CHIRPS Reality Checks

From CHG-Wiki
Jump to navigationJump to search

CHIRPS v2.0 monthly Reality Checks

Background

A team of data analysts routinely quality check each month’s CHIRPS data before its release. This page documents major points of these Reality Checks. Rchecks country highlights has information that CHIRPS users may find helpful, for example, notes about major rainfall anomalies in the data. Notes on Rchecks resources is information for CHG to continually improve the Reality Checks process. Rcheck is a hands-on approach that helps enable a quality product for hazards monitoring and other scientific activities. In Reality Checks we examine the data visually via the Early Warning Explorer and separately using calculated statistics. Ancillary information, such as FEWS NET datasets, news reports, and government meteorological reports, are frequently used in the process. Rchecks has been successful in: 1) Validating anomalous wet and dry events around that world as shown by CHIRPS, 2) catching inaccurate station reports that would have otherwise negatively influenced the dataset, such as creating false droughts, 3) checking that the semi-automated flow CHIRPS data creation is working correctly, 4) identifying weaknesses and strengths of the algorithm and data inputs, which helps in planning improvements in future versions.

Helpful Links

The following may be helpful links in completing the monthly reality checks:

Global NOAA Climate Products [1]: Choose zone of the world for special products (Central Asia, South Asia, Africa, Central America & Caribbean)

Global FEWS Data Portal [2]

Africa FEWS Archived Hazards/Threats [3]

Central America FEWS Archived Hazards/Threats [4]

Brazil INMET[5]: Click the Mapas de Precipitacao tab and then the Plus button at the bottom of the map and an end date and time period can be selected (30 dias).


November 2015

Rchecks highlights

United States CHIRPS shows the very wet conditions that occurred across the South and southern Midwest. November rainfall was 100-200 mm above average in northeastern Texas, eastern Oklahoma, Arkansas, Missouri, and in Georgia, North Carolina, and South Carolina. Monthly totals in Arkansas and Missouri were the highest on record (wettest of 120 years) in November 2015. CHIRPS shows below average rainfall in coastal Oregon and coastal northern and southern California.

October 2015

Rchecks highlights

United States A storm complex that tapped into the moisture from Hurricane Joaquin off the south-eastern coast of the US hit the Carolina's, dumping 12-24 inches of rain (picked up in CHIRPS)[6]. Louisiana and Texas, in addition, were hit with the remnants of Hurricane Patricia, receiving a lot of rain (also visible in r-checks) [7].Finally, a dry-spell it the midwest showed up as well [8].

Mexico Hurricane Patricia hit the south-west coast of Mexico, and rapidly downgraded, as shown in CHIRPS and backed up by FEWS early warning data [9]. Two stations however reported very low values (4 and 5 mm) along the northern border of Patricia's path. One of them, 400141, is blocked by mountains so might have a rain shadow. The other, 400984, however, is on the coast and is a station to keep an eye on. The path of Patricia can be seen here [10].

El Salvador, Honduras, and Nicaragua The bay bordered by these three countries (where Choluteca is) has a very high station next to a low one. This is right where a severe gradient is that goes from high precipitation quickly to low. The mountains around this bay might act as a rain shadow. In El Salvador and Honduras, we see a similar pattern as the mountains block rain coming up from the south and a steep gradient occurs from high rain to very low.

Haiti Haiti is showing a positive rain anomaly in the north west. While this area has been doing well, other reports show average instead of a positive anomaly. Might be good to keep an eye on it. [11]

East Africa October CHIRPS is in general agreement with rainfall estimates from NOAA CPC ARC2 and RFE2 datasets in east Africa. Differences in magnitude occur in parts of Sudan and South Sudan, where CHIRPS estimates are approximately 30% lower than ARC2 and RFE. The products are all in agreement that October rainfall was above average in Sudan, South Sudan, and Uganda (CHIRPS, ARC2, and RFE2). The source of lower CHIRPS values may be (1) a drier climatology in October CHIRPS compared to ARC2 climatology and (2) features in the CHIRPS algorithm that lead to conservative estimates and low bias. Also, in an area of reported heavy rains with flooding (border of Ethiopia, Sudan, and South Sudan) there appears to be a drying influence of two GHCN-v2 stations in southeast Sudan used in CHIRPS- these reported low but reasonable values locally, but reduced the CHIRP estimate in the high rainfall area. Ethiopia contained 8 stations this month, a big improvement from the past few months.

