 3 area, we calculate for each tile in the level 3 the population of the level 3 area divided by number of house tiles in that level 3. If the level 3 area is not entirely contained in our area, we calculate population times the percentage of level 3 area contained in our area of interest.
Data Source

Evaluation

None

Summary statistics

Average: 17.79 / Standard deviation: 210.30 / Min: 0.00 / 25 percentile: 0.00 / 50 percentile: 0.00 / 75 percentile: 0.00 / Max: 4000.00

Dataset

Units: road or no road
Resolution: 90m

Introduction

Road data was downloaded from OSM in 2014.
https://data.humdata.org/dataset/papua-new-guinea-roads

Evaluation

None

Summary statistics

Less than 1% of land mass is road.

Dataset

Units: pH
Resolution: 30m

Introduction

Soil pH indicates the degree of acidity, which is related to crop production potential.

Sampling Method

Samples were taken at a depth of 0-15cm. Soil pH was measured after shaking a suspension of soil and deionized water at a ratio of 1:5 (weight/weight basis) at 25°C for 1 hour. The pH measurements were taken in a calibrated pH meter (Mettler Toledo Seven Multi pH meter).
We interpolate pH by weighing the pH of soil samples taken within 5km according to their inverse distances.

Summary statistics

Average: 7.45 / Standard deviation: 0.73 / Min: 6.15 / 25 percentile: 6.79 / 50 percentile: 7.56 / 75 percentile: 8.01 / Max: 8.97

Dataset

Units: %
Resolution: 30m

Introduction

Soil organic Carbon content is an indicator of soil fertility, biological activity, ability to cycle carbon, and potential availability of nitrogen and sulfur.

Sampling Method

Samples were taken at a depth of 0-15cm. A finely-milled (< 0.2 mm) air-dry sample was analyzed for organic carbon content by the Walkley-Black’s wet oxidation method in a dichromate-concentrated sulphuric acid mixture. The reaction mixture was titrated against standard ferrous ammonium sulfate solution in the presence of a suitable redox indicator and concentrated phosphoric acid.
We interpolate organic Carbon content by weighing the pH of soil samples taken within 5km according to their inverse distances.

Summary statistics

Average: 2.73 / Standard deviation: 1.78 / Min: 0.54 / 25 percentile: 1.35 / 50 percentile: 2.28 / 75 percentile: 3.42 / Max: 8.47

Dataset

Units: %
Resolution: 30m

Introduction

Particle size analysis determines texture of the soil. Soil texture influences a range of physical properties such as soil bulk density, moisture retention, aggregate stability, infiltration and permeability. Soil texture also influences cation exchange capacity and nutrient retention characteristics.

Sampling Method

Samples were taken at a depth of 0-15cm. Particle size analysis was carried out with the Bouyoucos hydrometer method. A weighed portion of a previously moisture-determined soil was treated with hydrogen peroxide to decompose organic matter and then placed in a water bath to remove excess peroxide. Sodium-hexa-meta-phosphate solution was added and the mixture was stirred and made up to 1000 mL in a cylinder. A hydrometer was immersed in the suspension and density measurements were taken at certain time intervals. The time required to settle particles of a given size can be calculated using Stoke’s equation.
Soil particles were separated into three size categories.
- Sand: > 0.02 mm
- Silt: 0.002-0.02 mm
- Clay: < 0.002 mm
We interpolate clay percentage by weighing the clay percentage of soil samples taken within 5km according to their inverse distances.

Summary statistics

Average: 12.10 / Standard deviation: 8.56 / Min: 0.80 / 25 percentile: 4.00 / 50 percentile: 11.90 / 75 percentile: 17.08 / Max: 34.00

Dataset

Units: %
Resolution: 30m

Introduction

Particle size analysis determines texture of the soil. Soil texture influences a range of physical properties such as soil bulk density, moisture retention, aggregate stability, infiltration and permeability. Soil texture also influences cation exchange capacity and nutrient retention characteristics.

Sampling Method

Samples were taken at a depth of 0-15cm. Particle size analysis was carried out with the Bouyoucos hydrometer method. A weighed portion of a previously moisture-determined soil was treated with hydrogen peroxide to decompose organic matter and then placed in a water bath to remove excess peroxide. Sodium-hexa-meta-phosphate solution was added and the mixture was stirred and made up to 1000 mL in a cylinder. A hydrometer was immersed in the suspension and density measurements were taken at certain time intervals. The time required to settle particles of a given size can be calculated using Stoke’s equation.
Soil particles were separated into three size categories.
- Sand: > 0.02 mm
- Silt: 0.002-0.02 mm
- Clay: < 0.002 mm
We interpolate silt percentage by weighing the silt percentage of soil samples taken within 5km according to their inverse distances.

Summary statistics

Average: 29.26 / Standard deviation: 12.22 / Min: 4.00 / 25 percentile: 20.10 / 50 percentile: 27.70 / 75 percentile: 37.55 / Max: 54.40

Dataset

Units: %
Resolution: 30m

Introduction

Particle size analysis determines texture of the soil. Soil texture influences a range of physical properties such as soil bulk density, moisture retention, aggregate stability, infiltration and permeability. Soil texture also influences cation exchange capacity and nutrient retention characteristics.