West Africa In October 2015, most of the station reports that contribute to CHIRPS in West Africa are from GHCN-v2, a source of monthly data that is highly ranked due to its quality control. CHIRPS compared well with CPC 30 day anomaly from the FEWS Hazards Report, 10/29/15, except in Guinea and Sierra Leone where CHIRPS is much drier.

South Africa Stations between Lesthoso and Swaziland reported lower values than CHIRP estimated. The stations reduced the final CHIRPS which is in agreement with the FEWS Africa Hazards Report, 10/29/15, abnormal dryness polygon over the region.

Brazil CHIRPS reports below average rainfall across much of the nation, with exception of heavy rainfall that occurred in Porto Alegre, Rio Grande do Sul state. Major flooding occurred in that area [12]. The rainfall deficits during October covered much of the Amazon rain forest and are highly concerning given the magnitude and persistence of drought conditions since 2014. The Amazon drought is reported as the worst in the past 80-100 years [13] and has created water shortages in major cities and rural areas [14][15]. Reality Checks monthly CHIRPS comparison identified the South America mean rainfall in September 2015 and October 2015 as being the driest since 1981, when the CHIRPS record began, and Brazil deficits play a major role in these continental scale deficits.

Asia CHIRP did not perform well with Typhoon Mujigae, between Hong Kong and Hainan on Oct. 5-7th. The stations reported the precipitation amounts and CHIRPS estimates were increased in the area.

A GSOD station on norther border of Pakistan recorded 529 mm for the month. A neighboring station reported 179 mm. A news report claimed record rainfall in the area so we decided to keep the measurement.

Contributors: Laura Harrison (South America); Marty Landsfeld (Asia); Emily Williams (North and Central America, Hispanola); all-of-the-above and Shrad Shukla (Africa); 11/17/15

September 2015

Rchecks highlights

Vietnam Tropical Storm Vamco brought a lot of rainfall to central/northern Vietnam/Laos/Thailand (300+ mm with flooding and fatalities). CHIRPS picked up significant rainfall in Northern Vietnamn (around Hanoi), but CHIRPS's pattern is a bit north of the reports. [16] [17]

China and Taiwan Typhoon Dujuan brought torrential rainfall to Northern Taiwan and Eastern China; CHIRPS shows it in E. China but not much in Taiwan. [18] [19] Stations, though, seem to be reporting well so may be an interpolation problem with the climatology.

Japan Lots of rain and flooding along Western coast and center [20].

Malaysia, Indonesia, Papua Massive dry swaths; many large uncontrolled fires from slash-and-burn agriculture (could be good to keep an eye on for next month).

Columbia Station in Cali, Columbia looks low but looked at previous months values and it looks fine. Diego confirmed it has been dry there in July and August. The rains are returning now he says. SQL commands: select * from precip_monthly4 where station_seqnum=20777 select * from precip_monthly4 where station_seqnum= 205576

Brazil Near Sao Pablo all the stations are higher than CHIRP estimate. INMET rainfall map confirmed the higher values. Brazilian rainfall map can be found at: http://www.inmet.gov.br/portal/index.php?r=home2/index Click the Mapas de Precipitacao tab and then the Plus button at the bottom of the map and an end date and time period can be selected (30 dias).

*South America record rainfall deficit* Station comparison graphs showed a very low CHIRPS mean for South America. /home/CHIRPS/diagnostics.etc/v2.0/marty/chirps.s_amer.stats.09.png But examining the CHIRPS didn’t reveal any errors in processing and given such a strong El Nino and the warm water in the eastern Pacific to the north, it we decided it was a real phenomenon. Also, stations in NW Brazil and Columbia verify the low precip values.

Ethiopia Ethiopia contains only 3 stations. A special check was done on the contribution these stations had for CHIRPS data and anomalies. The 3 stations (2 were co-located) showed below average September rainfall and enhanced CHIRP deficits. As a result, CHIRPS September rainfall shows ~100 mm below average in Addis Ababa and Dire Dawa areas-- roughly 25% of average September rainfall. The CHIRPS September values were compared to data plotted with the Ethiopia Met Agency's MapRoom, which is described as being satellite estimates merged with ~600 stations for the country. Similar magnitude September 2015 anomalies are seen in CHIRPS and the Ethiopia data [21] for these areas. The stations were retained in CHIRPS.

Somalia, Uganda, Rwanda and Burundi All had 0 stations reporting

Contributors: Emily Williams (Asia, Australia, Pacific Islands), Marty Landsfeld (South America, Africa), Laura Harrison (North and Central America, Africa) 10/19/15 - 10/23/15

August 2015

Rchecks highlights

North Korea Removal of a GSOD station. DPRK experienced flooding associated with seasonal rains in early August, and from Tropical Cyclone Goni on 22-23 August, affecting South Hwanghae and North Hamgyong Provinces ([22],[23]) . The two GTS stations in the country are in these areas and reported heavy rains, which the CHIRPS reflects near these areas. GSOD stations (~12) in rest of the country report below average rain, making August an overall poor month for rain in DPRK. This has likely exacerbated the problems associated with late start to seasonal rainfall-- in June DPRK declared they were experiencing the 'worst drought in a century.' There have been major population impacts in the region, perhaps due to a combination of weather and political forces. Suggest removal of GSOD station #274228: Report is 9.19 mm in area with flooding. [24]

South Korea CHIRPS dry anomaly confirmed. Stations show low August rainfall (verified by news reports), which has created an overall poor season there [25]. Makes North Korea dryness believable also.

China Some cases of good stations not influencing CHIRPS local values. Saw several instances when neighboring stations swamped what looks like reasonable above average rainfall reports from some stations. Led to below average CHIRPS values in these areas. Examining monthly decorrelation distance maps may help identify scope of problem.

Ghana Concern about station reports and conditions. Several stations reported 0-10 mm in August. These had some influence on CHIRPS. Were deemed ok stations as they had reasonable values in earlier months and seasonal rains have been below average due to active ITCZ enhanced rains in an abnormally northern position. Note: In July 2015 Ghana had only 1 reporting station in CHIRPS, as compared to 5-13 in other months.

Ethiopia Exceptional dryness identified by ranked z scores. Z score=-2.6, station value=160 mm, Ethiopia highlands (10.33N, 37.740E). Determined CHIRPS value in area was representative of conditions. Also that this is an area of potential concern that needs highlighting on Hazards report-- GeoWRSI shows crops were in reproductive phase in August; Prelim CHIRPS shows September 1-10 was below average; ARC2 shows below average thru September there also.

India Incorrect wet anomaly in northern India. India's Met Department shows below average ([26]), but CHIRPS shows strongly above average. Due to combination of: 0 India stations in area + very wet observation in NE Pakistan (verified by reports) + wet station in China + CHIRP shows wet anomaly. Otherwise, CHIRPS correctly identified August rainfall deficits across most of India and surplus in Bhutan.

Chile Atacama desert, possible problem. Reports say significant rain in Atacama desert ([27] ,[28]) , which rounds to 15 mm instead of 5; CHIRPS isn’t really picking it up as it’s a very fine difference, but as that area is a desert area, that small increase in rainfall resulted in massive flooding and evacuation. CHIRPS showed significant precipitation inland of Concepcion, which agrees with rain and snow reports ([29]). Santiago: CHIRPS wet anomaly confirmed. Report ([30]) details increased rain, which is showing up in CHIRPS.