Sampling Method

Samples were taken at a depth of 0-15cm. Particle size analysis was carried out with the Bouyoucos hydrometer method. A weighed portion of a previously moisture-determined soil was treated with hydrogen peroxide to decompose organic matter and then placed in a water bath to remove excess peroxide. Sodium-hexa-meta-phosphate solution was added and the mixture was stirred and made up to 1000 mL in a cylinder. A hydrometer was immersed in the suspension and density measurements were taken at certain time intervals. The time required to settle particles of a given size can be calculated using Stoke’s equation.
Soil particles were separated into three size categories.
- Sand: > 0.02 mm
- Silt: 0.002-0.02 mm
- Clay: < 0.002 mm
We interpolate sand percentage by weighing the sand percentage of soil samples taken within 5km according to their inverse distances.

Summary statistics

Average: 58.65 / Standard deviation: 17.97 / Min: 14.80 / 25 percentile: 46.98 / 50 percentile: 58.90 / 75 percentile: 72.80 / Max: 94.80

Dataset

Units: mg/kg
Resolution: 30m

Introduction

Plant-available Phosphorus is an essential crop nutrient for oil seeds, legumes and forages. The Olgen test used here is suitable for measuring readily available forms of soil Phosphorus in slightly alkaline soils.

Sampling Method

Samples were taken at a depth of 0-15cm. Extractable Phosphorus content was determined according to the method described by Olsen. A portion of air-dried soil sample was extracted using a freshly prepared 0.5 M sodium bicarbonate buffered at a pH of 8.5. The soil extractant ratio was 1:20. Phosphorous content in the extract was measured by a molybdenum-blue color method at a wavelength of 882 nm in a spectrophotometer (SHIMADZU UV-1800).
We interpolate available Phosphorus by weighing the available Phosphorus of soil samples taken within 5km according to their inverse distances.

Summary statistics

Average: 10.33 / Standard deviation: 9.51 / Min: 0.08 / 25 percentile: 3.44 / 50 percentile: 6.50 / 75 percentile: 15.57 / Max: 42.20

Dataset

Units: cmol Ca++/kg
Resolution: 30m

Introduction

Calcium is generally the predominant basic cation in the soil. Soil base saturation is a prerequisite for maintaining fertility and productivity. Gradual loss of calcium may turn soil acidic leading to serious consequences to crops and animals. An ideal soil should have ca. 65% calcium as exchangeable cation.

Sampling Method

Samples were taken at a depth of 0-15cm. The soil and 1 M NH4Cl extractant solutions at a ratio of 1:20 were equilibrated at pH 7.0 for 1 hour with mechanical shaking. The suspensions were clarified by filtration and analyzed for exchangeable calcium content by Inductively Coupled Plasma-Optical Emission Spectrophotometer (Varian 725-ES). The method does not require pre-treatment for the removal of chlorides and is suitable for all soil types and provides a good estimate of exchangeable calcium in acidic and neutral, non-saline, non-calcareous and non-gypsiferous soils.
We interpolate exchangeable Calcium by weighing the exchangeable Calcium of soil samples taken within 5km according to their inverse distances.

Summary statistics

Average: 32.71 / Standard deviation: 11.35 / Min: 5.40 / 25 percentile: 25.24 / 50 percentile: 33.04 / 75 percentile: 41.32 / Max: 58.95

Dataset

Units: cmol/kg
Resolution: 30m

Introduction

Magnesium plays an important role in plant nutrition and physiology. The ratio of calcium to magnesium is also critical. An ideal soil should have ca. 10% magnesium as exchangeable cation.

Sampling Method

Samples were taken at a depth of 0-15cm. The soil and 1 M NH4Cl extractant solutions at a ratio of 1:20 were equilibrated with at pH 7.0 for 1 hour with mechanical shaking. The suspensions were clarified by filtration and analyzed for exchangeable magnesium content by Inductively Coupled Plasma-Optical Emission Spectrophotometer (Varian 725-ES). The method does not require pre-treatment for the removal of chlorides and is suitable for all soil types and provides a good estimate of exchangeable magnesium in acidic and neutral, non-saline, non-calcareous and non-gypsiferous soils.
We interpolate exchangeable Magnesium by weighing the exchangeable Magnesium of soil samples taken within 5km according to their inverse distances.

Summary statistics

Average: 5.89 / Standard deviation: 3.58 / Min: 1.47 / 25 percentile: 3.40 / 50 percentile: 5.38 / 75 percentile: 7.33 / Max: 18.06

Dataset

Units: cmol/kg
Resolution: 30m

Introduction

Crop requirement of potassium is quite high and hence potassium plays significant role as a fertilizer nutrient. An ideal soil should have ca. 5% potassium as exchangeable cation.

Sampling Method

Samples were taken at a depth of 0-15cm. • The soil and 1 M NH4Cl extractant solutions at a ratio of 1:20 were equilibrated with at pH 7.0 for 1 h with mechanical shaking. The suspensions were clarified by filtration and analyzed for exchangeable potassium content by Inductively Coupled Plasma-Optical Emission Spectrophotometer (Varian 725-ES). The method does not require pre-treatment for the removal of chlorides and is suitable for all soil types and provides a good estimate of exchangeable potassium in acidic and neutral, non-saline, non-calcareous and non-gypsiferous soils.
We interpolate exchangeable Potassium by weighing the exchangeable Potassium of soil samples taken within 5km according to their inverse distances.

Summary statistics

Average: 2.24 / Standard deviation: 1.69 / Min: 0.20 / 25 percentile: 0.98 / 50 percentile: 1.97 / 75 percentile: 2.91 / Max: 9.91