Uruguay CHIRPS wet anomaly confirmed. CHIRPS recorded above average rainfall for Uruguay for August 2015. The rainfall throughout the country, according to CHPClim, tends to be between 60 and 90 mm; however, CHIRPS reported it ranging from 190 to 240 mm, with the south-eastern coast receiving between 300 and 340mm for the month. This increase in rainfall is backed up by reports, including one from “floodlist” ([31]).

Burkina Faso (added 9/23/15) Station near capital (Ouagadougou) flooded in August, but CHIRPS didn’t pick it up. However, CHIRPS did pick up the rest of the flooding in the country.

Notes on Rchecks resources

Regarding the rchecks-2.files: Reminder. The zscores in rchecks-2.files are not station zscores. They are zscores of CHIRPS data that has been placed at that station's pixel. Using the zscores in these files is helpful for identifying extreme CHIRPS that are caused by extreme station values. Using the zscores is not helpful for identifying stations that do not have influence much on CHIRPS BUT give bad reports (had concerns about this for some stations in Ghana, see above). We discussed routinely isolating a value that indicates this, so that we know which stations have issues. Basically we would do a sorting of the country information based on this value, like we do with the zscores (check.txt). Something to look into more. Also, we corrected a bug that limited the number of recheck text files. Now should have all countries.

Contributors: Laura Harrison, Marty Landsfeld, Emily Williams; 9/17/15

July 2015

Rchecks highlights

Ethiopia July’s r-check had quite a few high z-scores. Upon analysis, these discrepancies were coming from the highlands, near Addis Ababa, and south of Addis Ababa. Station (10.33, 37.74; in town of Debre Markos) reported much lower than CHIRP and CHPClim, in the same way it reported low for August 2015. Station (8.86, 39.92; in town of Metehara Merti, near Addis Ababa) also reported much lower than CHIRP and CHPClim. ARC2 tells a similar story; ARC2 dictates the area having received 200-300mm[32]. ARC2’s anomaly map for July also roughly matched CHIRPS’ anomaly map, at a decrease of around -200m of normal[33]. One report stated “July’s seasonal rains did not come this year…” [34], while anther confirmed that “the rain condition was ok for the first ten days during the month of June. It gradually declined and we started experiencing shortage in rain in July. But conditions are good in August…it has occurred several times in the past, including in 2005.”[35]. The July rains (that didn’t come this year) tend to climb north-east across the highlands throughout June, July and August, the area showing a shortage in rain. It seems the stations are correct and Ethiopia experienced a drought this summer. However, due to the decreased number of stations since August 2014, it is possible that these stations are overestimating the level of drought.

Sudan CHIRPS and CHIRP both indicate that in southwest Sudan, June was had normal rainfall, July was dry, and August returned to normal rainfall (aka summer rains came late). The proximity of the area to the Ethiopian highlands means it followed similar trends as seen in Ethiopia, as backed up by previous reports. HOWEVER, CHIRPS completely missed mass flooding in Darfur in July[36]; for that same time of mass flooding, CHIRPS reported a decrease in rainfall.

Burkina Faso We may have a faulty station (or stations) in northeast Burkina Faso. The two stations are located at (14.03, -0.033; stn 277004); one reported 482 mm of rain, CHIRP reported 129, and CHPClim 117; the other station seems to bring the number ranges from 0-10. These stations combined, though, are driving up the CHIRPS rainfall estimates. One report does state that regular rain started mid-July [37]. However, most reports indicate that the northeastern area had decreased, not increased, rainfall in July [38]; NOAA/FEWS NET reports for July show dryness as well in the west and southwest [39] [40] [41] [42] [43].

Guatemala CHIRPS is fairly accurately picking up a drought in Guatemala, but may be overestimating it in some areas. Station (-88.59, 15.74) has a high value (154) compared to CHIRP (104), and yet CHIRPS is drug down for that pixel (84). There is a neighboring station in Honduras that has an extremely low station value that is most likely over-influencing CHIRPS, creating exaggerated estimates of low rainfall for Guatemala. Reports confirm the drought's severity: TRMM (anomaly and time series) shows drought in the north and west of the nation (-100- -200), in concurrence with CHIRPS (-200- - 300), and written articles confirm the precip data ; ftp://ftp.cpc.ncep.noaa.gov/fews/cent_amer_threats/camerhaz20150723.pdf. CHIRPS is picking up the drought around El Salvedor and in southern Honduras, but again might be overestimating the drought in Guatemala.

Flooding occurred in July in Chinaulta (southeast) [44] ; CHIRP and TRMM both reported higher than average rainfall in that area, but CHIRPS reported higher rainfall only along the coast and drought in Chinaulta. The underestimation of precipitation by CHIRPS might again be the influence of the station in the north of Honduras. It seems that stations are having too high an influence, especially when they are all reporting the same phenomenon and are then influencing another with the opposite (ex//drought all around, but flooding in Chinaulta).

Honduras see above CHIRPS shows a more extreme drought in the West than CHIRP shows, but TRMM reports similarly, suggesting CHIRPS is accurately reporting the drought and CHIRP underestimated it[45].


Contributors: Emily Williams; 9/23/15 - 9/24/15


June 2015

Rchecks highlights

Honduras CHIRPS reported Honduras having experienced flooding around the capital and the northern Caribbean coast. The flooding in the center was backed up by reports [46]. However CHIRP reporting higher rainfall than the stations. Mid-June, Hurricane Bill formed along Honduras's northern coast, which could account for the flooding. However, NOAA/FEWS does not report any flooding, and in fact reports dryness along the Southwest of the country [47].

Nicaragua CHIRPS picked up some significant rainfall (anomaly of +30-50mm) along Western Nicaragua. It was confirmed that the flooding seen in CHIRPS is roughly the pattern of flooding experienced on the ground [48]. NOAA/FEWS did not pick up this flooding in their reports [49].

Costa Rica Costa Rica similarly experienced significant flooding in much of the country [50], also missed by NOAA/FEWS. However, it was picked up by TRMM.

Ethiopia CHIRPS reporting lower-than-average rainfall for Ethiopia's highlands, which seems to be in line with the summer's drought. Compared to ARC1 estimates [51], CHIRPS might be overestimating the drought at this point, but it is nonetheless present. NOAA/FEWS have not captured the severity and distribution, and even declared "no drought" in Ethiopia at the end of June [52].

Kenya Flooding in Nairobi that CHIRPS didn't catch [53]. Other general trends in climate are echoed by ARC1 [54]. There is a station in the southeast of Kenya (in the town of Mombasa) which reported a higher-than-average value for CHIRPS, but a low station value; this same area was subject to flooding [55].

In May, flooding in Mombasa, picked up by CHIRPS [56].

Madagascar CHIRPS reporting higher-than-average rainfall along Madagascar's eastern coastline.

Senegal One station is reporting an abnormally low value (7mm) for June, when CHIRP puts it around 100mm. However, Arc2 shows a decline in rainfall for that area in June [57]. The station looks ok for other months; may be good to keep an eye on though.

MaliStation reporting a low value (99) in an area that tends to get 127; CHIRPS is reporting 139 for this area. The area is on the slope of a mountain, though, and on a rain gradient, so it's likely that the station is correct.

GhanaCHIRPS failed to pick up flooding in coastal town of Accra during "biggest storm in the past 20 years" [58]. The station failed to pick it up, which is probably why CHIRPS didn't get it (station said 154mm, while the report said 250 mm in only the first 3 days.


Contributors: Emily Williams; 9/28/15


May 2015

Rchecks highlights

Guatemala There is a station in the far east of Guatemala which is reporting a significantly high rainfall value, but CHIRPS is reporting a value closer to CHIRP and CHPClim. There is, however, a station very near in Honduras which has an extremely low value which could be over-influencing CHIRPS at our station's location. TRMM total rainfall anomolyis reporting lower-than-average rainfall for that period of time, but then a positive anomaly for June which aligns with our station. However, TRMM's total rainfall time series from station values agrees with our station that is reporting higher rain[59], and disagrees with the Honduras station.

Honduras The most notable station this month has a value of 1 in the North West of the country. However, the country as a whole had an anomaly of 100-200mm, lining up with TRMM values. The station is in a valley surrounded by mountains so it might be correct in that it received no rainfall there, but is something to keep an eye on.

Tajikistan CHIRPS to picked up flooding in the Khatlon province of Tajikistan [60].

Kenya CHIRPS is showing a high quantity of rain along the southern coast and boarder of Kenya. The ARC2 time series points back up this claim [61], in addition to the May estimates [62]. Additionally, the RFE anomoly backs up the claim of high rainfall in the south, and low in the northwest [63].

Tanzania CHIRPS and ARC2 both show an high positive anomaly of rain in Tanzania in the SE [64]. The RFE for the end of the month shows a negative anomaly for the same area [65], but a positive anomaly for the beginning [66]. Reports of flooding back up this claim [67].

Uganda CHIRPS shows dryness through much of the country except for a high incidence of rain in the south near the lake. RFE and ARC2 show similar trends. ARC2 time series shows both the dryness just north of the lake (in correspondence with the low station value) and then the wetness to the west of the lake [68]. CHIRPS is correct for May.

MadagascarDry station on east (central/north) coast; ARC2 time series shows same dryness [69]; ARC2 month shows the same [70]. However, CHIRPS showing a lot of wetness along the rest of the eastern coast, but not backed up by ARC2. It seems like the coastal stations have been overreporting rainfall for May-July. ARC2 timeseries and month-estimates, RFE, and reports [71] show less rainfall than CHIRPS.

Burkina Faso Country received high rainfall in West according to RFE and ARC2 [72] [73]. Station in CHIRPS shows the same story; however CHIRP is showing dryness across the country. ARC May estimates agree there should be more rainfall in the West. Seems like this station is being drowned out by dry stations and should have a higher influence on the area to make it come up as wet instead of dry/average.

Mali Station in south of country shows wet values but is also being drowned out by surrounding drought area, making the area look average instead of wet.

April 2015

Rchecks highlights

Cote D'Ivoire and Ghana CHIRPS is struggling again to pick up anomalous wet periods when surrounded by dry areas. There were stations reporting wet values surrounded by stations reporting dry values, and CHIRPS didn't allow for the high value to significantly impact the area around it; RFE [74] and ARC [75] both indicate wetness for the southern part of the country, which CHIRPS under-reports.

Eastern Africa CHIRPS did a good job in picking up the significant wetness stretching along the Eastern coast of Africa. ARC and CHIRPS are in agreement, where RFE is reporting much drier conditions throughout Kenya and parts of Tanzania.

Kenya Kenya was fairly wet in August. RFE and ARC show wet conditions in northern and western Kenya, though drier conditions in southern and eastern Kenya. CHIRPS isn't picking up the dryness. All the z-scores that stood out, however, are for stations that reported lower rainfall values than CHIRPS shows. Seems CHIRPS isn't heeding local station signals enough and relies too strongly to influences of surrounding precipitation conditions.

Honduras There is a station that received nearly zero rain in April, but this is backed up by the RFE time series points.

Dominican Republic CHIRPS picked up that very little rain fell along the northern and eastern coasts of the Dominican Republic.

Tajikistan CHIRPS shows lower-than-average rainfall in north along ridgeline, and higher-than-average rainfall in the western-central portion of the country. Stations are reporting higher values for rain than the weather records [76] (130 vs 71). Difficult to verify if CHIRPS is accurately reporting or not due to limited precipitation data to compare with. Would be good to have NOAA's RFE png blown up a bit to see better whether our results correlate [77].


Contributors : Emily Williams, 10/16/15